Professional Documents
Culture Documents
(Dr.-Ing. Ulrich Tietze, Dr.-Ing. Christoph Schenk (B-Ok - CC) PDF
(Dr.-Ing. Ulrich Tietze, Dr.-Ing. Christoph Schenk (B-Ok - CC) PDF
(Dr.-Ing. Ulrich Tietze, Dr.-Ing. Christoph Schenk (B-Ok - CC) PDF
Electronic Circuits
U. Tietze • Ch. Schenk • E. Gamm
Electronic Circuits
Handbook for Design and Application
2nd edition
www.tietze-schenk.com
mail@tietze-schenk.com
Translation of
Tietze, U.; Schenk, Ch.: Halbleiter-Schaltungstechnik. 12. edition, 2002
Existing translations:
Polish: Naukowo-Techniczne, Warsaw 1976, 1987, 1996
Hungarian: Müszaki, Budapest 1974, 1981, 1990
Russian: Mir, Moscow 1982, Dodeca Publishin, Moscow 2007
Spanish: Marcombo, Barcelona 1983
Chinese: (bootleg) 1985
English: Springer, Heidelberg 1978, 1991
The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply,
even in the absence of a specific statement, that such names are exempt from the relevant protective laws
and regulations and therefore free for general use.
987654321
springer.com
Preface
The purpose of this book is to help the reader to understand off-the-shelf circuits and to
enable him to design his or her own circuitry. The book is written for students, practicing
engineers and scientists. It covers all major aspects of analog and digital circuit design.
The book is a translation of the current 12th edition of the German bestseller Halbleiter-
Schaltungstechnik.
Part I describes semiconductor devices and their behavior with respect to the models
used in circuit simulation. This part introduces all major aspects of transistor level design
(IC-design). Basic circuits are analyzed in five steps: large-signal transfer characteristic,
small-signal response, frequency response and bandwidth, noise and distortion. Digital
circuits are covered starting with the internal circuitry of gates and flip-flops up to the
construction of combinatorial and sequential logic systems with PLDs and FPGAs. Design
examples and a short form guide for the digital synthesis tool ispLever are included on the
CD enclosed.
Part II is dedicated to board level design. The main chapters of this part describe
the use of operational amplifiers for signal conditioning including signal amplification,
filtering and AD-conversion. Further chapters cover power amplifiers, power supplies and
other important functional blocks of analog systems. The chapters are self-contained with
a minimum of cross-reference. This allows the advanced reader to familiarize himself
quickly with the various areas of applications. Each chapter offers a detailed overview of
various solutions to a given requirement. In order to enable the reader to proceed quickly
from an idea to a working circuit, we discuss only those solutions we have tested thoroughly
by simulation. Many of these simulation examples are included on the CD enclosed.
Part III describes circuits for analog and digital communication over wireless chan-
nels. The first chapter is dedicated to transmission channels, scattering parameters and
analog and digital modulations. Further chapters treat the architecture of transmitters and
receivers, the high frequency behavior of components, circuits for impedance matching,
high frequency amplifiers and mixers for frequency conversion.
To support analog circuit design, design examples and a short-form guide for the
well known circuit simulator PSpice are included on the CD. This package contains li-
braries with examples of scalable transistors for IC-like design. The library also supports
S-parameter and loop-gain simulations. An HTML-based index allows comfortable navi-
gation throughout the simulations.
Our homepage www.tietze-schenk.com offers updates, supplements and design exam-
ples. We encourage you to use our email-address mail@tietze-schenk.com for feedback
and comments.
We would like to thank Dr. Merkle at Springer Heidelberg for the administration,
Gerhard Büsching for the translation and Danny Lewis at PTP-Berlin for the assembly of
this book. In particular we like to thank Dr. Eberhard Gamm for the contribution of the first
four chapters of circuit design fundamentals in part I and the chapters of communications
in part III. We have added him as a young innovative author.
2. Bipolar Transistors 33
2.1 Performance of a Bipolar Transistor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.1.1 Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.1.2 Description by Way of Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.1.3 Characteristic of the Current Gain . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.1.4 Operating Point and Small-Signal Response . . . . . . . . . . . . . . . . 39
2.1.5 Limit Data and Reverse Currents . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.1.6 Thermal Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.1.7 Temperature Sensitivity of Transistor Parameters . . . . . . . . . . . . 53
2.2 Design of a Bipolar Transistor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.2.1 Discrete Transistors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.2.2 Integrated Transistors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
X Contents
4. Amplifiers 269
4.1 Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
4.1.1 Current Sources and Current Mirrors . . . . . . . . . . . . . . . . . . . . . . 277
4.1.2 Cascode circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
4.1.3 Differential Amplifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
4.1.4 Impedance Converters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
4.1.5 Circuits for Setting the Operating Point . . . . . . . . . . . . . . . . . . . . 395
4.2 Properties and Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
Contents XI
Bibliography 1525
Index 1529
Part I
The diode is a semiconductor component with two connections, which are called the anode
(A) and the cathode (K). Distinction has to be made between discrete diodes, which are
intended for installation on printed circuit boards and are contained in an individual case,
and integrated diodes, which are produced together with other semiconductor components
on a common semiconductor carrier (substrate). Integrated diodes have a third connection
resulting from the common carrier. It is called the substrate (S); it is of minor importance
for electrical functions.
A A A
p metal
n n
K K K
Graphical symbol pn diode Schottky diode
1.1
Performance of the Diode
The performance of a diode is described most clearly by its characteristic curve. This shows
the relation between current and voltage where all parameters are static which means that
they do not change over time or only very slowly. In addition, formulas that describe the
diode performance sufficiently accurately are required for mathematical calculations. In
most cases simple equations can be used. In addition, there is a model that correctly reflects
the dynamic performance when the diode is driven with sinusoidal or pulse-shaped signals.
This model is described in Sect. 1.3 and knowledge of it is not essential to understand the
fundamentals. The following Sections focus primarily on the performance of silicon pn
diodes.
1.1.1
Characteristic Curve
Connecting a silicon pn diode to a voltage VD = VAK and measuring the current ID
in a positive sense from A to K results in the characteristic curve shown in Fig. 1.2.
It should be noted that the positive voltage range has been enhanced considerably for
reasons of clarity. For VD > 0 V the diode operates in the forward mode, i.e. in the
conducting state. In this region the current rises exponentially with an increasing voltage.
When VD > 0.4 V, a considerable current flows. If −VBR < VD < 0 V the diode is
in the reverse-biased state and only a negligible current flows. This region is called the
reverse region. The breakthrough voltage VBR depends on the diode and for rectifier diodes
amounts to VBR = 50 . . . 1000 V. If VD < −VBR , the diode breaks through and a current
flows again. Only Zener diodes are operated permanently in this breakthrough region; with
all other diodes current flow with negative voltages is not desirable. With germanium and
Schottky diodes a considerable current flows in the forward region even for VD > 0.2 V,
and the breakthrough voltage VBR is 10 . . . 200 V.
In the forward region the voltage for typical currents remains almost constant due to the
pronounced rise of the characteristic curve. This voltage is called the forward voltage VF
ID
mA Schottky Silicon pn
ID
2.0
VD 1.5
1.0
0.5
– V BR
– 1.0
ID
µA
VD
–150 –100 – 50 V
–VBR
– 0.2
– 0.4
– 0.6
Fig. 1.3. Characteristic curve of a
– 0.8 small-signal diode in the reverse
region
and for both germanium and Schottky diodes lies at VF,Ge ≈ VF,Schottky ≈ 0.3 . . . 0.4 V
and for silicon pn diodes at VF,Si ≈ 0.6 . . . 0.7 V. With currents in the ampere range as used
in power diodes the voltage may be significantly higher since in addition to the internal
forward voltage a considerable voltage drop occurs across the spreading and connection
resistances of the diode: VF = VF,i + ID RB . In the borderline case of ID → ∞ the diode
acts like a very low resistance with RB ≈ 0.01 . . . 10 .
Figure 1.3 shows the enlarged reverse region. The reverse current IR = −ID is very
small with a low reverse voltage VR = −VD and increases slowly when the voltage
approaches the breakthrough voltage while it shoots up suddenly at the onset of the break-
through.
1.1.2
Description by Equations
Plotting the characteristic curve for the region VD > 0 in a semilogarithmic form results
approximately in a straight line (see Fig. 1.4); this means that there is an exponential relation
between ID and VD due to ln ID ∼ VD . The calculation on the basis of semiconductor
physics leads to [1.1]:
ID
A
1
100 m
10 m
1m
100 µ
10 µ
1µ
100 n
10 n
1n
0 0.5 1.0 VD
V
⎛ ⎞
VD
ID (VD ) = IS ⎝e VT − 1⎠ for VD ≥ 0
For the correct description of a real diode a correction factor is required which enables the
slope of the straight line in the semilogarithmic representation to be adapted [1.1]:
⎛ ⎞
VD
ID = IS ⎝e nVT − 1⎠ (1.1)
ID
A A RB = 0 RB > 0
ID
ID DVD
RB
(b) RB
VD VD VF DVD
(a)
(b) (a) VF VD
K K
a Diagram b Characteristic curve
Fig. 1.5. Simple equivalent circuit diagram for a diode without (-) and with (- -) spreading
resistance
The voltage VF is VF ≈ 0.6 V for silicon pn diodes and VF ≈ 0.3 V for Schottky diodes.
The corresponding circuit diagram and characteristic curve are shown in Fig. 1.5 as dashed
lines. Different cases must be distinguished for both variations, that is, it is necessary to
calculate with the switch open and closed and to determine the situation in which there
is no contradiction. The advantage is that either case leads to linear equations which are
easy to solve. In contrast, when using the e function according to (1.1), it is necessary to
cope with an implicit nonlinear equation that can only be solved numerically.
Example: Figure 1.6 shows a diode in a bridge circuit. To calculate the voltages V1 and V2
and the diode voltage VD = V1 − V2 it is assumed that the diode is in the reverse state,
that is, VD < VF = 0.6 V and the switch in the equivalent circuit is open. In this case, V1
and V2 can be determined by the voltage divider formula V1 = Vb R2 /(R1 + R2 ) = 3.75 V
and V2 = Vb R4 /(R3 + R4 ) = 2.5 V. This results in VD = 1.25 V, which does not comply
with the assumption. Consequently the diode is conductive and the switch in the equivalent
circuit is closed; this leads to VD = VF = 0.6 V and ID > 0. From the nodal equations
V1 Vb − V1 V2 Vb − V2
+ ID = , = ID +
R2 R1 R4 R3
it is possible to eliminate the unknown elements ID and V1 by adding the equations and
inserting V1 = V2 + VF ; this leads to:
1 1 1 1 1 1 1 1
V2 + + + = Vb + − VF +
R1 R2 R3 R4 R1 R3 R1 R2
This results in V2 = 2.76 V, V1 = V2 + VF = 3.36 V and in ID = 0.52 mA by substitution
in one of the nodal equations. The initial condition ID > 0 has been fulfilled, that is, there
is no contradiction and the solution has been found.
R1 R3
1kΩ ID 1kΩ
Vb 5V
V1 R2 VD R4 V2
3kΩ 1kΩ Fig. 1.6. Example for the demonstration of the use
of the equivalent circuit of Fig. 1.5
8 1 Diode
1.1.3
Switching Performance
In many applications the diodes operate alternately in the forward mode and in the reverse
mode, for example when rectifying alternating currents. The transition does not follow the
static characteristic curve as the parasitic capacitance of the diode stores a charge that builds
up in the forward state and is discharged in the reverse state. Figure 1.7 shows a circuit for
determining the switching performance with an ohmic load (L = 0) or an ohmic-inductive
load (L > 0). Applying a square wave produces the transitions shown in Fig. 1.8.
R L ID
+V
0 Vg VD Fig. 1.7. Circuit for determining
–V the switching performance
V
V
10
Vg
VF VD
60 70 80 90
0
10 20 30 40 t
1 ns
1, 2 3
3, 4 2
– 10
4
t=0
1 1N4148
L= 0
2 BAS40
– 20
3 1N4148
L = 5 µH
4 BAS40
ID
mA 15
1, 2
10
5 3, 4
4
60 70 80 90
0
0 10 20 30 40 t
ns
–5 2
3
– 10
1
Fig. 1.8. Switching performance of the silicon diode 1N4148 and the Schottky diode BAS40 in the
measuring circuit of Fig. 1.7 with V = 10 V, f = 10 MHz, R = 1 k and L = 0 or L = 5 mH
1.1 Performance of the Diode 9
ID VD
pin diode,
IF VFR IF high
tRR
IR VF
10 t
QRR
IR
t
Switching performance with ohmic load: With an ohmic load (L = 0) a current peak
caused by the charge built up in the capacitance of the diode occurs when the circuit is
activated. The voltage rises during this current peak from the previously existing reverse
voltage to the forward voltage VF which terminates the switch-on process. In pin diodes1
higher currents may cause a voltage overshoot (see Fig. 1.9b) as these diodes initially
have a higher spreading resistance RB at the switch-on point. Subsequently the voltage
declines to the static value in accordance with the decrease of RB . When switching off
there is a current in the opposite direction until the capacitance is discharged; then the
current returns to zero and the voltage drops to the reverse voltage. Since the capacitance
of Schottky diodes is much lower than that of silicon diodes of the same size, their turn-off
time is significantly shorter (see Fig. 1.8). Therefore, Schottky diodes are preferred for
rectifier diodes in switched power supplies with high cycle rates (f > 20 kHz), while the
lower priced silicon diodes are used in rectifiers for the mains voltage (f = 50 Hz). When
the frequency becomes so high that the capacitance discharge process is not completed
before the next conducting state starts, the rectification no longer takes place.
1 pin diodes have a nondoped (intrinsic) or slightly doped layer between the p and n layers in order
to achieve a higher breakthrough voltage.
2 With rectifier diodes the measurement is sometimes taken at 25 %.
10 1 Diode
the area below the x axis (see Fig. 1.9a). Both parameters depend on the previously flowing
forward current IF and the cutoff speed; therefore the data sheets show either information
on the measuring conditions or the measuring circuit. An approximation is QRR ∼ IF and
QRR ∼ |IR |tRR [1.2]; this means that in a first approximation the reverse recovery time is
proportional to the ratio of the forward and reverse current: tRR ∼ IF /|IR |. However, this
approximation only applies to |IR | < 3 . . . 5 · IF , in other words, tRR can not be reduced
endlessly. In pin diodes featuring a high breakdown voltage, the high cutoff speed may
even cause the breakdown to occur far below the static breakdown voltage VBR if the
reverse voltage at the diode increases sharply before the weakly doped i-layer is free of
charge carriers. With the transition to the forward state the forward recovery voltage VF R
occurs, which also depends on the actual switching conditions [1.3]; data sheets quote a
maximum value for VF R , typically VF R = 1 . . . 2.5 V.
1.1.4
Small-Signal Response
The performance of the diode when controlled by small signals around an operating point
characterized by VD,A and ID,A is called the small-signal response. In this case, the
nonlinear characteristic given in (1.1) can be replaced by a tangent to the operating point;
with the small-signal parameters
iD = ID − ID,A , vD = VD − VD,A
dID
1
iD =
vD = vD
dVD A rD
ID,A IS
dVD
nVT nVT
rD =
= ≈ (1.3)
dID A ID,A + IS ID,A
Thus, the equivalent small-signal circuit for the diode consists of a resistance with the value
rD ; with large currents rD becomes very small and an additional spreading resistance RB
must be introduced (see Fig. 1.10).
The equivalent circuit shown in Fig. 1.10 is only suitable for calculating the small-
signal response at low frequencies (0 . . . 10 kHz); therefore, it is called the DC small-
signal equivalent circuit. For higher frequencies it is necessary to use the AC small-signal
equivalent circuit given in Sect. 1.3.3.
rD RB
1.1.5
Limit Values and Reverse Currents
The data sheet for a diode shows limit values that must not be exceeded. These are the
limit voltages, limit currents and maximum power dissipation. In order to deal with positive
values for the limit data the reference arrows for the current and the voltage are reversed
in their direction for reverse-biased operation and the relevant values are given with the
index R (reverse); the index F (forward) is used for forward-biased operation.
Limit Voltages
Reaching the breakthrough voltage V(BR) or VBR causes the diode to break through in
the reverse mode and the reverse current rises sharply. Since the current already increases
markedly when approaching the breakthrough voltage, as shown in Fig. 1.3, a maximum
reverse voltage VR,max is specified up to which the reverse current remains below a limit
value in the mA range. Higher reverse voltages are permissible when driving the diode with
a pulse chain or a single pulse; they are called the repetitive peak reverse voltage VRRM
and the peak surge reverse voltage VRSM , respectively, and they are chosen so that the
diode remains undamaged. The pulse frequency is considered to be f = 50 Hz since it
is assumed that it will be used as a mains rectifier. Due to the reversed direction of the
reference arrow all voltages are positive and are related in the following way:
VR,max < VRRM < VRSM < V(BR)
Limit Currents
For forward-biased operation a maximum steady-state forward current IF,max is specified.
It applies to situations in which the diode case is kept at a temperature of T = 25 ◦ C; at
higher temperatures the permissible steady-state current is lower. Higher forward currents
are permissible when driving the diode with several pulses or a single pulse; they are called
the repetitive peak forward current IF RM and the peak surge forward current IF SM ,
respectively, and they depend on the duty cycle or the pulse duration. The currents are
related:
IF,max < IF RM < IF SM
With very short single pulses IF SM ≈ 4 . . . 20 · IF,max . The current IF RM is of particular
importance for rectifier diodes because of their pulsating periodic current (see Sect. 16.2);
in this case the maximum value is much higher than the mean value.
For the breakthrough region a maximum current-time area I 2 t is quoted which may
occur at the breakthrough caused by a pulse:
I t =
2
IR2 dt
Reverse Current
The reverse current IR is measured at a reverse voltage below the breakthrough voltage
and depends largely on the reverse voltage and the temperature of the diode. At room
temperature IR = 0.01 . . . 1 mA for a small-signal silicon diode, IR = 1 . . . 10 mA for
12 1 Diode
a small-signal Schottky diode and a silicon rectifier diode in the Ampere range and IR >
10 mA for a Schottky rectifier diode; at a temperature of T = 150 ◦ C these values are
increased by a factor of 20 . . . 200.
1.1.6
Thermal Performance
The thermal performance of components is described in Sect. 2.1.6 for bipolar transistors;
the parameters and conditions described there also apply to the diode when PV is replaced
by the power dissipation of the diode.
1.1.7
Temperature Sensitivity of Diode Parameters
The characteristic curve of a diode is heavily dependent on the temperature; an explicit
statement of the temperature sensitivity means for the silicon pn diode [1.1]
V
D
ID (VD , T ) = IS (T ) e nVT (T ) − 1
with:
T =300 K
kT mV
VT (T ) = = 86.142 T ≈ 26 mV
q K
T VG (T ) xT ,I
−1 T n
IS (T ) = IS (T0 ) e T0 nVT (T ) with xT ,I ≈ 3 (1.4)
T0
Here, k = 1.38 · 10−23 VAs/K is Boltzmann’s constant, q = 1.602 · 10−19 As is the
elementary charge and VG = 1.12 V is the gap voltage of silicon; the low temperature
sensitivity of VG may be ignored. The temperature T0 with the respective current IS (T0 )
serves as a reference point; usually T0 = 300 K is used.
In reverse mode the reverse current IR = −ID ≈ Is flows; with xT ,I = 3 this yields
the temperature coefficient of the reverse current:
1 dIR 1 dIS 1 VG
≈ = 3+
IR dT IS dT nT VT
1.2 Construction of a Diode 13
1 VG − VD T =300 K
= 3+ ≈ 0.04 . . . 0.08 K−1
ID dT
VD =const. nT VT
By means of the total differential
∂ID ∂ID
dID = dVD + dT = 0
∂VD ∂T
the temperature-induced change of VD at constant current can be determined:
T =300 K
dVD
VD − VG − 3VT VD =0.7 V mV
= ≈ − 1.7 (1.5)
dT
ID =const. T K
This means that the forward voltage decreases when the temperature rises; a temperature
increase of 60 K causes a drop in VD of approximately 100 mV. This effect is used in
integrated circuits for measuring the temperature.
These results also apply to Schottky diodes when setting xT ,I ≈ 2 and replacing the
gap voltage VG by the voltage that describes the energy difference between the n and metal
regions: VMn = (WMetal − Wn-Si )/q; thus VMn ≈ 0.7 . . . 0.8 V [1.1].
1.2
Construction of a Diode
Diodes are manufactured in a multi-step process on a semiconductor wafer that is then cut
into small dies. On one chip there is either a discrete diode or an integrated circuit (IC),
comprising several components.
1.2.1
Discrete Diode
Internal design: Discrete diodes are mostly produced using epitaxial-planar technology.
Figure 1.11 illustrates the construction of a pn and a Schottky diode where the active areas
are particularly emphasized. Doping is heavy in the n+ layer, medium in the p layer and
low in the n− layer. The special arrangement of differently doped layers helps to minimize
the spreading resistance and to increase the breakthrough voltage. Almost all pn diodes are
designed as pin diodes, in other words, they feature a middle layer with little or no doping
14 1 Diode
A A A A
Al Al
SiO 2 SiO 2
p p metal –
– n
n
+
Si +
Si
n n n n
Al Al
K K K K
and with a thickness that is roughly proportional to the breakthrough voltage; in Fig. 1.11a
this is the n− layer. For practical purposes diodes are referred to as pin diodes only if the
lifetime of the charge carriers in the middle layer is very high, thus producing a particular
characteristic; this will be described in more detail in Sect. 1.4.2. In Schottky diodes the
weakly doped n− layer is required for the Schottky contact (see Fig. 1.11b); in contrast a
junction between metal and a layer of medium or heavy doping produces an inferior diode
effect or no effect at all, in which case it behaves rather like a resistor (ohmic contact).
Case: To mount a diode in a case the bottom side is soldered to the cathode terminal or
connected to a metal part of the case. The anode side is connected to the anode terminal
via a fine gold or aluminum bond wire. Finally the diode is sealed in a plastic compound
or mounted in a metal case with screw connector.
For the various diode sizes and applications there is a multitude of case designs that
differ in the maximum heat dissipation capacity or are adapted to special geometrical
requirements. Figure 1.12 shows a selection of common models. Power diodes are provided
with a heat sink for their installation; the larger the contact surface, the better the heat
dissipation. Rectifier diodes are often designed as bridge rectifiers consisting of four diodes
to serve as full-wave rectifiers in power supply units (see Sect. 1.4.4); the mixer described
in Sect. 1.4.5 is also made of four diodes. High-frequency diodes require special cases
because in the GHz frequency range their electrical performance depends on the case
geometry. Often, the case is omitted altogether and the diode chip is soldered or bonded
directly to the circuit.
1.2.2
Integrated Diode
Integrated diodes are also produced using epitaxial-planar technology. Here, all connec-
tions are located at the top of the chip and the diode is electrically isolated from other
components by a reverse-biased pn junction. The active region is located in a very thin
layer at the surface. The depth of the chip is called the substrate (S) and forms a common
connection for all components of the integrated circuit.
Internal construction: Figure 1.13 illustrates the design of an integrated pn diode. The
current flows from the p layer through the pn junction to the n− layer and from there via
the n+ layer to the cathode; a low spreading resistance is achieved by means of the heavily
doped n+ layer.
Substrate diode: The equivalent circuit diagram in Fig. 1.13 shows an additional sub-
strate diode located between the cathode and the substrate. The substrate is connected
to the negative supply voltage so that this diode is always in the reverse mode to act as
isolation relative to other components and the substrate.
A S A K
Al
SiO 2
1 p – +
n n
K p
+ 2 1
+
n
2
p 2
S
Fig. 1.13. Equivalent circuit and construction of an integrated pn diode with useful diode (1) and
parasitic substrate diode (2)
16 1 Diode
1.3
Model of a Diode
Section 1.1.2 describes the static performance of the diode using an exponential function;
but this neglects the breakthrough and the second-order effects in the forward operation.
For computer-aided circuit design a model is required that considers all of these effects
and, in addition, correctly reflects the dynamic performance. The dynamic small-signal
model is derived from this large-signal model by linearization.
1.3.1
Static Performance
The description is based on the ideal diode equation given in (1.1) and also takes other
effects into account. A standardized diode model like the the Gummel-Poon model for
bipolar transistors does not exist; some of the CAD programs therefore have to use several
diode models to describe a real diode with all of its current components. The diode model
is almost unnecessary for the design of integrated circuits since here the base-emitter diode
of a bipolar transistor is usually used as a diode.
The model parameters are the saturation reverse current IS and the emission coefficient
n. For the ideal diode n = 1; for real diodes n ≈ 1 . . . 2. This range is called the diffusion
range.
In Schottky diodes the emission current takes the place of the diffusion current. But
since both current conducting mechanisms lead to the same characteristic curve (1.6) can
also be used for Schottky diodes [1.1, 1.3].
Other Effects
With very small and very high forward currents as well as in reverse operation there are
deviations from the ideal performance according to (1.6):
– High forward currents produce the high-current effect, which is caused by a sharp rise
in the charge carrier concentration at the edge of the depletion layer [1.1]; this is also
referred to as a strong injection. This also affects the diffusion current and is described
by an extension to (1.6).
– Because of the recombination of charge carriers in the depletion layer a leakage or
recombination current IDR occurs in addition to the diffusion current which is described
by a separate equation [1.1].
– The application of high reverse voltages causes the diode to break through. The break-
through current IDBR is also described in an additional equation.
1.3 Model of a Diode 17
High-current effect: The high-current effect causes the emission coefficient to rise from
n in the medium current range to 2n for ID → ∞; it can be described by an extension to
(1.6) [1.4]: ⎛ ⎞
VD
IS ⎝e nVT − 1⎠ ⎧ VD VD
⎪
⎨
IS e nVT for IS e nVT < IK
IDD = ⎛ ⎞ ≈ ⎪ (1.8)
⎩ √
VD VD
VD
IS ⎝ nV IS IK e 2nVT for IS e nVT > IK
1 + e T − 1⎠
IK
An additional parameter is the knee-point current IK , which marks the beginning of the
high-current region.
Leakage current: Based on the ideal diode theory the following is applicable to the
leakage current [1.1]:
⎛ ⎞
VD
IDR = IS,R ⎝e nR VT − 1⎠
This equation only describes the recombination current accurately enough for forward
operation. Setting VD → −∞ yields a constant current IDR = −IS,R in the reverse region,
while in a real diode the recombination current rises with an increasing reverse voltage. A
more accurate description is achieved by taking into account the voltage sensitivity of the
width of the depletion layer [1.4]:
⎛ ⎞
mJ
VD 2 2
V
IDR = IS,R ⎝e nR VT − 1⎠
D
1− + 0.005 (1.9)
VDiff
Additional parameters are the leakage saturation reverse current IS,R , the emission coef-
ficient nR ≥ 2, the diffusion voltage VDiff ≈ 0.5 . . . 1 V and the capacitance coefficient
mJ ≈ 1/3 . . . 1/2.3 From (1.9) it follows that:
|VD | mJ
IDR ≈ − IS,R for VD < − VDiff
VDiff
The magnitude of the current rises as the reverse voltage increases; its actual curve depends
on the capacitance coefficient mJ . In the forward mode the additional factor given in (1.9)
has almost no effect since in this case the exponential dependence of VD is dominant.
Since IS,R IS , the recombination current is larger than the diffusion current at low
positive voltages; this region is called the recombination region. For
nnR IS,R
VD,RD = VT ln
nR − n IS
both currents have the same value. With larger voltages the diffusion current becomes
dominant and the diode operates in the diffusion region.
3V
Diff and mJ are primarily used to describe the depletion layer capacitance of the diode (see
Sect. 1.3.2).
18 1 Diode
I D [log]
IK
IS I K
I S,R
IS
I II III
Fig. 1.14. Semi-logarithmic diagram of
ID in forward mode: (I) recombination,
VD,RD VD (II) diffusion, (III) high-current regions
Figure 1.14 is the semilogarithmic presentation of ID in the forward region and shows
the importance of parameters IS , IS,R and IK . In some diodes the emission coefficients
n and nR are almost identical. In such cases the semilogarithmic characteristic curve has
the same slope in the recombination and diffusion regions and can be described for both
regions using one exponential function.4
Breakthrough: For VD < −VBR the diode breaks through; the flowing current can be
approximated by an exponential function [1.5]:
VD +VBR
−
IDBR = − IBR e nBR VT (1.10)
For this, the breakthrough voltage VBR ≈ 50 . . . 1000 V, the breakthrough knee-point
current IBR and the breakthrough emission coefficient nBR ≈ 1 are required. For nBR = 1
and VT ≈ 26 mV the current is:5
− IBR for VD = − VBR
ID ≈ IDBR =
− 1010 IBR for VD = − VBR − 0.6 V
Quoting IBR and VBR is not a clear definition since the same curve can be described with
different value sets (VBR , IBR ); therefore, the model for a certain diode may have different
parameters.
Spreading Resistance
The spreading resistance RB is necessary for the full description of the static performance;
according to Fig. 1.15 it is comprized of the resistances of the various layers and it is
represented in the model by a series resistor. A distinction has to be made between the
internal diode voltage VD
and the external diode voltage
VD = VD
+ ID RB . (1.11)
In the equations for IDD , IDR and IDBR voltage VD must be replaced by VD
. The spreading
resistance is between 0.01 for power diodes and 10 for small-signal diodes.
4 Figure 1.4 shows the characteristic curve of such a diode.
5 Based on 10V ln 10 = 0.6 V.
T
1.3 Model of a Diode 19
A
A
p RB1
V ´D
–
n RB2 VD
RB
+
n RB3
K
K
Fig. 1.15. Spreading resistance of
a In the diode b In the model a diode
1.3.2
Dynamic Performance
The response to pulsating or sinusoidal signals is called the dynamic performance, and it
cannot be derived from the characteristic curves. The reasons for this are the nonlinear
junction capacitance of the pn or metal-semiconductor junction and the diffusion charge
that is stored in the pn junction and determined by the diffusion capacitance, which is also
nonlinear.
Junction Capacitance
A pn or metal–semiconductor junction has a voltage-dependent junction capacitance CJ
that is influenced by the doping of the adjacent layers, the doping profile, the area of the
junction and the applied voltage VD
. The junction can be visualized as a plate capacitor
with the capacitance C = A/d; where A represents the junction area and d the junction
width. A simplified view of the pn junction gives d(V ) ∼ (1 − V /VDiff )mJ [1.1] and thus:
CJ 0
CJ (VD
) = mJ for VD
< VDiff (1.12)
V
1− D
VDiff
⎪
⎪
⎪
⎪ 1−
⎪
⎨ VDiff
CJ (VD
) = CJ 0 (1.13)
⎪
⎪ mJ VD
⎪
⎪ − + +
⎪
⎪ 1 fC (1 m J )
⎪
⎪ VDiff
⎪
⎪ for VD
> fC VDiff
⎩
(1 − fC )(1+mJ )
20 1 Diode
where fC ≈ 0.4 . . . 0.7. Figure 2.32 on page 70 shows the curve of CJ for mJ = 1/2 and
mJ = 1/3.
Diffusion Capacitance
In forward operation the pn junction contains a stored diffusion charge QD that is propor-
tional to the diffusion current flowing through the pn junction [1.2]:
QD = τT IDD
The parameter τT is the transit time. Differentiation of (1.8) produces the diffusion capac-
itance:
V
IS nVD
1+ e T
dQD τT IDD 2IK
CD,D (VD
) =
= (1.14)
dVD nVT V
IS nVD
1+ e T
IK
For the diffusion region IDD IDR and thus ID ≈ IDD , meaning that the diffusion
capacitance can be approximated by:
ID
1+ ID IK
τT ID 2IK τT ID
CD,D ≈ ≈ (1.15)
nVT ID nVT
1+
IK
In silicon pn diodes τT ≈ 1 . . . 100 ns; in Schottky diodes the diffusion charge is negligible,
since τT ≈ 10 . . . 100 ps.
Fig. 1.16. Full model of a diode Fig. 1.17. Variables of the diode model
1.3.3
Small-Signal Model
The linear small-signal model is derived from the nonlinear model by linearization at an
operating point. The static small-signal model describes the small-signal response at low
frequencies and is therefore called the DC small-signal equivalent circuit. The dynamic
small-signal model also describes the dynamic small-signal response and is required for
calculating the frequency response of a circuit; it is called the AC small-signal equivalent
circuit.
dVD
dVD
= + RB = rD + RB
dI
D A I
D A
It is made up of the spreading resistance RB and the differential resistance rD of the inner
diode (see Fig. 1.10). Resistance rD comprises three portions corresponding to the three
current components IDD , IDR and IDBR :
1 dID
dIDD
dIDR
dIDBR
= = + +
rD dVD
A dVD
A dVD
A dVD
A
The differentiation of (1.6), (1.9) and (1.10) produces complex expressions; for practical
purposes the following approximations may be used:
IDD,A
1+ IS IDD,A IK
1 dIDD
⎨ n R VT
=
≈
rDR dVD A ⎪
⎪ IS,R
⎪
⎩ for IDR,A < 0
mJ
1 dIDBR
IDBR,A
= = −
rDBR dVD
A nBR VT
Thus, the differential resistance rD is:
rD = rDD ||rDR ||rDBR
For operating points that are in the diffusion region and below the high-current region
ID,A ≈ IDD,A and ID,A < IK ;7 the following approximation can be used:
nVT
rD = rDD ≈ . (1.16)
ID,A
RB rD LG RB rD
CD CD
CG
With small-signal diodes in reverse mode the diffusion resistance is rD ≈ 106 . . . 109 ;
in the Ampere region of rectifier diodes this value is reduced by a factor of 10 . . . 100.
The small-signal resistance in the breakthrough region is required only for Zener diodes
since only in Zener diodes an operating point in the breakthrough range is permissible; the
resistance is therefore called rz . For ID,A ≈ IDBR,A its value is:
nBR VT
rZ = rDBR = (1.17)
|ID,A |
CD = CJ (VD
) + CD,D (VD
)
In high-frequency diodes the additional parasitic influences of the case must be taken
into consideration: Figure 1.20b shows the extended model with a case inductivity LG ≈
1 . . . 100 nH and a case capacitance of CG ≈ 0.1 . . . 1 pF [1.6].
Simplified model: For practical calculations the spreading resistance RB can be ignored
and approximations can be used for rD and CD . From (1.15), (1.16) and the estimation
CJ (VD
) ≈ 2CJ 0 the values for forward operation are:
nVT
rD ≈ (1.18)
ID,A
τT ID,A τT
CD ≈ + 2CJ 0 = + 2CJ 0 (1.19)
nVT rD
For reverse operation rD is ignored, that is, rD → ∞ and CD ≈ CJ 0 .
24 1 Diode
1.4
Special Diodes and Their Application
1.4.1
Zener Diode
A Zener diode has a precisely specified breakthrough voltage that is rated for continuous
operation in the breakthrough region; it is used for voltage stabilization or limitation. In
Zener diodes the breakthrough voltage VBR is called the Zener voltage VZ and amounts to
VZ ≈ 3 . . . 300 V in standard Zener diodes. Figure 1.21 shows the graphic symbol and the
characteristic for a Zener diode. The current in the breakthrough region is given by (1.10):
VD +VZ
−
ID ≈ IDBR = − IBR e nBR VT
dVZ
TC =
dT
T =300 K,ID =const.
determines the voltage variation at a constant current:
VZ (T ) = VZ (T0 ) (1 + T C (T − T0 )) with T0 = 300 K
When the Zener voltage is below 5 V, the Zener effect dominates with a negative tem-
perature coefficient, while higher voltages produce the avalanche effect with a positive
temperature coefficient; typical values are T C ≈ −6 · 10−4 K −1 for VZ = 3.3 V, T C ≈ 0
for VZ = 5.1 V and T C ≈ 10−3 K −1 for VZ = 47 V.
The differential resistance in the breakthrough region is denoted by rZ and corresponds
to the reciprocal of the slope of the characteristic; from (1.17) it follows that:
dVD nBR VT nBR VT VD
rZ = = = − ≈
dID |ID | ID ID
The differential resistance depends largely on the emission coefficient nBR that reaches
a minimum of nBR ≈ 1 . . . 2 with VZ ≈ 8 V and increases with lower or higher Zener
voltages; typical values are nBR ≈ 10 . . . 20 for VZ = 3.3 V and nBR ≈ 4 . . . 8 for
ID
A
– VZ
VD VF VD
ID
I D
VD
rZ ≈
K VD I D
Vo
RV
VZ
ID
Vi
RL Vo
RV Vi
(
VZ 1+ R
L
(
a Circuit diagram b Characteristic
VZ = 47 V. The voltage-stabilizing effect of the Zener diode is based on the fact that the
characteristic is very steep in the breakthrough region so that the differential resistance is
very low; Zener diodes are best suited with VZ ≈ 8 V since here the characteristic shows
the steepest slope due to the minimum value of nBR . For |ID | = 5 mA the resistance is
rZ ≈ 5 . . . 10 for VZ = 8.2 V and rZ ≈ 50 . . . 100 for VZ = 3.3 V.
Figure 1.22a displays a typical circuit for voltage stabilization. For 0 ≤ Vo < VZ the
Zener diode is reverse-biased and the output voltage is generated by voltage division with
resistors RV and RL :
RL
Vo = Vi
RV + R L
Vo ≈ VZ applies to the Zener diode in the conductive state. For the characteristic curve
shown in Fig. 1.22b this means that:
⎧
⎪ RL RV
⎪
⎨ iV for Vi < VZ 1 +
RV + R L RL
Vo ≈
⎪
⎪ R V
⎩ VZ for Vi > VZ 1 +
RL
In order to render the stabilization effective, the operating point must be in the region in
which the characteristic is almost horizontal. From the nodal equation
Vi − Vo Vo
+ ID =
RV RL
differentiation by Vo generates the smoothing factor
dVi RV RV rZ RV ,RL RV
G = = 1+ + ≈ (1.20)
dVo rZ RL rZ
and the stabilization factor [1.7]:
dVi
Vi Vo dVi Vo Vo RV
S = = = G ≈
dVo Vi dVo Vi Vi rZ
Vo
Vo
RV
VZ
Vi Vo
– VF Vi
this circuit section as a resistor RL = VA /IA = 5.1 k and use the Zener diode circuit
in Fig. 1.22 with VZ = 5.1 V if Vi = Vb and Vo = VA . The series resistor RV must be
selected in such a way that G = dVi /dVo > 1 V/10 mV = 100; therefore from (1.20) it
follows that RV ≈ GrZ ≥ 100rZ . The nodal equation leads to
Vi − Vo Vo Vb − VA
− ID = − = − IA
RV RL RV
and (1.17) leads to −ID = nBR VT /rZ ; by setting RV = GrZ , G = 100 and nBR = 2 the
resistor RV is:
Vb − VA − GnBR VT
RV = = 1.7 k
IA
Then the currents are IV = (Vb − VA )/RV = 4.06 mA and |ID | = IV − IA = 3.06 mA.
It can be seen that the Zener diode causes the current to be much higher than the current
consumption IA for the circuit section to be supplied. Therefore, this type of voltage
stabilization is suitable only for partial circuits with a low current input. Circuits with a
higher current input require a voltage regulator that may be more expensive but, as well as
lower power losses, it also offers a better stabilization effect.
The circuit shown in Fig. 1.22a can also be used for voltage limitation. Removing the
resistor RL in Fig. 1.22a leads to the circuit in Fig. 1.23a with the characteristic shown in
Fig. 1.23b:
⎧
⎪ − VF for Vi ≤ − VF
⎨
Vo ≈ Vi for − VF < Vi < VZ
⎪
⎩
VZ for Vi ≥ VZ
In the medium range the diode is reverse-biased, that is, Vo = Vi . For Vi ≥ VZ the diode
breaks through and limits the output voltage to VZ . For Vi ≤ −VF ≈ 0.6 V the diode
operates in the forward mode and limits negative voltages to the forward voltage VF . The
circuit in Fig. 1.24a allows a symmetrical limitation with |Vo | ≤ VZ + VF ; in the event of
limitation one of the diodes is forward-biased and the other breaks through.
1.4 Special Diodes and Their Application 27
Vo
RV
VZ + V F
Vi Vo
Vi
– V Z – VF
I0
R1 R1
Vi Vo Vi Vo
rD
ID = I 0
Fig. 1.25. Voltage divider for alternating voltages with pin diode
1.4.2
Pin Diode
In pin diodes8 the life cycle τ of the charge carriers in the nondoped i layer is particularly
long. Since a transition from the forward-biased to the reverse-biased mode occurs only
after recombination of almost all charge carriers in the i layer, a conductive pin diode
remains in the forward mode even with short negative voltage pulses of a pulse duration
tP τ . The diode then acts as an ohmic resistor, with a value that is proportional to the
charge in the i layer and thus proportional to the mean current I D,pin [1.8]:
nVT
rD,pin ≈ with n ≈ 1 . . . 2
I D,pin
On the basis of this property the pin diode may be used with alternating voltages of a
frequency f 1/τ as a DC-controlled AC resistance. Figure 1.25 shows the circuit and
the small-signal equivalent circuit of a simple variable voltage divider using a pin diode. In
high frequency circuits mostly π attenuators with three pin diodes are used (see Fig. 1.26);
a variable attenuation and a matching of both sides to a certain resistance of usually 50
is then achieved by means of suitable control signals. The capacitances and inductances
in Fig. 1.26 result in a separation of the DC and AC circuit paths. Typical pin diodes have
τ ≈ 0.1 . . . 5 ms; this makes the circuit suitable for frequencies f > 2 . . . 100 MHz 1/τ .
8 Most pn diodes are designed as pin diodes, so that a high reverse voltage is reached across the
i layer. The term pin diode is used only for diodes with lower impurity concentrations and a
correspondingly higher life cycle of the charge carriers in the i layer.
28 1 Diode
V1
V2
Another important feature of pin diodes is the low junction capacitance due to a rel-
atively thick i layer. This allows pin diodes to be used also for high frequency switches
that provide a good off-state attenuation because of the low junction capacitance when the
switch is open (I D,pin = 0). The typical circuit of an RF switch corresponds largely to
the attenuator circuit shown in Fig. 1.26 which is designed as a short-series-short-switch
with a particularly high off-state attenuation.
1.4.3
Varactor Diodes
Due to the voltage sensitivity of the junction capacitance a diode can be used as a variable
capacitor (varactor); in this case the diode is operated in reverse mode and the junction
capacitance is controlled by the reverse voltage. Equation (1.12) shows that the region in
which the capacitance can be varied depends to a large degree on the capacitance coefficient
mJ and increases as mJ increases. A particularly large range of 1 : 3 . . . 10 is reached in
diodes with hyperabrupt doping (mJ ≈ 0.5 . . . 1) in which the impurity concentration
increases close to the pn border just at the junction to the other region [1.8]. Diodes with
this doping profile are called variable-capacitance diodes (tuning diodes, varicap) and are
used predominantly for frequency tuning in LC oscillator circuits. Figure 1.27 shows the
graphic symbol of a varactor diode and the curve of the junction capacitance CJ for some
typical diodes. Although the curves are similar only diode BB512 shows the particular
characteristic of a steeply decreasing junction capacitance. The capacitance coefficient
mJ can be derived from the slope in the double logarithmic diagram; therefore Fig. 1.27
also depicts the slopes for mJ = 0.5 and mJ = 1.
In addition to the curve of the junction capacitance CJ , the quality factor Q is an
important measure for the performance of a varactor diode. From the quality definition9
|Im {Z} |
Q =
Re {Z}
and the impedance of the diode
1 s=j ω 1
Z(s) = RB + = RB +
sCJ j ωCJ
CJ
pF
1000
500 BB512
200
100 BB814
50 1
mJ =
2
BB535
20 mJ = 1
10
BBY51
5
0.5 1 2 5 10 20 – VD
V
Q is derived as [1.8]:
1
Q =
ωCJ RB
For a given frequency, Q is inversely proportional to the spreading resistance RB . There-
fore, a high performance level is equivalent to a low spreading resistance and corresponds
to low losses and a low damping when used in resonant circuits. Typical diodes have a
quality factor of Q ≈ 50 . . . 500. As it is principally the spreading resistance that is needed
for simple calculations and for circuit simulations new data sheets often specify RB only.
In most cases, the circuits shown in Fig. 1.28 are used for frequency tuning in LC
resonant circuits. In the circuit depicted in Fig. 1.28a both the junction capacitance CJ of
the diode and the coupling capacitance CK are connected in series and arranged in parallel
with the parallel resonant circuit consisting of L and C. The tuning voltage VA > 0 is
provided via the inductivity LB ; with respect to the AC voltage this isolates the resonant
circuit from the voltage source VA and prevents the resonant circuit from being short-
circuited by the voltage source. It is essential that LB L is chosen to ensure that LB
does not affect the resonant frequency. The tuning voltage may also be provided via a
resistor which, however, is an additional load to the resonant circuit and thus reduces the
D1
LB CK LB
C L
VA VA
D1 C L D2
quality of the circuit. The coupling capacitance CK prevents the voltage source VA from
being short-circuited by the inductance L of the resonant circuit. Provided that LB L,
the resonant frequency is:
CK CJ (VA )
1 1
ωR = 2πfR = ≈
CJ (VA ) CK L (C + CJ (VA ))
L C+
CJ (VA ) + CK
The tuning range depends on the characteristic of the junction capacitance and its relation
to the resonant circuit capacitance C. The maximum tuning range is achieved with C = 0
and CK CJ .
In the circuit depicted in Fig. 1.28b a series connection of two junction capacitances is
arranged in parallel to the resonant circuit. Here, too, the inductivity LB L prevents a
high-frequency short-circuit of the resonant circuit by the voltage source VA . A coupling
capacitance is not required since both diodes are in reverse mode so that no DC current
can flow into the resonant circuit. In this case, the resonant frequency is:
1
ωR = 2πfR =
CJ (VA )
L C+
2
Here, again, the tuning range is maximum for C = 0; however, only half the junction ca-
pacitance is effective so that compared to the circuit shown in Fig. 1.28a either the junction
capacitance or the inductance must be twice as high for the same resonant frequency. A
material advantage of the symmetrical diode arrangement is the improved linearity with
high amplitudes in the resonant circuit; this largely offsets the decrease in the resonant
frequency with increasing amplitudes that is caused by the nonlinearity of the junction
capacitance [1.3].
1.4.4
Bridge Rectifier
The circuit shown in Fig. 1.29 made up of four diodes is called a bridge rectifier and
is used for full-way rectification in power supplies and AC voltmeters. Bridge rectifiers
for power supplies are divided into high-voltage bridge rectifiers, which are used for
direct rectification of the mains voltage and must therefore have a high breakdown voltage
(VBR ≥ 350 V), and low-voltage bridge rectifiers, which are used on the secondary side
of a line transformer; Sect. 16.5 describes this in more detail. Of the four connections two
are marked with ∼ and one each with + and −.
Ii ~
D4 D1
Io
Vi – +
D3 D2 Vo
~
Vo Io
2VF
Vi Ii
With a positive input voltage D1 and D3 are conductive while D2 and D4 are reverse-
biased; with a negative input voltage D2 and D4 are conductive while D1 and D3 are
reverse-biased. Since at any given moment the current flows through two conductive diodes,
the rectified output voltage is lower (by 2VF ≈ 1.2 . . . 2 V) than the magnitude of the input
voltage:
0 for |Vi | ≤ 2VF
Vo ≈
|Vi | − 2VF for |Vi | > 2VF
Figure 1.30a shows the voltage characteristic. A peak reverse voltage of |VD |max =
|Vi |max , which must be lower than the breakthrough voltage of the diodes, occurs across
the diodes in reverse mode.
Unlike the voltages the magnitudes of the currents are in a linear relationship (see
Fig. 1.30b):
Io = |Ii |
This fact is used in meter rectifiers; the AC voltage to be measured is fed through a voltage-
to-current converter and the resulting current is rectified in a bridge rectifier.
1.4.5
Mixer
Mixers are used in communication systems for frequency conversion. There are passive
mixers, which use diodes or other passive components, and active mixers, which use
transistors. In the case of passive mixers, the ring modulator consisting of four diodes
and two transformers with centre tabs is most frequently used. Figure 1.31 shows a ring
modulator in downconverter configuration with diodes D1 . . . D4 and transformers L1 −L2
and L3 − L4 [1.9]. The circuit converts the input signal VRF with the frequency fRF
by means of the local oscillator voltage VLO with a frequency fLO to an intermediate
frequency fI F = |fRF − fLO |. The output voltage VI F is supplied to a resonant circuit in
tune with the intermediate frequency in order to strip the signal from additional frequency
components generated in the conversion process. The local oscillator provides a sinusoidal
or rectangular voltage with an amplitude v̂LO ; VRF and VI F are sinusoidal voltages of the
amplitudes v̂RF and v̂I F respectively. In normal operation v̂LO v̂RF > v̂I F applies; in
other words, the voltage of the local oscillator determines which diodes are conductive;
32 1 Diode
the following applies when using a 1:1 transformer with L4 = L3a + L3b :
⎫ ⎧
VLO ≥ 2VF ⎬ ⎨ D1 and D2 are conductive
− 2VF < VLO < 2VF ⇒ No diode is conductive
⎭ ⎩
VLO < − 2VF D3 and D4 are conductive
VF is the forward voltage of the diodes. Due to their better switching performance Schottky
diodes with VF ≈ 0.3 V are used exclusively; the current through the diodes is limited by
the internal resistance RLO of the local oscillator.
When D1 and D2 are conductive a current caused by VRF flows through L2a and
D1 − L3a or D2 − L3b in the IF resonant circuit; when D3 and D4 are conductive, the
current flows through L2b and D3 − L3b or D4 − L3a . The polarity of VI F is different
from that of VRF so that the local oscillator and the diodes cause a polarity change at the
frequency fLO (see Fig. 1.32). If VLO is a square wave signal with v̂LO > 2VF the change
in polarity occurs suddenly; that is, the ring modulator multiplies the input signal with the
square wave signal. The IF filter extracts the desired components with m = 1, n = −1 or
m = −1 and n = 1 from the generated frequency components in the form |mfLO + nfRF |
with any integer value for m and n = ±1.
The ring modulator is available as a component with six connections, two each at the
RF, LO, and IF sides [1.9]. Furthermore, there are integrated circuits containing only the
diodes, and therefore have only four connections. In this context it must be noted that,
despite their similarity in form, the mixer and the bridge rectifier differ from one another
in terms of the arrangement of the diodes, as shown by a comparison of Figs. 1.31 and 1.29.
VRF L R C VIF
f L0
The bipolar transistor is a semiconductor component with three terminals that are known
as the base (B), emitter (E) and collector (C). There are discrete transistors that are used
for mounting to printed circuit boards and are contained in their own individual case and
integrated transistors that are produced together with other semiconductor elements on a
common substrate. Integrated transistors feature a fourth connection called the substrate
(S), which represents the common carrier; it is of secondary importance for the transistor’s
electrical function.
Equivalent circuits with diodes: Bipolar transistors consist of two anti-serially con-
nected pn diodes that have a common p or n region. Fig. 2.1 shows the graphic symbol
and the equivalent diode circuits of an npn transistor with a common p region and a pnp
transistor with a common n region. The equivalent diode circuit diagrams, however, do not
correctly reflect the bipolar transistor function but allow an overview of operating modes
and illustrate how the type (npn or pnp) and the base terminal of an unknown transistor
can be determined with the help of a continuity tester; due to the symmetrical design it is
not easy to distinguish the collector from the emitter.
Operating modes: The bipolar transistor is used to amplify and to switch signals and is
usually operated in normal mode (forward region), meaning that the emitter diode junction
(BE diode) is biased in the forward direction and the collector diode junction (BC diode)
is reverse-biased. In some applications the BC diode may also be operated temporarily
in the forward region, which is called the saturation region. Interchanging the emitter
and collector makes the transistor operate in the reverse region; this operating mode is
advantageous only in exceptional cases. In the cutoff region both diodes are nonconductive.
Figure 2.2 shows the polarity of voltages and currents for pnp and npn transistors in normal
operating mode.
C C C C
n p
B B p B B n
n p
E E E E
a npn transistor b pnp transistor
IC > 0 IC < 0
IB > 0 IB < 0
VBE > 0 IE < 0 VCE > 0 VBE < 0 IE > 0 VCE < 0
2.1
Performance of a Bipolar Transistor
The easiest way to demonstrate the behavior of a bipolar transistor is to look at its charac-
teristics, which describe the relationship between the currents and voltages in the transistor,
provided that all parameters are static, that is, not or only slowly variable over time. To
reflect its behavior with sufficient accuracy, the calculatory description of the bipolar tran-
sistor requires equations. Considering solely its normal operation, which is of particular
importance in practical applications, and neglecting any secondary effects makes the equa-
tions very simple. But when the proper functionality of a circuit is being evaluated by means
of computer simulation it is important to consider the influence of secondary effects as well.
For this purpose there are sophisticated models that are capable of correctly demonstrating
the dynamic performance when applying sinusoidal or pulsed signals. These models are
described in Sect. 2.3 and are not required to understand the basics. The following section
describes the behavior of an npn transistor; for pnp transistors all voltages and currents
have the opposite polarity.
2.1.1
Characteristics
Family of output characteristics: The application of various base–emitter voltages
(VBE ) to the circuit shown in Fig. 2.2a and measurement of the collector current IC as a
function of the collector–emitter voltage VCE produces the family of output characteris-
tics illustrated in Fig. 2.3. With the exception of a small region close to the IC -axis the
characteristic curves depend only slightly on VCE and the transistor is in normal mode;
that is, the BE diode is forward-biased and the BC diode is reverse-biased. Close to the
IC -axis voltage VCE is so small that the BC diode is in the forward region and the transistor
enters the saturation region. At this border, which is characterized by the saturation voltage
VCE,sat , there is a sharp bend in the curves and they run approximately through the origin
of the characteristics.
IC
mA VCE,sat
0.72
10
VBE
8 V
6
0.70
4
0.68
2
0.66
0 1 2 3 4 5 6 7 8 9 10 VCE
V
Current gain: A comparison of the transfer characteristics in Fig. 2.4a and the input
characteristics in Fig. 2.4b shows a striking similarity in shape. This means that in normal
operating mode the collector current IC is approximately proportional to the base current
IB . The proportionality constant B is called the current gain:
IC
B = (2.1)
IB
IC IB
mA µA
VCE VCE
10 25
8 20
6 15
4 10
2 5
0 0.2 0.4 0.6 0.8 1.0 VBE 0 0.2 0.4 0.6 0.8 1.0 VBE
V V
a Transfer characteristics b Input characteristics
2.1.2
Description by Way of Equations
The equations required for a mathematical description are based on the fact that a transis-
tor’s behavior can be explained for the main part by the characteristic function of the BE
diode. The exponential relationship between current and voltage that is typical of a diode
is reflected in the transfer and input characteristics of the transistor by the exponential
dependence of the currents IB and IC on the voltage VBE . Taking the general formulas
IC = IC (VBE , VCE ) and IB = IB (VBE , VCE ) as a basis gives us the equations for normal
operation [2.1]:
VBE
VCE
IC = IS e VT 1+ (2.2)
VA
IC
IB = with B = B(VBE , VCE ) (2.3)
B
In this case IS ≈ 10−16 …10−12 A is the saturation reverse current of the transistor and
VT is the temperature voltage (the temperature equivalent of thermal energy); therefore,
at room temperature VT ≈ 26 mV.
Early effect: The dependence on VCE is caused by the Early effect and is empirically
described by the terms on the right-hand side of (2.2). The basis for this description is the
observation that the extrapolated characteristic curves of the output characteristics intersect
approximately at one point [2.2]; Fig. 2.5 illustrates this relationship. The constant VA is
called Early voltage and is in the range of VA,npn ≈ 30 . . . 150 V for npn transistors and
VA,pnp ≈ 30 . . . 75 V for pnp transistors. Section 2.3.1 explains the Early effect in more
detail; the empirical description is sufficient for the normal operation considered here.
Base current and current gain: The base current IB is related to IC ; this results in
the current gain B being the constant of proportionality. This method is chosen because
the dependence of the current gain on VBE and VCE can be neglected in many simple
calculations; B then represents an independent constant. However, in most cases the de-
pendence on VCE is taken into consideration since it is also caused by the Early effect [2.2].
Therefore:
VCE
B(VBE , VCE ) = B0 (VBE ) 1 + (2.4)
VA
IC
– VA VCE
Fig. 2.5. Early effect and Early voltage VA in the output characteristics
2.1 Performance of a Bipolar Transistor 37
B0 (VBE ) is the extrapolated current gain for VCE = 0 V. The extrapolation is necessary
because for VCE = 0 V the transistor no longer operates in normal mode.
Large-signal equations: Inserting (2.4) into (2.3) leads to the following large-signal
equations for the bipolar transistor:
VBE
VCE
IC = IS e VT 1+ (2.5)
VA
VBE
IS V
IB = e T (2.6)
B0
2.1.3
Characteristic of the Current Gain
The Gummel-plot: The current gain B(VBE , VCE ) will be examined more closely in
the section below. Because of the exponential dependence of the currents IB and IC on
VBE , the semilogarithmic plot over VBE with VCE as a parameter seems to be the obvious
choice. This diagram, which is illustrated in Fig. 2.6, is called the Gummel plot and shows
that the exponential curves in (2.5) and (2.6) become straight lines if B0 is assumed to be
constant:
IC VBE VCE
ln = + ln 1 +
IS VT VA
IB VBE
ln = − ln(B0 )
IS VT
IC IB
,
A A IC
1 IB
100 m
10 m
1m
100 µ B
10 µ
1µ
100 n
VCE
10 n
1n
100 p
Fig. 2.6. Semilogarithmic plot of the currents IB and IC in normal mode (a Gummel plot)
38 2 Bipolar Transistor
In Fig. 2.6 the straight lines for two values of VCE are shown as broken lines. The
current gain B is seen as a shift in the y direction:
IC VCE
ln(B) = ln = ln(B0 ) + ln 1 +
IB VA
The real curves are also plotted in Fig. 2.6. In one large region they correspond to the
straight lines, which means that B0 can be assumed to have a constant value. However, in
two regions there are deviations [2.2]:
– In the event of very low collector currents the base current is higher than the value for
a constant B0 given in (2.6). This deviation is caused by additional portions in the base
current and results in a decrease of B or B0 . The large-signal (2.5) and (2.6) are valid
in this region as well.
– With very high collector currents the collector current is lower than the value given by
(2.5). This deviation is caused by the high-current effect and results in a decrease in B
or B0 . In this region the large-signal (2.5) and (2.6) are no longer valid since, according
to these equations, the decrease in B0 leads to an increase in IB and not, as required, to
a decrease in IC . This region is used in power transistors only.
Curve description: For practical purposes the current gain B is expressed as a function
of IC and VCE ; that is, B(VBE , VCE ) is replaced by B(IC , VCE ) by utilizing the relationship
between IC and VBE for a constant VCE in order to change the variables. Similarly B0 (VBE )
is replaced by B0 (IC ). This change in the expression facilitates the dimensioning of circuit
components because when setting the operating point, first IC and VCE are chosen and
then the relevant base current is determined by means of B(IC , VCE ); this is the procedure
for setting the operating point of the basic circuits shown in Sect. 2.4.
Figure 2.7 shows the curves of the current gain B and the differential current gain β
versus IC for two different values of VCE .
dIC
β = (2.7)
dIB
VCE =const.
B is called the large-signal current gain and β is the small-signal current gain.
V
V
Fig. 2.7. Curves of the large-signal current gain B and the small-signal current gain β in normal
operation
2.1 Performance of a Bipolar Transistor 39
These curves are typical of low-power transistors in which the maximum of the current
gain is achieved for IC ≈ 1…10 mA. For power transistors this maximum moves into the
ampere range. In practice the transistor is operated in the region of its maximum or to the
left of it; that is, with lower collector currents. The region to the right of the maximum is
avoided, where possible, since the high-current effect not only reduces B but also lowers
the transistor’s response time and cutoff frequencies; Sects. 2.3.2 and 2.3.3 explain this in
more detail.
The small-signal current gain β is required to describe the small-signal response in the
next section. On the basis of (2.7) the following equation
IC
∂
1 dIB
B(IC , VCE )
= =
β dIC
VCE =const. ∂IC
B
β =
IC ∂B
1−
B ∂IC
In the region to the left of the maximum B the derivative (∂ B/∂ IC ) is positive, so that
β > B. At the maximum the derivative (∂ B/∂ IC ) = 0, so that β = B. To the right of the
maximum B the derivative (∂ B/∂ IC ) is negative and thus β < B.
Determining the values: When the transistor is operated with a collector current in the
region of the maximum of the current gain B, the following approximation can be applied:
Here, Bmax (VCE ) is the maximum of B depending on VCE , as shown in Fig. 2.7.
If a transistor data sheet shows the curve for B in the form of a diagram as in Fig. 2.7,
then B(IC , VCE ) can be taken from the diagram and the approximation given in (2.8) can
be used if the β curves are missing. If only one value for B is stated on the data sheet,
it can be used as a replacement value for B and β. Typical values are B ≈ 100 . . . 500
for low-power transistors and B ≈ 10 . . . 100 for power transistors. Darlington transistors
consist of two transistor elements connected internally so that B ≈ 500 . . . 10.000 is
reached, depending on the power rating. Section 2.4.4 describes the Darlington circuit in
more detail.
2.1.4
Operating Point and Small-Signal Response
One area of application for bipolar transistors is the linear amplification of signals in small-
signal operation. The transistor is operated at an operating point A and driven by small
signals around the operating point. In this case the nonlinear characteristics can be replaced
by a line that is tangential to the operating point, and that describes an approximately linear
response.
40 2 Bipolar Transistor
VB1 VB2 IB
R1 R2
VB1
I1 I2 Io R1
I B, A
IC
Ii IB
VCE
VBE
VBE,A VB1 VBE
Graphical solution: Besides the numeric solution, a graphic solution is also possible
by drawing the straight lines under load in the diagram of the family of characteristics and
determining the intersections. As the family of characteristics comprises essentially only
one curve as a result of the very low dependence on VCE , there is only one intersection
according to Fig. 2.8b so that VBE,A and IB,A can be determined directly. In the diagram
of the output characteristics VCE,A and IC,A can be determined from the point at which
the straight line intersects the output characteristic for VBE,A (see Fig. 2.9).
Setting the operating point: Both the numeric and the graphic way of determining
the operating point are analytical procedures; that is, the operating point can be deter-
mined when the external circuit components are known. But for circuit design synthetic
2.1 Performance of a Bipolar Transistor 41
V V
V V V
Fig. 2.9. An example for determining the operating point from the output characteristics
procedures are required that are suitable for making decisions on the circuit components
necessary to achieve a certain operating point. These procedures are described, together
with the basic circuits, in Sect. 2.4.
Linearization: The characteristics are replaced by their tangents to the operating point;
that is, they are linearized. For this purpose, a Taylor series expansion is perfomed at the
operating point and is interrupted after the linear term:
∂IB
∂IB
= vBE + vCE + . . .
∂VBE
A ∂VCE
A
∂IC
∂IC
= v + vCE + . . .
∂VBE
A ∂VCE
A
BE
Figure 2.10 illustrates the linearization using the transfer characteristic as an example; for
this purpose the region around the operating point is shown in an enlarged format. The
change in current iC is determined along the characteristic curve from the voltage change
vBE ; the current change iC,lin is taken from the tangent. iC = iC,lin can be used with small
values of vBE .
Small-signal equations: The partial derivatives at the operating point are called small-
signal parameters. After introducing specific designations, the small-signal equations for
the bipolar transistor are:
42 2 Bipolar Transistor
IC
I C,lin I C,lin
iC i C,lin
I C,A
uvBE
1
iB = vBE + gm,r vCE (2.9)
rBE
1
iC = gm vBE + vCE (2.10)
rCE
∂IC
IC,A
gm = = (2.11)
∂VBE
A VT
The small-signal input resistance rBE characterizes the change in the base–emitter
voltage VBE in response to a change in the base current IB at the operating point. It can
be determined using the input characteristics according to Fig. 2.4b from the inverse value
of the slope of the tangent. Use of the following relationship renders the differentiation of
the large-signal (2.6) unnecessary:
∂VBE
∂VBE
∂IC
rBE = =
∂IB
A ∂IC
A ∂IB
A
This makes it possible to calculate rBE from the transconductance gm given in (2.11) and
the small-signal current gain β according to (2.7):
∂VBE
β
rBE =
= (2.12)
∂IB A gm
The small-signal output resistance rCE characterizes the change in the collector–
emitter voltage VCE in response to a change in the collector current IC at the operat-
ing point. It can be determined using the output characteristics shown in Fig. 2.3 from
2.1 Performance of a Bipolar Transistor 43
IB IC IC
VBE ,A
I C,A ∆ IC
∆VCE
∆ IB ∆ IC
IB,A IC,A
∆VBE ∆VBE
∆VBE ∆ IC ∆VCE
rBE = gm = r CE =
∆ IB ∆VBE ∆ IC
Fig. 2.11. Determination of the small-signal parameters from the family of characteristics
the inverse value of the slope of the tangent. It can be calculated by differentiating the
large-signal (2.5):
VCE,A VA
∂VCE
VA + VCE,A VA
rCE =
= ≈ (2.13)
∂IC A IC,A IC,A
∂IB
gm,r = ≈ 0 (2.14)
∂VCE
A
The small-signal parameters may also be derived from the characteristic curves; draw-
ing the tangent to the operating point makes it possible to determine the transconductance
values (see Fig. 2.11). In practice, however, this procedure is rarely used due to its low
accuracy; furthermore, the characteristics are not usually shown on a transistor’s data
sheets.
iB iC
frequency response and the cutoff frequency of a transistor circuit can only be obtained by
using the AC small-signal equivalent circuit, which will be described in Sect. 2.3.3.
Network Matrices
The small-signal equations can also be expressed in matrix notation as:
⎡ ⎤
1
iB ⎢ gm,r ⎥ v
= ⎢⎣
rBE ⎥ BE
iC 1 ⎦ vCE
gm
rCE
This corresponds to the conductivity presentation of a network and thus relates to network
theory. The conductivity presentation describes a network by the Y matrix Ye :
iB vBE y 11,e y 12,e vBE
= Ye =
iC vCE y 21,e y 22,e vCE
The subscript e indicates that the transistor is operated in the common-emitter mode which
means that the emitter connection is jointly used for the input and output ports according to
the through connection in the small-signal equivalent circuit shown in Fig. 2.12. Section 2.4
describes the common-emitter operating mode in more detail.
The hybrid presentation based on the H matrix He is also commonly used:
vBE iB h 11,e h 12,e iB
= He =
iC vCE h 21,e h 22,e vCE
A comparison reveals the following relationships:
1 y 21,e
rBE = h 11,e = , β = h 21,e =
y 11,e y 11,e
h 21,e h 12,e
gm = = y 21,e , gm,r = − = y 12,e
h 11,e h 11,e
h 11,e 1
rCE = =
h 11,e h 22,e − h 12,e h 21,e y 22,e
Section 4.2.3 describes this in more detail. The small-signal equivalent circuit diagram
is derived from the linear term of the interrupted Taylor series expansion. If subsequent
terms of the Taylor expansion are taken into account, for constant VCE the small-signal
collector current becomes [2.1]:
∂IC
1 ∂ 2 IC
2 1 ∂ 3 IC
3
iC = vBE + 2
vBE + 3
vBE + . . .
∂VBE
A 2 ∂VBE 6 ∂VBE
A A
IC,A IC,A 2 IC,A 3
= vBE + v +
2 BE
vBE + . . .
VT 2VT 6VT3
For a harmonic input signal with vBE = v̂BE cos ωt, this leads to:
! " ! "
iC 1 v̂BE 2 v̂BE 1 v̂BE 3
= + ... + + + . . . cos ωt
IC,A 4 VT VT 8 VT
! " ! "
1 v̂BE 2 1 v̂BE 3
+ + . . . cos 2ωt + + . . . cos 3ωt
4 VT 24 VT
+ ...
Polynomes with even or uneven exponents are contained in the brackets. With little input
signals – that is, not considering higher exponents – the ratio of the first harmonic with
2ωt to the fundamental with ωt leads to an approximation of the distortion factor k [2.1]:
iC,2ωt v̂BE
k ≈ ≈ (2.15)
iC,ωt 4VT
If k is to be kept below 1 % , then v̂BE < 0.04VT ≈ 1 mV. This means that in this case
only very small input signals are permissible.
2.1.5
Limit Data and Reverse Currents
For a transistor, various ratings are specified which must not be exceeded. They are clas-
sified in terms of limit voltages, limit currents and maximum power dissipation. The
following description again applies to npn transistors; for pnp transistors the polarity of
all voltages and currents has to be reversed.
Breakdown Voltages
BE diode: At the emitter–base breakdown voltage V(BR)EBO , the emitter diode breaks
through in the cutoff mode. The addition “(BR)” means breakdown; the subscript O indi-
cates that the third terminal, in this case the collector, is open. V(BR)EBO ≈ 5…7 V applies
to almost all transistors; thus, V(BR)EBO is the smallest of the limit voltages. It is of minor
importance, since transistors are rarely operated with negative base–emitter voltages.
BC diode: At the collector–base breakdown voltage V(BR)CBO the collector diode breaks
through in the cutoff mode. Since the collector diode is reverse-biased in normal mode
V(BR)CBO is an important upper limit of the collector–base voltage in practice. In low
46 2 Bipolar Transistor
IC
Primary breakdown
Secondary breakdown
IB > 0 IB = 0 R VBE = 0
Secondary Breakdown
In addition to the normal or primary breakdown described so far, there is a second or
secondary breakdown; this is characterized by localized overheating due to a nonhomoge-
nous current distribution (contraction) which results in localized melting and thus the
2.1 Performance of a Bipolar Transistor 47
Limit Currents
A distinction is made between maximum continuous currents and maximum peak currents.
For maximum continuous currents, the data sheets contain no specific identifiers; they are
called IC,max , IB,max and IE,max . The maximum peak currents are called ICM , IBM and
IEM in data sheets, and refer to pulsed operation with a predetermined pulse duration and
repetition rate; compared to continuous currents they are higher by a factor of 1.2…2.
Cut-Off Currents
For emitter and collector diodes data sheets specify not only the breakdown voltages
V(BR)EBO and V(BR)CBO but also the cutoff currents IEBO and ICBO , which are measured
at a voltage below their relevant breakdown voltage. Similarly, for the collector–emitter
portion, the cutoff currents ICO and ICES are specified; these are measured with the
base open or short-circuited at a voltage below V(BR)CEO and V(BR)CES , respectively. In
general, the following applies:
ICES < ICEO
IC
mA tP
IC
300 1 µs
mA
I C, max
100 10 µs
100 Ptot
100 µs
75
30
1 ms
50
Secondary
SOA breakdown 10
10 ms
25
DC
Ptot IC
W mA tP
I CM
1.5 300 1 µs
TC
I C,max
TA 100 10 µs
1.0
100 µs
30
1 ms
0.5
10 10 ms
DC
pulse durations. If the pulse duration is very short and the duty cycle is low, the transistor
can be operated at the maximum voltage V(BR)CEO and the maximum collector current
ICM at the same time; in this case, the SOA is a square. For this reason, the transistor can
switch loads the power of which is high compared to the maximum power dissipation;
Sect. 2.1.6 describes this in more detail.
Figure 2.15b shows the SOA of a high-voltage switching transistor that comes in
three different versions with V(BR)CEO = 160 V, 250 V or 300 V. The maximum contin-
uous current is IC,max = 100 mA and the maximum permissible peak current for one
pulse of 1 ms duration is ICM = 300 mA. If the pulse duration is shorter than 1 ms,
the SOA is a square. It allows the switching of loads with a power dissipation of up to
P = V(BR)CEO IC,max = 90W Ptot = 1.5 W.
2.1.6
Thermal Performance
The arrangement shown in Fig. 2.16 is used to explain the thermal performance. The
elements shown are provided with insulation on the outside and have the temperatures T1 ,
T2 and T3 ; Cth,2 is the thermal capacity (thermal storage capacity) of the centre element.
The differences in temperature cause the heat flows P12 and P23 ,1 which can be calculated
using the thermal resistances Rth,12 and Rth,23 of the junctions:
T1 − T2 T2 − T3
P 12 = ; P 23 =
Rth,12 Rth,23
1 In thermodynamics, is the symbol used to detote heat flux. Here, we use P , since in electrical
components heat flows are caused by the power dissipation PV .
50 2 Bipolar Transistor
C th,
P P
T T T
Balancing the heat flows makes it possible to determine the thermal quantity Qth,2
stored in the centre element and the temperature T2 :
Qth,2 = Cth,2 T2
dQth,2 dT2 P 12 − P 23
= P 12 − P 23 ⇒ =
dt dt Cth,2
If the temperatures T1 and T3 are constant, temperature T2 changes until P12 = P23 ; in this
state the heat flows into and out of the centre element balance and T2 remains constant. If the
supplied heat flow P12 is constant and the right-hand element represents the ambient air,
with an ambient temperature of T3 = TA , the centre element warms up to the temperature
T2 = T3 + Rth,23 P23 ; here, too, the steady state is P12 = P23 .
Thermal equivalent circuit: For the thermal performance an electrical equivalent circuit
diagram can be used. The parameters heat flow, thermal resistance, thermal capacity
and temperature correspond to the electrical parameters current, resistance, capacity and
voltage. For a transistor the elements junction (J ), case (C), ambient air (A) and, if used,
heat sink (H ) are important. The heat dissipation PV is assigned to the junction as s heat
flow; the ambient temperature TA is constant. The resulting thermal equivalent circuit
shown in Fig. 2.17 allows the calculation of the time characteristics of the temperatures
TJ , TC and TH using the known time characteristic of PV .
Operation without a heat sink: Where no heat sink is used, Rth,CH , Rth,H A and Cth,H
are replaced by the heat resistance Rth,CA between the case and the ambient air. On the data
sheet for a transistor, the thermal resistance Rth,J A between the junction and ambient air
is often quoted with the transistor installed upright on a printed circuit board and operated
without a heat sink:
Rth,J A = Rth,J C + Rth,CA
2°/ W 3° / W 5° / W
PV Cth, J TJ Cth, C TC Cth, H TH TA
10 W
Operation with a heat sink: The thermal resistance Rth,H A of the heat sink is stated
on the data sheet for the heat sink; it depends on the size, design and installation position.
The thermal resistance Rth,CH depends on the transistor’s installation on the heat sink; in
order to prevent any loss in the efficiency of the heat sink it must be kept low through a
specific heat transfer compound (thermolube). The use of insulating washers for electrical
isolation between a transistor and a heat sink may cause Rth,CH to become so high that the
efficiency of a large heat sink with a low Rth,H A value is drastically reduced; the relation
Rth,CH < Rth,H A should always be maintained. Therefore:
Rth,J A = Rth,J C + Rth,CH + Rth,H A
If several transistors are mounted on a common heat sink, the equivalent circuit contains
several junctions and cases that are connected to the heat sink node.
SMD transistors: Transistors that employ SMD technology dissipate the heat to the
circuit board via the connecting wires. The thermal resistance between the junction and
the soldering point is called Rth,J S on the data sheet; the subscript S indicates the soldering
point. This results in:
Rth,J A = Rth,J S + Rth,SA
TJ,limit − TA,max
P V ,max(stat) = (2.18)
Rth,J A
TJ,limit = 150 ◦ C (300 ◦ F) is used for calculations with silicon transistors. TA,max must
be given for the specific application and determines the maximum ambient temperature
allowed for circuit operation.
The data sheet of a transistor quotes PV ,max(stat) as a function of TA and/or TC ;
Fig. 2.15a shows the power derating curves. Their down-sloping portion is described
by (2.18) when the respective values for T and Rth are inserted:
TJ,limit − TA
P V ,max(stat) (TA ) =
Rth,J A
TJ,limit − TC
P V ,max(stat) (TC ) =
Rth,J C
Therefore, the thermal resistances Rth,J A and Rth,J C can also be determined from the
negative slope of these curves.
52 2 Bipolar Transistor
– First, the maximum static power dissipation PV ,max(stat) is calculated using (2.18) and
from this the value of PV ,max(puls) ; for this purpose the data sheet shows the plot of the
ratio PV ,max(puls) /PV ,max(stat) for several values of D versus tp (see Fig. 2.18a). With
a declining pulse duration tp the amplitude of the sawtooth-shaped portion of the TJ
curve decreases more and more; tp → 0 results in TJ = TJ,max and thus:
P V ,max(puls) 1
lim =
tP →0 P V ,max(stat) D
These limit values can be taken from the vertical axis of the graph in Fig. 2.18a: for
D = 0.5, and with a very short pulse duration the ratio is PV ,max(puls) = 2PV ,max(stat) ,
and so on.
PV,max(puls) tP Rth,JA(puls) tP
PV,max(stat)
tP T T
D
T
tP
D
T
tP tP
PV,max(puls) /PV,max(stat) Rth,JA(puls)
– The data sheet contains values for the thermal resistance in pulsed operation from which
PV ,max(puls) can be calculated directly:
TJ,limit − TA,max
P V ,max(puls) (tP , D) = (2.19)
Rth,J A(puls) (tP , D)
The data sheet shows a plot of Rth,J A(puls) for different values of D versus tP (see
Fig. 2.18b).
The two procedures are equivalent. Apart from a constant factor, the ratio PV ,max(puls) /
PV ,max(stat) is basically the reciprocal value of Rth,J A(puls) :
P V ,max(puls) TJ,limit − TA,max 1 1
= ∼
P V ,max(stat) Rth,J A(puls) P V ,max(stat) Rth,J A(puls)
2.1.7
Temperature Sensitivity of Transistor Parameters
The characteristics of a bipolar transistor are highly sensitive to temperature. Of particular
importance is the temperature-dependent relationship between IC and VBE . Where the
dependence of VBE and of temperature T are explicitly quoted, the current is:
VBE
VCE
IC (VBE , T ) = IS (T ) e VT (T ) 1 +
VA
The reason for the temperature sensitivity of IC is the temperature sensitivity of the reverse
current IS and the temperature voltage VT [2.2], [2.4]:
kT µV
VT (T ) = = 86.142 T
q K
T VG (T ) xT ,I
−1 T
IS (T ) = IS (T0 ) e T0 VT (T ) with xT ,I ≈ 3 (2.20)
T0
Here, k = 1.38 · 10−23 VAs/K is Boltzman’s constant, q = 1.602 · 10−19 As is the
elementary charge and VG = 1.12 V is the gap voltage of silicon; the low-temperature
sensitivity of VG is negligible.
The relative change in IS is derived by differentiation of IS (T ):
1 dIS 1 VG T =300 K
= 3+ ≈ 0.15 K−1
IS dT T VT
If the temperature increases by 1 K then IS rises by 15 %. The change in IC can be calculated
accordingly:
T =300 K
1 dIC
T =300 K
dVBE
The voltage Vdot is a matter constant and amounts to approx 44 mV for npn transistors
made of silicon. Differentiation leads to:
T =300 K
1 dB Vdot
= ≈ 5.6 · 10−3 K −1
B dT VT T
In practice a simplified relationship is often used [2.4]:
xT ,B
T
B(T ) = B(T0 ) with xT ,B ≈ 1.5 (2.22)
T0
For the region used in practice the same temperature sensitivity is obtained:
T =300 K
1 dB xT ,B
= ≈ 5 · 10−3 K −1 (2.23)
B dT T
With a temperature increase of 1 K, the current gain increases by about 0.5 %. This, how-
ever, is of secondary practical importance, since the deviations in current gain resulting
from the manufacturing process are much higher. It is meaningful only in differential con-
siderations – such as, for example, in the calculation of the temperature coefficient of a
circuit.
2.2
Design of a Bipolar Transistor
In general, the bipolar transistor has an asymmetric design. This clearly enables a distinc-
tion to be made between the collector and the emitter and causes the differing behavior in
the normal and inverse modes as described below. Discrete and integrated transistors are
comprised of more than three regions, with the collector region consisting of at least two
sub-regions. Therefore, the type designations “npn” and “pnp” only describe the sequence
of the active inner regions. Transistors are produced in a multi-stage process on a wafer that
is subsequently cut by sawing into little chips. Each chip carries either a single transistor
or an array of several integrated transistors and other components; that is, an integrated
circuit (IC).
2.2 Design of a Bipolar Transistor 55
E E B E E B
n +
p +
n p p n
B p – B n –
n p
+ +
n n p p
C C C C
2.2.1
Discrete Transistors
Internal design: Discrete transistors are manufactured predominantly by using the epi-
taxial planar technique. Fig. 2.19 shows the construction principles of npn and pnp transis-
tors with emphasis on the active region. Doping is heavy in the n+ and p + layers, medium
in the n and p layers and weak in the n− and p − layers. This particular layering of differ-
ently doped regions improves the electrical properties of the transistor. The bottom layer
of the chip forms the collector while the base and emitter are at the top.
Case: The transistor is mounted in a case by soldering the underside to the collector
wire or a metallic part of the case. The two other connections are bonded with fine gold
or aluminum wires to the relevant connecting wires. Figure 2.20 shows a low-power and
a power transistor after soldering and bonding. Finally the low-power transistor is cast in
a plastic material; the case of the power transistor is closed off with a lid.
For the various designs and applications there is a multitude of cases that differ in their
maximum heat dissipation capacity and that are adapted to specific geometric requirements.
Figure 2.21 shows a selection of the most common models. Power transistors have cases
intended for installation in heat sinks; they have a large contact surface to enhance the
heat flow. SMD transistors for high power ratings are provided with two collector wires
for improved heat conductance to the circuit board. Special case models are used for high-
frequency transistors, as the electrical behavior at frequencies in the GHz range is heavily
dependent on the geometry; some cases have two emitter connections in order to increase
the contact to earth.
2.2.2
Integrated Transistors
Integrated transistors are also produced using the epitaxial planar technique. Here, too,
the collector terminal is placed on the upper side of the chip. The individual transistors
are electrically isolated from each other by reverse-biased pn junctions. Only a very thin
layer close to the surface forms the active region of the transistors. The depth of the chip
is called the substrate (S) and represents a fourth connection common to all transistors; it
is also arranged on the top. As npn and pnp transistors have to be produced in the same
process, the two types deviate considerably in construction and electrical data.
Internal design: Since npn transistors are built as vertical transistors according to
Fig. 2.22, the current flows vertically from the collector to the emitter – in other words,
its flow is perpendicular to the plane of the chip. In contrast, pnp transistors are usually
2.2 Design of a Bipolar Transistor 57
C
2 S C B E B C
S
+ + +
n n p n
+ 2 –
p 1 n
+
n
B
p 2
1
Fig. 2.22. Equivalent diode circuit and design of an integrated vertical npn transistor
constructed as lateral transistors according to Fig. 2.23; the current flows laterally – in
other words, parallel to the surface of the chip.
Substrate diodes: The equivalent diode circuits illustrated in Figs. 2.22 and 2.23 show
an additional substrate diode arranged between the collector and substrate of the vertical
npn transistor or between the base and substrate of the lateral pnp transistor. The substrate
is connected to the negative supply voltage so that the diodes are always reverse-biased
and act as isolators between the transistors themselves and between the transistors and the
substrate.
Differences between vertical and lateral transistors: As the thickness of the base re-
gion can be kept thinner in a vertical transistor, the current gain exceeds that of a lateral
transistor by a factor of 3…10; the response time and the cutoff frequencies are also signif-
icantly higher in the vertical transistor. Therefore, more and more vertical pnp transistors
are now being produced. Their design corresponds to that of a vertical npn transistor if the
n and p doping are interchanged in every region. Isolation from the substrate is achieved
by embedding the transistor in a tub of n-doped material that is connected to the positive
supply voltage. In this case, the npn and pnp transistors are called complementary – even
if their electrical data do not conform as well as is the case in complementary discrete
transistors.
C S B C E C B
+ +
2 n p p p n
+ 2 –
p 1 1 n
B S +
n
1 p 2
Fig. 2.23. Equivalent diode circuit and design of an integrated lateral pnp transistor
58 2 Bipolar Transistor
2.3
Models of Bipolar Transistors
Section 2.1.2 described the static performance of a bipolar transistor in normal mode by
means of large-signal (2.5) and (2.6); secondary effects were not taken into account or
were only considered qualitatively as in the description of the current gain in Sect. 2.1.3.
Computer-aided circuit design using CAD programs requires a model that allows for all
effects, applies to all operating modes and, in addition, correctly reflects the dynamic
performance. Linearization at the operating point of this large-signal model leads to the
dynamic small-signal model that is necessary for calculating the frequency response of a
circuit.
2.3.1
Static Performance
The static performance is demonstrated for an npn transistor; for the pnp transistor all
currents and voltages have the opposite polarity. The simplest model of a bipolar transistor
is the Ebers–Moll model, based on the equivalent diode circuit. This model has only
three parameters and describes all primary effects. More accurate modelling necessitates
a conversion that first leads to the transport model and, after adding other parameters to
describe the secondary effects, to the Gummel–Poon model; the latter is capable of a very
accurate description of the static performance and is used in CAD programs.
Ebers–Moll Model
An npn transistor consists of two pn diodes in anti-serial connection with a common p
region. The diodes are called emitter or BE diode and collector or BC diode. The func-
tionality of a bipolar transistor depends on the fact that large portions of the diode currents
can be drained through the third connection because of the very thin common-base region.
Therefore, the Ebers–Moll model in Fig. 2.24 comprises the two diodes of the equiva-
lent diode circuit and two current-controlled current sources that describe the current flow
through the base. The controlling factors of the controlled sources are called AN for nor-
mal operation and AI for inverse operation; typical values are AN ≈ 0.98…0.998 and
AI ≈ 0.5…0.9. The different values for AN and AI are a consequence of the asymmetric
construction explained in Sect. 2.2.
the currents through the terminals can be calculated according to Fig. 2.24 [2.5]:
⎛ ⎞ ⎛ ⎞
VBE VBC
IC = AN IS,N ⎝e VT − 1⎠ − IS,I ⎝e VT − 1⎠
2.3 Models of Bipolar Transistors 59
IC
C
VBC A N I D,N
I D,I
B = B
IB
I D,N
VBE A I I D,I
E
IE
⎛ ⎞ ⎛ ⎞
VBE VBC
IE = − IS,N ⎝e VT − 1⎠ + AI IS,I ⎝e VT − 1⎠
⎛ ⎞ ⎛ ⎞
VBE VBC
IB = (1 − AN )IS,N ⎝e VT − 1⎠ + (1 − AI )IS,I ⎝e VT − 1⎠
The following equation can be deduced from the theorem on reciprocal networks:
AN IS,N = AI IS,I = IS
Normal operation: In normal mode the BC diode is reverse-biased since VBC < 0;
together with the related controlled source this can be neglected since ID,I ≈ −IS,I ≈ 0.
In addition, for VBE VT the term −1 can be neglected compared with the exponential
function, which leads to:
VBE
IC = IS e VT
VBE
1
IE = − IS e VT
AN
VBE VBE
1 − AN 1
IB = IS e VT = IS e VT
AN BN
Figure 2.25a shows the reduced model with the most significant relationships; in this case,
AN is the current gain in the common-base circuit and BN is the current gain in the
60 2 Bipolar Transistor
C E
I C = – AN IE = BN IB I E = – AI IC = BI IB
IB IB
B B
VBE VBC
I E = – ID,N I C = – ID,I
E C
common-emitter circuit:2
IC
AN = −
IE
AN IC
BN = =
1 − AN IB
Typical values are AN ≈ 0.98…0.998 and BN ≈ 50…500.
Inverse operation: The reduced model shown in Fig. 2.25b for inverse operation is
arrived at in a similar way; the current gains are:
IE
AI = −
IC
AI IE
BI = =
1 − AI IB
Typical values are AI ≈ 0.5…0.9 and BI ≈ 1…10.
Saturation voltage: The transistor used as a switch is driven from normal operation into
saturation; what is interesting here is the achievable minimum collector–emitter voltage
VCE,sat (IB , IC ). This voltage is:
BN (1 + BI ) (BI IB + IC )
VCE,sat = VT ln
BI2 (BN IB − IC )
For 0 < IC < BN IB the voltage is VCE,sat ≈ 20…200 mV.
The minimum of VCE,sat is reached when IC = 0:
1
VCE,sat (IC = 0) = VT ln 1 + = − VT ln AI
BI
2 For the current gains a distinction must be made between model parameters and measurable
external current gains. In the Ebers–Moll model the model parameters AN and BN of normal
operation and AI and BI of inverse operation are identical to the external current gains; therefore,
they may be defined as external currents.
2.3 Models of Bipolar Transistors 61
After exchanging emitter and collector, the voltage for IE = 0 when switching from inverse
operation into saturation is:
1
VEC,sat (IE = 0) = VT ln 1 + = − VT ln AN
BN
Due to AI < AN < 1 the following applies: VEC,sat (IE = 0) < VCE,sat (IC = 0).
Typical values are VCE,sat (IC = 0) ≈ 2…20 mV and VEC,sat (IE = 0) ≈ 0.05…0.5 mV.
Transport Model
An equivalence conversion transforms the Ebers–Moll model into the transport model [2.5]
shown in Fig. 2.26; it has only one controlled source and it forms the basis for modelling
other effects, as described in the next section.
IC
C
VBC
I B,I
B = B
IB
I T = B N I B,N – BI I B,I
I B,N
VBE
E
IE
I B = I B,N
B
VBE I C = BN I B
I E = – (1+ BN ) I B
Fig. 2.27. Reduced transport model for normal
E operation
⎛ ⎞
VBE VBC
1 1 ⎠
IE = IS ⎝− 1 + e VT + e VT +
BN BN
Normal operation: In normal operation, if the reverse currents are neglected, the fol-
lowing equations are obtained:
VBE
IS V
IB = e T
BN
VBE
IC = IS e VT
Taking the relationship between AN and BN into consideration these equations are identical
to those of the Ebers-Moll model. Figure 2.27 shows the reduced transport model for normal
operation.
Properties: The transport model describes the primary DC current response of the bipo-
lar transistor under the assumption that the emitter and collector diodes are ideal. An
important property of the model is that the transport current I T flowing through the base
region occurs separately; this is not the case in the Ebers–Moll model. As in the Ebers–Moll
model three parameters are necessary for the description: IS , BN and BI [2.5].
Other Effects
The transport model can be expanded for a more accurate description of the static per-
formance. The resulting modeled effects have already been described qualitatively in
Sects. 2.1.2 and 2.1.3:
– The recombination of the charge carriers in the pn junctions generates additional leakage
currents in the emitter and collector diodes; these currents add to the base current and
have no influence on the transport current IT .
– With large currents the transport current IT is smaller than the value resulting from
(2.26). This high current effect is caused by the significantly increased charge carrier
concentration in the base region; this is also called strong injection.
– The voltages VBE and VBC influence the effective thickness of the base region and thus
also influence the transport current IT ; this is called the Early effect.
2.3 Models of Bipolar Transistors 63
Leakage currents: In order to take the leakage currents into account, the transport model
is expanded by two more diodes with the currents [2.5]:
⎛ ⎞
VBE
IB,E = IS,E ⎝e nE VT − 1⎠ (2.27)
⎛ ⎞
VBC
IB,C = IS,C ⎝e nC VT − 1⎠ (2.28)
This requires four additional model parameters: The leakage saturation reverse currents
IS,E and IS,C and the emission coefficients nE ≈ 1.5 and nC ≈ 2.
High-current effect and the Early effect: The influence of the high-current and Early
effects on the transport current IT is reflected by the nondimensional quantity qB [2.5]:
⎛ ⎞
VBE VBC
BN IB,N − BI IB,I IS ⎝ V
IT = = e T − e VT ⎠ (2.29)
qB qB
General equations: The currents IB,N and IB,I are available from (2.24) and (2.25).
Figure 2.28 shows the expanded model. The currents are:
Definition of qB : The quantity qB is a measure for the relative majority carrier charge
in the base and is made up of the quantities q1 , which describes the Early effect, and q2 ,
VBC
VBE
The Early voltages VA,N and VA,I and the knee-point currents for strong injection IK,N
and IK,I are required as further model parameters. The Early voltages range between 30 V
and 150 V, but lower values are possible with integrated and high-frequency transistors.
The knee-point currents depend on the size of the transistor; they are in the milliampere
range for low power transistors and in the ampere range for power transistors.
3 In the specialist literature (for example [2.5]) another expression for q is often found; the
B
expression quoted here is used by Spice [2.4, 2.6].
4 The large-signal equations in Sect. 2.1.2 only apply to normal operation; additional identification
by a subscript N is therefore not necessary.
2.3 Models of Bipolar Transistors 65
I K,N IB
IS IK ,N
IB,E
IS
IS,E
I B,N
IS
BN Fig. 2.29. A semilogarithmic plot of currents
IB and IC in normal operation (a Gummel
VBE plot)
Figure 2.29 shows the curves of IC and IB in the semilogarithmic plot and illustrate
the importance of parameters IK,N and IS,E . When the reverse currents are not taken into
consideration, IB is:
VBE VBE
IS V
IB = e T + IS,E e nE VT (2.32)
BN
Comparing the curves in Fig. 2.29 with the measured values plotted in Fig. 2.6 on page 37
shows that parameters IK,N , IS,E and nE offer a very good description of the true behavior
in normal operation; the same is true of the parameters IK,I , IS,C and nC in inverse
operation.
Course of the current gain: The formula B = B(IC , VCE ) is better suited in practice
but does not allow a full description of B. Instead, three portions can be distinguished:
– For small collector currents the leakage current IB,E is the dominant component of the
base current so that IB ≈ IB,E ; for qB ≈ q1 this leads to
1
1−
nE 1−
1 1
IC nE VCE nE
B ≈ ∼ IC 1+
1 VA,N
q1 nE
IS,E
IS
66 2 Bipolar Transistor
nE –1
I C [log]
( BN IS,E ) nE –1 I S nE –1 I K,N
Fig. 2.30. Dependence of the large-signal current gain B on the collector current
1/3
Using nE ≈ 1.5 results in B ∼ IC . In this portion, B is smaller than for medium
collector currents and increases with a rising collector current. This portion is called the
leakage current range.
– IB ≈ IB,N for medium collector currents and thus:
VCE
B ≈ BN 1 + (2.33)
VA,N
In this portion, B reaches a maximum and depends only to a small extent on IC . This
portion is called the normal range.
– For large collector currents the high-current effect sets in; for IB ≈ IB,N we obtain:
2
BN IK,N VCE
B ≈ ≈ BN 1+
qB IC VA,N
In this portion B is proportional to the reciprocal value of IC and thus decreases rapidly
with an increasing collector current. This portion is called the high-current range.
Figure 2.30 shows a plot of B with both axes plotted using a logarithmic scale; the
approximations for the three portions change to straight lines with slopes of 1/3, 0 and
−1. The boundaries of the portions are also shown:
−1
% & nE
Normal range ↔ leakage current region : IC = BN IS,E nE −1 IS nE −1
Normal range ↔ high current region : IC = IK,N
Current gain maximum: The maximum value of B at a constant voltage VCE is called
Bmax (VCE ) (see Fig. 2.7 on page 38 and (2.8)). For transistors with a low leakage current
IS,E and a high knee-point current IK,N , the normal range is so wide that B is almost
tangential to the horizontal approximation given in (2.33). In this case, Bmax (VCE ) is
determined by (2.33), and the maximum value B0,max extrapolated for VCE = 0 is deter-
mined by BN . For transistors with a high leakage current and a low knee-point current the
normal range may be very narrow or even nonexistent. In this case B is below the straight
line of (2.33) and thus does not come up to the value indicated by this line; therefore
B0,max < BN .
2.3 Models of Bipolar Transistors 67
Substrate Diodes
Integrated transistors have a substrate diode that lies between the substrate and the collector
in the vertical npn transistor and between the substrate and the base in the lateral pnp
transistor (see Figs. 2.22 and 2.23). The current flowing through these diodes is expressed
using the simple diode equation – for example, for vertical npn transistors:
⎛ ⎞
VSC
ID,S = IS,S ⎝e VT − 1⎠ (2.34)
A further parameter is the substrate saturation reverse current IS,S . As these diodes are
normally nonconductive, more accurate modelling is not required; the only thing that is
important is the fact that a current can flow if a certain – that is, wrong – circuitry is
connected to the substrate or the surrounding tub. For lateral pnp transistors it is necessary
to replace VSC by VSB .
Spreading Resistances
To describe the static performance fully the spreading resistances must also be taken into
account. In Fig. 2.31a a discrete transistor is used to illustrate these resistances:
• The emitter spreading resistance RE is of a low value because of the high doping (n+ )
and the small longitudinal/cross section ratio of the emitter region; typical values are
RE ≈ 0.1…1 for low-power transistors and RE ≈ 0.01…0.1 for power transistors.
• The collector spreading resistance RC is caused predominantly by the low doping (n− )
of a section in the collector region; typical values are RC ≈ 1…10 for low-power
transistors and RC ≈ 0.1…1 for power transistors.
• The base spreading resistance RB is formed by the external base spreading resistance
RBe between the base connection and the active base region plus the internal base
spreading resistance RBi throughout the active base region. With high currents RBi
has only a small effect as the current flows predominantly in a region close to the base
connection due to the current displacement (emitter edge displacement). The Early
E B C
RC
C'
RE RBe VB' C'
+
n
RB I B,C I B,I
B'
RBi p B IT
– I B,E I B,N
RC n
+ VB' E '
n
E'
RE
C E
a Inside the transistor b Included in the model
effect on the thickness of the base region has an additional influence. These effects can
be described by the quantity qB according to (2.30)5 :
RBi
RB = RBe + (2.35)
qB
Consequently, for normal operation the following applies:
⎧
⎨ VCE
RBe + RBi 1 + for IC < IK,N
RB = VA,N
⎩
RBe for IC → ∞
Typical values are RBe ≈ 10…100 for low-power transistors and RBe ≈ 1…10
for power transistors; the values of RBi are higher by a factor of 3…10.
Figure 2.31b shows the relevant expanded model. A distinction must now be made
between the external connections B, C and E and the internal connections B’, C’ and E’;
in other words, all diode currents and the transport current IT no longer depend on VBE ,
VBC and VSC , but on VB
E
, VB
C
and VSC
.
Effects of the spreading resistances: The voltages across the spreading resistances are
very low in low-power transistors; therefore, the emitter and collector spreading resistances
are often neglected. The base spreading resistance cannot be neglected because, even if it
is very low, it influences the speed of operation and the cutoff frequencies. The voltage
drop at RB is only 1 mV for RB = 100 and IB = 10 mA which are typical values for
low-power transistors; the cutoff frequencies, however, are clearly reduced in most circuits.
Therefore, the dependence of RB on the operating point according to (2.35) must only be
taken into account to reflect the dynamic performance exactly.
With power transistors working with high currents all spreading resistances must be
taken into account; for IB = IC /B and IE ≈ −IC , the voltages are:
RB
VBE ≈ VB
E
+ IC + RE
B
VCE ≈ VC
E
+ IC (RC + RE )
Here, the external voltages VBE and VCE may deviate sharply from the internal voltages
VB
E
and VC
E
. If a power transistor is used as a switch in the saturation region with
IC = 5 A and B = 10, the external voltages are VBE = 1.5 V and VCE,sat = 1.85 V with
VB
E
= 0.75 V, VC
E
,sat = 0.1 V, RB = 1 , RE = 0.05 and RC = 0.3 . Due to the
spreading resistances the values for VBE and VCE,sat can be relatively high.
2.3.2
Dynamic Performance
The response of a transistor to pulsed or sinusoidal signals is called the dynamic perfor-
mance and cannot be determined from the characteristic curves. This is because of the
nonlinear junction capacitances of the emitter, the collector and, with integrated transis-
tors, the substrate diode as well as the diffusion charge, stored in the base region, which is
also described by the nonlinear diffusion capacitances.
5 This equation is used by PSpice as a standard [2.6]; there is, however, an alternative expression
for RB [2.4, 2.6], which is not described here.
2.3 Models of Bipolar Transistors 69
Junction Capacitances
A pn junction has a junction capacitance CJ that depends on the doping rate of the adjacent
regions, the doping profile, the junction surface and the applied voltage V ; a simplified
expression is provided by [2.2]:
CJ 0
CJ (V ) = mJ for V < VDiff (2.36)
V
1−
VDiff
The zero capacitance CJ 0 = CJ (V = 0 V) is proportional to the junction surface and
rises with an increase in doping. The diffusion voltage VDiff also depends on the doping
and increases with it; the range is VDiff ≈ 0.5…1 V. With the capacitance coefficient mJ ,
the doping profile of the junction is taken into account; mJ ≈ 1/2 for abrupt junctions
with a doping jump, while mJ ≈ 1/3 for linear junctions.
The simplifying assumptions leading to (2.36) are no longer fulfilled for V → VDiff .
A more accurate calculation shows that (2.36) can be used up to about 0.5 VDiff only;
with higher values of V the zero capacitance CJ rises only slightly compared to (2.36).
A sufficiently accurate description is reached when the curve of CJ for V > fC VDiff is
replaced by a tangent to point fC VDiff :
dCJ
% &
CJ (V > fC VDiff ) = CJ (fC VDiff ) +
V − fC VDiff
dV V =fC VDiff
CJ
mJ = 1
V 2
mJ = 1
3
mJ = 1 CJ0
3
mJ = 1
2
f C VDiff VDiff V
Fig. 2.32. Curves of the junction capacitance CJ for mJ = 1/2 and mJ = 1/3 according to (2.36)
(broken line) and (2.37)
In discrete transistors CJ,Ce is mostly lower than CJ,Ci ; that is, xCJ C ≈ 0.5…1. For
integrated transistors xCJ C < 0.5.
– In integrated transistors there is an additional junction capacitance CJ,S of the substrate
diode with parameters CJ,S , mJ 0,S and VDiff ,S . In vertical npn transistors the junction
capacitance acts on the internal collector C
– that is, CJ,S = CJ,S (VSC
) – and in lateral
pnp transistors it acts on the internal base B
– that is, CJ,S = CJ,S (VSB
).
Extended model: Figure 2.33 shows the static model of an npn transistor extended by
the junction capacitances CJ,E , CJ,Ci , CJ,Ce and CJ,S ; also included are the diffusion
capacitances CD,N and CD,I , which are described in the next section.
Diffusion Capacitances
A pn junction contains a diffusion capacitance QD , which is in the first approximation,
proportional to the ideal current through the pn junction. In a transistor QD,N is the
diffusion charge of the emitter diode and QD,I is the diffusion charge of the collector
C
CJ,S
RC
C'
S
RB I B,C I B,I
B'
B IT
I B,E I B,N
CJ,E CD,N
E'
RE
diode; both are related to the respective portions of the ideal transport current IT according
to (2.26) – that is, to BN IB,N or BI IB,I [2.5]:
⎛ ⎞
VB
E
QD,N = τN BN IB,N = τN IS ⎝e VT − 1⎠
⎛ ⎞
VB
C
QD,I = τI BI IB,I = τI IS ⎝e VT − 1⎠
Parameters τN and τI are called the transit times. Differentiation leads to the diffusion
capacitances CD,N and CD,I [2.5]:
VB
E
dQD,N τN IS
CD,N (VB
E
) = = e VT (2.40)
dVB
E
VT
VB
C
dQD,I τI IS
CD,I (VB
C
) = = e VT (2.41)
dVB
C
VT
Figure 2.33 presents the model with the two capacitances CD,N and CD,I .
Normal operation: The diffusion capacitances CD,N and CD,I are parallel to the junc-
tion capacitances CJ,E and CJ,Ci (see Fig. 2.33). In normal operation the collector diffusion
capacitance CD,I is very small due to VB
C
< 0 and may be ignored with respect to the
collector junction capacitance CJ,Ci arranged in parallel; therefore, CD,I can be described
by a constant transit time τI = τ0,I . For small currents the emitter diffusion capacitance
CD,N is smaller than the emitter junction capacitance CJ,E , while for higher currents it is
larger. A more accurate description of the dynamic performance with high currents requires
a more detailed model for τN .
Current dependence of the transit time: With high currents the diffusion charge in-
creases disproportionately due to the high-current effect. In this region the transit time τN
is no longer constant but increases with rising currents. There is also an influence from the
Early effect, which changes the effective thickness of the base region and thus the stored
charge. However, even with the parameters already introduced, IK,N for the high current
effect and VA,N for the Early effect, it is not possible to give a satisfactory description; an
empirical equation is therefore used [2.6]:
⎛ ⎞
# $ VB
C
with
⎛ ⎞
VB
E
IS ⎝e VT − 1⎠
BN IB,N
x= = ⎛ ⎞ (2.42)
BN IB,N + Iτ,N VB
E
IS ⎝e VT − 1⎠ + Iτ,N
72 2 Bipolar Transistor
τN
τ 0,N
100
0 VB' C'
– Vτ,N
– 2Vτ,N
10
1
0.01 0.1 1 10 BN I B,N
I τ,N
The new model parameters are the ideal transit time τ0,N , the coefficient for the transit time
xτ,N , the transit time knee-point current Iτ,N and the transit time voltage Vτ,N . Coefficient
xτ,N determines the maximum increase in τN for VB
C
= 0:
% &
lim τN
= τ0,N 1 + xτ,N
IB,N →∞ VB
C
=0
Gummel–Poon Model
Figure 2.33 shows the complete model of an npn transistor; it is called the Gummel–Poon
model and is used for circuit simulations in CAD programs. Figure 2.35 lists the variables
and equations of the model. Figure 2.36 lists the parameters together with the names of
the parameters in the PSpice circuit simulator6 .
Figure 2.37 lists the parameters of some selected transistors taken from the PSpice com-
ponent library; it only contains the parameters for normal operation in the forward region.
Parameters not specified are treated by PSpice as follows:
– A standard value is used:
IS = 10−16 A, BN = 100, BI = 1, nE = 1.5, nC = 2, xT ,I = 3, fC = 0.5
VDiff ,E = VDiff ,C = VDiff ,S = 0.75 V, mJ,E = mJ,C = 0.333, xCJ C = 1
2.3.3
Small-Signal Model
By linearization at an operating point the nonlinear Gummel–Poon model becomes a
linear small-signal model. For practical purposes the operating point is selected so that
the transistor operates in forward mode; therefore, the small-signal models described here
only apply to this operating mode. Similarly, one can set up small-signal models for other
operating modes, although these are of secondary importance.
The static small-signal model describes the small-signal characteristics at low fre-
quencies and for that reason is called the DC small-signal equivalent circuit. The dynamic
small-signal model also describes the dynamic small-signal characteristics and is required
for calculating the frequency response of circuits; it is called the AC small-signal equivalent
circuit.
(IB,I = IB,C = ID,S = 0) results in the static Gummel–Poon model for forward oper-
ation as shown in Fig. 2.38a. The nonlinear variables IB = IB,N (VB
E
) + IB,E (VB
E
)
and IC = IT (VB
E
, VC
E
) are linearized at the operating point A:
∂IC
IC,A VT ∂qB
gm = = 1−
∂VB
E
A VT qB ∂VB
E
A
VB
E
,A VB
E
,A
1 ∂IB
IS IS,E
= = e VT + e nE VT
rBE ∂VB
E
A B N VT nE VT
2.3 Models of Bipolar Transistors 75
1 ∂IC
IC,A
= =
rCE ∂VC
E
A VA,N
VA,N + VC
E
,A − VB
E
,A 1 +
VA,I
∂IC
∂VB
E
∂VB
E
∂IC
β
rBE = = =
∂IB
A ∂IC
A ∂IB
A gm
VC
E
,A VA,N
∂VC
E
VA,N + VC
E
,A VA,N
rCE =
≈ ≈
∂IC A IC,A IC,A
76 2 Bipolar Transistor
C C
IC iC
RC RC
IB RB iB RB
B' C' B' C'
B B
I B,E I B,N
IT rBE vB' E ' gmvB' E ' rCE
E' E'
RE RE
E E
a Before linearisation b After linearisation
Fig. 2.38. Static small-signal model as derived from the static Gummel–Poon model by
linearization
The approximations for rBE and rCE are in accordance with (2.12) and (2.13) given
in Sect. 2.1.4. In order to determine rBE it is necessary to know the small-signal current
gain β or to assume a reasonable value.
The equation for transconductance gm is derived by the approximate evaluation of the
entire expression; (2.11) is thus extended by one term to describe the high-current effect.
The high-current effect causes a relative reduction of gm with high collector currents; that
is, to two-thirds of IC,A /VT for IC,A = IK,N and to half of IC,A /VT for IC,A → ∞. If
the reduction is to remain below 10 %, then IC,A < IK,N /8 must be chosen.
DC small-signal equivalent circuit: Figure 2.38b shows the resulting static small-signal
model. For almost all practical calculations the spreading resistances RB , RC and RE are
neglected; this leads to the small-signal equivalent circuit described in Sect. 2.1.4 and
shown again in Fig. 2.39a.
If the Early effect is also disregarded (rCE → ∞), the alternative version shown in
Fig. 2.39b can also be used in addition to the reduced equivalent circuit of Fig. 2.39a. The
following thus applies:
1 1 β
rE = ≈ ; α = = g m rE
1 gm 1+β
gm +
rBE
B C B C
rBE vBE gmvBE rCE α iE
rE
iE
E E
a After ignoring the spreading b Alternative circuit with the
resistances Early effect ignored
( rCE → ∞)
CCe
RB CCi RC
B' C'
B C
E'
RE
7 Some literature shows a variation with an additional resistor r between the base and the collector.
C
This results from linearization of the collector–base diode of the Ebers–Moll model, which may
not be ignored in this case, and is therefore not used for modelling the Early effect as often
assumed. For this reason, this variation is not an equivalent to the simplified model shown in
Fig. 2.39a.
78 2 Bipolar Transistor
v gmvB'E S
the calculations in the following Sections. The practical determination of the capacitances
CE and CC is described in more detail in the next Section.
8 The static small-signal current gain in a common-emitter circuit which has previously been called
β is now named β 0 to distinguish it from the inversely Laplace-transformed β = L−1 {β(s)}; the
subscript zero means that the frequency is zero with the consequence β0 = |β(j 0)|.
2.3 Models of Bipolar Transistors 79
RB CC
B B' C
quency response |β(j ω)| for β0 = 100 with the zero considered; at the β cutoff
frequency
1
ωβ = 2πfβ ≈ (2.43)
rBE (CE + CC )
Transit frequency: The frequency at which |β(j ω)| decreases to a value of one is called
the transit frequency fT ; therefore [2.7]:
gm
ωT = 2πfT = β0 ωβ ≈ (2.44)
CE + C C
Due to the approximations in the small-signal model and in calculating β(s) the transit
frequency according to (2.44) is not identical to the real transit frequency of the transistor;
therefore it is also called the extrapolated transit frequency because it can be derived by
extrapolating the declining portion of |β(j ω)| similar to a lowpass filter of first degree.
Transistor data sheets always state the extrapolated transit frequency.
The transit frequency depends on the operating point; outside the high-current region:
IC,A τN IC,A
gm = , CE = + CJ,E , CC = CJ,C
VT VT
⏐a⏐,⏐b⏐
a0 ≈ 1
⏐b⏐ b 0 = 100
100
70
10
⏐a⏐
1 ⏐a⏐
0.7
⏐b⏐
fb f T fa f [log]
Fig. 2.43. Absolute-value frequency responses |α(j ω)| and |β(j ω)|
80 2 Bipolar Transistor
f T [log]
1
τ 0,N
VT I C,A [log]
(C + CJ,C )
τ 0,N J,E
Fig. 2.44. Dependence of the transit frequency on the collector current IC,A
1
ωT ≈
IC,A % &
τN + CJ,E + CJ,C
VT
Figure 2.44 shows the dependence of the transit frequency on the collector current IC,A .
Three regions can be distinguished:
IC,A VT % &
ωT ≈ % & ∼ IC,A for IC,A < CJ,E + CJ,C
VT CJ,E + CJ,C τ0,N
1 1 VT % &
ωT ≈ ≈ for CJ,E + CJ,C < IC,A Iτ,N
τN τ0,N τ0,N
determined; the base and the collector are connected to earth, the latter due to vBC =
VBC − VBC,A = 0. rCE → ∞ and α0 = gm rE 9 result in:
R B CC r E C E RB C C
1+s + s2
i α0 α0
α(s) = − C = α0
iE (1 + s rE CE ) (1 + sRB CC )
The transfer function comprises two poles and two zeros; the magnitude frequency
response |α(j ω)| is shown in Fig. 2.43 [2.8]. In general, RB CC rE CE , so that the
following approximation can be used:
α0
α(s) ≈
1 + s r E CE
This leads to the α cutoff frequency:
1
ωα = 2πfα ≈ (2.45)
rE C E
1
ωY 21e = 2πfY 21e ≈ (2.46)
RB (CE + CC )
The transconductance cutoff frequency depends on the operating point; however, its depen-
dence on IC,A is not easy to describe since RB depends on the operating point in a highly
nonlinear manner. The general tendency is that the transconductance frequency decreases
with an increase in the collector current IC,A .
control generally allows a higher bandwidth (see Sect. 2.4.1); this applies in the same way
to the common-collector circuit (see Sect. 2.4.2).
The highest bandwidth is reached with the common-base circuit; with the general
condition Ri > rE the transistor is current-controlled and the bandwidth of the circuit is
limited to a maximum by α cutoff frequency fα (see Sect. 2.4.3).
Selection of the operating point: Among other things, the bandwidth of a circuit de-
pends on the operating point of the transistor. The common-emitter circuit with current
control and the common-base circuit yield the maximum bandwidth when the collector
current IC,A is chosen in such a way that the transit frequency fT is at a maximum. In
the common-emitter circuit with voltage control the situation is more complicated; the
transconductance cutoff frequency fY 21e decreases with an increasing IC,A , but with the
same gain the resistance of the circuitry at the collector node is lower so that the bandwidth
on the output side increases (see Sect. 2.4.1).
Determination of the small-signal capacitances: The data sheet for a transistor contains
the transit frequency fT and the output capacitance Cobo in a common-base circuit (output,
grounded base, open emitter); Cobo corresponds to the collector–base capacitance. Using
(2.44) the following can be calculated:
CC ≈ Cobo
gm
CE ≈ − Cobo
ωT
2.3.4
Noise
In resistors and pn junctions there are noise voltages or noise currents the generation
of which is attributed to thermal agitation of the charge carriers in the resistors and to
discontinuous current flow due to the cross-over of individual charge carriers in the pn
junctions.
Noise Densities
Since noise is a stochastic occurrence, it is not possible to calculate in the normal way
using voltages and currents. A noise voltage vr is described by means of the noise voltage
density |v r (f )|2 and a noise current ir by means of the noise current density |i r (f )|2 ; the
densities reflect the spectral distribution of the effective values vreff or ireff 10
10 In this case the unilateral frequency f with 0 < f < ∞ is used instead of the bilateral angular
frequency ω or j ω with −∞ < ω < ∞ as the frequency variable. |v r (f )|2 = 4π|v r (j ω)|2 is
applicable; here, factor 4π is made up of the factor 2π according to ω = 2πf and the factor 2
for the transfer to the unilateral frequency variable.
2.3 Models of Bipolar Transistors 83
IC,A
gm Transconductance gm = for VT ≈ 26 mV at T = 300 K
VT
Small-signal directly from the data sheet or indirectly from the data
(β)
current gain sheet using β ≈ B
or a reasonable assumption (β ≈ 50 . . . 500)
Small-signal β
rBE rBE =
input resistance gm
2
d(vreff )
|v r (f )| =
2
df
2
d(ireff )
|i r (f )|2 =
df
The effective values can be determined from the noise densities by integration [2.9]:
∞
vreff = |v r (f )|2 df
0
∞
ireff = |i r (f )|2 df
0
If the noise densities are constant the noise signal is called white noise. A noise signal
can be white only in a certain region; for f → ∞ in particular, the noise density must
approach zero so that the integrals remain finite.
Transfer of noise densities in circuits: If a noise voltage vr,e with a noise voltage
density |v r (f )|2 exists at a point e, it is possible to calculate the resulting noise voltage
vr,a with noise voltage density |v r,a (f )|2 at a given point a by means of the transfer
function H (s) = v r,a (s)/v r,e (s) [2.9]:
R ID
R CR D D i D,r
vR,r
a Resistor b pn junction
With several noise sources the noise densities at any point can be added if the noise
sources are noncorrelated; that is, independent of each other – this is generally the case.
For instance, if a noise voltage source has the density |v r (f )|2 and a noise current source
has the density |i r (f )|2 , the situation at point a can be calculated with the use of Ha (s) =
v r,a (s)/v r (s) and Z a (s) = v r,a (s)/i r (s):
Noise of a resistor: A resistor R generates a noise voltage vR,r , with the noise voltage
density [2.9]:
|v R,r (f )|2 = 4kT R
Here, k = 1.38 · 10−23 VAs/K is Boltzmann’s constant and T is the resistor temperature
in Kelvin. This noise is called thermal noise, since it is caused by the thermal agitation of
the charge carriers; the noise voltage density is thus proportional to the temperature.√ For
R = 1 and T = 300 K, |v R,r (f )|2 ≈ 1.66 · 10−20 V2 /Hz or |v R,r (f )| ≈ 0.13 nV/ Hz.
Figure 2.46a shows a noise voltage source for noise modelling; the arrow pointing in
both directions marks the source as a noise source. It is white noise because of the constant
noise voltage density; therefore, the result of the calculated effective value is ∞. However,
this result is not correct since for f → ∞ the parasitic capacitance CR of the resistor must
be taken into account; this is shown in Fig. 2.46a. Using the equation
v R,r (s)
v
R,r (s) =
1 + sRCR
for the noise voltage vR,r across the resistor leads to the expression:
|v R,r (f )|2
|v
R,r (f )|2 =
1 + (2πf RCR )2
Its integration results in a finite effective value [2.10]:
kT
vR,reff =
CR
Noise of a pn junction: A pn junction – that is, an ideal diode – generates a noise current
iD,r with a noise current density [2.9]:
|i D,r (f )|2 = 2qID
2.3 Models of Bipolar Transistors 85
Here, q = 1.602 · 10−19 As is the elementary charge. The noise current density is propor-
tional to the current ID flowing through the pn junction. This noise is called√shot noise.
ID = 1 mA results in |i D,r (f )|2 ≈ 3.2 · 10−22 A2 /Hz or |i D,r (f )| ≈ 18 pA/ Hz.
Figure 2.46b shows a noise current source for noise modelling; here, too, an arrow
pointing in both directions characterizes the source as a noise source. It is white noise as
in the case of the resistor; the same considerations apply to the effective value – that is, for
f → ∞ the capacitance of the pn junction must be taken into consideration.
1/f noise: Resistors and pn junctions produce an additional 1/f noise with a noise
density that is inversely proportional to the frequency. For resistors this portion is usually
negligible; for a pn junctions it is
γ
k(1/f ) ID(1/f )
|i D,r(1/f ) (f )|2 =
f
with the experimental constants k(1/f ) and γ(1/f ) ≈ 1…2 [2.10].
When calculating the effective value the result is ∞ if f = 0 is used as the lower
limit for the integration. But since in practice this process can be observed for a finite time
only it is the reciprocal value of the observation time that is used for the lower limit. With
measuring instruments the frequency portions below the reciprocal value of the measuring
time are called drift instead of noise.
vRB,r
RB CC
B'
B C
Fig. 2.47. Small-signal model of a bipolar transistor with the original (top) and the equivalent
noise sources (bottom)
|i C,r (f )|2
|i r,0 (f )|2 = |i B,r (f )|2 +
|β(j 2πf )|2
Using β/gm = rBE > RB , B ≈ β 1 and γ(1/f ) = 1 leads to [2.10]:
2
1 RB2 fg(1/f ) f
|v r,0 (f )| = 2qIC,A
2
2
+ 1+ + RB
2
+ 4kT RB
gm β f fT
(2.47)
2
1 ffg(1/f )
|i r,0 (f )|2 = 2qIC,A 1+ + (2.48)
β fT f
√
In the frequency range fg(1/f ) < f < fT / β the equivalent noise densities are
constant; that is, the noise is white noise. For gm = IC,A /VT :
2kT VT 2qRB2 IC,A
|v r,0 (f )|2 = + 4kT RB + (2.49)
IC,A β
2.3 Models of Bipolar Transistors 87
⏐v r,0⏐ ⏐i r,0⏐
nV/ Hz pA/ Hz
50
⏐v r,0⏐ ⏐i r,0⏐
10
20
5
10
2 5
1
I1 I2
Fig. 2.48. Depencence of the equivalent noise densities on the operating point for RB = 60 : an
asymptotic curve for β = 100 (broken line) and the actual curve with a β which depends on the
operating point and with βmax = 100
2qIC,A
|i r,0 (f )|2 = (2.50)
β
√
For f < fg(1/f ) and f > fT / β the noise densities
√ increase. For low-noise low-power
transistors we have fg(1/f ) ≈ 100 Hz and fT / β ≈ 10 MHz.
Operating point dependence: Figure 2.48 shows the dependence of the equiv-
alent noise densities √ on the operating point current IC,A for the frequency range
fg(1/f ) < f < fT / β. The noise current density |i r,0 (f )|2 for β = const. is propor-
tional to IC,A ; this relation is plotted in Fig. 2.48 as an asymptotic curve (broken line).
With low and high collector currents the real curve runs above the asymptote due to the
decrease in β. A distinction has to be made between three regions for the noise voltage
density |v r,0 (f )|2 :
⎧
⎪
⎪
2kT VT VT
= I1
⎪
⎪ for IC,A <
⎪
⎪ I C,A 2R B
⎨ VT 2β VT
|v r,0 (f )|2 ≈ 4kT RB for < IC,A <
⎪
⎪ 2R B RB
⎪ 2qR 2 I
⎪
⎪
⎪ C,A 2β V T
⎩ B
for IC,A > = I2
β RB
The three portions of the curve are shown in Fig. 2.48 as a asymptotic curve (broken line)
for β = constant. With high collector currents the real curve runs above the asymptote due
to the decrease in β.
vg i r,0
= vg
a With noise source of signal generator and b With equivalent noise source
equivalent noise sources of the transistor
The noise source of the signal generator can be combined with the equivalent noise sources
of the transistor into an equivalent noise source vr , as shown in Fig. 2.49b. The noise voltage
density is:
|v r (f )|2 = |v r,g (f )|2 + |v r,0 (f )|2 + Rg2 |i r,0 (f )|2 (2.51)
One assumes that the noise of the transistor is caused by the signal generator, and the ratio
of the noise density due to the equivalent noise source to the noise density of the signal
generator is called the spectral noise figure [2.10]:
The transistor intensifies the noise density by the spectral noise figure F (f ); consequently
the signal-to-noise-ratio decreases to:
2 2
vgeff vgeff
SNR = fb
= fb
|v r (f )|2 df F (f )|v r,g (f )|2 df
fa fa
If one assumes that the noise of the signal generator is caused by the thermal noise of
the internal resistance Rg so that |v r,g (f )|2 = 4kT Rg , this expression can be placed in
front of the integral, which leads to:
fb
1
F = F (f )df
fb − fa fa
In this case the mean noise figure F is achieved by averaging over the spectral noise figure
F (f ). Often F (f ) is constant for the frequency interval observed; then F = F (f ), which
is simply called the noise figure F.
The noise figure is usually given in decibels: FdB = 10 log F . Figure 2.51 shows the
noise figure of a low-power transistor as a function of the operating point current IC,A
for various internal resistances Rg of the signal generator. Figure 2.51a contains the plots
for a frequency above the 1/f cutoff frequency fg(1/f ) ; here, (2.53) applies, which means
that the noise figure is independent of the frequency. Figure 2.51b contains the plots for
100
1
~
f ~f 2
30
10
1 f1 1k 1M f2 1G f
Hz
fg ( 1/f ) fT / fT
Fig. 2.50. Plots of the spectral noise figure F (f ) of a bipolar transistor with IC,A = 1 mA,
β = 100, RB = 60 , Rg = 1 k, fg(1/f ) = 100 Hz and fT = 100 MHz
90 2 Bipolar Transistor
F F
dB dB
Rg 1 M Ω
Rg 1 M Ω
100 k Ω
15 15
100 k Ω
10 k Ω 10 k Ω
10 10
1 kΩ 1 kΩ
5 5
0 0
1µ 10 µ 100 µ 1m 10 m I C 1µ 10 µ 100 µ 1m 10 m I C
A A
a fg(1/f ) < f = 100 kHz < fT / β b f = 1 Hz < f g(1/f )
a frequency below fg(1/f ) ; here, the noise figure depends on the frequency, which means
that the plots are true only for the frequencies indicated.
Minimizing the noise figure: We can see in Fig. 2.51a that the noise figure reaches a
minimum under certain conditions; the optimum operating point current IC,Aopt can be
taken directly from the respective Rg plot. This is presented more clearly in Fig. 2.52,
which shows the contours of the noise figure in a log–log diagram of IC,A and Rg . Using
∂F
= 0
∂IC,A
1M
Rg 100 k
Ω 100
10 k
10
3
1k
F = 1.2 1.5
100
3
10
10
100
1
1µ 10 µ 100 µ 1m 10 m 100 m
I C,A
A
Fig. 2.52. Contours of the noise figure in the IC,A − Rg plane for RB = 60 and β = 100
2.3 Models of Bipolar Transistors 91
the optimum operating point current IC,Aopt can be calculated with (2.53) if Rg is given:
⎧
⎪
⎪ V T β
⎪
⎨ R for Rg < RB
VT β B
IC,A opt = ' ≈ (2.54)
⎪
⎪
Rg2 + RB2 ⎪
⎩
VT β
for Rg > RB
Rg
For low-resistance signal generators with Rg < RB the optimum operating point current
IC,Aopt is determined by RB , β and VT ; that is, it does not depend on Rg . The current
IC,Aopt ≈ 1…50 mA if RB ≈ 10…300 and β ≈ 100…400. However, in practice this
case seldom occurs. For signal generators with Rg > RB , the current IC,A opt is inversely
proportional to Rg ; for low-power transistors the following estimation can be used:
0.3 V
IC,A opt ≈ for Rg ≥ 1 k (2.55)
Rg
Rgopt
Ω
1M
100 k
–1
~ I C,A
10 k
1k –1/2
~ I C,A
100
I1 I2
Fig. 2.53. Interdependence of the operating point and the optimum internal resistance Rgopt for
RB = 60 : an asymptotic curve for β = 100 (broken line) and the real curve shape with β,
depending on the operating point and βmax = 100
Noise figure in the range of 1/f noise: If f < fg(1/f) , substituting (2.47) and (2.48) into
(2.52) leads to:
1 fg(1/f ) VT R 2 IC,A IC,A Rg fg(1/f )
F (f ) = 1 + RB + + B +
Rg 2f IC,A β VT 2β VT f
The noise figure increases for f → 0. Likewise in the range of 1/f noise the optimum
operating point current IC,Aopt is determined by (2.54); that is, it is not dependent on the
frequency. This means that with a given internal resistance Rg the optimum
√ noise figure is
achieved with IC,Aopt according to (2.54) at any frequency f < fT / β. In contrast, with
a given IC,A the optimum internal resistance Rgopt , (1/f) depends on the frequency:
β VT VT 2RB f
Rgopt,(1/f ) = RB2 + +
IC,A IC,A fg(1/f )
RB fg(1/f ) RB 2 Rg >RB
RB fg(1/f )
Fopt,(1/f ) = 1+ + 1+ ≈ 1+ +
Rg βf Rg Rg βf
Notes about minimizing the noise figure: Various aspects should be taken into consid-
eration in order to minimize the noise figure:
– Minimizing the noise figure does not cause an absolute minimum of the noise signal;
rather, as can be seen from the definition of the noise figure it is the decrease in the signal
noise ratio SNR that is minimized. The minimum absolute noise – that is, the minimum
noise density |v r (f )|2 of the equivalent noise source – is achieved for Rg = 0 according
to (2.51). Which of the parameters needs to be minimized depends on the application
involved: in a circuit for signal transmission it is necessary to minimize the noise figure
in order to achieve an optimum SNR at the output; on the other hand, in a circuit not
intended for signal transfer – for example, a current source to set the operating point –
it is necessary to minimize the absolute noise at the output. Therefore, the noise figure
is relevant only for signal transmission systems.
– The absolute minimum of the noise figure is achieved with a high internal resistance√Rg
and a small operating point current IC,A . However, the result is true only for f < fT / β.
For IC,A ≈ 1 mA a typical low-power transistor with a maximum transit frequency of
300 MHz and a maximum small-signal current gain of 400 achieves only fT ≈ 200 kHz
and β ≈ 100; therefore, this consideration only applies if f < 20 kHz. Thus it is not
possible to decrease IC,A to an arbitrary low value; its lower limit is determined by the
required bandwidth of the circuit.
– In most cases the internal resistance Rg is given and IC,Aopt can be calculated using
(2.54) or estimated using (2.55). If, in the event of very high quality requirements, the
value determined by these methods should prove unsatisfactory, a transformer as shown
in Fig. 2.54 can be used to transform the internal resistance. This method is employed
with very low internal resistances since the optimum noise figure according to (2.57)
94 2 Bipolar Transistor
1:n
Rg = 2
n Rg LTr
Fig. 2.54. Transforming the internal resistance of a signal generator by use of a transformer
is relatively high in such cases. The transformer converts the internal resistance to a
higher value n2 Rg for which a lower optimum noise figure can be achieved. Due to the
inductance LT r of the transformer the resulting highpass filter has a cutoff frequency
fT r = n2 Rg /(2πLT r ); fT r must be lower than the minimum signal frequency that is
of interest.
In general, this is not the case. Optimization of the noise figure by partial differentiation
of (2.52) is, however, independent of |v r,g (f )|2 , as the constant 1 is eliminated by
differentiation and the remaining expression is only scaled by |v r,g (f )|2 . This causes
Fopt to change, but the corresponding values of Rgopt and IC,Aopt remain constant.
2.4
Basic Circuits
Basic circuits using bipolar transistors: There are three basic circuits in which bipolar
transistors can be used: the common-emitter circuit, the common-collector circuit and the
common-base circuit. The name reflects the node of the transistor that is connected as
a common reference point for the input and output of the circuit; this is illustrated in
Fig. 2.55.
In many circuits this condition is not met strictly, so a less stringent criterion must be
used:
The designation reflects the node of the transistor that does not serve either as
the input or the output of the circuit.
Example: Figure 2.56 shows a three-stage amplifier with feedback. The first stage consists
of the npn transistor T1 . The base connection serves as the input to the stage with the input
voltage Vi across R1 and the feedback output voltage Vo across R2 , while the collector
serves as the output; thus, T1 is operated in a common-emitter circuit. A difference com-
pared to the stringent criterion is that despite the designation as a common-emitter circuit,
Vo Vo Vo
Vi Vi Vi
R4 R5
T2
R6
T1 T5
R1 R2
T3 T4
Vi Vo
D1
R3 R8 R7 R9
it is not the emitter but the ground that is used as the common reference for the input and
output of the stage. The output of the first stage is connected to the input of the second
stage, which consists of a pnp transistor T2 . Here, the emitter is the input and the collector
the output; therefore, T2 is operated in a common-base circuit. Again, the base is not used
as the reference point. The third stage is made up of the npn transistor T5 . The base is the
input while the emitter forms the output of that stage and, at the same time, the output of
the entire circuit; consequently T5 is operated in a common-collector circuit. Transistors
T3 and T4 serve as current sources and provide the bias currents for T2 and T5 .
Basic circuits with several transistors: There are several configurations with two or
more transistors. They are used so often that they, too, must be regarded as basic circuits
which serve, for example, as differential amplifiers or current mirrors; such circuits are
described in Sect. 4.1. A special configuration is the Darlington circuit, which uses two
transistors in such a way that they can be treated as one transistor (see Sect. 2.4.4).
Polarity: Since their electrical characteristics are better suited, npn transistors are used
predominantly in all these circuits; this applies especially to integrated circuits. In principle,
npn and pnp transistors can be interchanged in all configurations by reversing the polarity
of the supply voltages, the electrolytic capacitors and the diodes.
2.4.1
Common-Emitter Circuit
Figure 2.57a shows a common-emitter circuit consisting of the transistor, the collector
resistance RC , the supply voltage source Vb and the signal voltage source Vg with the
internal resistance Rg . In what follows, we assume that Vb = 5 V and RC = Rg = 1 k
so that typical numeric results can be quoted in addition to the formulas.
2.4 Basic Circuits 97
Io
Vo Ii
Vo
V V V V
Vi
Normal mode: Figure 2.57b shows the equivalent circuit for normal mode, for which
the simplified transport model according to Fig. 2.27 replaces the transistor; thus:
VBE
IC = BIB = IS e VT
This equation is derived from the basic (2.5) and (2.6) by disregarding the Early effect and
assuming a constant large-signal current gain which leads to B = B0 = β.
V
Normal mode Saturation mode
V
5 Vo
2 Vg
IB Rg
1 Vi
Vi ≈ Vg
1 2 Vg
V
Saturation mode: The transistor enters the saturation mode when VCE reaches the sat-
uration voltage VCE,sat ; then with VCE,sat ≈ 0.1 V:
Vb − VCE,sat IC
IC = = 4.9 mA ⇒ IB = = 12.25 mA
RC B
IC
⇒ Vi = VBE = VT ln = 709 mV ⇒ Vg = Vi + IB Rg = 721 mV
IS
For Vg > 0.72 V the transistor operates in the saturation region which means that the
collector diode is forward-biased. In this region all variables with the exception of the base
current are approximately constant:
IC ≈ 4.9 mA , Vi = VBE ≈ 0.72 V , Vo = VCE,sat ≈ 0.1 V
The base current is
Vg − VBE Vg − 0.72 V
IB = ≈
Rg Rg
and is the sum of the currents through the emitter and the collector diode. In this case the
internal resistance Rg has to limit the base current to acceptable values. In Fig. 2.58 it was
assumed that Vg,max = 2 V; for Rg = 1 k it follows that IB,max ≈ 1.28 mA which is a
permissible value for low-power transistors.
11 Typical values for the npn low-power transistor BC547B
2.4 Basic Circuits 99
Calculation from the characteristics: The small-signal voltage gain corresponds to the
slope of the transfer characteristic (see Fig. 2.59); differentiation of (2.59) leads to:
∂Vo
∂IC
IC,A RC
A =
= −
RC = − = − gm RC
∂Vi A ∂VBE A VT
When gm = IC,A /VT = 77 mS and RC = 1 k, then A = −77. This gain is also called
the no-load gain because it describes operation without any load (Io = 0). Furthermore,
it is obvious that the small-signal voltage gain is proportional to the voltage drop IC,A RC
across the collector resistance RC . Due to IC,A RC < Vb , the possible maximum gain when
using an ohmic collector resistance RC is proportional to the supply voltage Vb .
The small-signal input resistance results from the input characteristic:
∂Vi
∂VBE
ri = = = rBE
∂Ii
A ∂IB
A
If rBE = β/gm and β = 400, then ri = 5.2 k.
The small-signal output resistance can be determined from (2.59):
∂Vo
ro = = RC
∂Io
A
In this case ro = 1 k.
Vo A
V
0.5 1.0
5 0
Vi
4 –20 V
3 –40
2 –60
1 –80
–100
0.5 1.0 V i
V
a Transfer characteristic b Gain = slope of the transfer characteristic
Rg ii = iB iC io
Calculation from the small-signal equivalent circuit: Figure 2.60 shows the small-
signal equivalent circuit of the common-emitter configuration which is gained by inserting
the small-signal equivalent circuit according to Fig. 2.12 or Fig. 2.39a, short-circuiting the
DC voltage sources, omitting the DC current sources and changing over to the small-signal
values:13
vi = Vi − Vi,A , ii = Ii − Ii,A
vo = Vo − Vo,A , io = Io − Io,A
vg = Vg − Vg,A , iC = IC − IC,A
Omitting the load RL leads to the common-emitter circuit as shown in Fig. 2.60:
Common-emitter circuit
rCE RC
vo
A =
= − gm (RC ||rCE ) ≈ − g m RC (2.61)
vi io =0
vi
ri = = rBE (2.62)
ii
rCE RC
vo
ro = = RC ||rCE ≈ RC (2.63)
io
If we take into consideratin that the Early effect has been disregarded, which means
that rCE → ∞ was assumed, the results are the same as those determined from the
characteristics. If rCE = VA /IC,A and VA ≈ 100 V, then A = −75, ri = 5.2 k and
ro = 980 .
A, ri and ro fully describe the common-emitter circuit; Fig. 2.61 presents the corre-
sponding equivalent circuit. The load resistance RL may be an ohmic resistance or an
equivalent element for the input resistance of a circuit connected to the output. In this
situation it is important that the operating point is not shifted by RL ; in other words, no or
only a minute direct current is allowed to flow through RL . This will be outlined in more
detail together with the method for setting the operating point.
12 The output resistance r
CE of the transistor is of no significance here since when the characteristics
curves were established the Early effect was omitted; that is, by assuming rCE → ∞.
13 Changing over to the small-signal values by subtracting the operating point values corresponds
to short-circuiting the DC voltage sources or omitting the DC current sources since the operating
point values are DC voltages and DC currents.
2.4 Basic Circuits 101
Rg ii ro io
vi vo
vg ri Av i RL
Maximum gain µ and β −VA product: The gain of the common-emitter circuit reaches
its maximum with RC → ∞; the maximum gain is derived from (2.61):
IC,A VA VA
µ = lim |A| = gm rCE = =
RC →∞ VT IC,A VT
It is difficult to reach this borderline case with an ohmic collector resistance RC , since
RC → ∞ causes RC rCE so that the voltage drop across RC needs to be much larger
than the Early voltage VA ≈ 100 V because IC,A RC IC,A rCE = VA . However, this
borderline situation can be achieved by replacing the collector resistance by a constant
current source with the current I0 = IC,A ; this results in very high small-signal resistances
even with low voltages.
In practice µ is very seldom stated because it is only a substitute for the Early voltage
VA . One can summarize by saying that the possible maximum gain of a bipolar transistor
is proportional to VA . With npn transistors VA ≈ 30…150 V and thus µ ≈ 1000…6000
while in pnp transistors the voltage VA ≈ 30…75 V results in µ ≈ 1000…3000.
The maximum gain µ is reached only in the no-load condition. In many circuits –
especially in integrated circuits – the input resistance of the subsequent stage acts as a load
which, in common-emitter and common-collector circuits, is proportional to the current
gain β. Thus, the gain actually achieved depends on both VA and β; therefore it is often
the β − VA product that is used as a quality criterion for bipolar transistors. Typical values
range between 1000…60000.
Temperature sensitivity: The temperature dependence can be seen from (2.21); this
shows that with a constant collector current IC , the base–emitter voltage VBE declines at a
rate of 1.7 mV/K. This means that the input voltage must be reduced by 1.7 mV/K in order
to keep the operating point of the circuit IC = IC,A constant. If, on the other hand, the
102 2 Bipolar Transistor
input voltage is kept at a constant level, then a temperature increase acts like an increase
in the input voltage of dVi /dT = 1.7 mV/K; the temperature drift of the output voltage
can thus be calculated using the gain:
dVo
∂Vo
dVi
= ≈ A · 1.7 mV/K (2.65)
dT
A ∂Vi
A dT
In the numeric example this leads to (dVo /dT )|A ≈ −127 mV/K.
We can see that a temperature change of only a few Kelvin results in a marked shift of the
operating point; this is accompanied by a change in A, ri and ro due to the altered operating
point and a change in A and an additional change in A and ri due to the temperature
sensitivity of gm and/or VT and β. As, in practice, temperatures may change by 50 K
and above, it is essential that the operating point be stabilized; this can be achieved, for
example, by feedback.
Normal mode: Figure 2.62b shows the equivalent circuit for normal operation. The
voltages are:
Io =0
Vo = Vb + (Io − IC ) RC = Vb − IC RC (2.66)
Vi = VBE + VE = VBE + (IC + IB ) RE ≈ VBE + IC RE (2.67)
Vi = Vg − IB Rg ≈ Vg (2.68)
Vb Vb
2k Io
Ii
Vo Vo
V
Vi
Vg Vg
VE
1k
V
Normal mode Saturation mode
V
5
Vo
4 Vg
IB Rg
3 Vi
2
VBE
Vi ≈ Vg
1
VE
1 2 3 4 Vg
V
In (2.67) the base current IB is very small compared to the collector current IC because
B 1. It is assumed in (2.68) that the voltage drop across Rg can be disregarded. In
(2.67) the current feedback is demonstrated by the fact that the voltage VBE is reduced
to VBE ≈ Vi − IC RE by the collector current IC , unlike in the common-emitter circuit
without negative feedback where VBE = Vi (see (2.60)).
If 0.8 V < Vg < 2.2 V, then VBE ≈ 0.7 V; thus, from (2.67) and (2.68) it follows that
Vg − 0.7 V
IC ≈
RE
which can be substituted in (2.66):
RC % &
Vo ≈ Vb − Vg − 0.7 V (2.69)
RE
This linear relationship is shown as a broken line in Fig. 2.63 and corresponds closely to
the transfer characteristic for 0.8 V < Vg < 2.2 V; the difference compared to the transfer
characteristic in this range is now only dependent on RC and RE . The feedback therefore
has the effect that the behavior of the circuit in its first approximation no longer depends on
the nonlinear properties of the transistor, but solely on the linear resistances; in addition,
inter-component deviations of transistor parameters have virtually no influence.
A point located in the centre of the down-sloping segment of the transfer characteristic
is selected for the operating point; this allows maximum output signal. With the sample
operating point marked in Fig. 2.63 for Vb = 5 V, IS = 7 fA, B = β = 400, RC = Rg =
1 k and RE = 500 we achieve:
Vb − Vo IC
Vo = 3.5 V ⇒ IC = = 1.5 mA ⇒ IB = = 3.75 mA
RC B
⇒ VE = (IC + IB ) RE = 752 mV
IC
⇒ Vi = VBE + VE = VT ln + VE = 1430 mV
IS
104 2 Bipolar Transistor
⇒ Vg = Vi + IB Rg = 1434 mV
Saturation mode: The transistor enters the saturation region when VCE reaches the
saturation voltage VCE,sat ; if VE ≈ Vg − 0.7 V, (2.69) leads to:
RC % &
VCE ≈ Vo − VE = Vb − 1 + Vg − 0.7 V
RE
Inserting VCE = VCE,sat ≈ 0.1 V and solving the equation with respect to Vg leads to
Vg ≈ 2.3 V. For Vg > 2.3 V the collector diode is forward-biased and a base current flows
through the emitter diode and the collector diode; this current increases with Vg and is
limited by Rg (see Fig. 2.63). Since the base current flows through RE the voltages Vi ,
Vo and VE cannot be considered to be constant values as in the common-emitter circuit
without feedback, but they increase with Vg .
Small-signal response: The voltage gain A corresponds to the slope of the transfer
characteristic (see Fig. 2.64); it is almost constant in the region of linear approximation
according to (2.69). Calculation of A is carried out by using the small-signal equivalent
circuit shown in Fig. 2.65. From the nodal equations
v i − vE vo − vE vE
+ gm vBE + =
rBE rCE RE
vo − vE vo
gm vBE + + = io
rCE RC
Vo A
V
1 2
5 0
Vi
4 V
3
–1
2
–2
1 2 Vi
V
a Transfer characteristic b Gain = slope of the transfer
characteristic
Fig. 2.64. Gain of the common-emitter circuit with negative current feedback
2.4 Basic Circuits 105
Rg ii io
Fig. 2.65. Small-signal equivalent circuit for common-emitter configuration with negative current
feedback
β rCE
A = = −
vi
io =0 1 RC 1 RC
1 + R E gm 1 + + + +
β β rCE rCE rCE
If gm RE 1, the gain only depends on RC and RE . When a load resistance RL is used the
relevant operating gain AB can be calculated by replacing RC by the parallel resistors RC
and RL (see Fig. 2.65). For the sample operating point selected we obtain an exact value
of A = −1.927 with gm = 57.7 mS, rBE = 6.9 k, RC = Rg = 1 k and Ri = 500 ;
the first approximation yields A = −1.933 and the second approximation A = −2.
The input resistance is:
vi
(1 + β) rCE + RC
ri = = rBE + RE
ii
io =0 rCE + RE + RC
It depends on the load resistance for which in this case the open-circuit input resistance
is given due to io = 0 (RL → ∞). The input resistance for other values of RL can
be calculated by replacing RC with the two resistors RC and RL connected in parallel;
by setting RL = RC = 0 we obtain the short-circuit input resistance. However, the
dependence on RL is so slight that it is eliminated by the approximation. For the operating
point used as an example ri,o = 202.1 k is the exact open-circuit input resistance and
ri,s = 205 k the exact short-circuit input resistance; the approximation gives us ri =
206.9 k.
The output resistance depends on the internal resistance Rg ; here we will look at the
borderline cases only. The short-circuit output resistance applies to a short-circuited input
106 2 Bipolar Transistor
with vi = 0 and Rg = 0:
⎛ ⎞
rBE
β+
vo
⎜ rCE ⎟
ro,s = = RC || rCE ⎜
⎝1 +
⎟
i
rBE ⎠
o vi =0 1+
RE
rCE rBE
rCE RC
β1 βRE + rBE
≈ RC || rCE ≈ RC
RE + rBE
If ii = 0 or Rg → ∞, the open-circuit output resistance is:
rCE RC
vo
ro,o =
= RC || (RE + rCE ) ≈ RC
io ii =0
Here, too, the dependence on Rg is so insignificant that it is negligible in practical appli-
cations. In our example we have ro = RC = 1 k.
For rCE RC , RE , β 1 and no load resistance RL , the following are obtained for
the common-emitter circuit with current feedback:
Common-emitter circuit with current feedback
gm RE 1
vo
g m RC RC
A =
≈ − ≈ − (2.70)
vi io =0 1 + g m RE RE
vi
ri = ≈ rBE + βRE = rBE (1 + gm RE ) (2.71)
ii
vo
ro = ≈ RC (2.72)
io
Insertion of the operating point, changing to small-signal values and serial expansion
results in
iC
vi = iC RE + VT ln 1 +
IC,A
iC VT i C 2 VT iC 3
= iC RE + VT − + − ···
IC,A 2 IC,A 3 IC,A
and yields after, inverting the series,
! 2 "
iC 1 vi 1 vi
= + + ···
IC,A 1 + g m RE VT 2 (1 + gm RE )2 VT
For input signal vi = v̂i cos ωt the ratio of the first harmonic with 2ωt to the fundamental
with ωt represents the approximate distortion factor k when using low amplitudes; that is,
when neglecting higher exponential powers:
vo,2ωt iC,2ωt v̂i
k ≈ ≈ ≈ (2.74)
vo,ωt iC,ωt 4VT (1 + gm RE )2
If a maximum is specified for k, then v̂i < 4kVT (1 + gm RE )2 must be applied. With
v̂o = |A|v̂i this leads to the maximum output amplitude. For the numeric example we thus
have v̂i < k · 93 V and, if A ≈ −1.93, v̂o < k · 179 V.
A comparison with (2.15) shows that due to the negative feedback the permissible input
amplitude v̂i is increased by the square of the feedback factor (1 + gm Re ). As the gain
is reduced by the feedback factor at the same time, the permissible output amplitude is
increased by the feedback factor as long as the distortion factor remains the same and the
transistor is not overloaded or driven into saturation; that is, as long as the range of validity
of the series expansion is not exceeded. With the same output amplitude the distortion
factor is reduced by the feedback factor.
14 The common-emitter circuit without feedback according to Fig. 2.57a requires the internal resis-
tance Rg of the signal voltage source in order to limit the base current in the saturation mode; in
the present case the base current is limited by R1 , i.e. we may set Rg = 0 and use the voltage
source Vi = Vg to drive the circuit. This approach is chosen in order to prevent the characteristics
from depending on Rg in normal mode.
108 2 Bipolar Transistor
5V Vb Vb
RC RC
2 k Ω R2 1k Ω R2 Io
Ii R1 IB
IC
R1
Vo Vo
1k Ω VBE B IB
Vi Vi
collector current increases together with Vi and the output voltage decreases accordingly;
this segment of the characteristic is almost linear because of the feedback. Up to here the
transistor operates in normal mode. It enters the saturation region if Vi > 1 V and we have
Vo = VCE,sat .
Normal mode: Figure 2.66b presents the equivalent circuit for the normal mode. From
the nodal equations
Vi − VBE Vo − VBE IC
+ = IB =
R1 R2 B
Vo Vi
VBE
Vi
Fig. 2.67. Characteristics of the common-emitter circuit with negative current feedback
2.4 Basic Circuits 109
Vb − V o Vo − VBE
+ Io = + IC
RC R2
For operation without any load – that is, Io = 0 – it follows that:
Vb R2 − IC RC R2 + VBE RC
Vo = (2.75)
R2 + R C
IC R 1 R1 R1
Vi = + VBE 1 + − Vo (2.76)
B R2 R2
Solving (2.75) with respect to IC and substituting into (2.76) leads to the following
equation if B 1 and BRC R2 :
Vb R2 R2 R2
Vo ≈ + 1+ VBE − Vi (2.77)
BRC R1 R1
If −0.6 V ≤ Vi ≤ 0.9 V, then VBE ≈ 0.7 V; thus (2.77) describes a linear relation between
Vo and Vi , as indicated as a broken line in Fig. 2.67, which corresponds very well to the
transfer characteristic. Thus, voltage feedback causes the transfer line in this region to
depend only on R1 and R2 as a first approximation.
Vi,A = 0 V is selected for the operating point; this point is located approximately in
the middle of the linear segment. In this case it is not possible to successively calculate
the operating point values because only implicit equations can be derived from (2.75) and
(2.76). However, the operating point can still be determined very accurately with the help
of approximations and an iterative approach; this is based on estimated values that become
more and more accurate in the course of the calculation. For R1 = 1 k, R2 = 2 k,
B = β = 400, Vi = 0 and the estimated value VBE ≈ 0.7 we obtain from (2.76)
From the nodal equation at the output for Vb = 5 V and RC = 1 k it follows that:
Vb − Vo Vo − VBE
IC = − ≈ 2.2 mA
RC R2
The estimated value for IC and IS = 7 fA make it possible to determine VBE more
accurately:
IC
VBE = VT ln ≈ 688 mV
IS
Repeating the calculation with this new value leads to:
Saturation mode: The transistor enters the saturation region when Vo reaches the sat-
uration voltage VCE,sat ; inserting Vo = VCE,sat ≈ 0.1 V and VBE ≈ 0.7 V into (2.77)
leads to Vi ≈ 1 V. For Vi > 1 V the collector diode is operated in the forward region.
110 2 Bipolar Transistor
Vo A
V
–1 1
5
Vi
4 V
3
–1
2
–2
–1 0 1 Vi
V
a Transfer characteristic b Gain = slope of the transfer
characteristic
Fig. 2.68. Gain of the common-emitter circuit with negative current feedback
Small-signal response: The voltage gain A corresponds to the slope of the transfer
characteristic (see Fig. 2.68); it is approximately constant in the region for which the
linear approximation according to (2.77) applies. Calculation of A is carried out with the
aid of the small-signal equivalent circuit shown in Fig. 2.69. From the nodal equations
vi − vBE vo − vBE vBE
+ =
R1 R2 rBE
vo − vBE vo vo
gm vBE + + + = io
R2 rCE RC
and RC
= RC ||rCE it follows that:
vo
− gm R2 + 1
A = =
vi
io =0 1 1 R2 R1
1 + R 1 gm 1 + +
+
1+
β RC RC rBE
rCE RC
β1 − gm R2 + 1
≈
R1 R2 R1
1 + g m R1 + + 1+
RC RC rBE
rBE R1
R1 ,R2 1/gm gm RC 1+R2 /R1
R2 R2
≈ − ≈ −
R1 + R2 R1
R1 +
gm R C
ii R1 R2 iC io
vo
vi vBE rBE gmvBE rCE RC RL
Fig. 2.69. Small-signal equivalent circuit for the common-emitter configuration with negative
current feedback
2.4 Basic Circuits 111
If all conditions are met then A depends only on R1 and R2 ; the last condition means
that the gain without feedback – that is, −gm RC – must be much higher than the ideal
gain with feedback – that is, −R2 /R1 . If the circuit is operated with a load resistance
RL the operating gain AB can be calculated by replacing RC by the parallel resistors RC
and RL (see Fig. 2.69). For the operating point selected as an example the exact value is
A = −1.885 for gm = 86.2 mS, rBE = 4.6 k, rCE = 45 k, RC = R1 = 1 k and
R2 = 2 k; while the first approximation is A = −1.912, the second approximation is
A = −1.933 and the third approximation is A = −2.
For the open-circuit input resistance it follows with RC
= RC || rCE that:
% &
vi
rBE RC
+ R2
ri,o = = R1 +
ii
io =0 rBE + (1 + β) RC
+ R2
rCE RC
β1 rBE (RC + R2 )
≈ R1 +
rBE + βRC + R2
βRC rBE ,R2
1 R2
≈ R1 + 1+
gm RC
gm RC R2 /R1
1 gm R1 1
≈ R1 + ≈ R1
gm
Here io = 0; that is, RL → ∞. For other values of RL the input resistance is calculated
by replacing RC by the parallel arrangement of RC and RL . Inserting RL = RC = 0 leads
to the short-circuit input resistance:
vi
ri,s = = R1 + rBE || R2
ii
vo =0
With the sample operating point the open-circuit input resistance is exactly ri,o = 1034 ;
the first approximation also provides ri,o = 1034 , the second approximation ri,o =
1035 , the third approximation ri,o = 1012 and the fourth approximation ri,o = 1 k.
The short-circuit input resistance is ri,s = 2.4 k.
With RC
= RC ||rCE the short-circuit output resistance is:
vo
rBE (R1 + R2 ) + R1 R2
ro,s = = RC
||
i
o vi =0 r + R (1 + β)
BE 1
rCE RC
β1 rBE (R1 + R2 ) + R1 R2
≈ RC ||
rBE + βR1
βR1 rBE
1 R2 R2
≈ RC || 1+ +
gm R1 β
For R1 → ∞ the open-circuit output resistance is:
rCE RC
vo
rBE + R2 β1 1 R2
ro,o =
= RC
|| ≈ RC || +
io ii =0 1+β gm β
For the sample operating point the short-circuit output resistance is exactly ro,s = 37.5 ;
the first approximation is also ro,s = 37.5 , and the second approximation ro,s = 38.3 .
112 2 Bipolar Transistor
The open-circuit output resistance is exactly ro,o = 16.2 ; the approximation is ro,o =
16.3 .
In the first approximation the following equations apply to the
Common-emitter circuit with voltage feedback
gm RC 1+R2 /R1
vo
R2 R2
A =
≈ − ≈ − (2.78)
vi io =0 R1 + R2 R1
R1 +
gm R C
vi
ri = ≈ R1 (2.79)
ii
vo 1 R2 R2
ro = ≈ RC || 1+ + (2.80)
io gm R1 β
Non-linearity: The voltage feedback significantly reduces the nonlinearity of the transfer
characteristic. The distortion factor of the circuit can be determined approximately by a
series expansion of the characteristic at the operating point. Insertion of the operating point
into (2.75) and (2.76) leads to:
RC iC
vo = − R2 iC + VT ln 1 +
R2 + R C IC,A
R1 R1 iC R1
vi = iC + 1 + VT ln 1 + − vo
β R2 IC,A R2
From the series expansion and by eliminating iC it follows for β 1 and gm R2 1 that:
R2 1 1 2 R2 VT R2
vo ≈ − vi + + 1+ 2
vi + · · ·
2
R1 R2 RC R1 2IC,A R1
For signals vi = v̂i cos ωt an approximation of the distortion factor k can be obtained
from the ratio of the first harmonic with 2ωt to the fundamental with ωt when using low
amplitudes; in other words, when disregarding higher order terms:
R2 R2
1+
vo,2ωt v̂i R1 R1
k ≈ ≈
vo,ωt 4VT gm 2
(R2 || RC )2
If a maximum value is specified for k then
2
gm (R2 || RC )2
v̂i < 4kVT
R2 R2
1+
R1 R1
The maximum output amplitude is obtained using v̂o = |A|v̂i . In the numeric example
above the values are as follows: v̂i < k · 57 V and for A ≈ −1.89 we have v̂o < k · 108 V.
Temperature sensitivity: The base–emitter voltage VBE decreases at a rate of 1.7 mV/K
according to (2.21). The resulting temperature drift of the output voltage can be calcu-
lated using the small-signal response by adding a voltage source vT D with dvT D /dT =
−1.7 mV/K in series with rBE (see Fig. 2.70) and calculating their effect on the output
2.4 Basic Circuits 113
R1 R2
vTD
vo
gmvBE rCE RC
v BE rBE
R1 R2
R1 vTD vo
vTD (1+ ) vBE rBE gmvBE rCE RC
R2 R2
Fig. 2.70. Small-signal equivalent circuit for calculating the temperature drift of the common-
emitter circuit with voltage feedback: with voltage source vT D (above) and after shifting the
source (below)
voltage. The calculation can be considerably simplified by shifting the voltage source in
a suitable manner: replacing the voltage source by two sources in series with R1 and R2 ,
converting the latter into two current sources vT D /R2 at the base and at the collector nodes
and then reconverting the source at the base node into a voltage source vT D R1 /R2 leads to
the equivalent small-signal circuit diagram shown in Fig. 2.70. Use of the values already
defined for A and ro,s results in:
dVo
R1 ro,s dvT D R1 mV
= − 1 + A + ≈ 1 + A · 1.7
dT A R2 R2 dT R2 K
For the sample operating point we arrive at a temperature drift (dVo /dT )|A ≈ −4.8 mV/K
for A = 1.885 and ro = ro,s = 37.5 .
βRC R2
B1 R2
≈ V b − R 2 Ii + V i
BRC
For Vi = VBE ≈ 0.7 V the approximation is Vo ≈ 0.72 V − 2 k · Ii .
The small-signal response of the current–voltage converter can be determined from two
equations for the common-emitter circuit with voltage feedback. The transfer resistance
15 The term transimpendence amplifier is also used for operational amplifiers with current feedback
and voltage output (CV-OPA).
114 2 Bipolar Transistor
Vb V
V
RC
R2 3
Vo
2
Vi
Vo 1
Ii Vi
–1.5 –1.0 – 0.5 0 0.5 Ii
mA
a Circuit b Characteristics
(transimpedance) takes the place of the gain; from (2.78) it follows that:
vo
vo
RT = = lim R1 = lim R1 A
ii
io =0 R1 →∞ vi
io =0 R1 →∞
− g m R2 + 1
=
1 1 R2
gm 1+ +
1+
β RC rBE
The input resistance can be calculated from the equations for the common-emitter cir-
cuit with voltage feedback by setting R1 = 0. The output resistance corresponds to the
open-circuit output resistance of the common-emitter circuit with voltage feedback. To
summarize, the equations for the current–voltage converter in common-emitter configu-
ration are:
Current-to-voltage converter
vo
RT = ≈ − R2 (2.81)
ii
io =0
vi 1 R2
ri = ≈ 1+ (2.82)
ii gm RC
vo 1 R2
ro = ≈ RC || + (2.83)
io gm β
2.4 Basic Circuits 115
B IS
Temperature coefficient + 0.5 %/K + 15 %/K
Variance − 30/ + 50 % − 70/ + 200 %
There are two fundamentally different methods for setting the operating point: AC
coupling and DC coupling.
Setting the Operating Point with AC Coupling: In AC coupling the amplifier or the
amplifier stage is connected to the signal source and the load via coupling capacitors (see
Fig. 2.72). This makes it possible to set the operating point independent of the DC voltage of
the signal source and independent of the load; the charge of the coupling capacitors reflects
the voltage difference. Since no DC current can flow through the coupling capacitors one
can connect any signal source or load without risking a shift in the operating point. With
multi-stage amplifiers the operating point can be set for every stage separately.
Each coupling capacitor forms a highpass filter together with the input or output re-
sistance of the coupled stage and with the signal source or the load. Figure 2.73 shows a
portion of the small-signal equivalent circuit of a multi-stage amplifier; the small-signal
equivalent circuit according to Fig. 2.61 with parameters A, ri and ro is used for every stage.
Vb Vb Vb Vb
R1 RC R1 RC
Co Co
Ci Ci
Vo Vo
Vi Vi
R2
Fig. 2.73. Small-signal equivalent circuit of a multi-stage amplifier to calculate the highpass filters
for AC coupling
116 2 Bipolar Transistor
vo
R1 R2 vTD vBE rBE gmvBE rCE RC
From the small-signal equivalent circuits it is possible to calculate the cutoff frequencies
of the highpass filters. The dimensions of the coupling capacitors must be selected so that
the lowest signal frequency of interest is still transmitted in full. DC voltages cannot be
transmitted.
In a common-emitter circuit the operating point may be set by voltage or current
biasing; VBE,A or IB,A is specified to produce the desired collector current IC,A and thus
the desired output voltage Vo,A . Due to
IC,A IC,A
VBE,A (T , C) = VT (T ) ln , IB,A (T , C) =
IS (T , C) B(T , C)
VBE,A and IB,A depend on the temperature T and on the individual component C.
Voltage biasing: For voltage biasing according to Fig. 2.72a the voltage VBE,A is ad-
justed using the resistances R1 and R2 . When a current through the resistances is much
higher than IB,A , the operating point is no longer influenced by a change in IB,A . The
dependency on the individual component can be eliminated by replacing R2 by a poten-
tiometer, in order to adjust the operating point. In order to calculate the temperature drift of
the output voltage caused by VBE a voltage source vT D with dvT D /dT = −1.7 mV/K is
introduced in the small-signal equivalent circuit (see Fig. 2.74). It acts like a signal voltage
source vg = −vT D with an internal resistance Rg = R1 ||R2 as shown by a comparison
with Fig. 2.60. Thus:
dVo
ri dvT D rBE mV
= − A = A · 1.7 (2.84)
dT A r +R
i g dT r + (R || R )
BE 1 2 K
Example: If A = −75 and R1 ||R2 = rBE , then (dVo /dT )|A ≈ −64 mV/K.
Due to its high temperature drift this method of setting the operating point is not used
in practice.
Current biasing: For current biasing according to Fig. 2.72b the base current IB,A is
set by resistance R1 :
Vb − VBE,A Vb − 0.7 V
R1 = ≈
IB,A IB,A
If Vb VBE,A , then a change of VBE,A has practically no influence on IB,A ; on the basis
of Vo = Vb − IC RC , it follows that:
dVo
dIC
dB IC,A RC VT dB
≈ − RC
= − IB RC = −
dT A dT IB =const. dT V B dT T
VT dB (2.23) mV
≈ A ≈ A · 0.13 (2.85)
B dT K
Vb Vb Vb
R1 RC RC
Co Co
Ci
Vo Vo
Vi Vi
R2 RE CE RE CE
– Vb
Even though the temperature drift is lower than with voltage biasing, it is still too high
for practical applications. Due to the large variance of β it is necessary to replace R1 by a
potentiometer in order to adjust the operating point. For this reason this method of setting
the operating point is not used in practice.
Setting the operating point by DC current feedback: The temperature drift is pro-
portional to the gain (see (2.84) and (2.85); therefore, the stability of the operating point
can be improved by reducing the gain. Since the temperature drift is a slow process it
is sufficient to reduce the DC voltage gain ADC ; the AC voltage gain AAC may remain
unaltered. This can be achieved by a frequency-dependent feedback which is effective only
for zero-frequency and frequencies below the lowest signal frequency of interest, while it
is totally or partially ineffective for higher frequencies. This is the principle on which the
setting of the operating point by means of DC current feedback according to Fig. 2.75a is
based; here, voltage biasing is combined with a current feedback through resistance RE .
With increasing frequencies the capacitance CE causes RE to be short-circuited and thus
renders the feedback ineffective for higher frequencies.
The voltage
% &
VB,A = IC,A + IB,A RE + VBE,A ≈ IC,A RE + 0.7 V
that is required at the base of the transistor for operation at the operating point is adjusted
using R1 and R2 ; the cross-current through the resistances is thus selected to be much
higher than IB,A , so that the operating point no longer depends on IB,A . With the signal
source providing a suitable DC voltage and the required base current IB,A , the resistances
and the coupling capacitor Ci may be omitted and a direct coupling is possible; then VB,A
can be tuned by adjusting RE to the existing DC input voltage. RE must not be made too
small since then the feedback would not be effective and the stability of the operating point
would be reduced. For low positive and negative DC input voltages it is possible to achieve
direct coupling by an additional negative supply voltage (see Fig. 2.75b).
The temperature drift of the output voltage follows from (2.84) by inserting the values
of the common-emitter circuit with current feedback according to (2.70) and (2.71) instead
of A and ri ; then A = ADC . ri R1 ||R2 results in the worst-case scenario:
dVo
mV gm RE 1 RC mV
≈ ADC · 1.7 ≈ − · 1.7
dT A K RE K
118 2 Bipolar Transistor
In addition, RE should be as high as possible to achieve a low DC voltage gain ADC and
thus a low temperature drift. In practice values in the range of RC /RE ≈ 1…10 are used.
The frequency response of the gain can be determined by means of the small-signal
equivalent circuit shown in Fig. 2.76 or by means of (2.70) through substitution of
RE ||(1/sCE ) for RE :
gm RE 1
gm RC (1 + sCE RE ) RC 1 + sCE RE
A(s) ≈ − ≈ −
1 + gm RE + sCE RE RE CE
1+s
gm
Figure 2.77 shows the absolute value of the frequency response A = |A(j 2πf )| and the
corner frequencies f1 and f2 ; in this case:
1 gm
ω1 = 2πf1 = , ω2 = 2πf2 ≈
CE RE CE
For f < f1 the feedback is fully effective; we have A ≈ ADC ≈ −RC /RE . For f > f2
the feedback is ineffective and we have A ≈ AAC ≈ −gm RC . The cross-over region lies
between these two values. The capacitance CE must be dimensioned so that f2 is lower
than the lowest signal frequency of interest.
The small-signal equivalent circuit according to Fig. 2.76 further shows that R1 and R2
are connected in parallel at the input and must be taken into consideration when calculating
the input resistance ri ; for f > f2 the input resistance is:
ri = rBE || R1 || R2
R1 and R2 must not be too small since otherwise the input resistance will drop sharply.
A [log]
– g mR C
RC
–
RE
1 gm f [log]
f1 = f2 ≈
2 π RE CE 2 π CE
Vb Vb Vb
RC RC RC
RE1
RE2 RE2
RE1 I0
r0
RE2 CE CE CE
a b c
Fig. 2.78. Setting the operating point with DC and AC current feedback
If, for example, in order to reduce the nonlinear distortions, a current feedback is
desired for AC voltages, which means for f > f2 , and if the AC voltage gain is to be
higher than the DC voltage gain, the circuits shown in Fig. 2.78 can be used. Figure 2.79
lists the characteristic parameters.
In the circuit shown in Fig. 2.78c a constant current source with current I0 and the
internal resistance r0 is used to set the operating point; thus IC,A ≈ I0 . Due to r0 RC ,
the DC voltage gain ADC , and hence the temperature drift caused by the transistor, is very
small; in this case the temperature drift of the circuit depends on the temperature drift of
the constant current source:
r0 RC
dVo
RC mV dI0 dI0
≈ − · 1.7 − RC ≈ − RC
dT A r0 K dT dT
Fig. 2.79. Characteristic parameters of the common-emitter circuit with DC current feedback
120 2 Bipolar Transistor
Vb Vb
R1 RC Co
Vb = 12 V
560 k Ω 6.8 k Ω 1µF
Rg Ci
10 k Ω 270 nF
T1
RE1 Vo RL
Vi R2 120 Ω 10 k Ω
Vg
240 k Ω
RE2 CE
4.7 k Ω 100 µF
temperature drift is noticed only as a shift in the operating point in the respective stage and
is not transmitted to the next stage as is the case with direct coupling.
Despite the advantages that AC coupling offers for straight AC amplifiers it is avoided
in practice if possible, because of the need for additional capacitors and resistors. This
is particularly the case with low-frequency amplifiers since their high capacitances make
it necessary to use electrolytic capacitors which are large and expensive and have a high
failure rate. AC coupling is widely found in high-frequency amplifiers; here ceramic capac-
itors in the picofarad range can be used – these are small and comparatively inexpensive.
In integrated circuits AC coupling is only found in exceptional cases, because capacitors
are very difficult to integrate. Where capacitors are still required, they are often connected
externally.
Setting the operating point with DC coupling: In DC coupling, which is also known
as direct or galvanic coupling, the amplifier or the amplifier stage is directly connected to
the signal source and the load. This requires adaptation of the DC voltages at the input and
the output at the operating point – that is, Vi,A and Vo,A – to the DC voltage of the signal
source and the load. With multi-stage amplifiers the operating point can no longer be set
separately for each stage.
In multi-stage amplifiers DC coupling is almost always used in combination with a
feedback across all stages; this means that the individual stages are coupled directly and
the operating point is set by the feedback. Vi,A = Vo,A is a frequent requirement which
means that the amplifier is not meant to alter the DC voltage component of the signal.
Example: Figure 2.81 shows an amplifier with DC voltage coupling that is made up of two
stages in common-emitter configuration and features feedback across both stages. The first
stage consists of npn transistor T1 and resistor R1 , the second stage of pnp transistor T2
and resistor R2 ; the resistors R3 , R4 and R5 make up the feedback for setting the operating
point and the gain. The amplifier is rated for Vi,A = Vo,A = 2.5 V and A = 10. For a
common-emitter circuit with an npn transistor working at the operating point, the output
voltage is higher than the input voltage while it is lower in a common-emitter circuit with
a pnp transistor. Therefore, it is useful to use a pnp transistor in the second stage in order
122 2 Bipolar Transistor
Vb Vb Vb
R1
10 k Ω VEB2 Vb = 5V
R5
10 k Ω T2
I C1 I B2
I C2
T1
R3
VBE1
33 k Ω
Vi Vo
R4 R2
VE
4.7 k Ω 1.8 k Ω
Fig. 2.81. Example of an amplifier with DC coupling featuring two stages in common-emitter
configuration and feedback
to meet the requirement Vi,A = Vo,A . The dimensioning of the resistors is not described
here.
Calculation of the operating point is based on Vo,A = 2.5 V. Disregarding the current
through R3 leads to IC2,A ≈ −Vo,A /R2 ≈ −1.4 mA. If IS2 = 1 fA and β2 = 300,17 then
VEB2,A = VT ln(−IC2,A /IS2 ) ≈ 0.73 V and IB2,A ≈ −4.7 mA. From this it follows that
IC1,A = VEB2,A /R1 − IB2,A ≈ 78 mA. From the nodal equation
VE,A Vo,A − VE,A Vb − VE,A
= + + IC1,A
R4 R3 R5
at the emitter connection of T1 we have VE,A = 1.9 V. If IS1 = 7 fA, then VBE1,A =
VT ln(IC1,A /IS2 ) ≈ 0.6 V which leads to Vi,A = VBE1,A + VE,A ≈ 2.5 V. Finally a check
must be made as to whether it is permissible to disregard the current through R3 in the
calculation of IC2,A : IR3 = (Vo,A − VE,A )/R3 ≈ 18 mA |IC2,A |. This calculation again
illustrates the procedure for calculating operating points.
Rg' = Rg + RB
RC' = rCE ⏐⏐RC ⏐⏐RL
Rg RB CC
vo
vg vi rBE CE vBE gmvBE rCE RC RL
Fig. 2.82. Dynamic small-signal equivalent circuit for the common-emitter circuit without
feedback
the capacitances of the transistor. To obtain information on the frequency response and
the upper cutoff frequency it is essential to use the dynamic small-signal model of the
transistor according to Fig. 2.41 on page 78 as a basis for calculation; this model takes
into account the base spreading resistance RB as well as the emitter capacitance CE and
the collector capacitance CC .
Common-emitter circuit without feedback: Figure 2.82 shows the dynamic small-
signal equivalent circuit of the common-emitter circuit without feedback. The operating
gain AB (s) = v o (s)/v g (s) with Rg
= Rg + RB and RC
= RL ||RC ||rCE is:
(gm − sCC ) RC
AB (s) = − # # $$
Rg
1+ + s CE Rg
+ CC Rg
+ RC
+ gm RC
Rg
+ s 2 CE CC Rg
RC
rBE
(2.86)
Figure 2.83 shows the magnitude of the frequency response with the corner frequencies
fP 1 and fP 2 of the two poles and the corner frequency fN of the zero. The zero can be
−1 = (2πf )−1 . The two poles are
disregarded on the basis of its short time constant CC gm N
real and far apart. Therefore, the frequency response can be described approximately by
⏐A B⏐ [log]
A0
f –3dB ≈ f P1 f P2 fN f [log]
Fig. 2.83. Magnitude frequency response |AB | of the common-emitter circuit: (a) calculation with
(2.86) and (b) approximation (2.88)
124 2 Bipolar Transistor
a lowpass filter of first order, after the s 2 term in the denominator has been removed.19
Using the low-frequency gain
rBE
A0 = AB (0) = − gm RC
(2.87)
rBE + Rg
results in:
A0
AB (s) ≈
(2.88)
R
# $
1 + s CE + C C 1 + gm RC
+ C
rBE || Rg
Rg
Figure 2.83 shows the magnitude frequency responses of the approximation given in (2.88)
and of the full expression (2.86).
From (2.88) we obtain the approximation for the −3dB cutoff frequency f-3dB at which
the magnitude of the gain declines by 3 dB:
1
ω-3dB = 2πf-3dB ≈
(2.89)
R
# $
CE + C C 1 + gm RC
+ C
rBE || Rg
Rg
In most cases we have RC
, Rg
1/gm :
1
ω-3dB = 2πf-3dB ≈ % &# $ (2.90)
CE + CC gm RC
rBE || Rg
The upper cutoff frequency depends on the low-frequency gain A0 . Under the assump-
tion that a change in A0 is induced by a change of RC
and that all other parameters remain
constant, solving (2.87) for RC
and substituting an expression with two time constants that
ar independent of A0 into (2.89) leads to:
1
ω-3dB (A0 ) ≈ (2.91)
T1 + T2 |A0 |
# $
T1 = (CE + CC ) rBE || Rg
(2.92)
1
T2 = CC Rg
+ (2.93)
gm
Two regions can be distinguished:
– If |A0 | T1 /T2 , then ω-3dB ≈ T1−1 ; that is, the upper cutoff frequency is independent
of the gain. For the borderline case of A0 → 0 and Rg = 0 we obtain the maximum
upper cutoff frequency:
rBE RB
1 1
ω-3dB,max ≈ ≈
(CE + CC ) (rBE || RB ) (CE + CC ) RB
19 This approach corresponds to a well-known procedure from control engineering in which several
poles are considered to be one pole with the sum of the time constants: (1+sT1 )(1+sT2 ) . . . (1+
sTn ) ≈ 1 + s(T1 + T2 + . . . + Tn ). The coefficient of s is the sum of the time constants. The
combination is thus achieved by omitting the higher powers of s.
2.4 Basic Circuits 125
1
GBW = f-3dB |A0 | ≈ (2.94)
2π T2
Z C (s) = RC
|| = (2.95)
sCL 1 + sCL RC
as shown in Fig. 2.84. When inserting Z C (s) into (2.86), neglecting according to (2.88)
and determining the time constants T1 and T2 , it becomes obvious that T1 does not change,
while T2 can be expressed as:
CL CC + CL
T2 = CC + Rg
+ (2.96)
β gm
Due to the load capacitance CL the gain bandwidth product GBW is reduced in accordance
with the increase of T2 (see (2.94).
ZC
Rg CC
vo
vg rBE CE vBE gmvBE RC CL
Rg ro
vg vi ri Ci Av i Co vo CL RL
Example: IC,A = 2 mA was selected for the numeric example used in the common-
emitter circuit without feedback according to Fig. 2.57a. For β = 400, VA = 100 V,
Cobo = 3.5 pF and fT = 160 MHz we obtain from Fig. 2.45 on page 83 the small-signal
parameters gm = 77 mS, rBE = 5.2 k, rCE = 50 k, CC = 3.5 pF and CE = 73 pF. For
Rg = RC = 1 k, RL → ∞ and Rg
≈ Rg it follows from (2.87) that A0 ≈ −63, from
(2.89) that f−3dB ≈ 543 kHz and from (2.90) that f−3dB ≈ 554 kHz. It follows from (2.92)
that T1 ≈ 64 ns, from (2.93) that T2 ≈ 3.55 ns and from (2.94) that GBW ≈ 45 MHz.
With a load capacitance CL = 1 nF we obtain from (2.96) T2 ≈ 19 ns, from (2.91)
f−3dB ≈ 126 kHz and from (2.94) GBW ≈ 8.4 MHz.
Common-emitter circuit with current feedback: The frequency response and the up-
per cutoff frequency of the common-emitter circuit with current feedback according to
Fig. 2.62a can be derived from the corresponding parameters of the common-emitter cir-
cuit without feedback. Figure 2.86a shows a portion of the small-signal equivalent circuit
in Fig. 2.82 with the additional resistance RE of the current feedback; here resistance rCE
is disregarded. This portion may be converted to the circuit shown in Fig.2.86b,20 which
20 This is not an equivalent conversion since it is based on the neglection of one pole in the Y matrix.
For any value of RE , however, the cutoff frequency of this pole is above the transistor transit
frequency fT and thus in a range that cannot be applied to the small-signal model of the transistor
anyway; therefore, the conversion is virtually equivalent [2.11].
2.4 Basic Circuits 127
CC CC
rBE
CE vBE gmvBE rBE CE vBE g'mvBE
RE
Fig. 2.86. Conversion of the small-signal equivalent circuit of the common-emitter circuit with
current feedback
leads back to the original small-signal equivalent circuit of Fig. 2.82; consequently:
since in this case they are only dependent on Rg and CC . Consequently, in the region of
|A0 | > T1 /T2 with constant GBW the upper cutoff frequency increases at exactly the
same rate as the gain decreases due to the current feedback. Therefore, the upper cutoff
frequency can be increased by means of a current feedback at the cost of the gain, while
the product of both does not increase.
The influence of load capacitance CL can be determined from (2.96) by inserting
the equivalent value gm
to replace g . With high current feedback small values of C
m L
already have a comparably high bearing, since T2 increases rapidly due to gm
g ; the
m
gain–bandwidth product GBW decreases accordingly.
The common-emitter circuit with current feedback can be described approximately
by the equivalent circuit shown in Fig. 2.85. The input capacitance Ci and the output
capacitance Co are determined from (2.97) and (2.98) by inserting the equivalent values
rBE and CE
for rBE and CE ; A, ri and ro are given by (2.70)–(2.72).
Example: IC,A = 1.5 mA is selected for the numeric example of the common-emitter
circuit with current feedback shown in Fig. 2.62a. For β = 400, Cobo = 3.5 pF and
fT = 150 MHz the small-signal parameters gm = 58 mS, rBE = 6.9 k, CC = 3.5 pF
and CE = 58 pF are determined from Fig. 2.45 on page 83; rCE is neglected. Conversion
to (2.99)–(2.101) provides for RE = 500 the equivalent values rBE = 207 k, gm
=
R1 R2
Rg R1 CC
RB
vo
vg vi rBE CE vBE gmvBE RC
Fig. 2.87. Dynamic small-signal equivalent circuit of the common-emitter circuit with voltage
feedback
it follows from (2.96) that T2 ≈ 526 ns, from (2.91) that f−3dB ≈ 156 kHz and from (2.94)
that GBW ≈ 303 kHz.
A comparison with the example of the common-emitter circuit without feedback on
page 126 shows that the gain bandwidth product GBW without load capacitance is identi-
cal; therefore, the upper cutoff frequency is reduced by a factor of 30 due to the gain which
is 30 times higher. For CL = 1 nF the upper cutoff frequency is about the same despite the
difference in gain; in this case the influence of T2 is predominant and the result for both
circuits is (ω−3dB )−1 ≈ T2 |A0 | ≈ CL RC
≈ 1 ms.
Common-emitter circuit with voltage feedback: Figure 2.87 shows the small-signal
equivalent circuit for the common-emitter configuration with voltage feedback; here RC
=
rCE ||RC ||RL still applies. The calculation of AB (s) is complicated. However, it is possible
to use the results from the common-emitter circuit by neglecting – that is, short-circuiting –
the base spreading resistance RB as shown in Fig. 2.87 and substituting into (2.86) the
parallel arrangement of CC and R2 for CC and resistance R1
= R1 + Rg for Rg . For R1
,
R2 , R0
1/gm and rBE R1
, the resulting approximation is sufficiently accurate for
practical purposes:
R
R
gm RC
R2 1 2 R2
A0 ≈ − ≈ − (2.102)
R2 R1
R1
+
gm RC
A0
AB (s) ≈ (2.103)
CE R2 C E CC R2
1+s 1 +
+ C C R2 + s 2
gm RC gm
Even though the two poles are not as far apart as in the common-emitter circuit without
feedback and the common-emitter circuit with current feedback, the upper cutoff frequency
can be estimated with sufficient accuracy by neglecting the s 2 term in the denominator of
AB (s):
1
ω-3dB = 2πf-3dB ≈ (2.104)
CE R2
1 +
+ C C R2
gm RC
This value depends on A0 . Considering A0 ≈ −R2 /R1 , and assuming a change in A0
induced by a change in R2 and a constant R1
, leads to a simple explicit expression with
two time constants that are independent of A0 :
1
ω-3dB (A0 ) ≈ (2.105)
T1 + T2 |A0 |
2.4 Basic Circuits 129
CE
T1 = (2.106)
gm
CE
T2 = + CC R1
(2.107)
gm RC
Example: IC,A = 2.24 mA is chosen for the numeric example of the common-emitter
circuit with voltage feedback shown in Fig. 2.66a. For β = 400, Cobo = 3.5 pF and
fT = 160 MHz the small-signal parameters gm = 86 mS, rBE = 4.6 k, CC = 3.5 pF
and CE = 82 pF are determined from Fig. 2.45 on page 83; rCE is disregarded. For
RC = R1 = 1 k, R2 = 2 k, RL → ∞ and Rg = 0 from (2.102) we have A0 ≈ −1.96,
from (2.106) T1 ≈ 0.95 ns, from (2.107) T2 ≈ 4.45 ns, from (2.105) f−3dB ≈ 16 MHz
and from (2.94) GBW ≈ 36 MHz. With a load capacitance CL = 1 nF, from (2.108) we
obtain T1 ≈ 12.6 ns, from (2.109) T2 ≈ 16.1 ns, from (2.105) f−3dB ≈ 3.6 MHz and from
(2.94) GBW ≈ 9.9 MHz.
A comparison with the example of the common-emitter circuit with current feedback
on page 127 shows that in both circuits approximately the same upper cutoff frequency is
achieved without load capacitance. With a load capacitance of CL = 1 nF the common-
emitter circuit with voltage feedback produces an upper cutoff frequency that is 20 times
higher; this is caused by the significantly lower output resistance ro . For this reason voltage
feedback should be preferred to the current feedback where large load capacitances are
involved.
21 In practical circuit applications there is a parasitic stray capacitance of some pF caused by the
physical circuit arrangement.
130 2 Bipolar Transistor
Summary
The common-emitter circuit can be operated without feedback, with current feedback or
with voltage feedback. Figure 2.88 shows the three variations; Fig. 2.89 lists the most
important characteristic values.
The gain of the common-emitter circuit without feedback is heavily dependent on the
operating point; therefore an accurate and temperature-stable setting of the operating point
Vb Vb Vb
RC RC RC
R2
R1
Vo Vo Vo
Vi Vi Vi
RE
RC R2
A − gm RC − −
RE R1
gm gm 2π
+ CC R1
gm RC
for Rg
= Rg + RB for Rg
= Rg + RB for R1
= R1 + Rg
according to (2.100)
and gm and RC
= R ||R
C L
2.4.2
Common-Collector Circuit
Figure 2.90a shows the common-collector configuration consisting of the transistor, the
emitter resistance RE , the supply voltage source Vb and the signal voltage source Vg with
the internal resistance Rg . The following explanations are based on the assumption that
Vb = 5 V and RE = Rg = 1 k.
Vb Vb
Rg IB IC
Rg
VBE B IB
Vi
Vg Vg Io
Vo Vo
RE RE
V
V
5
Vi ≈ V g
4
3 Vo
VBE
2
1 2 4 3 5 Vg
V
Vb
V
V
Vg Vi ≈ Vg
Vo
VBE
Vb
Vo
Vg
V
Fig. 2.92. Characteristics of the common-collector circuit with an additional negative supply
voltage and the load RL
Here, the voltage drop across Rg is only 5 mV and can be neglected; therefore in Fig. 2.91
we have Vi ≈ Vg .
Feeding the common-collector circuit with an additional negative supply voltage −Vb
and placing a load RL between output and earth as shown in Fig. 2.92 enables the generation
of negative output voltages. The transfer characteristic then depends on the ratio between
resistances RE and RL as the minimum output voltage Vo,min is determined by the voltage
divider formed by RL and RE :
Vb RL
Vo,min = −
RE + R L
This means that a large signal range is only achieved if |Vo,min | is large; this requires
RL > RE . For Vg < Vo,min the transistor is in reverse mode because VBE < 0 and there-
fore Vo = Vo,min . For Vg ≥ Vo,min the transistor is in forward mode and the characteristic
is as shown in Fig. 2.91. In Fig. 2.92 the supply voltages are symmetrical; in other words,
the positive and negative supply voltages have the same absolute value. This is typical in
practice, but, as a general rule, the negative supply voltage can be chosen independent of
the positive supply voltage.
Rg ii
vg vi
io
vo
RE RL
Vi,A , Vo,A , Ii,A = IB,A and IC,A ; as an example we use the operating point as determined
above with Vi,A = 2.69 V, Vo,A = 2 V, IB,A = 5 mA and IC,A = 2 mA.
The small-signal voltage gain corresponds to the slope of the transfer characteristic. As
the output voltage follows the input voltage the differentiation of (2.113) leads as expected
to the approximation:
∂Vo
A = ≈ 1
∂V
i A
A more accurate calculation of A is possible using the small-signal equivalent circuit shown
in Fig. 2.93. The nodal equation
vi − vo 1 1
+ gm vBE = + vo
rBE RE rCE
= R ||r
for vBE = vi − vo and RE E CE leads to:
1
1+ gm RE
vo
β
A = =
vi
io =0 1
1+ gm RE +1
β
rCE RE
β1 gm RE 1
g m RE
≈ ≈ 1
gm R E + 1
For gm = IC,A /VT = 77 mS, β = 400, RE = 1 k and rCE = VA /IC,A = 50 k
we achieve for the selected operating point the exact value and the first approximation
A = 0.987.
The small-signal input resistance is:
rCE RE
gm RE 1
vi
β1
ri = = rBE + (1 + β) RE ≈ rBE + βRE ≈ βRE
ii
io =0
This depends on the load resistance which in our case is the open-circuit input resistance
since io = 0 (RL → ∞). The input resistance for other values of RL can be calculated by
replacing RE with the parallel arrangement of RE and RL (see Fig. 2.93; with RL < RE ,
which is often found in practical applications, it thus depends to a high degree on RL . For
rBE = β/gm and RL → ∞ the input resistance for the chosen operating point is exactly
ri = 398 k; the first and second approximations are ri = 405 k and ri = 400 k,
respectively.
2.4 Basic Circuits 135
ro [log]
RE
Rg
1
gm
rBE RE Rg [log]
Fig. 2.94. Plot of the small-signal output resistance ro of the common-collector circuit in its
dependence on the internal resistance Rg of the signal generator
vg vg
Vo
Vg Vo Vg'
Vg,' A
Vg, A
gm RE
A = ≈ ≈ 1 (2.114)
vi
io =0 1 + g m RE
gm RE 1
vi
ri = ≈ r + βR ≈ βRE (2.115)
ii
io =0
BE E
vo Rg 1
ro = ≈ RE || + (2.116)
io β gm
Rg 1
A≈1 , ri ≈ β(RE || RL ) , ro ≈ + (2.117)
β gm
The relevant equivalent circuit with signal generator and load is shown in Fig. 2.96.
One can see that the common-collector circuit features a strong coupling between input
and output since in this case, unlike in the common-emitter circuit, the input resistance ri
depends on the load RL at the output and the output resistance ro depends on the source
resistance Rg of the signal generator at the input.
Figure 2.96 enables the small-signal operating gain to be calculated:
vo ri RL
AB = =
vg ri + R g RL + r o
In most cases ri Rg and RL ro ; this results in AB ≈ 1.
Rg 1
ro ≈ +
Rg gm
vi vo
vg ri ≈ ( RE RL ) vi RL
23 Compare Fig. 2.91 on page 132 with Fig. 2.58 on page 97.
24 Values for temperature sensitivity and variations are given on page 115.
138 2 Bipolar Transistor
R1
Co Co
Ci
Co
I0
R2 RE RE
r0
When setting the operating point a distinction is made between AC coupling and DC
coupling. In addition to the purely AC or DC coupling in many common-collector circuits
a DC coupling at the input is combined with an AC coupling at the output.
Setting the operating point in the case of AC coupling: Figure 2.97a shows AC cou-
pling. The signal source and the load are connected via coupling capacitors and the oper-
ating point voltages can be selected independent of the DC voltages of the signal source
and of the load; other properties are described on page 115. The voltage
% &
VB,A = IC,A + IB,A RE + VBE,A ≈ IC,A RE + 0.7 V
required at the operating point at the base of the transistor is determined by R1 and R2 ;
the current through the resistors is selected to be much higher than the base current IB,A
so that the operating point is not influenced by IB,A .
In practice pure AC coupling is rarely used because in most cases DC coupling is
possible at the input at least; this means that resistors R1 and R2 and coupling capacitor
Ci can be omitted.
Setting the operating point by DC coupling at the input: Figure 2.97b shows the
common-collector circuit with DC coupling at the input and DC or AC coupling at the
output. The input voltage Vi,A at the base of the transistor is given by the output voltage
of the signal source when it is assumed that the voltage drop IB,A Rg caused by the base
current IB,A through the internal resistance of the signal source can be neglected. With AC
coupling at the output the collector current at the operating point can be set by resistance
RE , according to
Vi,A − VBE,A Vi,A − 0.7 V
IC,A ≈ ≈ (2.120)
RE RE
or by a current source; Fig. 2.97b shows both possibilities. If a current source is used, then
IC,A ≈ I0 ; furthermore, in the small-signal calculation the internal resistance r0 of the
current source must be used instead of resistance RE . In addition, for DC coupling at the
output the output current Io,A through the load must be taken into account.
Example: In the example on page 120 a common-emitter circuit is dimensioned for a load
RL = 10 k (see Fig. 2.80 on page 121). Now the circuit should be operated with a load
2.4 Basic Circuits 139
Vb Vb Vb
RC
R1 6.8 k Ω Vb = 12 V
560 k Ω
T2 Co
Ci
15 µF
270 nF
T1
RE1 Vi
R2 120 Ω RE Vo RL
240 kΩ 7.5 k Ω 1 kΩ
RE2 CE
4.7 k Ω 100 µF
Using AC and/or DC coupling: The most important considerations for the use of AC
and/or DC coupling are described on pages 120 and 122. Usually the use of DC coupling
at the output is complicated by the fact that low-resistance loads cause relatively high DC
currents to flow even with low DC voltages at the output.
1+β + + sc1 + s 2 CE CC Rg
rBE
RL
rBE Rg
c1 = CE rBE + (CE + CC ) + CC Rg
(1 + β)
RL
1+
βRL
(2.122)
CE Rg
C C R
2 E C g
1+s 1 +
+ CC Rg + s
gm RL gm
Both poles are real and due to
gm
fN = > fC
2π CE
the corner frequency at the zero is above the transit frequency fT of the transistor as can be
seen by comparison with (2.44). The frequency response can be described in an approxi-
Rg' = Rg + RB CE
RL' = rCE ⏐⏐RC ⏐⏐RL
Rg RB rBE
vBE vo
vg vi CC gmvBE rCE RC RL
Fig. 2.99. Dynamic small-signal equivalent circuit for the common-collector configuration
2.4 Basic Circuits 141
mation by a lowpass filter of first order after removing the s 2 term from the denominator
and subtracting the linear terms:
A0
AB (s) ≈
CE
1+s + C C Rg
gm RL
This results in an approximation of the upper −3dB cutoff frequency f-3dB at which the
magnitude of the gain has declined by 3 dB:
1
ω-3dB = 2πf-3dB ≈ (2.123)
CE
+ C C Rg
gm RL
Due to Rg
= Rg + RB ≈ Rg this value is proportional to the internal resistance Rg of
the signal generator. The maximum upper cutoff frequency is achieved with Rg → 0 and
RL
→ ∞
1
ω-3dB,max ≈
CC RB
which is usually higher than the transit frequency fT of the transistor.
If the load also contains a capacitive portion in addition to the resistive portion – that
is, if a load capacitance CL exists in parallel to the load resistance RL , we obtain by
substituting
1 RL
Z L (s) = RL
|| =
sCL 1 + sCL RL
instead of RL
:
CE
A0 1 + s
gm
AB (s) ≈ (2.124)
1 + sc1 + s 2 c2
CE Rg
1 Rg
c1 = 1 +
+ CC Rg + CL +
gm RL gm β
Rg
c2 = (CC CE + CL (CC + CE ))
gm
In this case the poles can be real or complex conjugate. An approximation by a lowpass
filter of first order yields a reasonable estimation of the upper cutoff frequency for real
poles only:
1
ω-3dB = 2πf-3dB ≈ (2.125)
CE CL
CL
+ C + R +
gm RL
C g
β gm
For complex conjugate poles the following estimation must be used:
1
ω-3dB = 2πf-3dB ≈ √ (2.126)
c2
142 2 Bipolar Transistor
CE
Rg RB rBE
vBE vo
vg Cg CC gmvBE rCE RE RL CL
CE
Rg rBE
vBE vo
vg Cg gmvBE RL CL
Fig. 2.100. Small-signal equivalent circuit for calculating the range of complex conjugate poles:
accurate (top) and simplified (bottom)
From (2.124) it follows that the common-collector circuit is always stable26 ; that is,
with complex conjugate poles there is still an oscillation in the step-response, although
this does die down again. But in practical applications this circuit may become unstable,
which means that a sustained oscillation occurs that, due to overload effects, stabilizes
at a certain amplitude and under unfavorable conditions may destroy the transistor. This
instability is caused by second order effects that are not taken into consideration by the
small-signal equivalent circuit of the transistor used in this case.27
26 A second order transfer function with positive coefficients in the denominator is stable.
27 Due to the transit time in the base region of the transistor there is an additional time constant; in
the small-signal equivalent circuit of the transistor this effect can be simulated by an inductance
in series with the base spreading resistance RB . This results in a third order transfer function that
may become instable with a capacitive load.
2.4 Basic Circuits 143
% & kg
c1 = T E 1 + k g + T g + T L kS +
β (2.129)
c2 = T g T E + T g T L kS + T L T E kg
This enables the quality to be calculated:
√
c2
Q = (2.130)
c1
and under the condition Q > 0.5, the range of complex conjugate poles can be determined.
In Fig. 2.101 this range is shown for β = 50 and β = 500 as a function of the normalized
1000
= 50
Tg
TE
100
10
Rg
kg =
RL
1 10 10 1
0.1
1000
= 500
Tg
TE
100
10
Rg
kg =
RL
1 10 100 100 1 10
0.1
1 10 100 1000 10000 100000
TL
TE
Fig. 2.101. Range of complex conjugate poles for β = 50 and β = 500 with gray background
144 2 Bipolar Transistor
signal source time constant Tg /TE and the normalized load time constant TL /TE for
various values of kg ; kS = 0.01 is used.
Figure 2.101 shows that no complex conjugate poles occur for very small and very
large load capacitances CL (TL /TE small or large) and with a sufficiently large output
capacitance Cg of the signal generator (Tg /TE large). The range of complex conjugate
poles is heavily dependent on kg . The regions for kg < 1 are within the region for kg = 1;
no complex conjugate poles occur for kg > β. The dependence on kS has an influence only
in the event of a large load capacitance (TL /TE large), a high current gain β and small
internal resistance Rg of the signal generator; this is responsible for the dent in the right
portion of the plot shown in Fig. 2.101 for β = 500 and kg = 1.
If Rg , Cg , RL and CL are given and complex conjugate poles occur, there are four
different ways to avoid this range:
1. Tg can be increased in order to leave the range of complex conjugate poles toward the
top. This requires the insertion of an additional capacitor connected between the input
of the common-collector circuit and ground or to a supply voltage; in the small-signal
equivalent circuit it is arranged in parallel with Cg and causes an increase in Tg . This
method can always be used; for this reason it is very common in practical applications.
2. TE can be increased in order to leave the range toward the lower left if operating close
to the left boundary of this range. This requires the use of a slower transistor with a
higher time constant TE ; that is, a lower transit frequency fT .
3. TE can be reduced in order to leave the range toward the top right if operating close
to the right boundary of this range. This requires the use of a faster transistor with
a shorter time constant TE ; that is, a higher transit frequency fT . This method is
used, for example, in power supplies because of their high load capacitance as a
result of the capacitor at the output which shifts the operating point toward the right
boundary; the use of a faster transistor in this case leads to an improved transient
response.
4. TL can be increased in order to leave the range toward the right if operating close to
the right boundary of this range. This requires an enlargement of the load capacitance
CL by adding an additional capacitor in parallel. This possibility is also used in power
supplies; it increases the capacitor at the output accordingly.
The above four methods are indicated in Fig. 2.102. A fifth method is the reduction
of TL , which is seldom used in practical applications since with given values for RL and
CL it can be achieved only by connecting a resistor in parallel, which means adding an
additional load to the output. All of these methods cause a reduction in the upper cutoff
frequency. In order to keep this reduction within acceptable limits it is necessary to leave
the range of complex conjugate poles in the shortest way possible.
Tg
lg
TE
Tg
TE
TE
TL
TL
TL
lg
TE
Rg ro Lo
vi vo
vg Cg ri Ci vi Co RL CL
CE rBE + CL RL
ri = βRL
+ rBE , Ci =
βRL
+ rBE
Rg
1 βCg
Rg
ro = + , Co =
β gm Rg
+ rBE
CE Rg
Lo =
gm
This indicates that in addition to the resistances ri and ro the capacitances Ci and Co as
well as the inductance Lo are also heavily dependent on the signal source and the load;
this means that there is strong coupling between the input and the output.
Example: IC,A = 2 mA was selected for the numeric example of Fig. 2.90a. With β = 400,
VA = 100 V, Cobo = 3.5 pF and fT = 160 MHz we obtain from Fig. 2.45 on page 83
the small-signal parameters gm = 77 mS, rBE = 5.2 k, rCE = 50 k, CC = 3.5 pF
and CE = 73 pF. For Rg = RE = 1 k, RL → ∞ and Rg
≈ Rg and for RL
=
RL ||RE ||rCE = 980 it follows from (2.121) that A0 = 0.984 ≈ 1 and from (2.123)
that f−3dB ≈ 36 MHz. With a load capacitance CL = 1 nF we achieve using (2.125)
f−3dB ≈ 8 MHz and using (2.126) f−3dB ≈ 5 MHz. From (2.127) and (2.128) we obtain
Tg = 3.5 ns, TL = 980 ns, TE = 0.95 ns, rg = 0.98 and rS = 0.013 and thus from
(2.129) c1 = 20.6 ns and c2 = 979 (ns)2 . From (2.130) it follows that Q = 1.52; that
is, there are complex conjugate poles. This result is also achieved by means of Fig. 2.101
since point TL /TE ≈ 1000, Tg /TE ≈ 4, kg ≈ 1 is within the range of complex conjugate
poles; the region for β = 500 is used because β = 400. In this case, leaving the range of
146 2 Bipolar Transistor
ZBE iB ZBE
v BE
Zg g m v BE ZL iB
Zi Zo
complex conjugate poles is possible only by increasing Tg to Tg /TE ≈ 75; this means that
Cg
≈ 71 pF is required; that is, a capacitor of Cg = Cg
− CC ≈ 68 pF must be connected
between the base of the transistor and ground. This causes a reduction in the upper cutoff
frequency; from (2.125) we obtain f−3dB ≈ 1.8 MHz if CC is replaced by Cg
= 71 pF. A
smaller Cg could be chosen if a weak complex conjugate pole pair is allowed with resulting
overshoots when applying square-wave signals; then the upper cutoff frequency does not
drop as much.
28 If C = 0 then ω−1 = C r
C β E BE , see (2.43); furthermore, β0 = |β(j 0)| = gm rBE .
2.4 Basic Circuits 147
Vb Vb
Zg Zo ZL Zi
Figure 2.105 illustrates this relationship. Often it is possible to neglect Z BE (s) and to use
the simplified transformation formulas:
Z g (s)
Z i (s) ≈ β(s)Z L (s) , Z o (s) ≈
β(s)
Figure 2.106 shows some selected examples. Particularly noteworthy are the cases of
Z g (s) = sL and Z L (s) = 1/(sC) in which the transformation produces a frequency-
dependent negative resistance; in this case Z o (s) and Z i (s) are no longer passive and
unfavorable circuitry can render the circuit instable. This means for practical purposes that
inductances in the base circuit and/or capacitances in the emitter circuit of a transistor
may produce unwanted oscillations; an example of this is the common-collector circuit
with capacitive load. The RC parallel circuit with the secondary condition ωβ RC = 1, as
shown in the lower left part of Fig. 2.106 causes a purely ohmic output impedance; here an
Zg Zo ZL Zi
1
C b0 C w C
T
L b0 L wT L
R R
R b0 wT 1
R b0 R
wT R
2
L w L
L b0 wT C wT
C
b0 w2R
R R
b0 R
C b0 R
L
wb CR = 1 wb L = R
Vb Vb
RC B IB RC
IC Io
IB
Vo Vo
Vi
RBV Vi VBE RBV VB
additional capacitance at the output will not produce complex conjugate poles – in other
words, no oscillations will occur.
2.4.3
Common-Base Circuit
Figure 2.107a shows the common-base circuit consisting of the transistor, the collector
resistance RC , the supply voltage source Vb and the signal voltage source Vi .29 The resis-
tance RBV acts as a limiter for the base current in the event of overload; in normal mode
it has no noticeable influence. The following considerations are based on Vb = 5 V and
RC = RBV = 1 k.
Normal mode: Figure 2.107b shows the equivalent circuit for the normal mode in which
the transistor is replaced by the simplified transport model according to Fig. 2.27 with:
VBE
IC = BIB = IS e VT
From Fig. 2.107b it follows that:
Io =0
Vo = Vb + (Io − IC ) RC = Vb − IC RC (2.131)
IC RBV
Vi = − VBE − IB RBV = − VBE − ≈ − VBE (2.132)
B
29 In contrast to the procedure with common-emitter and common-collector circuits, here a voltage
source without internal resistance is used; from Rg = 0 we have Vi = Vg as a comparison with
Fig. 2.57b and Fig. 2.90b shows. This approach was selected in order to make the characteristics
of normal operation independent of Rg .
2.4 Basic Circuits 149
V
V
5
Vo 4
1
– 2.0 – 1.5 – 1.0 – 0.5
Vi
VB
Vi V
VBE –1
It is assumed in (2.132) that the voltage drop across RBV can be neglected if B is sufficiently
large and RBV is sufficiently small.
A point in the centre of the declining region of the transfer characteristic is selected for
the operating point; this allows maximum output amplitude. When setting B = β = 400
and IS = 7 fA30 the operating point as shown in Fig. 2.108 under the assumptions Vb = 5 V
and RC = RBV = 1 k will be:
Vb − Vo IC
Vo = 2.5 V ⇒ IC = = 2.5 mA ⇒ IB = = 6.25 mA
RC B
IC
⇒ VBE = VT ln = 692 mV ⇒ Vi = − VBE − IB RBV = − 698 mV
IS
In this case the voltage drop across RBV is only 6.25 mV and may be neglected; that is,
the voltage at the base of the transistor is VB ≈ 0.
Saturation mode: For Vi < −0.72 V the transistor enters the saturation region, which
means that the collector diode is forward-biased. In this region VCE = VCE,sat and Vo =
Vi + VCE,sat and the base current that flows must be limited to acceptable values by
resistance RBV :
Vi + VBE Vi + 0.72 V
IB = − ≈ −
RBV RBV
Transfer characteristic when driven by a current source: It is also possible to drive
the circuit by a current source Ii (see Fig. 2.109); for Vb = 5 V and RC = RBV =
1 k the circuit operates as a current–voltage converter or a transimpedance amplifier for
V
Vb V
RC 5
Vo 4
Ii Vi Vo
3
VB RBV
1
–8 –7 –6 –5 –4 –3 –2 –1 1
VB Ii
VBE –1 mA
Vi
Fig. 2.109. Circuit and characteristics of the common-base circuit when driven by a current source
−5.5 mA ≤ Ii ≤ 0:31
B
Vo = Vb − IC RC = Vb + IE RC ≈ Vb + Ii RC (2.133)
1+B
Ii
Vi = − VBE − IB RBV ≈ − VBE ≈ − VT ln − (2.134)
IS
Here, Ii = IE ≈ −IC is used. In this region the transistor operates in normal mode and
the transfer characteristic is almost linear. For Ii > 0 the transistor is in the reverse region
and for Ii < −5.5 mA it enters the saturation region.
In most practical applications a common-emitter circuit with open collector or a current
mirror is used as input current source; the section on setting the operating point outlines
this in more detail.
31 The name transimpedance amplifier is also used for operational amplifiers with current feedback
and voltage output (CV-OPV).
2.4 Basic Circuits 151
rCE
Rg ii io
β 1
A = = + (RC || rCE )
vi
io =0 rBE + RBV rCE
rCE RC
β rCE rBE +RBV rBE RBV
βRC
≈ ≈ g m RC
rBE + RBV
The maximum gain is achieved with RBV = 0; this requires the base of the transistor to
be connected either directly or via a capacitor to earth. The following section on setting
the operating point describes this in more detail. When operating with a load resistance
RL , it is possible to calculate the corresponding operating gain AB by replacing RC by
RL and RC connected in parallel (see Fig. 2.110). For gm = IC,A /VT = 96 mS, β = 400,
rBE = 4160 , rCE = VA /IC,A = 40 k, and RBV = 1 k, we obtain the exact value
and the first approximation A = 76; the result A = 96 of the second approximation is
very inaccurate because the condition rBE RBV is only insufficiently met.
For the small-signal input resistance we obtain:
vi
RC + rCE
ri =
= (rBE + RBV ) ||
ii io =0 β rCE
1+
rBE + RBV
β1
rCE RC
β rCE rBE +RBV 1 RBV rBE RBV 1
≈ + ≈
gm β gm
It depends on the load resistance. Here, ri is the open-circuit input resistance because of
io = 0 (RL → ∞). The input resistance for other values of RL is calculated by replacing
RC with RC and RL connected in parallel; the short-circuit input resistance is obtained by
setting RL = RC = 0. However, the influence of RL is so insignificant that it is eliminated
by the approximation. For the sample operating point we obtain exactly ri = 13.2 ; the
approximation is ri = 12.9 .
The small-signal output resistance is:
vo Rg β rCE + rBE + RBV
ro = = RC || rCE 1 +
io rCE rBE + RBV + Rg
152 2 Bipolar Transistor
βRC
A =
≈ ≈ g m RC (2.135)
vi io =0 rBE + RBV
rBE RBV
vi 1 RBV 1
ri = ≈ + ≈ (2.136)
ii gm β gm
vo
ro = ≈ RC (2.137)
io
vo
vo
vi
RT = =
ii
io =0 vi
io =0 ii
io =0
2.4 Basic Circuits 153
vo
RT = ≈ RC
ii
io =0
(2.138)
The input and output resistances are according to (2.136) and (2.137).
dVo
∂Vo
dVi
= ≈ − A · 1.7 mV/K
dT
A ∂Vi
A dT
For the numeric example we obtain (dVo /dT )|A ≈ −129 mV/K.
If a current source is used to drive the circuit, then from (2.133) it follows that:
dVo
dIC
IC,A dB B dIi,A
= − RC = − RC +
dT
A dT
A (1 + B) B dT 1 + B dT
In the numeric example, and with an input current independent of temperature, we ob-
tain from (2.23) a temperature drift of (dVo /dT )|A ≈ −31 mV/K; in this case only the
temperature sensitivity of the current gain B has an effect.
Vb Vb Vb
R1 RC RC
Co Co
RBV
Ci
Ci
Cb
RBV R2 RE RE
– Vb
a With base voltage divider b With base connected to ground
Setting the operating point by AC coupling: Figure 2.111 shows two versions of AC
coupling in which the signal source and the load are connected through coupling capacitors;
the other properties are described on page 115. In both versions the operating point is set
by the DC current feedback, which is used in the same way in the common-emitter circuit
(see Fig. 2.75 on page 117).
In the circuit shown in Fig. 2.111a, the voltage required at the base of the transistor
% &
VB,A = IC,A + IB,A RE + VBE,A ≈ IC,A RE + 0.7 V
is adjusted with R1 and R2 ; the current through the resistors is chosen to be significantly
higher than IB,A to prevent the operating point from being influenced by IB,A . The temper-
ature stability of the operating point depends to a high degree on the ratio of the resistances
RC and RE :
dVo
RC mV
≈ − · 1.7
dT
A R E K
RE must be selected to be as large as possible to minimize the temperature drift; ratios used
in practical applications are RC /RE ≈ 1…10. In the small-signal equivalent circuit RE is
connected in parallel to the input resistance ri , but can be neglected because RE ri =
1/gm . The parallel connection of R1 and R2 replaces resistance RBV of Fig. 2.107a:32
RBV = R1 || R2
The maximum gain is achieved only with a low-resistance base circuit; (2.135) requires
the condition RBV rBE . In practice, it is seldom possible to make R1 and R2 small
enough to meet this requirement, since otherwise the current through R1 and R2 is too
high.
32 In Fig. 2.107a the base connection of the transistor is connected to earth via resistance R
BV ;
RBV may be regarded as the internal resistance of a voltage source with V = 0. The equivalent
voltage source for the base voltage divider in Fig. 2.111a features the internal resistance R1 R2
and the open-circuit voltage V = Vb R2 /(R1 + R2 ).
2.4 Basic Circuits 155
Vb Vb Vb Vb
RC RC R1
RBV
T1 T2 T2
Cb
T1
RE R2 RBV
– Vb
Example: If IC,A = 1 mA and β = 400, then RBV rBE = 10.4 k; if we select
R1 = 3 k and R2 = 1.5 k – that is, RBV = 1 k – then for Vb = 5 V we obtain a current
that is higher than IC,A : IQ = Vb /(R1 + R2 ) ≈ 1.1 mA. However, the requirement that
the current must be significantly higher than the base current is already met for IQ = 25 mA
because IB,A = IC,A /β = 2.5 mA.
Therefore, the current must only be significantly higher than the base current and the
requirement for a low-resistance base circuit is met for AC voltages only by putting a
capacitor Cb between the base connection and ground (see Fig. 2.111a);33 Cb must be
selected so that 1/(2πfL Cb ) rBE is still the case with the smallest signal frequency of
interest fL .
If there is an additional negative supply voltage, the base connection of the transistor
can be connected directly to earth (see Fig. 2.111b), and the operating point can be adjusted
with RE :
Vb − VBE,A Vb − 0.7 V
IC,A ≈ − IE,A = ≈
RE RE
In both versions resistance RE can be replaced by a current source with current I0 ;
then IC,A ≈ I0 . In this case the temperature drift is determined by the temperature drift of
the current source.
Setting the operating point by DC coupling: Figure 2.112 shows two versions of DC
coupling. In Fig. 2.112a the common-base circuit (T2 ) is driven by a common-collector
circuit (T1 ); this is in fact a voltage control because the common-collector circuit has a
low output resistance. The operating point current IC,A is the same in both transistors and,
as shown, is adjusted by resistance RE or a current source. The circuit can be regarded as
an asymmetrically operated differential amplifier as can be seen from a comparison with
Fig. 4.54c on page 330.
Figure 2.112b shows a cascode circuit in which one transistor (T2 ) in common-base
configuration is driven by a common-emitter circuit (T1 ); this is a current control situation.
The operating point of the common-base circuit is determined by resistors R1 and R2 and
by the operating point of the common-emitter circuit. In Fig. 2.112b the common-emitter
circuit is shown symbolically only, since the circuitry required to set the operating point
is missing. The cascode circuit is described in more detail in Sect. 4.1.2.
Preventing high-frequency oscillations: Due to the high upper cutoff frequency high-
frequency oscillations may occur at the operating point; the circuit then operates as an
oscillator. This phenomenon occurs especially in circuits where the base of the transistor
is connected to ground either directly or via a capacitor Cb . The reason for this is a
parasitic inductance in the base circuit that is caused by transfer time effects in the base
region of the transistor and by conductor inductancees. Together with the input capacitance
of the transistor and/or the capacitor Cb this parasitic inductance forms a series resonant
circuit which, given sufficiently high quality, may run the risk of self-excitation. In order
to prevent this the quality of the resonant circuit must be reduced by adding a dumping
resistor. This is the purpose of RBV , which is drawn using dotted lines in Figs. 2.111
and 2.112. The resistors used in practical applications are in the range of 10…100 or
even higher in exception cases. They should be connected with short leads to ground to
keep the inductance low.
Voltage source as input signal: Figure 2.113 shows the dynamic small-signal equivalent
diagram for the common-base circuit driven by a signal voltage source with source resis-
tance Rg . The accurate calculation of the operating gain AB (s) = v o (s)/v g (s) is difficult
and leads to complicated expressions. A sufficiently accurate approximation is reached by
neglecting resistance rCE and assuming β 1; for RBV
= RBV + RB , RC
= RC ||RL
and the low-frequency gain
βRC
A0 = AB (0) ≈
(2.139)
βRg + RBV + rBE
rCE ∞ RC
Rg
vi vo
vg RB RC RL
RBV
RBV
it follows that:
CE CC RBV
1 + sCC RBV + s2
gm
AB (s) ≈ A0
1 + sc1 + s c2
2
& %
% % & & % &&
CE rBE Rg + RBV + CC RBV β Rg + RC
+ rBE + RC
βRg + rBE
c1 =
c2 =
+ rBE
1+s
+ rBE
The upper cutoff frequency depends on the low-frequency gain A0 ; from (2.139)
and (2.141) an expression with two time constants independent of A0 is derived:
1
ω-3dB (A0 ) ≈ (2.142)
T1 + T 2 A0
%
&
rBE Rg + RBV
T1 = CE
(2.143)
βRg + RBV + rBE
1
T2 = CC Rg + RBV + (2.144)
gm
Here, too, there is a close similarity to the common-emitter circuit as can be seen from
a comparison of (2.142)–(2.144) and (2.91)–(2.93). The information regarding the gain
bandwidth product GBW including (2.94) outlined on page 125 applies here as well.
If the load comprises both a resistive and a capacitive component, which means that
there is a load resistance RL and a load capacitance CL in parallel, then
1 CL
are obtained on the basis of the requirement that a calculation of AB (s) after eliminating
the s 2 term must result in (2.140):
%
& R
R ,r
rBE Rg + RBV BV g BE
C i ≈ CE %
& ≈ CE
Rg rBE + RBV
%
&
Current source as input signal: When a current source is used to drive the circuit it is
interesting to know the frequency response of the transimpedance Z T (s). On the basis of
(2.140) one can describe an approximation by a lowpass filter of first order:
v o (s) RC
gm
The upper cutoff frequency is:
1
ω-3dB = 2πf-3dB ≈ (2.147)
CE
+ CC RC
gm
Example: Using the numeric example for the common-base circuit in Fig. 2.107a IC,A =
2.5 mA was chosen. For β = 400, Cobo = 3.5 pF and fT = 160 MHz we obtain from
Fig. 2.45 on page 83 the small-signal parameters gm = 96 mS, rBE = 4160 , CC =
3.5 pF and CE = 92 pF. For RBV = RC = 1 k, RBV
≈ RBV , RL → ∞ and Rg = 0 it
follows from (2.139) that A0 ≈ 77.5 and from (2.141) f−3dB ≈ 457 kHz. The comparably
low upper cutoff frequency is caused by the resistance RBV . A much higher upper cutoff
frequency can be achieved by making RBV smaller or by removing it, as long as this does
A 0 = gmRC RC
1 r A 0 vi
vi CE = BE CC vo
gm
A 0 = – gm RC RC
vi CE + (1+ A0 ) CC rBE A 0 vi CC vo
Fig. 2.114. Equivalent circuit of the common-base circuit (top) and the common-emitter circuit
(bottom)
follows from (2.139) that A0 ≈ 49 and from (2.141) f−3dB ≈ 25.9 MHz. From (2.143) it
follows that T1 ≈ 0.94 ns, from (2.144) T2 ≈ 107 ps and from (2.94) GBW ≈ 1.5 GHz.
These values are heavily dependent on RB ; if RB = 100 then A0 ≈ 48, f−3dB ≈
6.2 MHz, T1 ≈ 5.1 ns, T2 ≈ 421 ps and GBW = 378 MHz. For a load capacitance CL =
1 nF and RB = 10 we obtain from (2.145) T2 ≈ 20.5 ns, from (2.142) f−3dB ≈ 158 kHz
and from (2.94) GBW ≈ 7.74 MHz.
If the circuit is driven by a current source and RL → ∞, then from (2.146) we
obtain RT = Z T (0) ≈ RC = 1 k and from (2.147) f−3dB = 35.7 MHz. In this case,
resistance RBV has no effect. For a load capacitance CL = 1 nF we obtain from (2.147)
f−3dB ≈ 159 kHz when CC is replaced by CC + CL .
2.4.4
Darlington Circuit
In some applications the current gain of a single transistor is not sufficient; in such cases
a Darlington circuit can be used. This consists of two transistors and provides a current
gain that is approximately equal to the product of the current gains of the two individual
transistors:
B ≈ B1 B2 (2.148)
Under the name Darlington transistor the Darlington circuit is available as a component in
its own case for installation on a circuit board; the connections are called the base, emitter
and collector as for a discrete transistor. Of course, the Darlington circuit can be made up
of discrete components. Here, the Darlington transistor is an integrated circuit that consists
solely of one Darlington circuit.
Figure 2.115 shows the circuit and the graphic symbol for an npn Darlington transistor
that consists of two npn transistors and a resistor to improve the switching performance.
In general, it can be used like an npn transistor. The pnp Darlington transistor, which can
be used as a pnp transistor, comes in two versions (see Fig. 2.116):
160 2 Bipolar Transistor
C
B T1
T2 = B
Fig. 2.115. Circuit and graphic symbol of the npn Darlington transistor
– The normal pnp Darlington consists of two pnp transistors and is directly complementary
to the npn Darlington. It is commonly known as the pnp Darlington; that is, without the
word normal.
– The complementary pnp Darlington consists of a pnp and an npn transistor and is in-
directly complementary to the npn Darlington as the pnp transistor T1 determines the
polarity; the npn transistor T2 is only used for additional current amplification.
The current gain of a pnp Darlington is often much lower than that of a comparable
npn Darlington because the current gain of the pnp transistor is usually less than that
of an npn transistor, which is squared by the Darlington due to the multiplication of the
two individual gains. The solution can be the complementary pnp Darlington in which the
second pnp transistor is replaced by an npn transistor; thus only one pnp transistor provides
the lower current gain.
The following sections describe the npn Darlington, which is used more widely in
practice. However, the explanations also apply to the pnp Darlington when the signs of all
the currents and voltages are reversed. An exception is the complementary pnp Darlington,
which is described separately.
E E
B
T1
R
T2 T2
R
B T1
C C
a Normal b Complementary
IC V CE,sat 200
A
1.0 160
IB
0.8 120 µA
0.6
80
0.4
0.2
40
0 1 2 3 4 5 6 7 8 9 10 VCE
V
IC
mA 20
10.0
IB
7.5 19
µA
5.0
18
2.5 17
16
IB I B1 I C1 IC
B
VBE1 B 1 I B1
I B2 I C2 VCE
VBE
IR
VBE2 B 2 I B2
R
Description by Equations
Figure 2.119 shows the equivalent circuit of an npn Darlington transistor in normal mode
that combines the equivalent circuits of both transistors with the additional resistance R.
The currents are
IC = IC1 + IC2
IC1 = B1 IB1 = B1 IB (2.149)
IC2 = B2 IB2 = B2 (IC1 + IB − IR )
Here, IS1 and IS2 are the saturation reverse currents of T1 and T2 ; in most cases their ratio
is IS2 ≈ 2…3IS1 . Medium collector currents produce VBE ≈ 1.2…1.5 V.
1000
~ IC –3
~ IC
B0,1 =100
In this region the current gain is approximately proportional to the collector current.
This property is caused by resistance R as in this region a predominant portion of the
collector current IC1 flows through resistance R and only a small portion is available
as base current for T2 . However, an increase in IC1 causes a corresponding increase in
IB2 , because the current through the resistor R remains approximately constant due to
IR ≈ IR,max .
– For IC > B2 IR,max it follows from (2.150) that:
B ≈ B1 B2 ≈ B0,1 B0,2
This corresponds to (2.148), which was mentioned above. This region is the preferred
operating range for the Darlington transistor.
– With a further increase in the collector current, both transistors – first T2 and then T1 –
enter the high-current region. For
B0,1 B0,2
B1 = , B2 =
IC1 IC2
1+ 1+
IK,N1 IK,N2
it follows that:
B0,1 B0,2
B(IC ) = 2
IC IC IC
1+ + 1+
IK,N2 IK,N1 B0,2 IK,N2
Here, IK,N1 and IK,N2 are the knee-point currents for strong injection of T1 and T2 ; in
most cases the ratio is IK,N2 ≈ 2…3IK,N1 . In the high-current region the current gain
drops rapidly; this becomes obvious when a limit value condition is examined [2.8]:
2 2
B0,1 IK,N1 B0,2 IK,N2
lim B(IC ) =
IC →∞ IC3
In the Darlington the current gain declines at a rate of 1/IC3 for large currents, while for
the discrete transistor, it declines at a rate of 1/IC .
Small-Signal Response
In order to determine the small-signal response of the Darlington transistor at an operating
point A it is also necessary to know the operating point currents IB,A and IC,A , and the
internal currents IC1,A and IC2,A , which means that the distribution of the collector current
must be known; this gives us the small-signal parameters of the two transistors:
IC1/2,A β1/2 VA1/2
gm1/2 = , rBE 1/2 = , rCE1/2 =
VT gm1/2 IC1/2,A
The Early voltages are usually about the same so that an Early voltage VA ≈ VA1 ≈ VA2
can be used for calculations. An operating point in the region of large current gain is
selected; here IC2,A IC1,A , and the approximation IC2,A ≈ IC,A can be used, which
means that the collector current of the Darlington flows almost completely through T2 .
The upper part of Fig. 2.121 shows the complete small-signal equivalent circuit of a
Darlington transistor; it describes the npn and pnp Darlingtons but not the complementary
pnp Darlington. However, this comprehensive equivalent diagram is rarely used since, due
to its similarity with a discrete transistor, the Darlington can be described with sufficient
2.4 Basic Circuits 165
iB iC
R vBE1 rBE2
iB iC
Fig. 2.121. Small-signal equivalent circuit of a Darlington transistor: complete circuit (above) and
simplified circuit (below)
accuracy by the equivalent circuit of a discrete transistor (see Fig. 2.121); the parameters
gm , rBE and rCE can be determined either from the characteristics or by conversion us-
ing the complete equivalent circuit.35 For β1 , β2 1 the conversion of the parameters
provides
1 + gm2 (rBE 2 || R) RrBE 2
gm2
gm ≈ gm1 ≈
1 + gm1 (rBE 2 || R) 2
RrBE 2
rBE ≈ rBE 1 + β1 (rBE 2 || R) ≈ 2 rBE 1
1 + gm1 (rBE 2 || R) RrBE 2
2
rCE ≈ rCE2 || rCE1 ≈ rCE2
1 + gm2 (rBE 2 || R) 3
The small-signal current gain is:
RrBE 2
R
β = gm rBE ≈ β1 β2 ≈ β1 β2 (2.151)
rBE 2 + R
The requirement R rBE2 is fulfilled when the current through resistance R can be
neglected because of IB2 IR ; consequently:
IC,A
IC2,A ≈ IC,A , IC1,A ≈
B2
For this purpose the Darlington must be operated in the region of maximum power gain B;
that is, the condition IC,A B2 IR,max must be met (see Fig. 2.120). The following
equations apply to the region of maximum current gain of the Darlington transistor:
Darlington transistor
gm2 1 IC,A
gm ≈ ≈ (2.152)
2 2 VT
β β1 β2 VT
rBE = ≈ 2 (2.153)
gm IC,A
2 2 VA
rCE ≈ rCE2 ≈ (2.154)
3 3 IC,A
Switching Performance
The Darlington transistor is very often used as a switch; due to the high current gain, it
is possible to switch high-load currents with comparably low control currents. Switching
off the load is particularly critical: transistor T1 blocks relatively fast, while transistor
T2 remains conductive until the charge stored in the base is drained through resistance
R. Thus a short turn-off time is achieved only with a sufficiently low resistance R (see
Fig. 2.122). On the other hand, a small resistance R lowers the current gain. A compromise
must therefore be found; Darlingtons used in switching applications use smaller resistances
than Darlingtons used for general purposes.
In addition to the two transistors and resistance R, Darlington transistors used in
switching applications are provided with three additional diodes; Fig. 2.123 shows the
complete circuit diagram of such an npn Darlington. In order to shorten the turn-off time
the base current can be inverted; diodes D1 and D2 then limit the reverse voltage at the
base–emitter junctions. Diode D3 acts as a free-wheeling diode for inductive loads.
2.4 Basic Circuits 167
Vb Vg
10 V V 5
RL
10 Ω 0
Rg IC IC
1 kΩ A 1.0
R
0.5
Vg
0
1 2 3 4 5 t
µs
B T1
D1
T2 D3
D2 R
The field effect transistor (FET ) is a semiconductor component with three terminals, known
as the gate (G), source (S) and drain (D). There are discrete transistors that are used for
mounting on printed circuit boards, and are contained in their own housings, and integrated
field effect transistors that are produced together with other semiconductor elements on a
common substrate. Integrated field effect transistors feature a fourth terminal called the
substrate or bulk (B), which results from the common substrate.1 This terminal also exists
internally in discrete transistors, where it is not connected to the outside but to the source
terminal.
Mode of operation: In the field effect transistor, a control voltage between the gate and
the source is used to control the conductivity of the drain–source junction without the flow
of a control current; that is, they are controlled in a Watt-less fashion. Two different effects
are utilized:
– In the MOSFET (metal oxide semiconductor FET or insulated gate FET, IGFET ) the
gate is isolated from the channel by an oxide layer (SiO2 ) (see Fig. 3.1); this means
that the control voltage can have either polarity without a current flowing. The control
voltage influences the charge carrier density in the inversion layer beneath the gate,
which forms a conductive channel between the drain and the source, thus enabling
current flow. Without the inversion layer, at least one of the pn junctions between the
source and the substrate or the drain and the substrate is always reverse-biased and
no current can flow. Depending on the doping of the channel, there are “normally on”
(depletion) or “normally off ” (enhancement) MOSFETs; for VGS = 0 a drain current
flows through depletion MOSFETs but not enhancement MOSFETs. In addition to the
gate, the substrate B also has a slight controlling effect; this is described in more detail
in Sect. 3.3
VGS1 > Vth VDS > 0 VGS2 > VGS1 VDS > 0
Poly-Si Poly-Si
G G
+ + + + + SiO2 +++++++++++ SiO 2
S D Si S D
– – – – – –––––––––––
n n n n
ID ID
p p
B B
1 In a bipolar transistor, this terminal is called the substrate (S); since in a field effect transistor S
stands for source, bulk (B) is used for the substrate.
170 3 Field Effect Transistor
Vth < V GS1 < 0 VDS > 0 V th < VGS2 < VGS1 VDS > 0
p p
S G D S G D
n n
ID ID
p junction p junction
– In the junction FET (JFET or noninsulated gate FET, NIGFET ) the control voltage
influences the junction width of a pn junction operated in reverse mode. This influences
the cross-sectional area and thus the conductivity of the channel between the drain and
the source (see Fig. 3.2). As the gate is not isolated from the channel, it is possible to
use the pn junction in forward mode; but since this mode eliminates the advantage of
Watt-less control, it is not used in practice. In the MESFET (metal semiconductor FET )
a metal semiconductor junction (a Schottky junction) is used instead of a pn junction;
the functional principle is the same as for normal junction FETs. JFETs and MESFETs
are depletion FETs; that is, a drain current flows at a control voltage of VGS = 0.
It follows from Figs. 3.1 and 3.2 that MOSFETs and junction FETs are generally
symmetrical, which means that drain and source may be interchanged. But most discrete
FETs are not exactly symmetrical in their design, and in discrete MOSFETs the internal
connection between substrate and source makes the terminals noninterchangeable.
Both MOSFETs and junction FETs are available in n-channel and p-channel designs,
which means that there are a total of six types of field effect transistors; Fig. 3.3 shows
the graphic symbols together with a simplified plot of their characteristics. The polarities
stated in Fig. 3.4 are used in normal operation for the voltages VGS and VDS , the drain
current ID and the threshold voltage Vth .2
3.1
Behavior of a Field Effect Transistor
The behavior of a field effect transistor is most easily explained by means of its character-
istic curves. They describe the relations between the currents and voltages in the transistor
in the event of all values being static; in other words, not or only very slowly changing over
time. Field effect transistor calculations require simple equations that give a sufficiently
accurate description of the behavior of the field effect transistor. In order to check the
functionality of a circuit by computer simulation, however, the influence of secondary ef-
fects must be taken into consideration. For this purpose, sophisticated models are available
that also reflect the dynamic behavior when controlled with sinusoidal or pulsed signals.
These models are described in Sect. 3.3, but are not needed for a basic understanding. The
2 The threshold voltage V is usually used in relation to MOSFETs; in the case of junction FETs,
th
the pinch-off voltage VP replaces Vth . For the sake of a uniform designation, Vth is used for all
FETs.
3.1 Behavior of a Field Effect Transistor 171
n-MOSFET D ID ID
enhancement
ID
B
G
S Vth VGS VDS
p-MOSFET D ID ID
enhancement Vth V GS V DS
ID
B
G
ID ID
n-MOSFET D
depletion
ID
B
G
S V th V GS V DS
p-MOSFET D ID ID
depletion
ID Vth V GS V DS
B
G
n-JFET D ID ID
ID
p-JFET D
ID ID
ID Vth VGS VDS
text below primarily describes the behavior of an n-channel enhancement MOSFET; for
p-channel FETs, the polarity of all voltages and currents is reversed.
3.1.1
Characteristic Curves
Family of output characteristics: When one applies different gate–source voltages VGS
to an n-channel FET and measures the drain current ID as a function of the drain–source
voltage VDS , one arrives at the family of output characteristics illustrated in Fig. 3.5.
These characteristics are basically identical for all n-channel FETs, with the exception of
the gate–source voltages VGS for the individual characteristics, which are different for the
three n-channel types. A drain current flows only if VGS is higher than the threshold voltage
Vth . Two regions have to be distinguished:
– When VDS < VDS,po = VGS − Vth , the FET operates in the ohmic region (triode region);
this designation was chosen because for VDS = 0 the characteristics run almost linearly
through the origin, so that their behavior is identical to that of an ohmic resistance. When
approaching the limiting voltage VDS,po , the slope of the characteristics declines until
they run almost horizontally when VDS = VDS,po .
– If VDS ≥ VDS,po , then the characteristics are almost horizontal; this region is known as
the pinch-off region (saturation region).3
If VGS < Vth , then no current flows and the FET operates in the cutoff region.
Pinch-off region: The pinch-off region of the MOSFET is due to the decline of the charge
carrier concentration in the channel, which causes the channel to be pinched off ; with an
increasing voltage VDS this happens first on the drain side, since the voltage between the
gate and the channel is the lowest:
3 The term saturation region is not a very fortunate choice, since saturation has an entirely different
meaning in respect to bipolar transistors. The term pinch-off region is more neutral and is to be
given priority over the term saturation region, which is sometimes found in the literature.
3.1 Behavior of a Field Effect Transistor 173
ID
mA Ohmic region Pinch-off-region
4.0
10
V GS
8 VDS,po = VGS – Vth V
6
3.5
3.0
2
2.5
0 1 2 3 4 5 VDS
V
The pinch-off effect occurs exactly at the point at which VGD becomes smaller than Vth ;
the borderline between the ohmic region and the pinch-off region can thus be described
by:
A drain current still flows through the channel, as the charge carriers can cross the pinched-
off region, but an increase in the voltage VDS only has a slight effect on the nonpinched
portion of the channel; thus the drain current remains almost constant. The slight effect
of VDS in the pinched-off region is known as channel-length modulation and results in a
slight increase in the drain current with rising voltage VDS . In the cutoff region the channel
is also pinched off at the source side because VGS < Vth ; in this case, the flow of current is
no longer possible. Figure 3.6 shows the distribution of the charge carriers in the channel
for the three regions.
G G G
S D S +++++++++++++ D S +++++++ + D
–––––––––––––––––––––––– ––––––––––––––––––
n n n n n n
p p p
Cut-off-region Ohmic region Pinch-off region
In the junction FET, the pinch-off effect is due to the fact that the junctions are in contact
with one another and close off the channel; with an increasing voltage VDS this happens
first on the drain side, because this is where the voltage is highest across the junction. At the
border between the ohmic and the pinch-off region we have VDS,po = VGS − Vth , as with
the MOSFET. Here too, the drain current continues, since the charge carriers can cross the
pinched-off region. But a further increase of VDS has only a small effect. Figure 3.7 shows
the expansion of the junctions in the three regions.
Family of transfer characteristics: In the pinch-off region the drain current ID essen-
tially depends only on VGS . Plotting ID for various values of VDS as a function of VGS
that belong to the pinch-off region produces the family of transfer characteristics shown in
Fig. 3.8. Besides the characteristic of the enhancement MOSFET, the diagram also shows
those of the depletion MOSFET and the junction FET; apart from a shift along the VGS -
axis they have an identical shape. Due to their small dependence on VDS , the individual
characteristics are very close together. No current flows when VGS < Vth , as in this case
the channel is pinched off over its entire length.
ID
VDS VDS VDS
mA
6 Enhancement MOSFET
2
Vth Vth Vth
–2 –1 0 1 2 3 4 VGS
V
IG
– 40 V
40 V VGS
IG
– 0.6 V
30 V VGS
IG
– 50 V
0.6 V VGS
3.1.2
Description by Equations
On the basis of an ideal charge distribution in the channel, we can calculate the drain
current ID (VGS , VDS ); the equations for the junction FET and the MOSFET are different,
but an approximation by way of a simple equation causes no major errors [3.1]:
⎧
⎪
⎪
0 for VGS < Vth
⎪
⎪
⎨ VDS
ID = K VDS VGS − Vth − for VGS ≥ Vth , 0 ≤ VDS < VGS − Vth
⎪ 2
⎪
⎪ K
⎪
⎩ (VGS − Vth )2 for VGS ≥ Vth , VDS ≥ VGS − Vth
2
176 3 Field Effect Transistor
ID ID
The first equation describes the cutoff region, while the second is for the ohmic and the
third for the pinch-off regions. The transconductance coefficient K is a measure of the
slope of the transfer characteristic and is described in more detail below.
Curve description: The equation for the ohmic region is quadratic in VDS and therefore
describes a parabola in the family of output characteristics (see Fig. 3.10a). The peak of
the parabola is at VDS,po = VGS − Vth , that is, at the borderline of the pinch-off region,
and the validity of this equation ends here, since it only applies for 0 ≤ VDS < VDS,po .
For VDS ≥ VDS,po we have to use the equation for the pinch-off region, which does not
depend on VDS and thus describes parallels to the VDS -axis; Fig. 3.10a shows the related
characteristic as a dot–dash line.
The equation for the pinch-off region is quadratic in VGS and therefore describes a
parabola in the family of transfer characteristics (see Fig. 3.10b). The peak of the parabola
is at VGS = Vth ; the validity of the equation starts here and in n-channel FETs it applies to
VGS > Vth only.
All equations only apply in the first quadrant of the output characteristics; that is, for
VDS ≥ 0.4 In a symmetrical FET the characteristics in the third quadrant are symmetric to
those in the first quadrant; this is particularly true for integrated FETs. The equations can
also be applied to the third quadrant if the drain and source are interchanged; that is, using
VGD and VSD instead of VGS and VDS , respectively.5 Discrete MOSFETs, especially power
MOSFETs, are designed asymmetrically and display in the third quadrant a behavior that
is different from that in the first quadrant (see Sect. 3.2).
In order to simplify the description, we will use abbreviations for the operating regions
of the n-channel FET in the text below:
⎧ ⎫
CR : cutoff region ⎪ ⎪
V < Vth
⎨ GS ⎬
OR : ohmic region ⇒ VGS ≥ Vth , 0 ≤ VDS < VGS − Vth (3.1)
⎪
⎭ ⎪
⎩
PR : pinch-off region VGS ≥ Vth , VDS ≥ VGS − Vth
Also, taking the influence of channel-length modulation [3.2] into account and comple-
menting the equation for the gate current leads to the large-signal equations for a field
effect transistor:
⎧
⎪
⎪ 0 CR
⎪
⎪
⎪
⎨ VDS VDS
K VDS VGS − Vth − 1+ OR
ID = 2 VA
⎪
⎪ (3.2)
⎪
⎪ K VDS
⎪
⎩ (VGS − Vth )2 1 + PR
2 VA (3.3)
⎧
⎪ 0 MOSFET
⎪
⎨ ⎛ ⎞
VGS
IG =
⎪
⎪ ⎝ VT − 1⎠ junction FET
⎩ IG,S e
(3.4)
W
G
Poly-Si
S dox SiO2 D
L Si
Fig. 3.11. Geometric dimensions of a
MOSFET
6 The mobility depends on the doping of the channel and is significantly lower than the mobility
in undoped silicon (µn ≈ 0.14 m2 /Vs).
178 3 Field Effect Transistor
A WL
≈ 20 . . . 60
V2
The transconductance coefficient K is obtained from (3.5) by multiplying by the factor
W/L, which is a measure of the size of the MOSFET. Typical values for discrete transistors
are L ≈ 1 . . . 5 mm and from W ≈ 10 mm in small-signal MOSFETs up to W > 1 m 8 in
power MOSFETs; thus K ranges from approximately 40 mA/V2 up to 50 A/V2 .
In p-channel MOSFETs, the mobility of the charge carriers in the channel is µp ≈
0.015 . . . 0.03 m2 /Vs and is thus lower by about a factor of 2–3 than in n-channel MOS-
FETs; this leads to Kp
≈ 6 . . . 20 µA/V2 .
In junction FETs, K also depends on the geometric dimensions.9 An accurate illus-
tration is not provided here; refer to reference [3.1]. For small-signal applications with
K ≈ 0.5 . . . 10 mA/V2 , junction FETs are almost always discrete transistors.
Channel-length modulation: The dependence of the drain current on VDS in the pinch-
off region is caused by channel-length modulation and is empirically described by the
term on the right-hand side of (3.3). In order to achieve a continuous changeover from the
ohmic to the pinch-off region, this term must also be complemented in (3.2) [3.2]. This
description is based on the observation that the extrapolated curves of the family of output
characteristics intersect approximately at one point; Fig. 3.12 illustrates this relation. With
7 K
is inversely proportional to d , so that continuing miniaturization causes ever-increasing
n ox
values; for example, Kn
≈ 100 . . . 120 µA/V2 in 3.3 V CMOS circuits.
8 Section 3.2 describes how to arrive at these high values for W .
9 The transconductance coefficient of a junction FET is usually called β in the literature; here we
use K to maintain uniform identification and to avoid confusion with the current gain β of a
bipolar transistor.
3.1 Behavior of a Field Effect Transistor 179
ID
VDS
– VA = – 1
λ
reference to the bipolar transistor, the constant VA is called the Early voltage: for MOSFETs
it lies at VA ≈ 20 . . . 100 V and for junction FETs it is VA ≈ 30 . . . 200 V. The channel-
length modulation parameter is often used instead of the Early voltage:
1
λ = (3.6)
VA
3.1.3
Field Effect Transistor as an Adjustable Resistor
A field effect transistor operated in the ohmic region can be used as an adjustable resistor
(see Fig. 3.13a); in this case the control voltage Vctl = VGS changes the resistance of the
drain source junction. Differentiation of (3.2) leads to:
VDS VDS
R2
ID
R ( VGS ) R1
VGS
Vctl VGS
1 ∂ID
2VDS 2
K VDS
= = K (V − V − V ) 1 + +
∂VDS
OR
GS th DS
R(VGS ) VA 2VA
However, due to the dependence on VDS the resistance is not linear. Of particular interest
is the on resistance RDS,on for signals around the point VDS = 0:
∂VDS
1
RDS,on = = (3.7)
∂ID
VDS =0 K (VGS − Vth )
As the characteristics around VDS = 0 are almost linear, the turn-on resistance RDS,on
is independent of VDS , and the FET acts as an adjustable linear resistor when modulated
with small amplitudes.
The linearity can be improved by not feeding the control voltage directly to the gate but
by adding half of the drain–source voltage to it in advance; this can be achieved by using
the circuit shown in Fig. 3.13b, with a voltage divider consisting of two high-resistance
components R1 = R2 in the M range that generates
VDS R1 + Vctl R2 R1 =R2 VDS + Vctl
VGS = =
R1 + R 2 2
Inserting this expression into (3.2) leads to
Vctl VDS
ID = K VDS − Vth 1+
2 VA
Consequently:
VDS VA
1 Vctl 2VDS Vctl
= K − Vth 1+ ≈ K − Vth
R(Vctl ) 2 VA 2
A dependence on VDS remains but is, however, much smaller than that of the simple circuit
shown in Fig. 3.13a, a fact that is illustrated by the curves in Fig. 3.14. The nonlinearity
can be further reduced by finely tuning the voltage divider. Optimum values
R1 VA − 2Vctl + 2Vth
=
R2 VA − 2Vth
R
400
Simple
300
200
Linearised
100
Fig. 3.14. Comparison of the resistance curves for K = 5 mA/V2 , Vth = 2 V, VA = 100 V and
VGS = 4 V and Vctl = 8 V
3.1 Behavior of a Field Effect Transistor 181
3.1.4
Operating Point and Small-Signal Behavior
One field of field effect transistor application is the linear amplification of signals in small-
signal circuits. Here, the field effect transistor is modulated with small signals around an
operating point. In this case, the characteristics can be replaced by their tangents to the
operating point.
Operating Point
The operating point A is characterized by the voltages VDS,A and VGS,A and the current
ID,A , and is determined by external circuitry. For the appropriate use as an amplifier, the
operating point must be in the pinch-off region. For the six FET types, Fig. 3.15 shows the
n-Channel p-Channel
Enhancement
MOSFET 3 mA 5V
3.1 V
5V
3.1 V
3 mA
Depletion
MOSFET 3 mA 5V
0.1 V
5V
0.1 V
3 mA
Junction
MOSFET 3 mA
0.9 V 5V
0.9 V
5V 3 mA
Fig. 3.15. Operating point settings for ID,A = 3 mA in n-channel and p-channel FETs with
K = 5 mA/V2
182 3 Field Effect Transistor
setting of the operating point and the polarity of voltages and currents; for n-channel FETs,
according to the family of characteristics shown in Fig. 3.8 on page 174, a threshold voltage
Vth = − 2 / − 1 / 2 V and a transconductance coefficient K = 5 mA/V2 are assumed. The
chosen example of a current ID,A = 3 mA is achieved with VGS,A = Vth + 1.1 V: 10
K mA
ID ≈ (VGS − Vth )2 = 2,5 2 · 1.1 V2 ≈ 3 mA
2 V
In p-channel FETs the threshold voltage Vth has the opposite sign, producing ID = − 3 mA
at VGS,A = Vth − 1.1 V. Procedures for adjusting the operating point are described in
Sect. 3.4.
Linearization: The characteristics are replaced by the tangents to the operating point;
that is, they are linearized. For this purpose, a Taylor series expansion is performed at the
operating point and is interrupted after the linear term:
iD = ID (VGS,A + vGS , VDS,A + vDS ) − ID,A
∂ID
∂ID
= vGS + vDS + . . .
∂VGS
A ∂VDS
A
Small-signal equations: The partial derivatives at the operating point are called the
small-signal parameters. The introduction of specific designations produces small-signal
equations for the field effect transistor:
iG = 0
(3.8)
1
iD = gm vGS + vDS
rDS (3.9)
gm
mA
V
10
–2 –1 0 1 2 3 4 VGS
V
Fig. 3.16. Transconductance of n-channel FETs, with transfer characteristics from Fig. 3.8
(K = 5 mA/V2 )
characteristics are shown in Fig. 3.8 on page 174. There are straight lines with the x-axis
segment Vth and the slope K:
∂gm ∂ 2 ID
K = = 2
∂VGS ∂VGS
By solving (3.3) for VGS − Vth and substituting in (3.10), the transconductance gm can also
be described as a function of the drain current ID,A :
VDS,A VA
∂ID
VDS,A
gm = = 2KI 1 + ≈ 2KID,A (3.11)
∂VGS
A
D,A
VA
In contrast to the bipolar transistor, where only the collector current IC,A is required to
calculate the transconductance, for the field effect transistor the drain current ID,A and the
transconductance coefficient K must be known; the dependence on VA is low. In practice,
the approximation given in (3.11) is used. Data sheets specify the transconductance for
a certain drain current instead of K. The value of K can then be determined from the
transconductance:
2
gm
K ≈
2ID,A
The small-signal output resistance rDS describes the changes of the drain–source volt-
age VDS caused by a changing drain current ID at the operating point. It can be determined
from the reciprocal value of the slope of the tangent in the family of output characteristics
according to Fig. 3.5. Differentiation of the large-signal (3.3) leads to:
VDS,A VA
∂VDS
VA + VDS,A VA
rDS = = ≈ (3.12)
∂ID
A ID,A ID,A
Small-signal parameters in the ohmic region: For the ohmic region, VDS VA ;
differentiation of (3.2) thus leads to:
184 3 Field Effect Transistor
gm,OR ≈ K VDS,A
1
rDS,OR ≈ % &
K VGS,A − Vth − VDS,A
The transconductance and the output resistance in the ohmic region are smaller than in the
pinch-off region; clearly, lower levels of amplification can therefore be achieved.
Four-Pole Matrices
Small-signal equations can also be written in matrix form:
⎡ ⎤
iG ⎢ 0 0 ⎥ vGS
= ⎣ 1 ⎦
iD gm vDS
rDS
iG vGS y 11,s y 12,s vGS
= Ys =
iD vDS y 21,s y 22,s vDS
The subscript S indicates that the FET is operated in common-source configuration; that
is, the source terminal is used for both the input and output port according to the through
connection shown in the small-signal equivalent circuit in Fig. 3.17. Section 3.4.1 describes
the common-source circuit in more detail.
iG = 0 iD
Since IG = 0, the voltage VGS only depends on the circuit connected to the gate
and the equation vGS = vGS (iG , vDS ) thus does not exist, which means that a hybrid
representation with the H matrix, as for the bipolar transistor, is not possible for the field
effect transistor.
∂ID
1 ∂ 2 ID
2 1 ∂ 3 ID
3
iD = vGS + 2
vGS + 3
vGS + . . .
∂VGS
A 2 ∂VGS 6 ∂VGS
A A
K 2
= 2KID,A vGS + vGS
2
Due to the parabolic shape of the characteristic, the series ends after the second term. For
harmonic signals with vGS = v̂GS cos ωt, this leads to:
K 2 K 2
iD = v̂GS + 2KID,A v̂GS cos ωt + v̂GS cos 2ωt
4 4
The distortion factor k is determined from the ratio of the first harmonic with 2ωt to the
fundamental with ωt:
iD,2ωt v̂GS K v̂GS
k = = = % & (3.13)
iD,ωt 4 2ID,A 4 VGS,A − Vth
It is inversely proportional to ID,A or VGS,A − Vth , so that with the same signal it declines
with an increasing drain current. In discrete transistors, VGS,A − Vth ≈ 1 . . . 2 V; with
v̂GS < 40 . . . 80 mV, this results in a distortion factor of k < 1%. From a comparison with
(2.15) on page 45, it can be noted that with the same distortion factor the possible input
signal of the FET is much higher than that of the bipolar transistor, for which k < 1% can
only be achieved with v̂BE < 1 mV.
3.1.5
Maximum Ratings and Leakage Currents
Various limit data that must not be exceeded are specified for field effect transistors. These
are divided up into limit voltages, limit currents and maximum dissipation. Once again,
n-channel MOSFETs will be examined here; for p-channel MOSFETs, the voltage and
current signs have to be reversed.
186 3 Field Effect Transistor
Breakthrough Voltages
Gate breakthrough: At the gate–source breakthrough voltage V(BR)GS , the gate oxide
of the MOSFET breaks through on the source side; while at the drain–gate breakthrough
voltage V(BR)DG , the gate oxide breaks through on the drain side. This breakthrough is
irreversible and will destroy the MOSFET if it is not protected by Zener diodes. Therefore,
discrete MOSFETs without Zener diodes must be protected against static charges, and
must not be touched until potential equalization has been carried out.
The gate–source breakthrough is symmetrical, which means that it is independent
of the polarity of the gate–source voltage; for this reason, data sheets specify a plus-
or-minus value, such as V(BR)GS = ±20 V, or the absolute value of the breakthrough
voltage. Typical values are |V(BR)GS | ≈ 10 . . . 20 V for MOSFETs in integrated circuits
and |V(BR)GS | ≈ 10 . . . 40 V for discrete transistors.
The drain region of symmetrically designed MOSFETs is the same as the source
region, which means that |V(BR)DG | = |V(BR)GS |; this is particularly true of MOSFETs
in integrated circuits. In asymmetrically designed MOSFETs, |V(BR)DG | is significantly
higher than |V(BR)GS |, as a large portion of the voltage drops over a weakly doped layer
between the channel and the drain terminal (see Sect. 3.2). On data sheets this voltage is
called V(BR)DGR or VDGR , since the measurement is taken with a resistance R between
gate and source; the size of the resistance is stated. Since this is the breakthrough of the
junction between the substrate and the weakly doped portion of the drain region, a drain–
source breakthrough occurs at the same time; therefore, the same value is usually quoted
for V(BR)DG as for the drain–source breakthrough voltage V(BR)DSS that is described in
the following paragraphs.
In case of the junction FET, V(BR)GSS is the breakthrough voltage of the gate–channel
diode; it is measured with the short-circuited drain source path, that is, with VDS = 0, and
it is negative in n-channel junction FETs and positive in p-channel junction FETs. Typical
values of V(BR)GSS for n-channel FETs range from approximately −50 V to −20 V. Also
specified are the breakthrough voltages V(BR)GSO and V(BR)GDO on the source and drain
sides, respectively; the subscript O indicates that the third terminal is open. These voltages
are normally the same: V(BR)GSS = V(BR)GSO = V(BR)GDO . Since the voltages VGS and
VDS are of opposite polarity in the junction FET, VGD = VGS − VDS is the voltage of the
highest absolute value, so that V(BR)GDO is of particular importance for practical purposes.
Contrary to the MOSFET, the junction FET is not destroyed by the breakthrough as long
as the current is limited and overheating is prevented.
ID ID
small-signal short-circuit, which means biasing the gate with a voltage source that has
a negligibly low internal resistance. The values range from V(BR)DSS ≈ 10 . . . 40 V in
integrated FETs to V(BR)DSS = 1000 V in discrete FETs for switching applications.
Junction FETs feature no direct breakthrough between drain and source because of
the homogeneity of this region. Here, with the channel pinched off and the drain–source
voltage increasing, it is the junction between the drain and the gate that breaks through when
the above-mentioned breakthrough voltage V(BR)GDO is reached. Figure 3.18b shows the
characteristics for the breakthrough of a small-signal junction FET; it occurs suddenly.
Maximum Currents
Drain current: For the drain current one distinguishes between the maximum continuous
current and the maximum peak current. Data sheets have no specific designation for the
maximum continuous current; here, it is referred to as ID,max . The maximum peak current
is called IDM 11 on data sheets and is quoted for a pulsed current with a given pulse duration
and repetition rate; it is higher than the maximum continuous current by a factor of 2 . . . 5.
For junction FETs the drain saturation current IDSS 12 is specified instead of the
maximum continuous current ID,max ; it is measured with VGS = 0 in the pinch-off region
and thus represents the maximum possible drain current in normal operation.
Backward diode: Due to the connection between the source and the substrate in discrete
MOSFETs, there is a backward diode between the source and the drain (see Sect. 3.2).
Two currents are specified for this diode: the maximum continuous current IS,max and the
maximum peak current ISM . Due to the component construction, they are of the same
magnitude as the corresponding drain currents ID,max and IDM , so that the backward diode
can be used without limitations as a freewheeling or commutating diode.
Gate current: For junction FETs the maximum gate current IG,max in forward mode is
also specified; typical values are IG,max ≈ 5 . . . 50 mA. However, this information is of
minor importance, since the channel gate diode is usually operated in reverse mode.
11 For MOSFETs in switching applications, I
D,puls is often used instead of IDM .
12 I
DSS is also known as ID,S and corresponds to ID,0 = ID (VGS = 0) described for junction FETs
in Sect. 3.1.2.
188 3 Field Effect Transistor
Leakage Currents
Drain current: A small drain–source leakage current IDSS flows in enhancement MOS-
FETs with short-circuited gate and source; it corresponds to the leakage current of the drain
substrate junction and, as such, is very temperature sensitive. Typically, IDSS < 1 mA in
integrated MOSFETs and discrete MOSFETs, while IDSS = 1 . . . 100 mA for discrete
MOSFETs with currents in the ampere range. In depletion MOSFETs, IDSS is also mea-
sured in the cutoff region; this requires a gate–source voltage of VGS < Vth .
Note that the current IDSS is also specified for junction FETs, but has an entirely
different meaning here. In MOSFETs, IDSS is the minimum drain current that also flows
in the cutoff region and occurs in switching applications as a leakage current across the
open switch; in junction FETs, IDSS is the maximum drain current in the pinch-off region.
Despite the different meanings, it is specified on data sheets using the same designation.
P V = VDS ID
This mainly takes place in the channel and leads to an increase in the channel temperature.
Due to the temperature difference, the heat can be dissipated to the environment via the
housing. However, the temperature in the channel must not exceed a material-specific limit
value of 175 ◦ C for silicon; for safety reasons, a limit value of 150 ◦ C is used for calculations
in practice. The related maximum dissipation of discrete transistors depends on the given
construction and installation method; the data sheet specifies the total dissipation Ptot for
two situations:
– Operation in an upright mounting on the circuit board without further cooling measures,
at an ambient temperature (free-air temperature) of TA = 25 ◦ C.
– Operation with a case temperature of TC = 25 ◦ C.
The two maximum values are known as P V ,25(A) and P V ,25(C) . Only Ptot = P V ,25(A) is
specified for small-signal FETs that are used for upright mounting without a heat sink, while
for power MOSFETs that are designed exclusively for installation with heat sinks only
Ptot = P V ,25(C) is specified. However, the conditions TA = 25 ◦ C or TC = 25 ◦ C cannot
be guaranteed in real applications. Since Ptot decreases as the temperature increases, data
sheets often show a power derating curve that plots Ptot versus TA or TC (see Fig. 3.19a).
Section 2.1.6 on page 49 describes the temperature response of the bipolar transistor in
detail; naturally, the results apply equally to FETs.
60 10 1 ms
40 10 ms
20 1 DC
V(BR)DSS
Ptot
ID,max =
RDS,on
For FETs to be used in switching applications, additional limiting curves for pulsed
operation with various pulse durations are shown. If the pulse duration is very short and the
duty factor is small, the FET can be operated with a maximum voltage V(BR)DSS and the
maximum drain current IDM at the same time; in this case, the SOA shows a rectangle. A
FET enables loads with a power dissipation of up to P = V(BR)DSS ID,max to be switched.
This maximum switching capacity is high compared to the maximum dissipation Ptot ; from
Fig. 3.19 it follows that P = V(BR)DSS ID,max = 100 V · 30 A = 3 kW Ptot = 100 W.
3.1.6
Thermal Behavior
The thermal behavior of components is described in Sect. 2.1.6 using the example of the
bipolar transistor; the values and relations mentioned there apply likewise to a FET if P V
is replaced by the power dissipation of the FET.
3.1.7
Temperature Sensitivity of FET Parameters
MOSFETs and junction FETs show different temperature responses and must therefore
be considered separately in this respect.
MOSFET
In case of the MOSFET, the threshold voltage Vth and the transconductance coefficient K
are temperature-dependent; the temperature coefficient of the drain current in an n-channel
MOSFET in the pinch-off region is derived by differentiation of (3.3):
190 3 Field Effect Transistor
1 dID 1 dK 2 dVth
= − (3.14)
ID dT K dT VGS − Vth dT
From (3.5) and the temperature sensitivity of the mobility related to the reference point T0
[3.1], it follows that
mµ
T0
µ(T ) = µ(T0 ) with mµ ≈ 1.5
T
which means that the transconductance coefficient declines with an increase in temperature:
T =300 K
1 dK mµ
= − ≈ − 5 · 10−3 K −1
K dT T
The threshold voltage [3.1] is
Vth = VFB + Vinv + γ Vinv
where VFB is the flat-band voltage, Vinv is the inversion voltage and γ is the substrate
control factor. The flat-band voltage depends on the design of the gate and is not important
in this context, while the other parameters are described in Sect. 3.3. VFB and γ are not
temperature-dependent; therefore
dVth γ dVinv
= 1+
dT 2 Vinv dT
Typical values are as follows: Vinv ≈ 0.55 . . . 0.8√
V, dVinv /dT ranges from approximately
−2.3 mV/K to −1.7 mV/K and γ ≈ 0.3 . . . 0.8 V. Hence:
dVth mV
≈ − 3.5 . . . − 2
dT K
As the temperature coefficients of K and Vth are negative, the temperature coefficient of the
drain current may be positive or negative because of the subtraction in (3.14) depending
on the operating point. Therefore there is a temperature compensation point TC at which
the temperature coefficient becomes zero; for n-channel MOSFETs, solving (3.14) leads
to:
dVth
dT
VGS,TC = Vth + 2 ≈ Vth + 0.8 . . . 1.4 V
1 dK
K dT
ID,TC ≈ K · 0.3 . . . 1 V2
Figure 3.20a shows the transfer characteristic of an n-channel MOSFET at the temperature
compensation point. For p-channel MOSFETs the values are: VGS,TC = Vth − 0.8 . . . 1.4 V
and ID,TC = − K · 0.3 . . . 1 V2 .
These values are for integrated MOSFETs with single diffusion. In contrast, discrete
MOSFETs are almost always designed with double diffusion (see Sect. 3.2); in discrete
MOSFETs, dVth /dT ≈ − 5 mV/K and thus:
VGS,TC(DMOS) ≈ Vth + 2 V
ID,TC(DMOS) ≈ K · 2 V2
3.1 Behavior of a Field Effect Transistor 191
ID RDS,on
RDS,on(25)
T1 2
T2 > T1
1
I D,TC
In practical applications, most n-channel MOSFETs are used with VGS > VGS,TC ;
the temperature coefficient in this region is negative, that is, the drain current decreases
when the temperature increases. This negative thermal relation enables thermally stable
operation without any circuit modifications. Bipolar transistors, however, require negative
electrical feedback in order to prevent positive thermal feedback due to temperature-based
current increases that can raise the temperature in the transistor and lead to its destruction.
Of particular importance in the ohmic region is the turn-on resistance RDS,on . Differ-
entiation of (3.7) leads to:
1 dRDS,on 1 dVth 1 dK
= −
RDS,on dT VGS − Vth dT K dT
VGS Vth
1 dK
≈ − ≈ 5 · 10−3 K −1
K dT
This means that RDS,on almost doubles in value if the temperature increases from 25 ◦ C
to 150 ◦ C; Fig. 3.20b shows the resulting curve for RDS,on .
Junction FET
Equation (3.14) also applies to n-channel junction FETs. The transconductance coefficient
K is proportional to the conductivity σ of the channel; and since σ ∼ µ, the temperature
coefficient is the same as for the MOSFET:
1 dK
≈ − 5 · 10−3 K −1
K dT
The threshold voltage Vth is formed from a temperature-independent portion and the dif-
fusion voltage VDiff of the pn junction between the gate and the channel. This leads to:
dVth dVDiff
= ≈ − 2.5 . . . − 1.7 mV/K
dT dT
Consequently, the temperature compensation point of an n-channel junction FET is:
VGS,TC(Jfet) ≈ Vth + 0.7 . . . 1 V
ID,TC(Jfet) ≈ K · 0.25 . . . 0.5 V2
192 3 Field Effect Transistor
The transfer characteristic is like that of the MOSFET with the exception of a shift in
direction of VGS ; the turn-on resistance RDS,on is the same as that of the MOSFET.
3.2
Construction of the Field Effect Transistor
Simple types of MOSFETs and junction FETs are symmetrical. The illustrations in Figs. 3.1
and 3.2 basically reflect this simple construction, which is predominantly used in integrated
circuits; for this reason, we will first look at integrated transistors.
3.2.1
Integrated MOSFETs
Construction: Figure 3.21 shows the principle of construction of n-channel and p-channel
MOSFETs on a common semiconductor substrate; the drain, gate, source and bulk termi-
nals are marked with respective subscripts. In n-channel MOSFETs the p-doped semicon-
ductor substrate serves as the bulk, with terminal Bn . The p-channel MOSFET requires
an n-doped bulk region and must therefore be produced in an n-doped recession; Bp is
the corresponding bulk terminal. The drain and source regions have high n-doping in the
n-channel MOSFET and high p-doping in the p-channel MOSFET. The gates are made
of polysilicon and are insulated against the channel beneath by the thin gate oxide. In the
outer areas the insulation between the semiconductor regions and the aluminum conduc-
tors of the metal-coated layer is achieved by way of the much thicker field oxide. Since
polysilicon is a relatively good conductor, the leads to the gate can be made of polysilicon;
for this reason the metal coating on the gates, as shown in Fig. 3.21, is not essential.
The term MOS (metal-oxide semiconductor) goes back to the time when the gate
was made of a metal (aluminum) instead of polysilicon. The modern MOSFET with a
polysilicon gate should be correctly called an SOS (semiconductor-oxide semiconductor),
but the familiar designation has been retained.
CMOS: Circuits designed according to Fig. 3.21 are called CMOS circuits (comple-
mentary metal-oxide semiconductor circuits) because they contain complementary MOS-
FETs. NMOS and PMOS circuits, the obsolete predecessors of CMOS circuits, consisted
Bn Sn Gn Dn Dp Gp Sp Bp
SiO2 A
Poly-Si Poly-Si
+ + + +
n n p p
+ +
p n-channel- p-channel- n
MOSFET MOSFET n
p
Fig. 3.21. Construction of an n-channel and a p-channel MOSFET in an integrated CMOS circuit
3.2 Construction of the Field Effect Transistor 193
Bulk diodes: In the sequence of layers in a CMOS circuit there are several pn junctions
which must be operated in reverse mode; they are shown as diodes in Fig. 3.21. In order for
the diodes to block between the drain or source regions and the bulk regions underneath, the
respective voltages are VSB ≥ 0 and VDB ≥ 0 in the n-channel MOSFET and VSB ≤ 0 and
VDB ≤ 0 in the p-channel MOSFET: the subscript B denotes the respective bulk region;
that is, Bn for the n-channel and Bp for the p-channel MOSFET. Furthermore, VBn ≤ VBp
is necessary in order for the diode between the bulk regions to be nonconductive. Thus it
follows that all diodes are reverse-biased if Bn is connected to the negative and Bp to the
positive supply voltage of the circuit; all other voltages then lie in between.
Latch-up: In addition to the diodes, the CMOS circuit also has a parasitic thyristor
formed by the sequence of layers and the connections Bn − Sn and Bp − Sp ; Fig. 3.22
shows a simplified equivalent circuit for the thyristor, which is made up of two bipolar
transistors and two resistors. The bipolar transistors result from the layer sequence, and
Rn and Rp are the spreading resistances of the comparatively high-resistance bulk regions.
Normally, these transistors are nonconductive since the bases are connected to the emitters
via Rn or Rp , and no current flows through the bulk regions; the thyristor is nonconductive.
In the event of under- or over-voltages at one of the inputs of the CMOS circuit, currents
will flow to the bulk regions through the protective diodes described in Sect. 7.4.6. This
can cause the voltage drop across Rp or Rn to become so high that one of the transistors is
forward-biased. The current that then flows causes a voltage drop across the other resistor
so that the second transistor also conducts and, in turn, its current keeps the first transistor
conductive. Thus we have a positive feedback that short-circuits the supply voltage Vb :
the thyristor has fired. This fault situation is called latch-up and it almost always destroys
the circuit. Modern CMOS circuits feature a high latch-up resistance that is achieved by
a suitable arrangement of the regions and specific circuitry at the inputs. One specialty
is CMOS circuits with dielectric insulation, which means that the individual MOSFETs
are formed in separate recessions that are insulated by an oxide layer; there is no thyristor
formation and the circuit is safe from latch-up.
V– Vb > 0 V+
Bn Sn Bp Sp
+ +
n p
+ Rn +
p n
Rp
n
B S G D
SiO2 Al
Poly-Si
+ – +
n n n
+
p
MOSFETs for higher voltages: Due to the fact that K ∼ W/L, the transconductance
coefficient of a MOSFET is inversely proportional to the channel length L, which one then
tries to keep as short as possible by reducing the distance between the drain and source
regions. This, however, causes a reduction in the drain–source breakthrough voltage. If the
breakthrough voltage is to be high despite the short channel length, then a weakly doped
drift region, across which a major part of the drain–source voltage drops, is introduced
between the channel and the drain terminal; this is shown in the example of an n-channel
MOSFET in Fig. 3.23. The breakthrough voltage is approximately proportional to the
length of the drift region; this is why integrated high-voltage MOSFETs require a larger
area on the wafer.
3.2.2
Discrete MOSFETs
Construction: Unlike integrated MOSFETs, discrete MOSFETs are usually built verti-
cally; that is, the drain terminal is on the underside of the substrate. Figure 3.24 shows a
three-dimensional section through a vertical MOSFET . The weakly doped drift layer with
n− doping does not run laterally on the surface, as in the integrated MOSFET as seen in
Fig. 3.23, but vertically; this saves space on the surface while guaranteeing a comparatively
high breakthrough voltage that corresponds to the thickness of the n− region. Again, the
channel runs along the surface beneath the gate. The p-doped bulk region is not formed
by the substrate, but by diffusion in the n− substrate and is connected to the source via
a p + contact region. Since the n+ source regions are also generated by diffusion, these
MOSFETs are also known as double diffused MOSFETs (DMOS).
The cellular construction is also visible in Fig. 3.24. A vertical MOSFET consists of
a two-dimensional parallel circuitry of small cells, whose source regions are connected to
the surface by a metal coating that covers the entire area of the source. The cells are driven
by a common polysilicon gate arranged beneath the source metal coating, in the form of a
grid that is connected to the outer gate terminal at the edge of the wafer only; the underside
serves as the common-drain terminal. This design facilitates a very high channel width W
on a small surface and thus a high transconductance coefficient K ∼ W . Thus it is possible
to achieve a channel width of W = 0.2 mm for a wafer with a surface area of 2 × 2 mm2
and a cell size of 20 × 20 mm2 with WCell = 20 mm; if L = 2 mm and Kn
≈ 25 µA/V2 ,
then K = Kn
W/L = 2.5 A/V2 . Since, with an n-fold reduction of the geometric size of
the cells, their number increases by the factor n2 but the width W per cell is only reduced by
a factor of n, any miniaturization results in an increase in the channel width per unit area.
3.2 Construction of the Field Effect Transistor 195
Poly-Si G
+ + + +
+ n n + n n
p p p p p p
–
ID n ID
+
n
Parasitic elements: Due to the specific construction of vertical MOSFETs, there are sev-
eral parasitic elements, which are shown in Fig. 3.25 together with the resulting equivalent
circuit diagram:
– The large overlap of gate and source results in a high external gate–source capacitance
CGS . This is usually higher than the internal gate–source capacitance, which is described
in more detail in Sect. 3.3.2.
– The overlap between the gate and the n− region results in a relatively high external gate–
drain capacitance CGD , which adds to the internal drain–gate capacitance; a detailed
description of the latter is found in Sect. 3.3.2.
S
CGS
G S
CGD
+
n
RB p
+ CGS RB
TB
p G
TB D rev
CB CDS CDS
– D rev
n CGD CB
+
n
D
D
Fig. 3.25. Parasitic elements and an equivalent circuit for an n-channel DMOS-FET
196 3 Field Effect Transistor
– The drain–source capacitances CDS and CB lie between the bulk region and the drain
region; CDS is located directly between drain and source, while CB is in series with the
spreading resistance RB of the bulk region.
– Due to the sequence of layers, there is also a bipolar transistor TB whose base is connected
to the emitter via the spreading resistance RB ; TB is therefore normally reverse-biased.
If the drain–source voltage increases very rapidly, then the current I = CB dVDS /dt
through CB increases, causing a high voltage drop across RB , so that TB becomes
conductive. To prevent this from happening when DMOS power switches are turned
off, the slew rate of VDS must be limited by suitable control conditions or by a turn-off
relief circuit.
– Between the source and the drain there is a reverse diode Drev , which is forward-biased
when the drain–source voltage is negative. It can be used as a freewheeling diode when
switching inductive loads, but it leads to unwanted currents in push-pull circuits due to
their long reverse recovery time tRR , which is determined by the design of the MOSFET.
– With high currents, the influence of parasitic resistances in the source terminal becomes
noticeable. In this case the external gate–source voltage VGS across the terminals is
comprised of the internal gate–source voltage and the voltage drop across the source
resistance RS ; this linearizes the transfer characteristic for high currents (see Fig. 3.26a).
– In vertical MOSFETs the pinch-off voltage VDS,po is higher than VGS − Vth due to
an additional voltage drop in the drift region. This voltage drop can be described by
a nonlinear drain resistance, and it results in a shift in the output characteristics (see
Fig. 3.26b).
VS RS VS = ID RS
DMOS
DMOS
B S G D B S G D
Al
+ + + + +
n p n n n
+ – + –
p n p n
p p
3.2.3
Junction FETs
Figure 3.27 shows the construction of a normal n-channel junction FET, with a pn junction
between the gate and the channel, and an n-channel MESFET with a metal semiconductor
junction (a Schottky junction) between the gate and the channel. In integrated junction
FETs, the substrate terminals B are connected to the negative supply voltage to enable the
pn junctions between the substrate and the n− channel regions always to be operated in
reverse mode. Furthermore, every FET must be surrounded by a closed p+ ring, so that the
channel regions of the individual FETs are insulated from each other. In discrete junction
FETs, the substrate may also be connected to the gate; this provides a controlling effect for
the substrate/channel junction in addition to that of the gate/channel junction. The vertical
construction used in the MOSFET and the bipolar transistor is not possible in the case of
the junction FET.
3.2.4
Cases
Discrete MOSFETs and discrete junction FETs are contained in the same cases as bipolar
transistors; Fig. 2.21 on page 56 shows the most common cases. MOSFETs are available
in all power ratings and thus in all case sizes. Junction FETs are always small-signal
transistors in small cases; an exception is the power MESFET for high-frequency power
amplifiers, which comes in a special high-frequency housing for surface mounting. There
are also junction FETs with a separate bulk terminal, in cases with four terminals. Dual-
gate MOSFETs also require cases with four terminals; these are always high-frequency
transistors in special high-frequency cases.
3.3
Models of Field Effect Transistors
Section 3.1.2 describes the static behavior of a field effect transistor by way of large-
signal (3.2)–(3.4), and with secondary effects neglected. More accurate models that take
these effects into consideration and, furthermore, correctly describe the dynamic response
are required for computer-aided circuit design. The dynamic small-signal model that is
necessary for calculating the frequency response of circuits is derived from this large-
signal model by linearization.
198 3 Field Effect Transistor
D D
G I D,D
B VGS ID VDS B
G
I D,S
S S
3.3.1
Static Behavior
Contrary to the bipolar transistor, for which the Gummel–Poon model has proven its
general applicability, for FETs there are numerous models that have their advantages and
disadvantages for the various applications, and can become rather complex. The level-1
MOSFET model13 , which is available in almost all CAD programs for circuit simulation, is
described below. It is very suitable for discrete transistors with a comparably large channel
length and width, but not for integrated MOSFETs, with their small dimensions typical
of large-scale integrated circuits. Here, the more elaborate level-2 and level-3 models or
the BSIM models14 are necessary; they also take into account the short-channel, narrow-
channel and sub-threshold effects. Only a qualitative description of these effects is provided
here.
Junctions FETs have their own model, the static response of which corresponds to that
of the level-1 MOSFET model, despite the fact that CAD programs often use other param-
eters or other designations for parameters with the same meaning; this will be discussed
in more detail at the end of this section.
Drain current: The level-1 model uses (3.2) and (3.3) in combination with (3.5); from
VDS,po = VGS − Vth (3.15)
and K = Kn
W/L, we obtain:
13 This name is used in circuit simulators of the Spice family; for example, PSpice from OrCAD.
In the literature it is often called the Shichman–Hodges model, because major portions are taken
from a publication by H. Shichman and D.A. Hodges.
14 The BSIM models (Berkeley short-channel IGFET model) were developed at the University of
California, Berkeley, and are presently considered the most advanced models for short-channel
MOSFETs.
3.3 Models of Field Effect Transistors 199
⎧
⎪ 0 for VGS < Vth
⎪
⎪
⎪
⎪ Kn
W
⎪
⎪ VDS VDS
⎪
⎪ VDS VGS − Vth − 1+ for VGS ≥ Vth ,
⎨ L 2 VA
ID = 0 ≤ VDS < VDS,po (3.16)
⎪
⎪
⎪
⎪ Kn
W
⎪
⎪ − 2
+
VDS
for VGS ≥ Vth ,
⎪
⎪ (VGS Vth ) 1
⎪
⎩ 2L VA
VDS ≥ VDS,po
In discrete MOSFETs the oxide layer thickness dox ≈ 40 . . . 100 nm, while it is reduced
to 15 nm in large-scale integrated CMOS circuits.
Threshold voltage: The threshold voltage Vth is the gate–source voltage above which
the inversion channel is generated beneath the gate. As the channel is in the substrate
region, the inversion and thus the threshold voltage depend on the gate–substrate voltage
VGB . This effect is known as the substrate effect and is determined by the doping of the
substrate. Since the form Vth = Vth (VGB ) is not very illustrating, the source is taken as the
reference point, as in VGS and VDS , and replaces VGB = VGS − VBS by the bulk–source
voltage VBS . Consequently [3.1]:
# $
Vth = Vth,0 + γ Vinv − VBS − Vinv (3.18)
The parameters
√ are the zero threshold voltage Vth,0 , the substrate control factor γ ≈
0.3 . . . 0.8 V and the inversion voltage Vinv√≈ 0.55 . . . 0.8 V. Figure 3.29 shows the plot
of Vth versus VBS for Vth,0 = 1 V, γ = 0.55 V and Vinv = 0.7 V;15 this requires VBS ≤ 0
so that the bulk source diode is in reverse mode.
The substrate effect is noted especially in integrated circuits, where all n-channel MOS-
FETs have a common substrate region and are operated with different bulk–source voltages
depending on the operating point; integrated MOSFETs of the same geometric dimensions
thus have different characteristics if operated with different bulk–source voltages. In the
discrete MOSFET, with its internal connection between source and substrate, this effect
does not occur; here, VBS = 0 and Vth = Vth,0 .
As an alternative to γ and Vinv , the substrate doping density Nsub and the oxide thickness
dox may be given. Consequently [3.1]:
2q0 r,Si Nsub 2qr,Si Nsub dox
γ =
= (3.19)
Cox 0 r,ox
Vth
V
2
V th,0 = 1 V
g = 0.55 V
Vi nv = 0.7 V
1
~
~
–10 –5 0 V BS
V
Fig. 3.29. Dependence of the threshold voltage Vth on the bulk–source voltage VBS (substrate
effect)
Nsub
Vinv = 2VT ln (3.20)
ni
Inserting the constants q = 1.602 · 10−19 As, 0 = 8.85 · 10−12 As/Vm, r,ox = 3.9 and
r,Si = 11.9, as well as VT = 26 mV and ni = 1.45 · 1010 cm−3 for T = 300 K, we
obtain:
√ '
γ ≈ 1.7 · 10−10 V · Nsub /cm−3 · dox /nm
T =300 K
Nsub
Vinv ≈ 52 mV · ln
1.45 · 1010 cm−3
Typical values are Nsub ≈ 1 . . . 7 · 1016 cm−3 for integrated circuits and Nsub ≈ 5 ·
1014 . . . 1016 cm−3 for discrete MOSFETs.
Substrate diodes: Due to the construction of a MOSFET, there are substrate diodes
between bulk and source and bulk and drain; Fig. 3.28 shows the location and polarity
of these diodes in the equivalent circuit diagram for an n-channel MOSFET. The currents
through these diodes are described by the diode equations
V
BS
ID,S = IS,S e nVT − 1 (3.21)
VBD
ID,D = IS,D e nVT −1 (3.22)
with saturation reverse currents IS,S and IS,D and the emission factor n ≈ 1.
The reverse current density JS and the edge current density JS,SW may be given as an
alternative to IS,S and IS,D ; with the areas AS and AD and the edge length lS and lD for
the source and drain regions, it follows that:
IS,S = JS AS + JS,SW lS (3.23)
IS,D = JS AD + JS,SW lD (3.24)
In particular, this is used in CAD programs for the design of integrated circuits; in this
case, JS and JS,SW are parameters of the MOS process and are the same for all n-channel
3.3 Models of Field Effect Transistors 201
MOSFETs. After the sizes of the individual MOSFETs have been determined, only the
areas and edge lengths need to be determined; the CAD program then calculates IS,S
and IS,D .
In normal operation, the bulk terminal of an n-channel MOSFET has a lower, or at
most the same, potential as the drain and source; thus VBS , VBD ≤ 0 and the diodes are
operated in the reverse-biased mode. In discrete MOSFETs, with their internal connection
between source and bulk, this condition is automatically met as long as VDS > 0. In
integrated circuits the common bulk terminal of the n-channel MOSFET is connected to
the negative supply voltage, so that the diodes are always reverse-biased. In the case of
smaller MOSFETs, the reverse currents ID,S ≈ − IS,S and ID,D ≈ − IS,D are in the pA
region, while in power MOSFETs they are in the µA region; they can usually be neglected.
Other effects: There is a multitude of other effects that are not taken into consideration
by the level-1 model. The most important are briefly outlined below [3.2]:
– For short channel lengths L, the region beneath the channel is highly restricted by
the depletion layers of the bulk–source and the bulk–drain diodes. Space charges in
that area are increasingly compensated by charges in the source and drain regions,
resulting in a reduction of the gate charge; this causes the threshold voltage Vth to
drop. This is known as the short-channel effect and it depends on the voltages VBS
and VBD or VDS = VBS − VBD . With an increasing drain–source voltage, the threshold
voltage decreases with a corresponding rise in the drain current; the slope of the output
characteristics in the pinch-off region thus depends on VDS . The description of these
effects in the level-2/3 and the BSIM models can therefore be regarded as extended
channel-length modulation which, in this case, is no longer modeled by the Early voltage
VA or the channel-length modulation parameter λ, but by the threshold voltage
# $
Vth = Vth,0 + γ (1 − f (L, VDS , VBS )) Vinv − VBS − Vinv
The function f (L, VDS , VBS ) is detailed in reference [3.3]. Figure 3.30a shows the
dependence of the threshold voltage on the channel length in an integrated MOSFET.
– With a shrinking channel width W , the charge at the edges of the channel increases
compared to the charge within the channel and must be taken into consideration. It is
compensated by the charge of the gate, which causes the threshold voltage Vth to increase.
This is known as the narrow-channel effect and it can be described by expanding the
threshold voltage equation:
# $ Vinv − VBS
Vth = Vth,0 + γ ... +k
W
The factor k is described in more detail in reference [3.3]. Figure 3.30b shows the
dependence of the threshold voltage on the channel width in an integrated MOSFET.
– Free charges exist in the channel area even without the inversion channel; this allows a
small drain current to flow even with a voltage below the threshold voltage Vth . This effect
is called the sub-threshold effect and the current is called the sub-threshold current. In
this sub-threshold region, the characteristic is an exponential curve, and in the range of
202 3 Field Effect Transistor
0.7 0.7
0.6 0.6
~
~ ~
~
1 5 10 L 1 2 3 4 W
µm µm
a Short-channel effect b Narrow-channel effect
Here, nV ≈ 1.5 . . . 2.5 is the emission factor in the sub-threshold region. The
changeover takes place at VGS ≈ Vth + 3 . . . 5 · VT ≈ Vth + 78 . . . 130 mV. Figure 3.31
shows a plot of the drain current in the range of the threshold region on both linear and
logarithmic scales; the latter shows the exponential sub-threshold current as a straight
line. In integrated MOS circuits for battery-operated units, the MOSFETs are often op-
erated in this region; this allows the current consumption to be markedly reduced at the
expense of circuit bandwidth.
ID ID
VGS – V th
[log]
~e nVVT
VGS – Vth
~e nVVT
~ (VGS – Vth)2
~ (VGS – Vth)2
~
~
~
a Linear b Logarithmic
Kp W
⎪
⎪ VDS VDS
⎪
⎪ − VDS VGS − Vth − 1− for VGS ≤ Vth ,
⎨ L 2 VA
ID = VDS,po < VDS ≤ 0
⎪
⎪
⎪
⎪
⎪
⎪
Kp W VDS
⎪
⎪ − 2L (VGS − Vth ) 1 − V
2
for VGS ≤ Vth ,
⎪
⎪
⎩ A
VDS ≤ VDS,po
# $
Vth = Vth,0 − γ Vinv + VBS − Vinv
The parameters γ and Vinv are also determined for the p-channel MOSFET using (3.19) and
(3.20). The Early voltage VA is positive in both the n channel and the p channel MOSFET;
the relative transconductance coefficient is also positive:
µp 0 r,ox
Kp
=
dox
Here, µp = 0.015 . . . 0.025 m2 /Vs. The currents in the substrate diodes are:
V
BS
−
ID,S = − IS,S e nVT −1
VBD
−
ID,D = − IS,D e nVT −1
Spreading Resistances
For each terminal, there is a spreading resistance that is made up of the resistance of
the relevant region and the contact resistance of the metal zone. Figure 3.32a shows the
resistances RG , RS , RD and RB in a sample integrated n-channel MOSFET. In CAD
programs for circuit simulations, these resistances can be entered directly or by using the
sheet resistance Rsh and the multipliers nRG , nRS , nRD and nRB :
⎡ ⎤ ⎡ ⎤
RG nRG
⎢ RS ⎥ ⎢ ⎥
⎢ ⎥ = Rsh ⎢ nRS ⎥ (3.26)
⎣RD ⎦ ⎣ nRD ⎦
RB nRB
In this case, the sheet resistance is a property of the MOS process and is identical for all
n-channel MOSFETs in a given integrated circuit. Typical values are Rsh ≈ 20 . . . 50
in n-channel MOSFETs and Rsh ≈ 50 . . . 100 in p-channel MOSFETs.
204 3 Field Effect Transistor
B S G D D
RD
RG D' RB
RG B'
G' B
+ +
G S'
n RS RD n
+
p RS
RB
p
S
Figure 3.32b shows the extended model. We must distinguish between the external
terminals G, S, D and B and the internal terminals G
, S
, D
and B
, which means that
the drain current ID and the diode currents ID,S and ID,D depend on the internal voltages
VG
S
, VD
S
, . . . .
K 1+
VA
VA →∞
2ID
≈ Vth + + ID RS (3.27)
K
This equation is used for parameter extraction; on the basis of at least three pairs of values
(VGS , ID ) in the pinch-off region, it is possible to determine the three parameters Vth , K
and RS .16
The relations in the family of output characteristics shown in Fig. 3.26b are more
complicated. It is possible to describe the displacement by a resistance in the drain line
but, in contrast to the linear drain resistance RD in Fig. 3.32b, this resistance is nonlinear.
This is caused by the conductivity modulation in the drift region, which means that the
conductivity in the drift region increases with the current due to an increase in the charge
carrier density. The voltage drop VDrift is approximately [3.4]:
ID
VDrift = V0 1+2 −1 (3.28)
I0
16 In practice, a large number of pairs of values are used and the parameters are determined by means
of an orthogonal projection.
3.3 Models of Field Effect Transistors 205
VDrift
D
V
1.0 R= 1Ω ID
0.8
VDrift
0.6
0.4 D'
RG VD' S'
G'
0.2 G S'
RS VS
0 0.2 0.4 0.6 0.8 1.0 1.2 ID
A S
Here, V0 and I0 are the drift region parameters. Figure 3.33a shows the plot of the drift
voltage versus ID for a MOSFET with V0 = 1 V and I0 = 1 A. For small currents the
drift region acts like a linear resistor with R = V0 /I0 ; in Fig. 3.33a, R = 1 . For higher
currents, the conductivity increases and the voltage drop is less than when R = 1 .
The characteristic given by (3.28) corresponds to that of a depletion MOSFET with
the gate connected to the drain; if VGS = VDS and Vth < 0, then
2ID
VDrift = |Vth | 1+ −1
K|Vth |2
A comparison with (3.28) shows that a depletion MOSFET with Vth = −V0 and
K = I0 /V02 can be used to model the drift region; this leads to the equivalent circuit
shown in Fig. 3.33b, which features a depletion MOSFET in place of resistance RD as
compared to Fig. 3.32b.
Junction FETs
The model of a junction FET is derived from the model of a MOSFET by eliminating the
isolated gate, renaming the bulk gate and inserting β = K/2 in the equations; the result is
206 3 Field Effect Transistor
IG
= G ID
G
S
Fig. 3.34. Large-signal equivalent
S circuit for an n-channel JFET
the equivalent circuit diagram shown in Fig. 3.34, with the equations:
⎧
⎪
⎪
0 for VGS < Vth
⎪
⎪
⎪
⎪ V V
⎪ 2β VDS VGS − Vth − DS
⎪ 1+
DS
for VGS ≥ Vth ,
⎪
⎨ 2 VA
ID = 0 ≤ VDS < VGS − Vth
⎪
⎪
⎪
⎪ VDS
⎪
⎪
⎪ β (VGS − Vth ) 1 + VA for VGS ≥ Vth ,
2
⎪
⎪
⎩
VDS ≥ VGS − Vth
V V
GS GD
IG = IS e nVT + e nVT − 2 (3.29)
The parameters are the threshold voltage Vth , the JFET transconductance coefficient β, the
Early voltage VA , the saturation leakage current IS and the emission coefficient n.
As in the MOSFET, additional spreading resistances are allowed in the drain and
source lines; the corresponding parameters are RS and RD . In the JFET model there is no
gate resistance: however, this must be added externally for circuit simulations with CAD
programs if an accurate description of the high-frequency response is required.
Unlike the MOSFET model, the JFET model is not scalable; that is, there are no
geometric dimensions such as channel length and width. The JFET model is simple but
not very accurate.
3.3.2
Dynamic Behavior
The response to pulsed and sinusoidal signals is called the dynamic behavior and cannot be
determined from the characteristics. This is due to the capacitances between the different
regions of the MOSFET, as shown in Fig. 3.35. These may be classified into three groups:
– Channel capacitances CGS,ch and CGD,ch , which describe the capacitive effect between
the gate and the channel. They are active only with an existing channel; that is, when the
MOSFET is in the conductive state. Without a channel, there is a capacitance CGB,ch
between gate and bulk which is a component part of the gate–bulk capacitance CGB .
The channel capacitances are linear in the pinch-off region, but nonlinear in the ohmic
region.
– The linear overlap capacitances CGS,ov , CGD,ov and CGB,ov result from the geometric
overlap of the gate and the source, drain and bulk regions. CGB,ov results from the
overlap of the gate and the bulk at the sides of the channel and is part of CGB .
3.3 Models of Field Effect Transistors 207
S
SiO2
Poly-Si
CBS
– The nonlinear junction capacitances CBS and CBD are caused by the pn junctions
between bulk and source and between bulk and drain.
Channel capacitances: The gate together with the underlying channel form a plate
capacitor with an oxide capacity:
A WL
Cox = ox = 0 r,ox (3.31)
dox dox
In the cutoff region, that is, without a channel, this capacitance acts between gate and
bulk; consequently,
⎫
CGS,ch = 0 ⎪
⎬
CGD,ch = 0 for VG
S
< Vth (3.32)
⎪
⎭
CGB,ch = Cox
In the ohmic region the channel extends from the source to the drain region, and the
oxide capacitance is distributed along the channel according to the charge distribution.
For VD
S
= 0, the channel is symmetrical, resulting in CGS,ch = CGD,ch = Cox /2.
For VD
S
> 0, the channel is asymmetrical, resulting in CGS,ch > CGD,ch . Thus, the
capacitances depend on VD
S
and VG
S
, and may be described by the following equations
as an approximation [3.3]:
208 3 Field Effect Transistor
CGB,ch
CGS,ch
CGS,ch
CGD,ch
CGD,ch CGD,ch
Fig. 3.36. Schematic presentation of the channel capacitances for an n-channel MOSFET. In a real
MOSFET, the transitions are smooth
2
⎫
2 VG
S
− Vth − VD
S
⎪
⎪
CGS,ch = Cox 1 − ⎪
⎪
3 2 (VG
S
− Vth ) − VD
S
⎪
⎪
⎪
⎪
⎪
⎬
2
for VG
S
≥ Vth ,
2 VG
S
− Vth (3.33)
CGD,ch = Cox 1 − ⎪
⎪ V D
S
< VG
S
− Vth
2 (VG
S
− Vth ) − VD
S
⎪
⎪
3 ⎪
⎪
⎪
⎪
⎪
⎭
CGB,ch = 0
In the pinch-off region the channel is pinched-off at the drain side so that the connection
between the channel and the drain region no longer exists; thus CGD,ch = 0. Only CGS,ch
is left as a channel capacitance [3.3]:
⎫
2
CGS,ch = Cox ⎪ ⎪
⎪
⎬
3
CGD,ch = 0 for VG
S
≥ Vth , VD
S
≥ VG
S
− Vth (3.34)
⎪
⎪
⎪
⎭
CGB,ch = 0
Figure 3.36 shows the curves for the three capacitances. Note that the similarity to a
plate capacitor only exists where the charge distribution is uniform; only in this case does
CGS,ch + CGD,ch + CGB,ch = Cox . This is always true for the cutoff region, but only for
VD
S
= 0 in the ohmic region and never in the pinch-off region.
As Fig. 3.36 shows, the capacitance model provides an abrupt transition from CGB,ch
to CGS,ch at the boundary between the cutoff and the pinch-off regions, causing the total
capacitance to jump from Cox to 2Cox /3. This is only a very rough presentation of the
true relationship in this region. In a real MOSFET, the transitions are always smooth; the
corresponding curves are indicated by broken lines in Fig. 3.36.17
17 A relatively simple description of this transition is given in reference [3.3]. An additional problem
is charge conservation, consideration of which requires further modification of the equations; a
correspondingly expanded model is used in PSpice from OrCAD [3.5].
3.3 Models of Field Effect Transistors 209
Overlap capacitances: As the gate is usually larger than the channel18 , that is, wider
than the channel width W and longer than the channel length L, overlaps occur at the edges,
producing the overlap capacitances CGS,ov , CGD,ov and CGB,ov . But it is not possible to
obtain these capacitances from the related overlap areas by means of the plate capacitive
formula, as the field and charge distributions along the margins are not uniform. For this
, CGD,ov
CGS,ov W
CGD,ov =
CGD,ov W (3.35)
CGB,ov =
CGB,ov L
CGB,ov contains portions of both sides and therefore needs to be multiplied by the single
Junction capacitances: pn junctions between bulk and source and between bulk and
drain have a voltage-dependent junction capacitance, CBS or CBD respectively, which are
determined by the doping, the area of the junction and the applied voltage. The description
is like that of a diode; from (1.13) on page 19 follows:
CJ 0,S
CBS (VB
S
) = for VB
S
≤ 0 (3.36)
VB
S
mJ
1−
VDiff
CJ 0,D
CBD (VB
D
) = for VB
D
≤ 0 (3.37)
VB
D
mJ
1−
VDiff
with the zero capacitances CJ 0,S and CJ 0,D , the diffusion voltage VDiff and the capacitance
coefficient mJ ≈ 1/3 . . . 1/2.
As alternatives to CJ 0,S and CJ 0,D , one can use the junction capacitance per unit area
CJ
, the sidewall capacitance per unit length CJ,SW
, the sidewall diffusion voltage VDiff ,SW
and the sidewall capacitance coefficient mJ,SW ; for the areas AS and AD and the sidewall
length lS and lD of the source and drain regions, this results in:
CJ
AS CJ,SW lS
CBS = mJ + for VB
S
≤ 0 (3.38)
VB
S
VB
S
mJ,SW
1− 1−
VDiff VDiff ,SW
18 In order to allow the formation of a continuous channel, the size of the gate must be at least as
large as the channel region.
210 3 Field Effect Transistor
CJ
AD CJ,SW lD
CBD = mJ + for VB
D
≤ 0 (3.39)
VB
D
VB
D
mJ,SW
1− 1−
VDiff VDiff ,SW
CAD programs make use of this for the design of integrated circuits; in this case, CJ
, CJ,SW ,
VDiff , VDiff ,SW , mJ and mJ,SW are parameters of the MOS process and are the same for all
n-channel MOSFETs. If the parameters of the individual MOSFETs are given, then only
the areas and the sidewall length have to be determined; from these, the CAD program
calculates CBS and CBD .
Here, the range for the use of these equations is limited to VB
S
≤ 0 and VB
D
≤ 0.
If VB
S
> 0 and VB
D
> 0, then the pn junctions are forward biased and the diffusion
capacitance must be considered in addition to the junction capacitance. In other words, a
complete capacitance model, as for a diode, must be used (see Sect. 1.3.2 on page 19);
this includes the transit time τT as an additional parameter that is required for calculating
the diffusion capacitance. CAD programs use a complete capacitance model for every pn
junction.
RD
I D,D
D'
CGD CBD
RG RB
G G' ID B' B
CGS CBS
I D,S
S'
CGB
RS
NMOS CMOS
ParameterPSpice Enhancement Depletion n-channel p-channel Unit
Kn
, Kp
KP 37 33 69 23.5 mA/V2
Vth,0 VTO 1.1 −3.8 0.73 −0.75 V
√
γ GAMMA 0.41 0.92 0.73 0.56 V
λ LAMBDA 0.03 0.01 0.033 0.055 V−1
VA - 33 100 30 18 V
dox TOX 55 55 25 25 nm
µn UO 590 525 500 170 cm2 /Vs
Vinv PHI 0.62 0.7 0.76 0.73 V
Nsub NSUB 0.2 1 3 1.8 1016 /cm3
Rsh RSH 25 25 25 45
CJ
CJ 110 110 360 340 µF/m2
mJ MJ 0.5 0.5 0.4 0.5
VDiff PB 0.8 0.8 0.9 0.9 V
Where accuracy ranks higher, a macromodel must be used which, in addition to the
actual MOSFET model, includes other components to model specific characteristics. One
example is the static equivalent circuit for a DMOS FET, shown in Fig. 3.33b, for which an
additional MOSFET is used to model the nonlinear drain resistance. Similar expansions
are needed to describe the dynamic behavior of a DMOS FET, but there is no uniform
equivalent circuit diagram.
Even though in some cases the level-2 and level-3 models use different equations, their
parameters are largely identical. They use the following additional parameters [3.3]:
– Level-2 Model: UCRIT, UEXP and VMAX for the voltage dependence of the mobility,
and NEFF to describe the channel charge.
– Level-3 Model: THETA, ETA and KAPPA for empirical modelling of the static behavior.
– Both Models: DELTA for modelling the narrow-channel effect and XQC for the charge
distribution in the channel.
Both models describe channel-length modulation with the help of additional parameters;
this renders the channel-length modulation parameter λ unnecessary.
CGD RD
D
IG
G ID
S
CGS RS Fig. 3.42. Model of an n-channel junction FET
3.3.3
Small-Signal Model
A linear small-signal model can be derived from the level-1 MOSFET model by lineariza-
tion at an operation point. In practice, the operating point is selected so that the FET operates
in the pinch-off region; the small-signal models described here are thus applicable to this
operating mode only.
The static small-signal model describes the small-signal response at low frequencies
and is therefore also known as the DC small-signal equivalent circuit. The dynamic small-
signal model further describes the dynamic small-signal response and is needed to calculate
the frequency response of a circuit; it is also known as the AC small-signal equivalent
circuit.
Kn
W % & VD
S
,A
gm = = VG
S
,A − Vth 1 +
∂VG
S
A L VA
∂ID
∂ID
dVth
gm,B =
=
∂VB
S
A ∂Vth
A dVBS
γ Kn
W % & VD
S
,A
= VG S ,A − Vth 1 +
2 Vinv − VB
S
,A L VA
1 ∂ID
1 Kn
W % &2
=
= VG
S
,A − Vth
rDS ∂VD
S
A VA 2L
216 3 Field Effect Transistor
D D
RD RD
G RG G' B' RB B G = G' B = B'
D' D'
S' S'
RS RS
S S
a Before linearisation b After linearisation
Fig. 3.45. Generating the static small-signal model by linearization of the static level-1 MOSFET
model
∂ID
VD
S
,A VD
S
,A VA
gm = = 2K I 1 + ≈ 2K ID,A (3.40)
∂VG
S
A
D,A
VA
∂ID
γ gm
gm,B =
= (3.41)
∂VB
S
A 2 Vinv − VB
S
,A
∂VD
S
VA + VD
S
,A VD
S
,A VA VA
rDS = = ≈ (3.42)
∂ID
A ID,A ID,A
The approximations for gm and rDS correspond with (3.11) and (3.12) in Sect. 3.1.4. An
additional small-signal parameter is the substrate transconductance gm,B , which is only
effective with a small-signal voltage vBS = 0 between source and bulk.
ID,A
gm = for VGS < Vth + 2nV VT (3.43)
nV V T
Equations (3.41) and (3.42) for gm,B and rDS also apply to the sub-threshold region. For
nV ≈ 2, the borderline of the sub-threshold region is at VGS ≈ Vth + 4VT ≈ Vth + 100 mV
or ID ≈ 2K (nV VT )2 ≈ K ·0.005 V2 . The transconductance is continuous; in other words,
3.3 Models of Field Effect Transistors 217
D
G B
at the transition both (3.40) and (3.43) yield the same result:
ID,A =2K(nV VT )2 ID,A
2K ID,A =
nV V T
DC small-signal equivalent circuit: Figure 3.45b shows the resulting static small-signal
model. The spreading resistances RS and RD are disregarded in almost all practical calcu-
lations; this leads to the small-signal equivalent circuit shown in Fig. 3.46, which is derived
from the small-signal equivalent circuit described in Sect. 3.1.4 by adding the controlled
source with the substrate transconductance gm,B .
Small-signal equivalent circuit for junction FETs: Figure 3.46 also applies to junction
FETs if the controlled source with the substrate transconductance is removed; the small-
signal parameters are obtained using (3.29):
VD
S
,A VD
S
,A VA 2
gm = 2 β ID,A 1 + ≈ 2 β ID,A = ID,0 ID,A
VA |Vth |
VD
S
,A VA
VD
S
,A + VA VA
rDS = ≈ ,
ID,A ID,A
where ID,0 = ID (VGS = 0) = β Vth2 . By virtue of the relation K = 2β, this leads to the
same equations as those used for the MOSFET.
= CGD,ov = CGD,ov W
CGB
RD
CGD CBD
D'
C GB
S
The gate–source capacitance CGS comprises the channel capacitance in the pinch-
off region and the gate–source overlap capacitance; it is determined by the geometric
dimensions alone and is independent of the operating point voltages as long as one remains
in the pinch-off region. The gate–drain capacitance CGD and the gate–bulk capacitance
CGB are pure overlap capacitances and therefore are not influenced by the operating point,
while the junction capacitances CBS and CBD depend on the operating point voltages
VB
S
,A and VB
D
,A .
D
RG CGD CBD
G'
G B
CBS
CGS v G'S gmvG'S rDS gm,BvBS v BS
RG CGD RB CC
G G' B B'
D C
S S
Fig. 3.49. A comparison of the dynamic small-signal models for discrete MOSFET and bipolar
transistor
calculations can be used by exchanging the relevant variables, assuming rBE → ∞ and
inserting:20
. rDS
rCE =
1 + sCDS rDS
This model can also apply to integrated MOSFETs if the source and bulk in the small-signal
equivalent circuit are combined and connected to the small-signal ground.
1
ωY 21s = 2πfY 21s ≈ (3.46)
RG (CGS + CGD )
The transconductance cutoff frequency does not depend on the operating point as long as
one remains in the pinch-off region.
Transit frequency: The transit frequency fT is the frequency at which the absolute value
of the small-signal current gain is reduced to 1 when the circuit is operated in the pinch-off
a For calculating the transconductance cut-off frequency b For calculating the transit frequency
Fig. 3.50. Small-signal equivalent circuits for calculating the cutoff frequencies
|i D |
≡1
|i G |
s=j ωT
From the small-signal equivalent circuit shown in Fig. 3.50b, it follows that
iD gm − sCGD
=
iG s (CGS + CGD )
and thus:
gm
ωT = 2πfT ≈ (3.47)
CGS + CGD
Relationship and meaning of the cutoff frequencies: Comparing the cutoff frequencies
leads to the relationship:
gm RG <1
fT = fY 21s gm RG < fY 21s
Determining the small-signal capacitances from the cutoff frequencies: If the data
sheet for a FET specifies the transit frequency fT , the reverse capacitance Crss (reverse,
grounded source, gate shorted) and the output capacitance Coss (output, grounded source,
gate shorted), the capacitances of the equivalent circuit shown in Fig. 3.49a can be deter-
mined with the help of (3.47):
3.3 Models of Field Effect Transistors 221
gm
CGS ≈ − Crss
ωT
CGD ≈ Crss
CDS ≈ Coss − Crss
If the transconductance cutoff frequency fY 21s is also known, then the gate resistance can
be calculated:
fT
RG =
gm fY 21s
Output VA
rDS rDS =
resistance ID,A
(fT ) Transit frequency From the data sheet
Transconductance
(fY 21s ) From the data sheet
cutoff frequency
fT
Gate spreading RG =
RG gm fY 21s
resistance
or reasonable assumption (RG ≈ 1 . . . 100 )
Gate–drain
CGD
capacitance From the data sheet: CGD ≈ Crss
Gate–source gm
CGS
capacitance CGS ≈ − CGD
2πfT
Drain–source
CDS
capacitance From the data sheet: CDS ≈ Coss − Crss
Fig. 3.51. Method of determining small-signal parameters for a discrete FET (auxiliary parameters
in parentheses)
known; it is only necessary to evaluate (3.40)–(3.45). Figure 3.52 outlines the procedure
for determining the parameters for the small-signal model shown in Fig. 3.48.
3.3.4
Noise
The basics for describing noise and calculating the noise figure are outlined in Sect. 2.3.4
on page 82, using the example of a bipolar transistor. The same procedure can be applied
to the field effect transistor using the relevant noise sources.
3.3 Models of Field Effect Transistors 223
Output VA 1
rDS rDS = =
resistance ID,A λID,A
Gate spreading From the geometry: RG = nRG Rsh
RG
resistance or reasonable assumption (RG ≈ 1 . . . 100 )
Gate–source 2
2
CGS
capacitance CGS = Cox + CGS,ov W ≈ Cox
3 3
Gate–drain
CGD
capacitancet CGD = CGD,ov W
Fig. 3.52. Method of determining small-signal parameters for an integrated MOSFET (auxiliary
parameters in parentheses)
The noise figure of a field effect transistor is calculated below; first, we look at the noise
sources and the correlation between the noise sources. In general, a simplified description
is sufficient for practical applications and is given below.
S
v v
Fig. 3.53. Small-signal model of a FET with the original (top) and equivalent (bottom) noise
sources
neither homogenous nor in thermal equilibrium. Furthermore, there is 1/f noise with
the experimental parameters k(1/f ) und γ(1/f ) ≈ 1. At low frequencies, the 1/f portion
is predominant, while at medium and high frequencies the thermal portion dominates.
Making the portions equal leads to the 1/f cutoff frequency:
(γ
(1/f ) −1/2)
3 k(1/f ) ID,A γ(1/f ) =1 3 k(1/f ) ID,A
fg(1/f ) = √ =
8 kT K 8 kT K
This increases with a rising operating point current. For the MOSFET it is approximately
k(1/f ) ∼ 1/L2 , which means that the 1/f noise decreases as the channel length increases;
since MOSFETs in integrated circuits are scaled in correspondence with the current at
the operating point (ID,A ∼ K ∼ W/L) a large MOSFET has less 1/f noise than a small
one with the same current or the same transconductance. Typical values are fg(1/f ) ≈
100 kHz . . . 10 MHz for MOSFETs and fg(1/f ) ≈ 10 Hz . . . 1 kHz for junction FETs.
– Induced gate noise, with:
2
4 f
|i G,r (f )|2 = kT gm
3 fT
This noise current is also caused by the thermal noise in the channel, which is transmitted
to the gate by the capacitive coupling between the gate and the channel. Therefore, the
noise current sources iG,r and iD,r are not independent but correlated. This correlation
must be taken into account when calculating the noise figure.
The upper portion of Fig. 3.53 shows the small-signal model with noise sources vRG,r ,
iG,r and iD,r .
3.3 Models of Field Effect Transistors 225
and the impedances Z k and − Z k to describe the correlation. The correlation impedance23
√
j 2fT
Zk ≈ −
gm f
only influences the source vD,r ; for all other sources and signals, the parallel arrangement
of Z k und − Z k is ineffective.
Dependence on the operating point: The noise voltage density of the equivalent source
vD,r is inversely proportional to the transconductance;
in other words, it declines when
the
transconductance increases, such that gm = 2K ID,A leads to |v D,r (f )|2 ∼ 1/ ID,A .
For the noise current source iG,r , this means that |i G,r (f )|2 ∼ gm ∼ ID,A , so that the
noise current density increases with the transconductance. The noise of the gate resistance
RG is independent of the operating point.
Example: For a FET with K = 0.5 mA/V2 , RG = 100 and fT = 100 MHz we
obtain, for an operating point current ID,A = 1 mA, the transconductance gm = 1 mA/V
and thus, in the medium frequency range, that is, fg(1/f ) < f < fY 21s , the frequency-
√
independent noise voltage densities |v RG,r (f )| = 1.3 nV/ Hz and |v D,r (f )| =
√ √
3.3 nV/ Hz and the noise current density |i G,r (f )| = 2 pA/ Hz · f/fT , which is
√
proportional to the frequency; for f = 1 kHz, |i G,r (f )| = 0.02 fA/ Hz. For JFETs,
the 1/f cutoff frequency is relatively low, with fg(1/f ) ≈ 100 Hz; thus the calculated
value for |v D,r (f )| applies to a relatively wide frequency range of fg(1/f ) ≈ 100 Hz
to fY 21s = fT /(gm RG ) = 1 GHz. In contrast, fg(1/f ) ≈ 1 MHz applies to MOS-
FETs; due to the 1/f √ noise we thus achieve with f = 1 kHz a much higher value of
|v D,r (f )| = 105 nV/ Hz compared to the JFET.
F
Rg = 1MΩ Rg = 1k Ω
100
30 1 2 2
~ ~ f ~f
f
10 1
~
f
1
~
~
1 1k 1M 1G f
Hz
Fig. 3.54. Noise figure curve for a MOSFET with gm = 1 mA/V, RG = 100 , fT = 100 MHz
and fg(1/f ) = 1 MHz for Rg = 1 k und Rg = 1 M
Rg RG Rg2
≈ |v r,g (f )|2 + |v RG,r (f )|2 + 1 + |v D,r (f )|2
|Z k |2
RG 1/gm Rg 1/gm
RG 2 1 2 1
F ≈ 1+ + ≈ 1+ ≈ 1 (3.48)
Rg 3 gm Rg 3 gm Rg
With the gate resistance RG small and the source resistance Rg high compared to the
inverse value of the transconductance, we obtain the optimum noise factor F = 1 in
this area.
– With low frequencies the 1/f noise dominates; the noise figure in this area is inversely
proportional to the frequency:
3.3 Models of Field Effect Transistors 227
2 1 fg(1/f )
F (f ) ≈
3 gm Rg f
The boundary of the area for medium frequencies is at:
RG 1/gm Rg 1/gm
fg(1/f ) 2 fg(1/f )
f1 ≈ ≈
3 3 gm R g
1 + gm Rg
2
The noise figure and the cutoff frequency f1 are inversely proportional to the source re-
sistance Rg ; therefore the 1/f portion of the noise figure declines as the source resistance
increases (see Fig. 3.54).
– With high frequencies, the noise figure increases proportionally with the square of the
frequency:
2
2 f
F (f ) ≈ gm Rg
3 fT
The boundary of the area for medium frequencies is at:
Rg 1/gm
3 1 3 fT fT
f2 ≈ fT 1 + ≈ ≈
2 gm R g 2 gm R g gm R g
The noise figure increases with an increasing source resistance Rg , while the cutoff
frequency f2 decreases accordingly (see Fig. 3.54).
For JFETs, the 1/f cutoff frequency, and thus also the 1/f portion of the noise figure, is lower
than in MOSFETs by three or four orders of magnitude; therefore, with source resistances
in the M range the 1/f portion is virtually unnoticeable, since the cutoff frequency f1
drops below 1 Hz.
Minimizing the noise figure: Under certain conditions, the noise figure is reduced to a
minimum. With a given source resistance Rg , the optimum transconductance and thus the
optimum drain current at the operating point can be determined by evaluating:
∂F (f )
= 0
∂gm
It must be noted that according to (3.47) the transit frequency fT is proportional to the
transconductance: fT = gm /(2πC) with C = CGS + CGD ; inserting this leads to:
RG 2 1 1 fg(1/f ) fg(1/f )
F (f ) = 1 + + 1+ + 4π C Rg f f +
2 2
Rg 3 gm R g f 2
It is obvious that F (f ) declines with an increasing transconductance; this means that there
is no optimum, so that low-noise FET amplifiers must be operated with the highest possible
transconductance or the highest possible drain current.
The optimum source resistance Rgopt can be calculated by way of
∂F (f )
= 0
∂Rg
and by limiting to fg(1/f ) < f < fY 21s :
gm RG 1
fT 3 fT 1
Rgopt (f ) ≈ 1 + gm RG ≈ =
gm f 2 gm f 2πf (CGS + CGD )
228 3 Field Effect Transistor
Wide-band matching is not possible due to the frequency dependence of Rgopt . Inserting
Rgopt in F (f ) leads to the optimum spectral noise figure Fopt (f ). An approximation
is [3.6]:
Rg RG
RG 4 f 4 f
Fopt (f ) ≈ 1 + + ≈ 1+
Rg 3 fT 3 fT
Simplified Description
A simplified description is usually sufficient for practical applications. For this purpose,
the gate spreading resistance and the correlation between the channel noise and the induced
gate noise are neglected, and the frequency range is limited to medium frequencies; that
is, fg(1/f ) < f < fY 21s . An increase in the gate noise current density by a factor of 2 is
assumed in order to achieve an approximate compensation of the portion induced by the
correlation; for fg(1/f ) < f < fY 21s it follows that:
8
|i D,r (f )|2 ≈ kT gm
3
2
8 f
|i G,r (f )| ≈ kT gm
2
3 fT
With y 21,s (s) ≈ gm , this leads to the equivalent noise sources:
|i D,r (f )|2 8 kT
|v r,0 (f )|2 = ≈
|y 21,s (j 2πf )| 2 3 gm
2
8 f
|i r,0 (f )|2 = |i G,r (f )|2 ≈ kT gm
3 fT
Figure 3.55 shows the simplified small-signal model with equivalent noise sources.
The noise figure is:
2
|v r,0 (f )|2 + Rg2 |i r,0 (f )|2 2 1 f
F (f ) = 1 + ≈ 1+ + g m Rg
|v r,g (f )| 2 3 gm R g fT
With the exception of the missing 1/f portion, this yields the same values as in the detailed
calculation (see Fig. 3.54). From (∂F )/(∂Rg ) = 0, it follows for the optimum source
resistance Rgopt and the optimum noise figure Fopt that:
fT 1
Rgopt (f ) ≈ =
gm f 2πf (CGS + CGD )
v r,0
G' CGD
G D
Fig. 3.55. Simplified small-signal model of a FET with equivalent noise sources
3.4 Basic Circuits 229
4 f
Fopt (f ) ≈ 1 +
3 fT
These values coincide with the approximations of the detailed calculation, because the
assumption of a higher gate noise current density exactly replaces the portion caused by
the correlation.
3.4
Basic Circuits
Basic circuits using field effect transistors: There are three basic circuits in which FETs
can be used: the common-source circuit, the common-drain circuit and the common-gate
circuit. The name reflects the terminal of the FET that is connected as a common reference
point for the input and output of the circuit; Fig. 3.56 illustrates this relationship using an
enhancement n-channel MOSFET.
In many circuits this criterion does not apply in a strict sense, so that a less stringent
criterion must be used:
The designation reflects the terminal of the FET that is connected to form neither
the input nor the output of the circuit.
The substrate or bulk terminal has no influence on the classification of the basic circuits,
but influences their behavior. It is connected to the source terminal in discrete MOSFETs
and to ground or to a supply voltage source (= small-signal ground) in integrated circuits;
for the common-source circuit, both versions are identical, since in this case the source
terminal is connected to (small-signal) ground.
Basic circuits with several FETs: There are several configurations using two or more
FETs. They are so common that they too must be regarded as basic circuits, which serve,
230 3 Field Effect Transistor
Vo Vo Vo
Vi Vi
Vi
for example, as differential amplifiers or current mirrors; such circuits are described in
Sect. 4.1.
Polarity: n-channel MOSFETs are preferred in all of these configurations, as they feature
a higher transconductance coefficient than p-channel MOSFETs with the same channel
dimensions due to the higher mobility of charge carriers. Furthermore, enhancement MOS-
FETs are more common than depletion MOSFETs; this is particularly true for integrated
circuits. With respect to the small-signal response, there is no principal difference between
depletion MOSFETs and JFETs on the one hand and enhancement MOSFETs on the
other; only their operating point setting is different. All circuits can also be designed using
the respective p-channel FETs; but then the polarity of all supply voltages, electrolytic
capacitors and diodes must be reversed.
3.4.1
Common-Source Circuit
Figure 3.57a shows the common-source circuit, consisting of the MOSFET, the drain
resistance RD , the supply voltage source Vb and the signal voltage source Vg with the
internal resistance Rg . In what follows, we assume Vb = 5 V, RD = 1 k and K =
4 mA/V2 , Vth = 1 V for the MOSFET.
Vb
RD RD
Io
Rg IG = 0
Rg Vo Vo
VGS ID
Vg Vb Vg Vi
1
Vth
0 1 2 3 4 5 Vi
V
Operation in the pinch-off region: Figure 3.57b shows the equivalent circuit. Disre-
garding the Early effect leads to:
K
ID = (VGS − Vth )2
2
With Vg = Vi = VGS , the output voltage is:
Io = 0 RD K
Vo = VDS = Vb − ID RD = Vb − (Vi − Vth )2 (3.49)
2
The internal resistance Rg has no effect on the characteristic of a MOSFET due to
IG = 0; it only influences the dynamic behavior. But in JFETs there are gate leakage
currents in the pA and nA range that cause voltage drops across high internal resistances that
can no longer be neglected; MOSFETs are thus preferred for sources with Rg > 10 M.
A point at the centre of the declining segment of the transfer characteristic is selected as
the operating point; this allows maximum output signals. Setting Vb = 5 V, RD = 1 k,
K = 4 mA/V2 and Vth = 1 V for the operating point shown as an example in Fig. 3.58
results in:
Vb − Vo 2ID
Vo = 3 V ⇒ ID = = 2 mA ⇒ Vi = VGS = Vth + = 2V
RD K
232 3 Field Effect Transistor
Boundary of the ohmic region: With Vo = Vo,po = VDS,po , the MOSFET reaches
the boundary of the ohmic region. For VDS,po = VGS − Vth and Vi = VGS , the relation
Vo = Vi − Vth follows; inserting this into (3.49) leads to
1 # $ 2Vb RD K1 2Vb 1
Vo,po = 1 + 2Vb RD K − 1 ≈ −
RD K RD K RD K
and Vi,po = Vo,po + Vth . With the numeric example we obtain Vo,po = 1.35 V and
Vi,po = 2.35 V.
If the supply voltage is given, the product RD K must be increased in order to reduce
Vo,po and to increase the output voltage range. In practice, the control range is always
narrower than in the common-emitter configuration because a bipolar transistor can be
controlled down to VCE,sat ≈ 0.1 V almost independent of the external circuitry.
A =
= − gm (RD || rDS ) ≈ − g m RD (3.50)
vi io =0
vi
ri = = ∞ (3.51)
ii
rDS RD
vo
ro = = RD || rDS ≈ RD (3.52)
io
If K = 4 mA/V2 and VA = 50 V, then gm = 2K ID,A = 4 mS, rDS = VA /ID,A =
25 k, A = − 3.85 and ro = 960 . For comparison, at the same operating point of
the common-emitter circuit described in Sect. 2.4.1, with IC,A = ID,A = 2 mA and
Rg iD io
Rg ii ro io
vg vi ri Avi vo
RL
RC = RD = 1 k, the gain is A = − 75. The reason for the low gain of the MOSFET
is the low transconductance at the same current: gm = 4 mA/V for the MOSFET and
gm = 77 mA/V for the bipolar transistor.
The parameters A, ri and ro provide a complete description of the common-source
circuit; Fig. 3.60 shows the corresponding equivalent circuit. The load resistance RL may be
an ohmic resistance or an equivalent element for the input resistance of a circuit connected
to the output. It is important that the operating point is not shifted by RL , which means
that no or only a negligible DC current may flow through RL .
Using Fig. 3.60, it is possible to calculate the small-signal operating gain:
vo ri RL ri →∞ RL
AB = = A = A (3.53)
vg ri + R g RL + r o RL + r o
It consists of the circuit gain A and the voltage divider factor at the output.
Nonlinearity: In Sect. 3.1.4 the distortion factor k of the drain current is calculated
for sinusoidal signals with v̂i = v̂GS (see (3.13) on page 185); in the common-source
circuit it %is identical to
& the distortion factor of the output voltage
% vo . This& means that
v̂i < 4k VGS,A − Vth ; that is, for k < 1% we need v̂i < VGS,A − Vth /25. Using
VGS,A − Vth = 1 V in our numeric example, we obtain v̂i < 40 mV. Since v̂o = |A|v̂i ,
the corresponding output amplitude depends% & gain A; with A = − 3.85 in the
on the
above example, we obtain v̂o < 4k|A| VGS,A − Vth = k · 15.4 V. For comparison, in the
common-emitter circuit in Sect. 2.4.1 v̂o < k ·7.5 V, which means that the common-source
circuit achieves a higher output amplitude for the same distortion factor.
234 3 Field Effect Transistor
dID
1 dK 2 dVth
= − R = − I R −
dT
dT
D D,A D
A A K dT V −V
GS,A th dT
−3 −1 4...7V
≈ ID,A RD · 10 K 5−
VGS,A − Vth
The result from our numeric example is that (dVo /dT )|A ranges from approximately
≈ − 4 to +2 mV/K. The temperature drift is low, as the MOSFET operates close to the
temperature compensation point (see Sect. 3.1.7).
A temperature drift comparison between common-source and common-emitter circuits
makes sense only in respect to the gain; (dVo /dT )|A ranges from approximately − 1 to
+0.5 mV/K · |A| for the common-source circuit and (dVo /dT )|A ≈ − 1.7 mV/K · |A| for
the common-emitter circuit. This suggests that with the same gain the drift is lower in the
common-source circuit, especially if the operating point is located close to the temperature
compensation point.
Operation in the pinch-off region: Figure 3.61b shows the equivalent circuit diagram;
for Io = 0 the voltages in the pinch-off region are:
RD K
Vo = Vb − ID RD = Vb − (VGS − Vth )2 (3.54)
2
Vi = VGS + VS = VGS + ID RS (3.55)
For the sample operating point shown in Fig. 3.62 we obtain, with Vb = 5 V, K =
4 mA/V2 , RD = 1 k and RS = 200 , for the discrete MOSFET:
3.4 Basic Circuits 235
Vb Vb
RD RD
Io
Rg
VB
Rg
Vo VGS ID Vo
Vi
Vg
Vg RS VS RS
Vb − Vo
Vo = 3.5 V ⇒ ID = = 1.5 mA ⇒ VS = ID RS = 0.3 V
RD
2ID
⇒ VGS = Vth + = 1.866 V ⇒ Vi = VGS + VS = 2.166 V
K
For the integrated MOSFET it is necessary to take into consideration the dependence
BS , according to (3.18) on page 199. For the MOSFET we
of the threshold voltage on V√
assume Vth,0 = 1 V, γ = 0.5 V and Vinv = 0.6 V; consequently:
VBS = − VS = − 0.3 V
# $
⇒ Vth = Vth,0 + γ Vinv − VBS − Vinv ≈ 1.087 V
2ID
⇒ VGS = Vth + = 1.953 V ⇒ Vi = VGS + VS = 2.253 V
K
4 VB = 0
3
V BS = 0
2
1
Vth
0 1 2 3 4 5 Vi
V
Fig. 3.62. Transfer characteristic of the common-source circuit with current feedback for a discrete
MOSFET (VBS = 0) and an integrated MOSFET (VB = 0); the borderline between the pinch-off
and ohmic regions applies to the discrete MOSFET
236 3 Field Effect Transistor
Small-signal response: The small-signal equivalent circuit shown in Fig. 3.63 is used
for calculations. The nodal equation
vDS vo
gm vGS + gm,B vBS + + = 0
rDS RD
vo
gm RD
A = = −
vi
io =0 RD 1
1+ + gm + gm,B + RS
rDS rDS
rDS RD ,1/gm
gm RD
≈ − % &
1 + gm + gm,B RS
vBS =0 gm RS 1
gm RD RD
= − ≈ −
1 + g m RS RS
For a discrete MOSFET, that is, without a substrate effect (vBS = 0), and a high feedback
(gm RS 1), the gain only depends on RD and RS . However, because of the low maximum
gain of a MOSFET the feedback cannot be high, as otherwise the gain will become too low;
therefore the condition gm RS 1 is rarely met in practice. When using a load resistance
RL , the operating gain AB can be calculated by replacing RD by the parallel arrangement
of RD and RL (see Fig. 3.63). At the chosen sample operating point, the gain of the discrete
MOSFET with gm = 3.46 mS, rDS = 33 k, RD = 1 k and RS = 200 is exactly
A = − 2.002; the two first approximations yield A = − 2.045 and the third approximation √
cannot be used because gm RS < 1. For the integrated MOSFET, we assume γ = 0.5 V
and Vinv = 0.6 V; it follows from (3.41) that gm,B = 0.91 mS and we have the exact value
A = − 1.812; the first approximation yields A = − 1.846.
The input resistance is ri = ∞ and the output resistance:
rDS RD
1
ro = RD || rDS 1 + gm + gm,B + RS ≈ RD
rDS
For rDS RD , 1/gm and with no load resistance RL , we obtain for the:
Rg ii iD io
G D
vGS gm v GS rDS
gm ,B vBS
vi vo
vg S RD RL
vBS
vS RS
B
Fig. 3.63. Small-signal equivalent circuit for the common-source circuit with current feedback
3.4 Basic Circuits 237
vo
g m RD vBS =0 gm RD
A =
≈ − % & = − (3.56)
vi io =0 1 + gm + gm,B RS 1 + g m RS
ri = ∞ (3.57)
vo
ro = ≈ RD (3.58)
io
Nonlinearity: The current feedback also reduces the nonlinearity of the transfer char-
acteristic. The distortion factor of the circuit can be approximately determined by a series
expansion of the characteristic at the operating point. From (3.55) it follows that:
2ID
Vi = VGS + ID RS = Vth + + ID RS
K
By inserting the values for the operating point, changing to the small-signal values and
carrying out a series expansion, we achieve with (3.18), for VBS = − VS = − ID RS
R S iD
vi = γ Vinv + ID,A RS 1+ −1
Vinv + ID,A RS
2ID,A iD
+ 1+ − 1 + R S iD
K ID,A
1 % % & & 1 gm,B RS2 1
= 1 + gm + gm,B RS iD + + i + ···
2
gm 4 Vinv + ID,A RS ID,A D
and by inverting the series:
⎛ ⎞
gm gm gm,B RS2
⎜ + ⎟
gm ⎜ vi2 ID,A Vinv + ID,A RS ⎟
iD = % & ⎜vi + % % & &2 + · · ·⎟
1 + gm + gm,B RS ⎝ 4 1 + gm + gm,B RS ⎠
238 3 Field Effect Transistor
For input signals with vi = v̂i cos ωt, the distortion factor k is calculated approximately
from the ratio of the first harmonic with 2ωt to the fundamental with ωt at a low amplitude,
that is, by neglecting the higher powers:
gm gm gm,B RS2
+
ua,2ωt iD,2ωt v̂i ID,A Vinv + ID,A RS
k ≈ ≈ ≈ % % & &
ua,ωt iD,ωt 8 1 + gm + gm,B RS 2
vBS =0 v̂i
= % & (3.60)
4 VGS,A − Vth (1 + gm RS )2
The last expansion uses gm /ID,A = 2/(VGS,A − Vth ). With the numeric example, we have
v̂i < k · 11.5 V and, if A ≈ − 2, v̂o < k · 23 V.
A comparison with (3.13) shows that due to the feedback the permissible input am-
plitude v̂i is increased by the square of the feedback factor (1 + gm RS ). But since the
feedback simultaneously reduces the gain by the feedback factor, the permissible output
amplitude is increased by the feedback factor if the distortion factor is the same. If the
output amplitude is the same, then the distortion factor is reduced by the feedback factor.
A comparison with the common-emitter circuit with current feedback, described in
Sect. 2.4.1, shows that with the same gain (A ≈ − 2) and the same operating point current
(ID,A = IC,A = 1.5 mA) the distortion factor is higher: k ≈ v̂o /(23 V) in the common-
source circuit and k ≈ v̂o /(179 V) in the common-emitter circuit. The reason for this is
the low maximum gain of the MOSFET which, for the same circuit gain, results in a lower
feedback factor and therefore in a higher distortion factor. For very low operating point
currents, the maximum gain of the MOSFET increases and the distortion factor decreases;
under these conditions the same values are achieved as with the common-emitter circuit.
A special situation arises for cubic distortions. Due to the almost square characteristic
of the MOSFET without feedback, these distortions are very low and increase with the
feedback, while the dominating square distortions and thus the distortion factor k decrease
when the feedback increases. Figure 3.64 shows the dependence of the distortion factor k
k k3
% %
5 k3 0.10
4 0.08
3 0.06
k
2 0.04
1 0.02
Fig. 3.64. A plot of the distortion factor k and the cubic distortion factor k3 over the feedback
resistance RS with a constant amplitude at the output for the circuit shown in Fig. 3.61a
3.4 Basic Circuits 239
and the cubic distortion factor k3 on the feedback resistance RS with a constant amplitude
at the output. The data for this plot were determined by simulation with PSpice.
ID,A RD −3 −1 4...7V
≈ % & · 10 K 5−
dT
A 1 + gm + gm,B RS VGS,A − Vth
For the numeric example, we obtain the result that (dVo /dT )|A ranges from approximately
−3 mV/K to 0.4 mV/K.
Vb Vo
Pinch-off
V
RD region
5
R2 Io 4
3
ID
2
R1 Vo
1
VGS
Vi
–1 0 1 2 3 4 5 Vi
V
a Circuit b Characteristic
of the operating point voltage VGS,A for a given output voltage Vo,A . As an alternative,
an approximation can be used in which the current through the feedback resistance R2 is
neglected; with Vo,A = 2.5 V we obtain:
Vb − Vo,A
Vo,A = 2.5 V ⇒ ID,A ≈ ≈ 2.5 mA
RD
(3.62)
2ID,A
⇒ VGS,A = Vth + ≈ 2.12 V ⇒ Vi,A ≈ 2.06 V
K
Small-signal response: The calculation is carried out by means of the small-signal equiv-
alent circuit shown in Fig. 3.66. From the nodal equations
vi − vGS vo − vGS
+ = 0
R1 R2
vo − vGS vo vo
gm vGS + + + = io
R2 rDS RD
= R || r
it follows, with RD D DS , that:
rDS RD
vo
− gm R2 + 1 R1 ,R2 1/gm R2
A = = ≈ −
vi
io =0 R1 + R2 R1 + R2
1 + g m R1 +
R1 +
RD gm RD
If the gain without feedback is much higher than the feedback factor, that is, gm RD
1 + R2 /R1 , then A ≈ −R2 /R1 ; however, due to the low maximum gain of the FET this
condition is rarely met. Operating the circuit with a load resistance RL allows calculation
of the relevant operating gain AB by replacing RD with the parallel arrangement of RD and
RL (see Fig. 3.66). For the chosen sample operating point we obtain, with gm = 4.47 mS,
rDS = 20 k, RD = R1 = 1 k and R2 = 6.3 k, an exact value of A = − 2.067; the
approximation yields A = − 2.39.
The open-circuit input resistance is with RD
= R || r
D DS :
rDS RD 1/gm
vi
R2 + RD 1 R2
ri,o =
= R1 +
≈ R1 + 1+
ii io =0 1 + gm RD gm RD
This applies to io = 0; that is, RL → ∞. For other values of RL the input resistance can be
calculated by replacing RD with the parallel arrangement of RD and RL . For the sample
operating point selected we obtain an exact value of ri,o = 2.38 k; the approximation
yields ri,o = 2.63 k.
ii R1 R2 iD io
Fig. 3.66. Small-signal equivalent circuit for the common-source circuit with voltage feedback
3.4 Basic Circuits 241
= R || r
The short-circuit output resistance is with RD D DS :
rDS RD
vo
R1 + R 2 R1 1/gm 1 R2
ro,s =
= RD || ≈ RD || 1+
io vi =0 1 + g m R1 gm R1
For R1 → ∞, this leads to the open-circuit output resistance:
vo
1 rDS RD 1/gm 1
ro,o = = R || ≈
io
ii =0 D
gm gm
For the sample operating point selected, we obtain exact values of ro,s = 556 and
ro,o = 181 , while the approximation yields ro,s = 602 and ro,o = 223 .
The summarized equations for the common-source circuit with voltage feedback are:
common-source circuit with voltage feedback
vo
R2
A = ≈ − (3.63)
vi
io =0 R1 + R2
R1 +
gm RD
vi
1 R2
ri = ≈ R1 + 1+ (3.64)
ii
io =0 gm RD
vo
1 R2
ro = ≈ RD || 1+ (3.65)
io
vi =0 gm R1
Vb Vo
V
RD
5
R2 Io 4
3
ID
2
Vo
1
Ii VGS
–1 0 1 Ii
mA
a Circuit b Characteristic
24 The term transimpedance amplifier is also used for operational amplifiers with current input and
voltage output (CV-OPA).
242 3 Field Effect Transistor
Inserting the equations into one another leads to a quadratic equation in Vo and Ii , the
resolution of which is very extensive. When assuming |Ii RD | Vb − Vo and defining
Vo , VGS can be calculated from (3.66) and Ii from (3.66); for Vth = 1 V, K = 4 mA/V2 ,
RD = 1 k and R2 = 6.3 k, we obtain:
2 (Vb + Ii RD − Vo ) |Ii RD |Vb −Vo
Vo = 2.5 V ⇒ VGS ≈ Vth + ≈ 2.12 V
KRD
VGS − Vo K
⇒ Ii = ≈ − 60 mA und ID = (VGS − Vth )2 ≈ 2.509 mA
R2 2
With the iterative approach, one inserts the last value of Ii into (3.66) and calculates the
new values of VGS and Ii ; the next iteration yields VGS ≈ 2.105 V, Ii ≈ − 63 mA and
ID ≈ 2.44 mA, which represent almost the exact results.
The small-signal response of the current–to–voltage converter can be derived from the
equations for the common-source circuit with voltage feedback. The transfer resistance
(transimpedance) RT now takes the place of the gain; with RD
= R || r
D DS , we obtain:
vo
vo
RT =
= lim R1
= lim R1 A
ii io =0 R1 →∞ vi io =0 R1 →∞
gm R2 1
1 − g m R2 rDS RD gm R D
= RD
≈ − R2
1 + gm RD 1 + g m RD
The input resistance can be calculated using the equations for the common-source circuit
with voltage feedback by inserting R1 = 0, and the output resistance corresponds to the
open-circuit output resistance.
The summarized equations for the current–to–voltage converter in common-source
configuration are:
vo
gm R D
RT = ≈ − R2 (3.68)
ii
io =0 1 + g m RD
vi
1 R2
ri = ≈ 1 + (3.69)
ii
io =0 gm RD
vo 1
ro = ≈ RD || (3.70)
io gm
3.4 Basic Circuits 243
At the sample operating point selected, and for ID,A = 2.44 mA, K = 4 mA/V2 ,
RD = 1 k and R2 = 6.3 k, we obtain the values RT ≈ − 5.14 k, ri ≈ 1.65 k and
ro ≈ 185 .
The current–to–voltage converter is mostly used in photodiode receivers; the receiving
diode is operated in reverse-biased mode and therefore acts like a current source with a
very high internal resistance. The current–to–voltage converter transforms the current ii
into the voltage vo = RT ii . Due to the high internal resistance of the diode, the noise of
the circuit is mainly caused by the input noise current of the FET and the thermal noise of
the feedback resistance R2 ; the low input noise current of the FET, which is particularly
low compared to the bipolar transistor, results in a notably low noise figure.
Setting the operating point for AC voltage coupling: In AC voltage coupling, the am-
plifier is connected to the signal source and the load via coupling capacitors. For voltage am-
plifiers the voltage setting with DC feedback is generally used as shown in Fig. 3.68a; this
corresponds to the operating point setting in common-emitter circuits shown in Fig. 2.75a.
The variations of Figs. 2.75b and 2.78 can also be used with the FET; here, the extremely
high input resistance of the FET is only effective with direct coupling at the input, since
otherwise the voltage divider at the input determines the input resistance of the circuit.
Vb Vb Vb Vb
R1 RD R1 RD
Co Co
R3
Ci Ci
Vo Ck Vo
Vi Vi
R2 RS CS R2 RS
Vb Vb
RD RD
Co R2 Co
Ci Ci
R2 RS CS
A special case is current feedback with the bootstrap shown in Fig. 3.68b, which
reduces the voltage drop across R3 by feeding the signal back to the voltage divider so that
the input resistance increases accordingly: ri ≈ R3 (1 + gm RS ). This circuit only works
efficiently with high feedback (gm RS 1) and is therefore predominantly used in drain
circuits (see Sect. 3.4.2).
In addition, there are special circuits that are suitable for setting the operating point of
depletion FETs only. Since these can be operated with VG = 0, it is possible to remove the
resistance R1 from Fig. 3.68a, leading to the circuit shown in Fig. 3.69a; the same applies
to the bootstrap configuration. From the condition VGS = − ID RS and from the equation
for the pinch-off region, we obtain the calculation:
|Vth | 2ID,A |Vth | ID,A
RS = 1− = 1−
ID,A K Vth2 ID,A ID,A(max)
Here, ID,A(max) = K Vth2 /2 is the maximum possible operating point current. If the FET
is to be operated at the temperature compensation point with VGS,TC ≈ Vth + 1 V, then
ID,A ≈ K · 0.5 V2 and, consequently,
2|VGS,TC | |VGS,TC |
RS = % &2 ≈ for VGS,TC ≤ 0
K VGS,TC − Vth K · 0.5 V2
Enhancement MOSFETs can be operated in the pinch-off region with VGS = VDS (see
Fig. 3.69b); since there is no gate current, or only a very small one, resistance R2 can be so
high that the voltage feedback caused by R2 is negligible; in this case, the input resistance
can be determined from (3.69) on page 242.
The properties, advantages and disadvantages of AC voltage coupling are described in
detail on page 120 in respect to the common-emitter circuit.
Setting the operating point with DC voltage coupling: In DC voltage coupling, also
known as direct or galvanic coupling, the amplifier is connected directly to the signal
source and the load. The DC voltages at the input and output of the amplifier must be
adapted to the DC voltages of the signal source and load; therefore it is not possible, for
multi-stage amplifiers, to adjust the operating point for each stage separately.
3.4 Basic Circuits 245
Common-source circuit without feedback: Figure 3.70 shows the dynamic small-
signal equivalent circuit for the common-source circuit without feedback. With Rg
=
Rg + RG and RD
= R || R || r
L D DS , the operating gain AB (s) = v o (s)/v g (s) can be
expressed by:
(gm − sCGD ) RD
AB (s) = − (3.71)
1 + sc1 + s 2 c2
# $
c1 = CGS Rg
+ CGD Rg
+ RD
+ gm RD Rg
+ CDS RD
As with the common-emitter circuit, the frequency response can be described approxi-
mately by a first-degree lowpass filter, if the zero is neglected and the s 2 term is removed
from the denominator. With the low-frequency gain
A0 = AB (0) = − gm RD (3.72)
Rg RG CGD
Fig. 3.70. Dynamic small-signal equivalent circuit for the common-source circuit without
feedback
246 3 Field Effect Transistor
it follows that:
A0
AB (s) ≈ # # $ $ (3.73)
1+s CGS Rg
+ CGD Rg
+ RD
+ gm RD Rg
+ CDS RD
This gives an approximation for the −3 dB cutoff frequency f-3dB at which the magnitude
of the gain has dropped by 3 dB:
1
ω-3dB = 2πf-3dB ≈ # $ (3.74)
CGS Rg
+ CGD Rg
+ RD
+ gm RD Rg
+ CDS RD
, R
1/g ; this leads to:
In most cases RD g m
1
ω-3dB = 2πf-3dB ≈ (3.75)
CGS Rg
+ CGD gm RD Rg
+ CDS RD
1
GBW = f-3dB |A0 | ≈ (3.79)
2π T2
R'g ro
vi Ci Co vo CL
vg Avi RL
CGD + CDS + CL
T2 = CGD Rg
+ (3.80)
gm
The input capacitance Ci depends on the circuitry at the output, because A0 depends on
RL . The fact that CGD influences Ci with the factor (1 + |A0 |) is known as the Miller
effect and CGD is called the Miller capacitance. A and ro are given by (3.50) and (3.52),
and are independent of the circuitry. The gate spreading resistance RG is regarded as a
portion of the internal resistance of the signal source: Rg
= Rg + RG .
Example: ID,A = 2 mA was selected for the numeric example used in the common-
source circuit without feedback, according to Fig. 3.57a. For K = 4 mA/V2 , VA = 50 V,
Coss = 5 pF, Crss = 2 pF, fY 21s = 1 GHz and fT = 100 MHz, we obtain from Fig. 3.51
on page 222 the small-signal parameters gm = 4 mS, rDS = 25 k, RG = 25 , CGD =
2 pF, CGS = 4.4 pF and CDS = 3 pF. For Rg = RD = 1 k, RL → ∞ and Rg
≈ Rg , it
follows from (3.72) that A0 ≈ − 3.85, from (3.74) that f-3dB ≈ 8.43 MHz, and from (3.75)
that f-3dB ≈ 10.6 MHz. It follows from (3.77) that T1 ≈ 6.4 ns, from (3.78) that T2 ≈
3.25 ns, and from (3.79) that GBW ≈ 49 MHz. With a load capacitance CL = 1 nF we
obtain from (3.80) T2 ≈ 253 ns, from (3.76) f-3dB ≈ 162 kHz and from (3.79) GBW
≈ 630 kHz.
A comparison with the values for the common-emitter circuit described on page 126
makes sense only with regard to the gain–bandwidth product, as the low-frequency gains
are very different. It can be seen that for the common-source circuit without capacitive
load approximately the same GBW is reached as for the common-emitter circuit. However,
if a capacitive load is present, the GBW of the common-source circuit is markedly lower;
namely, in the borderline case of large load capacitances it is reduced by exactly the ratio
of the transconductances, as can be seen by a comparison of (3.80) and (2.96) on page 125.
In practice, this means that:
Due to the low transconductance of the FETs, the common-source circuit is poorly
suited for driving capacitive loads.
248 3 Field Effect Transistor
Common-source circuit with current feedback: The frequency response and the cutoff
frequency of the common-emitter circuit with current feedback, according to Fig. 3.61a,
can be derived from the corresponding parameters of the common-source circuit without
feedback. For this purpose, we shall use the conversion of the small-signal equivalent circuit
already carried out for the common-emitter circuit with current feedback (see Fig. 2.86 on
page 127). Figure 3.72 shows the small-signal equivalent circuit for the common-source
configuration with current feedback before and after conversion; here, the small-signal
parameters are converted into the equivalent values of a FET without current feedback:
⎡
⎤ ⎡ ⎤ ⎡ ⎤
gm gm gm
⎢
⎥ ⎢ ⎥ ⎢ ⎥
⎢CGS ⎥ ⎢CGS ⎥ ⎢CGS ⎥
⎢ ⎥ 1 ⎢ ⎥ vBS =0 1 ⎢ ⎥
⎢C
⎥ = % & ·⎢ ⎥ = ·⎢ ⎥ (3.83)
⎢ DS ⎥ 1 + gm + gm,B RS ⎢ ⎢ C DS ⎥ + ⎢ C DS ⎥
⎢ ⎥ ⎥ 1 g m S ⎢
R ⎥
⎣ 1 ⎦ ⎣ 1 ⎦ ⎣ 1 ⎦
The transconductance gm
corresponds to the reduced transconductance g
m,red that has
already been introduced in (3.59). The gate–drain capacitance CGD remains unchanged.
Now one can insert the equivalent values into (3.72) and (3.76)–(3.78) or (3.80) for the
common-source circuit without feedback; with Rg
= Rg + RG and RD
= r
|| R || R ,
DS D L
it follows:
1
ω-3dB (A0 ) ≈ (3.84)
T1 + T2 |A0 |
%
&
T1 = CGS + CGD Rg
(3.85)
CGD + CDS + CL
T2 = CGD Rg
+
(3.86)
gm
A0 = − gm RD (3.87)
Rg RG CGD
R'D = r'DS II RD II RL
R'g = Rg + RG CGD
Fig. 3.72. Dynamic small-signal equivalent circuit for the common-source circuit with current
feedback prior to (above) and after (below) conversion
3.4 Basic Circuits 249
From (3.86) it follows that a strong current feedback has a comparable high effect even
< g ; the
for a small capacitance CL , since T2 increases relatively strongly due to gm m
gain–bandwidth product GBW declines accordingly.
Example: ID,A = 1.5 mA is selected for the numeric example of the common-source
circuit with current feedback, as shown in Fig. 3.61a. For K = 4 mA/V2 and VA = 50 V,
we obtain from Fig. 3.51 on page 222 the parameter gm = 3.46 mS and rDS = 33.3 k.
The parameters RG = 25 , CGD = 2 pF, CGS = 4.4 pF and CDS = 3 pF are taken from
the example on page 247,25 and rDS is disregarded. For RS = 200 , the conversion to
= 2.04 mS, C
= 2.6 pF, C
Common-source circuit with voltage feedback: Figure 3.73 shows the small-signal
equivalent circuit in which the gate resistance RG of the FET is neglected. The results from
the common-emitter circuit with voltage feedback can be transferred to the common-source
circuit with voltage feedback if one takes into consideration the fact that the capacitance
CDS acts like a load capacitance. From (2.102), it follows, with R1
= R1 + Rg and
=r
RD DS || RD || RL , that:
1+R /R
gm RD
R2 2 1 R2
A0 ≈ −
≈ − (3.88)
R + R2 R1
R1
+ 1
gm RD
and, from (2.105)–(2.107):
1
ω-3dB (A0 ) ≈ (3.89)
T1 + T2 |A0 |
CGS + CDS + CL
T1 = (3.90)
gm
R2
R'1 = R1 + Rg
R'D = rDS II RD II RL
Rg R1 CGD
Fig. 3.73. Small-signal equivalent circuit for the common-source configuration with voltage
feedback
CGS
CDS + CL
T2 =
+ C GD R1 + (3.91)
gm RD gm
Strong voltage feedback can cause complex conjugate poles; in this case, the cutoff fre-
quency can be estimated only very roughly by (3.89)–(3.91).
The common-source circuit with voltage feedback can also be approximately described
by the equivalent circuit shown in Fig. 3.71; by analogy to the common-emitter circuit
with voltage feedback, and by taking into consideration the additional capacitance CDS
that occurs at the output, this leads to:
Ci = 0
1 1 % &
Co ≈ CGS +
+ CGD gm R1
|| R2 + CDS
R2 RD
Therefore, the input impedance is a purely ohmic resistance. A, ri and ro are given by
(3.63)–(3.65).
Example: ID,A = 2.5 mA was selected for the numeric example used in the common-
source circuit with voltage feedback, according to Fig. 3.65; for K = 4 mA/V2 and
VA = 50 V, we obtain, from Fig. 3.51 on page 222, gm = 4.47 mS and rDS = 20 k. The
parameters RG = 25 , CGD = 2 pF, CGS = 4.4 pF and CDS = 3 pF are adopted from
the example on page 247. For RD = R1 = 1 k, R2 = 6.3 k, RL → ∞, rDS RD and
Rg = 0, we obtain RD
≈ R = 1 k und R
= R = 1 k; therefore, it follows from
D 1 1
(3.88) that A0 ≈ − 2.6, from (3.90) that T1 ≈ 1.66 ns, from (3.91) that T2 ≈ 3.66 ns, from
(3.89) that f-3dB ≈ 14 MHz and from (3.79) that GBW ≈ 43 MHz. With a load capacitance
CL = 1 nF, we obtain from (3.90) T1 ≈ 225 ns, from (3.91) T2 ≈ 227 ns, from (3.89)
f-3dB ≈ 195 kHz and from (3.79) GBW ≈ 700 kHz.
Current-to-voltage converter: Figure 3.74 shows the small-signal equivalent circuit for
= R || R || r
the current–to–voltage converter of Fig. 3.67a; with RD D L DS , and removing
2
the s term from the denominator, it follows that
v o (s) gm RD R2 1
Z T (s) = ≈ −
% &
1 + gm RD
&
CGS R2 + RD + CDS RD
+ CGD R2
1 + gm RD
For rDS RD 1/gm and RL → ∞, we have:
1
ω-3dB = 2πf-3dB ≈ (3.92)
CGS R2 CDS
1+ + + CGD R2
gm RD gm
R2
R'D = rDS II RD II RL
CGD
rDS = 20.5 k. The parameters RG = 25 , CGD = 2 pF, CGS = 4.4 pF and CDS = 3 pF
are adopted from the example on page 247. With RD = 1 k, R2 = 6.3 k, RL → ∞
and rDS RD , we obtain from (3.92) f-3dB ≈ 7.75 MHz.
Summary
The common-source circuit can be operated without feedback, with current feedback or
with voltage feedback. Figure 3.75 shows the three variations and Fig. 3.76 lists the most
important characteristic values. The common-source circuit with voltage feedback is rarely
used, since it is not possible to make use of the high input resistance of the FET.
In practice, the common-source circuit without feedback and the common-source cir-
cuit with current feedback are used only if a high input resistance or a low noise figure with
high-resistance sources is required. In all other cases, the common-emitter circuit is supe-
rior due to the larger maximum gain, the much higher transconductance of the bipolar tran-
sistor for the same operating current and the lower noise figure for low-resistance sources.
An important role is played by the common-source circuit in integrated CMOS circuits,
because here no bipolar transistors are available. This applies in particular to large-scale
integrated mixed analog/digital circuits (mixed-mode ICs) that contain only a few analog
components in addition to many digital ones, and it means that they can be produced in
the comparably simple and reasonably priced CMOS digital process. However, there is
greater tendency toward BICMOS processes, which enable the production of MOSFETs
and bipolar transistors.
RD RD RD
R2
Vo Vo R1 Vo
Vi Vi Vi
RS
gm RD R2
A − gm RD − −
1 + g m RS R1 + R2
R1 +
gm RD
ri ∞ ∞ R1
1 R2
ro RD RD RD || 1+
gm R1
3.4.2
Common-Drain Circuit
Figure 3.77a shows the common-drain circuit, which consists of the MOSFET, the source
resistance RS , the supply voltage source Vb and the signal voltage source Vg with the
internal resistance Rg . The transfer characteristic and the small-signal response depend on
the circuitry at the bulk terminal, which is connected to the source in discrete MOSFETs and
to the most negative supply voltage (here, ground) in integrated MOSFETs. Assumptions
for the following examination are that Vb = 5 V and RS = Rg = 1 k; in addition,
K = 4 mA/V√
2 and U = 1 V for the discrete MOSFET and K = 4 mA/V2 , V = 1 V,
th th
γ = 0.5 V and Vinv = 0.6 V for the integrated MOSFET.
Vb Vb
Rg
Rg
VGS ID
Vi
Vg Vg Io
RS Vo Vo
RS
region as long as the signal voltage remains below the supply voltage, or exceeds it by a
maximum of Vth .
Figure 3.77b shows the equivalent circuit for the common-drain configuration. For
Vg ≥ Vth and Io = 0, the voltages are:
Vo = ID RS (3.93)
2ID
Vi = Vo + VGS = Vo + + Vth (3.94)
K
Equation (3.94) uses (3.3) solved for VGS for a current in the pinch-off region; the Early
effect is neglected. Inserting (3.93) into (3.94) leads to:
2Vo
Vi = Vo + + Vth (3.95)
KRS
This equation applies to both discrete and integrated MOSFETs, but with integrated MOS-
FETs the threshold voltage Vth depends on the bulk–source voltage VBS due to the substrate
effect; if VB = 0 then VBS = − Vo , so that applying (3.18) leads to:
2Vo # $
Vi = Vo + + Vth,0 + γ Vinv + Vo − Vinv (3.96)
KRS
Due to the approximately linear characteristic, the operating point can be selected over
a wide range; for the operating point marked on the characteristic curve of the discrete
MOSFET in Fig. 3.78, the voltages are:
Vo 2ID
Vo = 2 V ⇒ ID = = 2 mA ⇒ VGS = + Vth = 2 V
RS K
⇒ Vi = Vo + VGS = 4 V
With Vo = 2 V for the integrated MOSFET, we obtain from (3.96) Vi = 4.42 V.
Vo
Cutoff Pinch-off region
V region
3
VBS = 0
1
VB = 0
0 1 2 3 4 5 Vi
V
Fig. 3.78. Characteristic for the common-drain circuit with a discrete MOSFET (VBS = 0) and an
integrated MOSFET (VB = 0)
254 3 Field Effect Transistor
Rg ii G
Rg ii v GS io
vg vi gm vGS R'S vo RL
vo
gm RS
rDS 1/gm
gm RS vBS =0 g m RS
A =
=
≈ % & =
vi io =0 1 + gm RS 1 + gm + gm,B RS 1 + g m RS
√
For K = 4 mA/V2 , γ = 0.5 V, Vinv = 0.6 V and ID,A = 2 mA, it follows from Fig. 3.51
or Fig. 3.52 that gm = 4 mS and gm,B = 0.62 mS; for RS = 1 k we thus obtain A ≈ 0.8
26 g
m,B = 0 as a limiting condition would not be correct, as the substrate transconductance of the
discrete MOSFET is not zero but has no effect because vBS = 0; therefore, vBS = 0 is a correct
limitation and gm,B = 0 is its effect in the equations.
3.4 Basic Circuits 255
when using a discrete MOSFET and A ≈ 0.71 when using an integrated MOSFET. Due
to the relatively low transconductance, the gain is clearly below 1.
The small-signal input resistance is ri = ∞ and the small-signal output resistance is:
rDS 1/gm vBS =0
vo 1 1 1 1
ro = = || RS
≈ || || RS = || RS
io gm gm gm,B gm
For the numeric example, this results to ro ≈ 200 when using a discrete MOSFET and
ro ≈ 178 when using an integrated MOSFET.
With rDS 1/gm and without load resistance RL , the formulas for the common-drain
configuration are:
common-drain configuration
vo
gm RS vBS =0 g m RS
A =
≈ % & = (3.97)
vi io =0 1 + gm + gm,B RS 1 + g m RS
vi
ri = = ∞ (3.98)
i
i io =0
vo 1 1 vBS =0 1
ro = ≈ || || RS = || RS (3.99)
io gm gm,B gm
Maximum gain in integrated circuits: The maximum gain Amax is reached when the
source resistance RS is replaced by an ideal current source. In integrated circuits the
maximum gain is:
(3.41)
rDS 1/gm VBS =−Vo
gm 1
Amax = lim A ≈ =
RS →∞ gm + gm,B γ
1+
2 Vinv + Vo
√
For the numeric example with γ = 0.5 V, Vinv = 0.6 V and Vo,A = 2 V, we obtain
Amax = 0.87. With a discrete FET, we have Amax = 1.
gm gm gm,B RS2
+ vBS =0
v̂i ID,A Vinv + ID,A RS v̂i
k ≈ % % & & = % & (3.100)
8 1 + gm + gm,B RS 2 4 VGS,A − Vth (1 + gm RS )2
For the numeric example, we obtain v̂i < k · 100 V when using a discrete MOSFET and
v̂i < k · 85.5 V when using an integrated MOSFET.
256 3 Field Effect Transistor
dVo
dVo
dVGS
= = A =
dT
A dVGS
A dT
A dT
A gm dT
A
Inserting A according to (3.97) and dID /dT according to (3.14) on page 190, and using
the typical values, leads to:
dVo
ID,A RS −3 −1 4...7V
≈ % & · 10 K − 5
dT
A 1+ g +g m Rm,B S V −V GS,A th
For discrete MOSFETs, we set gm,B = 0. Using a discrete MOSFET in the numeric
example, we obtain the result that (dVo /dT )|A ranges from approximately − 0.4 mV/K to
+0.8 mV/K; the temperature drift is somewhat lower when using an integrated MOSFET.
Vb Vb
Vi Vi
Vo Vo
Resistance R results in an increase of the output resistance and for this reason should not
be too high.
gm RL
CGS + CDS Rg
c1
= + (CGS + CGD ) + CGD Rg
gm gm RL
Rg
c2
= (CGS CGD + CGS CDS + CGD CDS )
gm
The zero can be neglected, because the cutoff frequency
gm
fN = > fT
2π CGS
is higher than the transit frequency fT of the FET, as shown by comparison with (3.47).
With the low-frequency gain
gm RL
A0 = AB (0) = (3.101)
1 + gm RL
it follows that:
A0
AB (s) ≈ (3.102)
1 + sc1 + s 2 c2
(CGS + CDS ) RL
+ CGS Rg
c1 = + CGD Rg
(3.103)
1 + gm RL
c2 = (3.104)
1 + gm RL
R'g = Rg + RG 1
R'L = rDS II g II RS II RL
m,B
Rg RG CGS
vGS 1 vo
vg CGD gmvGS CDS rDS RS RL
gm,B
Fig. 3.81. Dynamic small-signal equivalent circuit for the common-drain circuit
258 3 Field Effect Transistor
Rg RG CGS
vGS
vg Cg CGD gm vGS CDS R'L CL vo
R'g CGS
vGS
vg C'g gm vGS R'L C'L vo
Fig. 3.82. Small-signal equivalent circuit for calculating the region of complex conjugate poles:
complete (above) and simplified (below)
1+s + + C GD Rg
gm gm RL
This provides an approximation for the −3 dB cutoff frequency f-3dB at which the absolute
gain has dropped by 3 dB:
1
gm RL
1 1
ω-3dB = 2πf-3dB ≈ ≈ (3.106)
c1 CGS + CDS CGS
+ + CGD Rg
gm gm RL
In the case of complex conjugate poles, that is, Q > 0.5, the following approximation
can be used:
1
ω-3dB = 2πf-3dB ≈ √ (3.107)
c2
√ √
For Q = 1/ 2 it produces
√ the exact result, while it is too high for 0.5 < Q < 1/ 2 and
too low for Q > 1/ 2.
If there is a load capacitance CL , it is in parallel with CDS and can be taken into account
by replacing CDS with CL + CDS .
adding the capacitance Cg of the signal generator and the load capacitance CL . The RC
elements Rg -Cg and RG -CGD can be combined to give one element with Rg
= Rg + RG
and Cg
= Cg + CGD , since Rg RG ; on the output side CL
= CL + CDS . After
introducing the time constants
CGS 1
Tg = Cg
Rg
, TL = CL
RL
, TGS = ≈ (3.108)
gm ωC
and the resistance ratios
Rg
1
kg = , kS = (3.109)
RL
gm RL
5
10
Tg
TGS
4
10
3
10
2
10
R'g
kg =
R'L
10
1
1 10 100 1000
–1 1 10 100 1000
10
2 3 4 5 6
1 10 10 10 10 10 10
TL
TGS
Vb
Tg ≈ RgCg
1
TGS ≈
T
Rg
TL ≈ RLCL
Vo
Vg Cg RL CL
Fig. 3.84. Circuit for the approximate calculation of the time constants
3. One can reduce TGS in order to leave the range toward the top right if operating close
to the right border of this range. This requires the use of a faster FET with a smaller
time constant TGS ; that is, a higher transit frequency fT .
4. One can increase TL in order to leave the range toward the right if operating close to
the right border of this range. This requires an enlargement of the load capacitance
CL by adding an additional capacitor in parallel.
Figure 3.85 indicates the four methods. The fifth method is the reduction of TL which
is rarely used in practical applications since, with given values for RL and CL , it can be
achieved only by connecting a resistor in parallel, which then puts an additional load on
the output. All of these methods cause a reduction in the lower cutoff frequency. In order to
3.4 Basic Circuits 261
Tg
lg T
GS
Tg
1
TGS
2
TGS 3
TL 5 4 TL
TL
lg
TGS
Fig. 3.85. Possibilities for leaving the complex conjugate pole region
keep this reduction within acceptable limits, it is necessary to leave the range of complex
conjugate poles by the shortest route possible.
Example: ID,A = 2 mA is selected for the numeric example, according to Fig. 3.77.
For K = 4 mA/V2 , VA = 50 V, Coss = 5 pF, Crss = 2 pF, fY 21s = 1 GHz and fT =
100 MHz, the parameters gm = 4 mS, rDS = 25 k, RG = 25 , CGD = 2 pF, CGS =
4.4 pF and CDS = 3 pF are determined from Fig. 3.51 on page 222. For Rg = RS = 1 k
and RL → ∞, we obtain Rg
= Rg + RG = 1025 , RL
= RL ||RS ||rDS = 960 and
thus, from (3.101), A0 = 0.793 and, from (3.106), the approximation f-3dB ≈ 31.4 MHz.
A more precise calculation with the help of (3.103)–(3.105) results in c1 = 4.45 ns,
c2 = 5.69 ns2 and Q ≈ 0.54; this means that complex conjugate poles exist, and (3.107)
√ f-3dB ≈ 67 MHz, which must be considered to be too high
provides the approximation
due to 0.5 < Q < 1/ 2. For a load capacitance CL = 1 nF, we obtain, from (3.108)
and (3.109), Tg = 2.05 ns, TL = 960 ns, TGS = 1.1 ns, kg = 1.07 und kS = 0.26,
so that (3.110) provides c1 = 202 ns and c2 = 1305 (ns)2 ; from (3.105) it follows that
Q = 0.179, that is, the poles are real, and from (3.106) we obtain f-3dB ≈ 788 kHz. The
indication of real poles can also be obtained without calculating c1 , c2 and Q on the basis
of Fig. 3.83, since point TL /TGS ≈ 1000, Tg /TGS ≈ 2, kg ≈ 1 is not in the region of
complex conjugate poles.
3.4.3
Common-Gate Circuit
Figure 3.86 shows a common-gate circuit that consists of the MOSFET, the drain resistance
RD , the supply voltage source Vb , the signal voltage source Vi 27 and the gate series
resistance RGV ; the latter has no effect on the transfer characteristic but influences the
frequency response and the bandwidth. The transfer characteristic and the small-signal
response depend on the circuitry at the bulk terminal, which is connected to the source
in discrete MOSFETs and to the negative supply voltage in integrated MOSFETs. Due to
the fact that the common-gate circuit shown in Fig. 3.86 is operated with a negative input
27 A voltage source without an internal resistance R is used in this case, so that the characteristics
g
are independent of Rg .
262 3 Field Effect Transistor
voltage, the bulk terminal of the integrated MOSFET must be connected to an additional
negative supply voltage VB that is lower than the minimum input voltage; this ensures
that the bulk–source diode is reverse-biased. In what follows, we assume that Vb = 5 V,
VB = − 5 V and RD = RGV = 1 k; in addition, we assume that K = 4 mA/V 2
√ and
Vth = 1 V for the discrete MOSFET and K = 4 mA/V , Vth,0 = 1 V, γ = 0.5 V and
2
By inserting (3.112) into (3.111) for the transfer characteristic, it follows that:
KRD KRD
Vo = Vb − (− Vi − Vth )2 = Vb − (Vi + Vth )2 (3.113)
2 2
Vb Vb
RD RD
ID Io ID Io
VBS
IG = 0 Vo IG = 0 Vo
Vi VB Vi
R GV R GV
Transfer characteristic when driven by a current source: This circuit may also be
driven by a current source Ii (see Fig. 3.88). With Ii < 0, the circuit then operates as a
current–to–voltage converter or a transimpendance amplifier:
3.4 Basic Circuits 263
Vo
V
5
VB = – 5 V 4
VBS = 0 3
–5 –4 –3 –2 –1 Vi
–1 V
–2
–3
–4
–5
V
Vb V
RD 5
Vo
ID Io
IG = 0 Vo 1
Ii Vi
RGV –5 –1 Ii
Vi –1
mA
Fig. 3.88. Circuit and characteristic of the common-gate circuit driven by a current source
ID =− Ii
Vo = Vb − ID RD =
Vb + Ii RD (3.114)
2ID ID =− Ii 2Ii
Vi = − VGS = − Vth − = − Vth − − (3.115)
K K
In practice, a common-source circuit with an open drain or a current mirror is used
for driving the current; this will be described in more detail in the section on setting the
operating point.
vGS gmvGS
vi vo
vBS
gm,BvBS
Fig. 3.89. Small-signal equivalent
circuit for the common-gate circuit
Figure 3.89 shows the small-signal equivalent circuit for the common-gate circuit. A
changeover from the integrated to the discrete MOSFET is based on the limitation of
vBS = 0; the equations then use gm,B = 0.28 From the nodal equation
vo vo − vi
+ + gm vGS + gm,B vBS = 0
RD rDS
it follows, for vi = − vGS = − vBS , that:
vo
1
A = = g + g + (RD || rDS )
vi
io =0
m m,B
rDS
rDS RD ,1/gm % & vBS =0
≈ gm + gm,B RD = g m RD
For ID,A = 2.5 mA, K = 4 mA/V2 and VA = 50 V, we obtain from Fig. 3.51 on page 222
the values gm = 4.47 mS and rDS = 20 k; for discrete MOSFETs, by inserting gm,B = 0
and RD = 1 k, we obtain the exact value A = 4.3 and in the first approximation
A = 4.47; while for integrated MOSFETs we obtain a slightly higher gain, because
gm,B > 0.
For the small-signal input resistance, it follows that:
rDS RD ,1/gm
vi
RD + rDS 1 vBS =0 1
ri =
= % & ≈ =
ii io =0 1 + gm + gm,B rDS gm + gm,B gm
This depends on the load resistance which, in this case, is the open-circuit input resistance
because io = 0 (RL → ∞). The input resistance for other values of RL is calculated by
replacing RD with the parallel arrangement of RD and RL ; inserting RL = RD = 0 leads
to the short-circuit input resistance. However, the dependence on RL is so low that it is
eliminated by the approximation. For the operating point taken as an example, we obtain,
for the discrete MOSFET, an exact value of ri = 232 ; the approximation is ri = 224 .
For the small-signal output resistance, it follows:
*% % & & + rDS RD
vo
ro = = RD || 1 + gm + gm,B Rg rDS + Rg ≈ RD
io
This depends on the internal resistance Rg of the signal generator; with Rg = 0, the
short-circuit output resistance is
ro,s = RD || rDS
28 g
m,B = 0 as a limiting condition would not be correct, as the substrate transconductance of the
discrete MOSFET is not zero, but has no effect because vBS = 0; therefore, vBS = 0 is a correct
limitation and gm,B = 0 is its effect in the equations.
3.4 Basic Circuits 265
vo
% & vBS =0
A =
≈ gm + gm,B RD = gm RD (3.116)
vi io =0
vi
1 vBS =0 1
ri =
≈ = (3.117)
ii io =0 gm + gm,B gm
vo
ro = ≈ RD (3.118)
io
When driven by a signal source with an internal resistance Rg and a load resistance
RL , the operating gain is:
ri RL gm (RD || RL ) vBS =0 gm (RD || RL )
AB = A ≈ % & = (3.119)
ri + R g ro + R L 1 + gm + gm,B Rg 1 + g m Rg
When the circuit is driven by a current source, the transfer resistance RT (the tran-
simpedance) takes the place of the gain; for the current–to–voltage converter in common-
base configuration this leads to:
current–to–voltage converter in common-base configuration
vo
vo
vi
RT = = = Ari = RD
ii
io =0 vi
io =0 ii
io =0 (3.120)
The input and output resistances are given by (3.117) and (3.118), respectively.
Nonlinearity: If the circuit is driven by a voltage source, then v̂GS = v̂i and (3.13)
on page 185 can be used: this describes the relationship between the amplitude v̂GS of
a sinusoidal signal with small amplitude and the distortion factor k of the drain current,
which in the common-gate
% circuit is &the same as the distortion factor of the output voltage.
Consequently, v̂i < 4k VGS,A − Vth . When driven by a current source the circuit operates
linear, which means that the distortion factor is zero.
Temperature sensitivity: The common-gate circuit has the same temperature drift as the
common-source circuit without feedback, because in both configurations there is a constant
input voltage between gate and source and the output voltage is given by Vo = Vb − ID RD .
This leads to:
dVo
dID
−3 −1 4...7V
= − R ≈ I R · 10 K 5 −
dT
A dT
A
D D,A D
VGS,A − Vth
266 3 Field Effect Transistor
Driving by a voltage source: Obtaining an exact calculation for the operating gain
AB (s) = v o (s)/v g (s) is a difficult process and leads to complex expressions. A sufficiently
accurate approximation is reached by neglecting the resistance rDS and the capacitance
CDS ; the latter only exists in discrete MOSFETs. Additional parameters that occur in the
integrated MOSFETs are the substrate transconductance gm,B and the bulk capacitances
CBS and CBD ; they are neglected here. This leads to the simplified small-signal equivalent
circuit for the discrete and the integrated MOSFET as shown in Fig. 3.91, which is to a large
extent similar to the small-signal equivalent circuit for the common-base configuration
shown in Fig. 2.113. Therefore, the results of the common-base circuit can be transferred to
the common-gate circuit by inserting the corresponding small-signal parameters in (2.139)
and (2.140) and taking the limit as β → ∞; with RGV
= RGV + RG and RD
= R || R
D L
and the low-frequency gain
gm RD
A0 = AB (0) ≈ (3.121)
1 + g m Rg
Vb Vb Vb Vb
RD RD R1
T1 T2 T2
T1
RS R2
– Vb
a Voltage-driven b Current-driven
R'D = RD II RL
Rg gm vGS
vg RG vo
RD RL
R'GV = RGV + RG
RGV
Fig. 3.91. Simplified dynamic small-signal equivalent circuit for the common-gate configuration
A0
AB (s) ≈ %
&
% %
&& (3.122)
CGS Rg + RGV + CGD RD 1 + gm Rg + RGV
1+s
1 + g m Rg
1 + g m Rg
ω-3dB ≈ %
&
% %
&& (3.123)
CGS Rg + RGV + CGD RD 1 + gm Rg + RGV
From (3.121) and (3.123), two time constants that are independent of the low-frequency
gain A0 can be calculated:29
1
ω-3dB (A0 ) ≈ (3.124)
T1 + T 2 A0
Rg + RGV
T1 = CGS (3.125)
1 + g m Rg
1
T2 = CGD Rg + RGV + (3.126)
gm
The explanation concerning the gain-bandwidth product GBW , including (3.79) given on
page 246, also applies to the common-gate circuit.
If there is a load capacitance CL in parallel with the load resistance RL , then
1 1
T2 = CGD Rg + RGV + + C L Rg + (3.127)
gm gm
CL has no influence on the time constant T1 .
Driving by a current source: When a current source is used to drive the circuit, the
frequency response of the transimpedance Z T (s) is interesting; on the basis of (3.122),
one can reach an approximation by a first-degree lowpass filter:
29 It is assumed that a change in A is caused by a variation in R
; the time constants are thus
0 D
.
independent of A0 if they are independent of RD
268 3 Field Effect Transistor
v o (s) RD
Z T (s) = = lim Rg AB (s) ≈ (3.128)
i i (s) Rg →∞ CGS
1+s + CGD RD
gm
In this case, the cutoff frequency is:
1
ω-3dB = 2πf-3dB ≈ (3.129)
CGS
+ CGD RD
gm
and RGV = RG = 25 ; from (3.121) it follows that A0 ≈ 4.47 and from (3.123) that
f-3dB ≈ 68 MHz. The cutoff frequency strongly depends on RGV ; with RGV = 1 k the
frequency drops to f-3dB ≈ 10 MHz.
When the circuit is driven by a current source and RL → ∞, then it follows from
(3.128) that RT = Z T (0) ≈ RD = 1 k and from (3.129) that f-3dB ≈ 53 MHz. The
resistance RGV is of no influence in this case.
Chapter 4:
Amplifiers
Amplifiers are important elements in analog signal processing. They amplify an input
signal of low amplitude to such a degree that it can be used to drive a subsequent unit.
For example, a microphone signal has to be amplified in several stages from the microvolt
(µV) range to the volt (V) range in order to feed a loudspeaker. Similarly, the signals of
thermocouples, photodiodes, magnetic reading heads, receiving antennas and many other
signal sources can only be processed after suitable amplification. Since digital circuits such
as microprocessors and digital signal processors (DSP) are increasingly being used in the
processing and evaluation of complex signals, a typical signal processing chain usually
consists of the following elements or stages:
1. A sensor for converting a physical unit such as pressure (microphone), temperature
(thermocouple), light (photodiode) or electromagnetc field (antenna) into an electrical
signal.
2. One or more amplifiers to amplify and filter the signal.
3. An analog-to-digital (A/D) converter for digitizing the signal.
4. A microprocessor, DSP or other digital circuit for processing the digitized signal.
5. A digital-to-analog (D/A) converter to produce an analog output signal.
6. One or more amplifiers to amplify and filter the signal to such a degree that it can be
used to drive an actuator.
7. An actuator to convert the signal into a physical unit such as pressure (loudspeaker),
temperature (heating rod), light (incandescent lamp) or electromagnetic field (transmit-
ting antenna).
Figure 4.1 illustrates the seven stages of a signal processing chain; one of the symbols
shown in Fig. 4.2 is used for the amplifiers.
The amplifiers of stage 2 process comparatively small signals and are therefore known
as small-signal amplifiers; in most cases, their output power is below 1 mW. In contrast,
Small-signal A/D
Sensor amplifier converter
Physical unit
DSP
Power D/A
Actuator amplifier converter
Physical unit
stage 6 requires power amplifiers that operate in a power range between a few mW (head-
sets, remote controls, etc.) to several kW (large loudspeaker systems, radio stations, etc.).
Power amplifiers are described in Chap. 15.
To filter the signals, passive filters are increasingly being replaced by active filters
that also contain amplifiers. Therefore, it is not possible to clearly distinguish between
the elements amplifier and filter, as every amplifier also acts as a filter due to its limited
bandwidth and each active filter may produce signal amplification.Active filters are covered
in Chap. 13.
Another distinctive feature is the frequency range of an amplifier. In the lower cut-
off frequency fL we differentiate between DC amplifiers and AC amplifiers, and in the
upper cutoff frequency fU between low-frequency (LF) amplifiers and high-frequency
(HF) amplifiers. With regard to the bandwidth B = fU − fL , we differentiate between
broadband amplifiers and narrowband or tuned amplifiers. With respect to the upper limit
frequency, another distinction is often made between audio or audio frequency ampli-
fiers (AF amplifiers), video amplifiers, intermediate frequency amplifiers (IF amplifiers)
and radio frequency (RF) amplifiers. While the division into AC and DC amplifiers re-
sults directly from the design of AC or DC voltage coupling, the distinction between LF
and HF amplifiers is not defined; often, 1 MHz is used as a limit value. Similarly, there
is no definition for broadband and narrowband amplifiers; the latter are usually charac-
terized by the mid-frequency fM = (fU + fL )/2 and the bandwidth B = fU − fL .
In narrowband amplifiers, the bandwidth is less than one tenth of the mid-frequency:
B < fM /10.
Despite this multitude of amplifier types, the circuit design used is almost identical,
because all amplifiers are based on standard transistor circuits that amplify DC voltages.
However, a distinction is made in respect to the coupling at the input and the output and
between the individual stages of multi-stage amplifiers: DC amplifiers use direct coupling
(DC coupling or galvanic coupling), AC amplifiers use capacitive coupling with coupling
capacitances (AC coupling) and narrowband amplifiers use selective coupling with LC
resonant circuits, ceramic resonators or surface acoustic wave filters. Figure 4.3 shows the
type of coupling and the frequency responses of the amplifiers mentioned, together with
the parameters fU , fL , fM and B.
Likewise, the distinction between LF and HF amplifiers is not so much the result of the
circuit design but, rather, of the transit frequencies of the transistors used. The quiescent
currents at the operating point also play a decisive role, since for small currents the transit
frequencies are approximately proportional to the quiescent current. Thus, a differential
amplifier that achieves a cutoff frequency of 10 MHz at a quiescent current of 1 mA may
yield a cutoff frequency of only 100 . . . 300 kHz at a quiescent current of 10 mA.
One speciality is operational amplifiers, which are very important as general-purpose
DC amplifiers for low frequencies. Operational amplifiers are used almost exclusively in
standard applications. Only if particular demands cannot be met by standard operational
4.1 Circuits 271
fU f
fL fU f
A B
fL fM fU f
Fig. 4.3. Coupling types and the frequency response of DC amplifiers (top), AC amplifiers (centre)
and narrowband amplifiers (bottom)
amplifiers, or those contained in amplifier libraries1 , may the amplifier circuit be made up
of discrete transistors, or a special integrated circuit be produced. Operational amplifiers
are described in Chap. 5.
4.1
Circuits
Amplifiers encompass one or more amplifier stages, where each stage is made up of one
or more coupled basic circuits with bipolar transistors or field effect transistors. Further
transistors may additionally be required to set the operating point. Going back to basic
circuits in many cases allows the use of the equations determined in Sects. 2.4 and 3.4.
Characteristics of transistors: The circuits below are described using bipolar transistors
and enhancement MOSFETs as far as this is possible and suitable; depletion MOSFETs
and JFETs are used in exception cases only. The basic equations, (2.2) and (2.3), or (3.3)
and (3.4), are used to calculate the characteristics and operating points:
VBE
VCE IC
npn transistor: IC = IS e VT 1 + , IB =
VA B
K VDS
n-channel MOSFET: ID = (VGS − Vth )2 1 + , IG = 0
2 VA
With MOSFETs the substrate effect must also be taken into account; (3.18) applies to the
n-channel MOSFET:
# $
Vth = Vth,0 + γ Vinv − VBS − Vinv
1 Predefined modules available from module libraries are used as far as possible when designing
integrated circuits.
272 4 Amplifiers
Scaling: The presentation is based on integrated circuit design, which makes particular
use of the almost unlimited scaling capabilities of transistors. In bipolar transistors the
reverse saturation current IS is scaled by varying the emitter surface, while in MOSFETs
the transconductance coefficient K is scaled by varying the channel width/length ratio
W/L. In MOSFETs it is usually the channel width W that is scaled, while the channel
length L remains constant.2
Scaling is generally done according to the quiescent currents at the operating point:
IS ∼ IC,A or W ∼ K ∼ ID,A (L = const.); this makes the current density equal in all
transistors. The result is that all npn transistors operate with the same base-emitter voltage
VBE,A at the operating point, with the exception of a small deviation caused by the Early
effect:
IC,A IC,A ∼IS
VBE,A ≈ VT ln = const. ≈ 0.7 V
IS
With MOSFETs the conditions are more complicated because of the substrate effect: two
MOSFETs with the same current density – if the Early effect is ignored – only have an
identical gate-source voltage VGS,A if the bulk-source voltages are the same:
ID,A ∼K∼W
2ID,A VBS,A =const.
VGS,A ≈ Vth (VBS,A ) + = const.
K
Normalization: The parameters of the various transistors are normalized to the size of
a reference transistor; it is given the relative size 1. Consequently, a bipolar transistor of
size 5 has five times the reverse saturation current IS and a MOSFET of size 5 has five
times the transconductance coefficient K than a transistor of size 1.
Often, the smallest transistor of a certain technology is used as the reference transistor;
in this case, there are only relative sizes, which are greater than or equal to one. For
bipolar transistors, the reference transistor has the smallest emitter surface and is thus the
smallest in terms of electrical properties – that is, with regard to IS – and in geometric
terms. For MOSFETs there is an additional degree of freedom, since the channel width
W and the channel length L may be freely selected. Since the short-channel and narrow-
channel effects are both undesirable in analog circuits, W and L must not be below certain
technology-dependent values: W ≥ Wmin and L ≥ Lmin . Selecting W = Wmin and
L = Lmin results in the geometrically smallest MOSFET, which represents reference
transistors with the relative size 1. Larger MOSFETs are created by increasing W and
keeping L = Lmin constant. However, one may also keep W = Wmin constant and enlarge
L; this creates a MOSFET that is electrically – that is, with regard to K ∼ W/L –
smaller but geometrically larger than the reference transistor. Therefore, it is necessary to
distinguish between the electrical size and the geometric size. In the following description,
the term size always refers to the electrical size. Proportionally enlarging W and L results
in a MOSFET of the same size; this method is, however, only used in exceptional cases,
because of the higher spatial requirements.3 Figure 4.4 uses bipolar transistors of sizes 1
2 In digital circuits, the channel lengths are mostly 0.2 . . . 0.5 mm, while analog circuits usually
have channel lengths of more than 1 mm. This is due to the fact that the Early voltage VA , and
thus the maximum gain, rise with an increasing channel length.
3 MOSFETs that are of the same electrical size but larger in terms of geometry generally feature
lower noise and higher Early voltage; the capacitances, on the other hand, increase.
4.1 Circuits 273
B E C S G D S G D
+ + + + + +
n n n n n n
p
n– p– p–
1 1 1/2
B E C S G D Wmin S G D
2 2 (1)
B E C S G D 2Wmin S G D
Lmin 2Lmin
and 2 and n-channel MOSFETs of sizes 1, 2 and 1/2 to illustrate transistor scaling and
normalization.
Complementary transistors: Most bipolar technologies only allow lateral pnp tran-
sistors, which have significantly poorer electrical properties than those of vertical npn
transistors; this is particularly true for the current gain and the transit frequency. These
technologies only use npn transistors in the signal path of an amplifier if possible; pnp tran-
sistors are only used for current sources or in common-collector or common-base circuits,
as in these applications their poorer properties are less noticeable. Vertical pnp transis-
tors with similar properties may be available in special complementary technologies, but
here too the npn transistors have somewhat better characteristics. The differences between
vertical and lateral bipolar transistors are described in more detail in Sect. 2.2.
MOS technologies are predominantly complementary technologies, i.e. CMOS. Here,
n-channel and p-channel MOSFETs with comparable properties are available. However,
the relative transconductance coefficient Kp
of p-channel MOSFETs is 2 . . . 3 times lower
than the relative transconductance coefficient Kn
of n-channel MOSFETs. This means
that a p-channel MOSFET with the same channel length L must have a channel width W
that is 2 . . . 3 times wider than in the n-channel MOSFET in order to achieve the same
transconductance coefficient K = Kn/p
W/L. But all that this means is that just the static
characteristics are almost the same. The dynamic characteristics of the p-channel MOSFET
are inferior, because the larger dimensions result in higher capacitances. Therefore, n-
channel MOSFETs are preferred. If both the static and the dynamic characteristics are to
be almost the same, then W and L for the n-channel MOSFET must be enlarged by a factor
274 4 Amplifiers
√ √
of between 2 and 3, to make the surfaces and thus the capacitances approximately
comparable to those of the p-channel MOSFET; this has no influence on the electrical
size of the n-channel MOSFET. As this method reduces the transit frequency of the n-
channel MOSFET to that of the p-channel MOSFET, it is only used if special symmetric
characteristics are required.
A A
A
K
K K
normal diode npn diode pnp diode
The circuits explained below are described on the basis of complementary bipolar
and CMOS technologies; the most important transistor parameters are listed in Figs. 4.5
and 4.6.
Diodes: In integrated circuits, diodes are realized with the help of transistors. An npn
or pnp transistor with a short-circuited base-collector junction is used for a bipolar diode
(see Fig. 4.7). This particular diode is called transdiode and is required in particular for
current scaling as described below; a collector or emitter diode is not suitable for this
purpose. Furthermore, a distinction must be made between npn and pnp diodes, since
they have different parameters. Scaling is done in the same way as in transistors; that is,
an npn diode of size 5 corresponds to an npn transistor of size 5 with a short-circuited
base-collector junction.
An important application of diodes is current-to-voltage conversion according to
Fig. 4.9a, in which the diodes provide an indication for the current:
V
I IS,D
I I
I = IS,D e − 1 VT ⇒ V = VT ln +1 ≈ VT ln
IS,D IS,D
Here, IS,D is the reverse saturation current of the diode. Feeding this voltage to the base-
emitter junction of a transistor with reverse saturation current IS,D leads to the following
equation, provided that the transistor is operating in normal mode and the base current is
negligible:
VBE I
VBE =V ln IS,T
IC ≈ IS,T e VT = IS,T e IS,D = I
IS,D
276 4 Amplifiers
A A
A
K
K K
normal diode n-channel diode p-channel diode
This means that the current is scaled in proportion to the reverse saturation currents. Defined
scaling is achieved only if an npn diode is combined with an npn transistor or a pnp diode
is combined with a pnp transistor; then the ratio of the reverse saturation currents is defined
by the dimensional proportions.
In MOS circuits the FET diodes shown in Fig. 4.8 may be used. The current-to-voltage
conversion according to Fig. 4.9b is:
KD 2I
I = (VGS − Vth )2 ⇒ V = Vth +
2 KD
Here, KD is the transconductance coefficient of the FET diode. Feeding this voltage to the
gate-source junction of a MOSFET with the transconductance coefficient KM leads to the
following equation, provided that the MOSFET is operated in the pinch-off region:
KM VGS =V KM
ID ≈ (VGS − Vth )2 = I
2 KD
Even in this case, an n-channel FET diode must be combined with an n-channel MOSFET
and a p-channel FET diode with a p-channel MOSFET in order to define the current scaling
by the dimensional proportions.
A A ID
IC
I I
IB ≈ 0 IG = 0
V V
K K
4.1.1
Current Sources and Current Mirrors
A current source supplies a constant output current and is used predominantly for setting
the operating point (biasing). A current mirror provides an amplified or attenuated copy
of the input current at the output; that is, it operates as a current-controlled current source.
Any current mirror may also be used as a current source by keeping the input current
constant; in this respect, the current source is a special application of the current mirror.
Stable operation also necessitates negative current feedback in order to keep the output
current constant despite variations in the transistor parameters caused by manufacturing
tolerances or temperature drift. This results in the circuits shown in Fig. 4.10. A load must
be connected to the output of the current source to allow the flow of current Io ; in Fig. 4.10,
resistance RL represents the load.
Output current: The current source with a bipolar transistor according to Fig. 4.10a is
described by the following mesh equation:
IC IB
VB = VBE + VR = VBE + (IC + IB ) RE ≈ VBE + IC RE
With IC = Io , it follows
Vb Vb
RL RL
Io Io
T1
T1
Vo Vo
VBE
VB VGS
VR VB VR
RE RS
Io operating area
RE
Vo,min Vo
Output characteristic: Plotting the output current Io versus Vo for several values of
RE produces the family of characteristics shown in Fig. 4.11, with the minimum output
voltage:
VCE,sat ≈0.2 V
VBE ≈0.7 V
Vo,min = VB − VBE + VCE,sat ≈ VB − 0.5 V
For Vo > Vo,min and VB = const., the circuit functions as a current source. Hereafter,
Vo,min will be referred to as the useful limit.
Output resistance: Besides the output current Io and the voltage limit Vo,min , the output
resistance
∂Vo
ro =
∂Io
VB =const.
in the operating range is also of particular interest; for an ideal current source this value is
ro = ∞, and therefore it should be as high as possible in any real current source. The finite
output resistance is caused by the Early effect and can be calculated using the small-signal
equivalent circuit. Since the circuit in Fig. 4.10a largely corresponds to the common-emitter
circuit with current feedback shown in Fig. 2.62a on page 102, the previous result can be
used once the resistances have been changed to Rg = 0 and RC → ∞;6 this leads to
6 In the common-emitter circuit with current feedback, R is assumed to be an inherent part of
C
the circuit and is thus included in the calculation of the output resistance; in the current source,
however, it is the output resistance of the collector without other circuit components that is of
interest. Resistance RC is removed by setting RC → ∞.
4.1 Circuits 279
ro [log]
rCE
rCE
Fig. 4.12. Output resistance of a current source with a bipolar transistor at a constant output current
rCE rBE
vo
βRE
ro = ≈ rCE 1+ (4.1)
io
VB =const. RE + rBE
The use of β 1 and rBE = β/gm leads to:
,
rCE (1 + gm RE ) for RE rBE
ro ≈
β rCE for RE rBE
Figure 4.12 shows the curve of ro versus RE with a constant output current.
Inserting rCE = VA /Io , gm = Io /VT , rBE = β VT /Io and VR ≈ Io RE shows the
dependence of the output resistance on the output current:
⎧
⎪ VA VA
⎪
⎨ I + V RE for VR β VT
o T
ro ≈
⎪
⎪ β VA
⎩ for VR β VT
Io
The maximum output current is achieved if a voltage drop VR across the feedback resistance
of more than β VT ≈ 2.6 V is selected. This results in a constant Io ro product:
VA ≈30...200 V
β≈50...500
Io ro ≈ β VA ≈ 1.5 . . . 100 kV
The product of the Early voltage VA and the current gain β is thus a crucial parameter for
evaluating the use of bipolar transistors in current sources.
Current source with a MOSFET: With Io = ID , the voltage for the MOSFET current
source in Fig. 4.10b is:
2Io
VB = VR + VGS = Io RS + VGS = Io RS + Vth +
K
Calculating the output current Io = ID is rather complex, since MOSFETs do not allow a
simple approximation for VGS , such as VBE ≈ 0.7 V for bipolar transistors. However, for
280 4 Amplifiers
Vb Vb Vb
R1 Io R1 Io R1 Io
Iq IB ≈ 0 Iq IB ≈ 0
T1 T1 T1
D1
D1
R2 R3 R2 R3 R3
Vb Vb
RV RV
Io Io
Ii Ii
Vo T1 T2 Vo
T1 T2
R1 R2 R1 R2
⎫
Vb − V D ⎬
Iq ≈ (Vb − VD ) R2
R1 + R 2 ⇒ Io ≈ with VD ≈ 0.7 V
⎭ (R1 + R2 ) R3
Iq R2 ≈ Io R3
The temperature sensitivity is:
dIo R2 dVD 2 mV/K R2 Io
= − ≈ ≈ 2 mV/K ·
dT (R1 + R2 ) R3 dT R3 R1 + R 2 Vb − V D
Compared to the circuit in Fig. 4.13a, this value is lower by the factor 1 + R1 /R2 and
it becomes zero when R1 is replaced by a (temperature-independent) current source with
current Iq .7
The current in the circuit of Fig. 4.13c is:
VZ − VBE VZ − 0.7 V
Io ≈ ≈
R3 R3
Here, VZ is the breakdown voltage of the Zener diode. The temperature sensitivity depends
on the temperature coefficient of the Zener diode. To keep the temperature sensitivity very
low, a normal diode can be connected in series as in Fig. 4.13b to compensate VBE ; then:
VZ
Io ≈
R3
Only the temperature coefficient of the Zener diode has any influence. The minimum
temperature sensitivity is achieved with VZ ≈ 5 . . . 6 V.
7 The transition to the current source is achieved by making R → ∞; at the same time this
1
necessitates that Vb → ∞ to keep the output current constant.
282 4 Amplifiers
Vb
RV Ii Io
I C1 I C2
I B1 I B2
T1 T2
Vi Vo
VBE1 VBE2
npn current mirror: Figure 4.15 shows the currents and voltages of the simple current
mirror with npn transistors, which is simply referred to as the npn current mirror. The
mesh equation for the base-emitter junction and the feedback resistances leads to:
In the normal operating range, both transistors are in normal mode and can be described
by the basic (2.2) and (2.3):
VBE 1
IC1
IC1 = IS1 e VT , IB1 =
B
(4.4)
VBE 2
VCE2 IC2
IC2 = IS2 e VT 1+ , IB2 =
VA B
For T1 , the Early effect is ignored, since VCE1 = VBE 1 VA . From Fig. 4.15, it thus
follows that:
B1,IS2 /IS1
Io 1 IS2
kI = ≈ ≈ (4.7)
Ii I S1 1 1 IS1
1+ +
IS2 B B
If the Early voltage VA and the current gain B are sufficiently high, and the size ratio IS2 /IS1
of the transistors is significantly lower than the current gain B, then the current ratio kI
corresponds approximately to the ratio of the transistor dimensions. If both transistors are
4.1 Circuits 283
Io Io
kI =
A Ii
1m
100µ 1
10 µ
1µ
1µ 10 µ 100µ 1m I i 1µ 10 µ 100µ 1m I i
A A
a Transfer characteristic b Currrent ratio
Fig. 4.16. The transfer characteristic of a current mirror with IS1 = IS2
Output characteristic: In current mirrors not only the current ratio is of interest; indeed,
the operating range and the small-signal output resistance in the operating range are also of
particular interest. Useful in this respect is the family of output characteristics that show Io
as a function of Vo using Ii as a parameter; usually only one characteristic is plotted for the
intended quiescent current Ii = Ii,A . Figure 4.17 shows the output characteristic of an npn
current mirror with kI = 1 for Ii = 100 mA; the characteristic of the n-channel current
mirror in Fig. 4.17 will be explained later. The characteristic corresponds to the output
characteristic of transistor T2 . For Vo > VCE,sat , transistor T2 operates in normal mode;
only in this operating range does the current mirror have the calculated current ratio. For
Vo ≤ VCE,sat , transistor T2 enters the saturation region and the current decreases. The
minimum output voltage Vo,min is an important parameter; in the npn current mirror it is:8
Vo,min = VCE,sat ≈ 0.2 V
The output resistance corresponds to the reciprocal value of the slope of the output
characteristic in the operating range. If only the approximations for the current gain are
carried out in (4.6) and the Early voltage is maintained, we obtain the following value
within the operating range:
8 A relatively high value of V
CE,sat ≈ 0.2 V is assumed for the collector–emitter saturation voltage,
since for this voltage the output characteristic of the transistor ought to be as horizontal as possible.
284 4 Amplifiers
Io
µA npn current mirror operating range
npn
100
60
n-channel
40
20
VCE,sat VDS,ab
0
0 0.5 1.0 1.5 2.0 Vo
V
Fig. 4.17. Output characteristics of an npn and an n-channel current mirror for R1 = R2 = 0
Io IS2 Vo
kI = ≈ 1+
Ii IS1 VA
This gives us the following value for the small-signal output resistance:
Vo VA
∂Vo
Vo + VA VA VA
ro = = ≈ = = rCE2
∂Io
Ii =const. Io Io IC2
Usually, the output resistance is calculated by using the small-signal equivalent circuit;
this will be explained in more detail further below.
npn current mirror with feedback: Feedback resistances can be used to stabilise the
current ratio and to increase the output resistance. Without feedback resistance the current
ratio only depends on the ratio of the transistor dimensions, but with feedback resistances
the ratio of the resistances R2 /R1 also has an influence. Inserting (4.4) into (4.3) and
neglecting the Early effect results in:
1 IC1 1 IC2
1+ R1 IC1 + VT ln = 1+ R2 IC2 + VT ln (4.9)
B IS1 B IS2
This equation cannot be solved easily, as the collector currents have a linear and a loga-
rithmic effect on the result. With sufficiently large resistances the linear terms dominate,
so that:
kI
Ii Io 2.0 2
R1
T1 T2 1.0 R2
1
1 1
0.5 0.5
R1 R2
0
100 n 1µ 10µ 100µ 1m I i
A
Fig. 4.18. Current dependence of the current ratio for transistors of equal size (IS2 /IS1 = 1) and
different R1 /R2 values
B1+R1 /R2
Io R1 R1
kI = ≈ ≈ (4.11)
Ii R1 + R2 R2
R2 +
B
In this case, the current ratio depends solely on the ratio of the resistances and no longer
on the dimensions of the transistors.
For integrated current mirrors, the ratio of the resistances is usually chosen to corre-
spond to the dimensional ratio of the transistors:
IS2 R1
≈
IS1 R2
In this case, the resistances have almost no influence on the current ratio but only cause
the output resistance to increase; this will be described in more detail below. For current
mirrors driven across a large current range, this condition is in fact mandatory, as the
ratio of linear and logarithmic terms in (4.9) depends on the current: for small currents,
the current ratio is determined by IS2 /IS1 , and for large currents by R1 /R2 . Figure 4.18
shows this dependence using a current mirror with transistors of equal size (IS2 /IS1 = 1)
and different values for R1 /R2 . A constant current ratio is achieved only in the case of
IS2 /IS1 = R1 /R2 .
With current mirrors made of discrete components feedback resistances must always be
used, as the tolerances of discrete transistors are so high that the ratio IS2 /IS1 is practically
undefined, even for transistors of the same type; this means that it is essential to adjust
the current ratio using the resistances. The required minimum resistance value can be
determined by differentiating both sides of (4.9) with respect to the actual current and
presuming that the influence of the terms with the resistances dominates:
1 VT 1 VT
1+ R1 , 1+ R2
B IC1 B IC2
It follows:
1 1
VR1 = 1+ R1 IC1 VT , VR2 = 1+ R2 IC2 VT
B B
286 4 Amplifiers
kI
Io
1
Ii
0.1
Vo –2
T1 T2 10
–3
10
R2 VR2
0 60 120 180 VR2
mV
a Circuit b Current ratio kI of identical
transistors ( IS1 = IS2 )
VR1 and VR2 represent the voltages across resistances R1 and R2 (see Fig. 4.15). As both
conditions are equivalent due to (4.10), and an approximate factor of 10 is required to meet
this condition, one must select
VR1 ≈ VR2 ≥ 10 VT ≈ 250 mV (4.12)
to ensure that the current ratio is only influenced by the resistances. In current mirrors
driven across a large current range, the condition given in (4.12) can usually not be met
across the entire range; in this case, the current ratio for decreasing currents is increasingly
determined by the unknown ratio IS2 /IS1 .
The feedback reduces the operating range, since the output voltage limit Vo,min is
increased by a value equal to the voltage across the resistances:
Vo,min = VCE,sat + VR2 ≥ 0.2 V + 0.25 V = 0.45 V
For this reason, the resistance values cannot be freely chosen.
Operation as a current source: The simple npn current mirror can be operated as a
current source by adding resistance RV as shown in Fig. 4.15; this makes the input current
constant. From Vi = VBE 1 + VR1 and Vb = Vi + Ii RV , it follows:
Vb = Ii RV + (IC1 + IB1 ) R1 + VBE 1
If the base currents of the transistors are neglected and we assume that VBE ≈ 0.7 V, then:
Vb − VBE 1 Vb − 0.7 V
Ii ≈ ≈
RV + R 1 RV + R 1
The output current is Io = kI Ii .
Widlar current mirror: Where very low current ratios are required, any adjustment by
means of the dimensional ratio of the transistors is unfavorable, as T2 can only be reduced
in size down to the basic size, causing T1 to become very large. In this situation, one
can use the Widlar current mirror shown in Fig. 4.19a, which only features the feedback
resistance R2 ; from (4.9), where R1 = 0 and B 1, it follows:
IC1 IC2
VT ln = R2 IC2 + VT ln
IS1 IS2
4.1 Circuits 287
V
Io IC2 IS2 − VR2
kI = ≈ = e T with VR2 = R2 IC2 (4.13)
Ii IC1 IS1
This value depends exponentially on the ratio VR2 /VT and decreases by a factor of 10 if
VR2 increases by VT ln 10 ≈ 60 mV; Fig. 4.19b illustrates this for identical transistors;
that is, for IS1 = IS2 . From (4.13) it can also be seen that the Widlar current mirror is only
suitable for constant currents, due to the high current dependency of the current ratio.
It would now seem that the same procedure could be used to realize very high current
ratios by only using resistance R1 in Fig. 4.14a. Theoretically, this is correct, but it is
not feasible in practice since the higher current at the output naturally requires a larger
transistor. This inverted Widlar current mirror can only be used if the current ratio is high
enough to warrant the use of a Widlar current mirror and the output current is so low that
a transistor of size 1 can also be used at the output; this case is, however, very rare.
Example: An output current Io = 10 mA is to be derived from an input current Ii =
1 mA. As in our sample technology a transistor of size 1 according to Fig. 4.5 is rated
for a current of 100 mA, we select size 10 for T1 and the minimum size 1 for T2 ; thus,
IS2 /IS1 = 0.1. For the desired current ratio kI = Io /Ii = 0.01, this means that the
exponential factor in (4.13) must also be 0.1; this leads to VR2 = VT ln 10 ≈ 60 mV and
R2 = VR2 /Io ≈ 6 k.
Three-transistor current mirror: A low current gain of the transistors has a negative
effect on the current ratio of the simple current mirror. Especially with large current ratios,
the base current of the output transistor can increase so much that the current ratio clearly
deviates from the dimensional ratio of the transistors. This means that the current ratio no
longer depends solely on the geometric sizes, but also to an increasing degree on the current
gain that is affected by imminent tolerances. This can be overcome by the three-transistor
current mirror shown in Fig. 4.20a, in which the base current for transistors T1 and T2 is
supplied via an additional transistor T3 . This, in turn, contributes to the input current Ii
Vb Vb Vb
RV
Io Ii
Ii
T3 T3
I o1 I o2
T2a T2b
T1 T2 Vo T1
R1 R2 R1 R2a R2b
with its very low base current, which significantly reduces the dependence on the current
gain.
Without feedback resistances – that is, with R1 = R2 = 0 – one arrives at VBE 1 =
VBE 2 , and by ignoring the Early effect at:
IC2 IS2
=
IC1 IS1
Inserting the nodal equations
Ii = IC1 + IB3 , IB1 + IB2 = IC3 + IB3 , Io = IC2
leads to the current ratio where IB1 = IC1 /B, IB2 = IC2 /B and IB3 = IC3 /B:
B2 + B B1
IS2
kI = # $ ≈ (4.14)
IS1 IS1
B2 + B + 1 + 1
IS2
n-channel current mirror: Figure 4.21 shows the currents and voltages in a simple cur-
rent mirror with n-channel MOSFETs, otherwise known as the n-channel current mirror.
In the normal operating range, both MOSFETs operate in the pinch-off region and the
basic (3.3) applies:
K1
ID1 = (VGS1 − Vth )2
2
(4.15)
K2 VDS2
ID2 = (VGS2 − Vth )2 1 +
2 VA
4.1 Circuits 289
Vb
RV I i = I D1 I= 0 I o = I D2
T1 T2
Vi Vo
VGS1 VGS2
Fig. 4.21. Currents and
R1 VR1 VR2 R2
voltages in the n-channel
current mirror
In T1 , the Early effect is neglected due to VDS1 = VGS1 VA . Since no gate current flows
in MOSFETs, the currents at the input and the output correspond to the drain currents:
Ii = ID1 , Io = ID2 (4.16)
The following mesh equation is also obtained from Fig. 4.21:
ID1 R1 + VGS1 = ID2 R2 + VGS2 (4.17)
If the Early voltage VA is sufficiently high, the current ratio is only influenced by the ratio
of the MOSFET sizes.
The output characteristic of the n-channel current mirror together with the output
characteristic of an npn current mirror of the same design is shown in Fig. 4.17 on page 284.
What can be seen here is that the operating range of the n-channel current mirror is narrower
due to Vo,min = VDS,po > VCE,sat . The output voltage limit, however, is not constant but
depends on the size of the MOSFETs, since:
VDS,po VA
2ID
Vo,min = VDS,po = VGS − Vth ≈
K
It is therefore possible to lower the output voltage limit by increasing the sizes of the
MOSFETs. In integrated analog circuits the operating points are usually given by VGS −
Vth ≈ 1 V; consequently, Vo,min ≈ 1 V. In order to achieve an output voltage limit of
Vo,min ≈ 0.1 . . . 0.2 V, as is the case in an npn current mirror, the MOSFET dimensions
would have to be increased by a factor of 25 . . . 100. In practice, this is only possible
in exceptional cases, because the gate capacitance increases by the same factor and the
transit frequency is reduced accordingly; when used as a current source, the higher output
capacitance is problematic.
n-channel current mirror with feedback: Calculating the current ratio is not easily
done in this case, as the voltages across resistances R1 and R2 not only exert an influence
on (4.17), but also cause a shift in the threshold voltages due to the substrate effect; this
is due to VBS1 = − VR1 and VBS2 = − VR2 . If both voltages are identical, the substrate
290 4 Amplifiers
effect has the same influence on both MOSFETs and both threshold voltages increase to
the same degree; this requires the resistances to be chosen according to the sizes of the
MOSFETs:
K2 R1
=
K1 R2
In this case, one arrives at the same current ratio as with the n-channel current mirror
without feedback.
The feedback increases the output resistance of the current mirror; this will be explained
in more detail below. In contrast, the output voltage limit is increased by the voltage drop
across the resistances:
ID2 =Io 2Io
Vo,min = VDS2,po + VR2 = VDS2,po + ID2 R2 = + I o R2
K2
Operation as a current source: The simple n-channel current mirror can be operated as
a current source by adding resistance RV as shown in Fig. 4.21; this makes for a constant
input current. From Vi = VGS1 + VR1 and Vb = Vi + Ii RV , it follows
Vb − VGS1
Ii =
RV + R 1
The output current is Io = kI Ii .
Output resistance: The output current of a current mirror should depend on the in-
put current alone and not on the output voltage; this means that the small-signal output
resistance
∂Vo
vo
ro = =
∂Io
Ii =const. io
ii =0
should be as high as possible. This value can be determined from the slope of the output
characteristic in the operating range or with the help of the small-signal equivalent circuit.
As the definition directly implies, the input is supplied by an ideal current source: Ii =
const. or ii = 0. Strictly speaking, this is the open-circuit output resistance. In the small-
signal equivalent circuit, the open-circuit condition at the input is expressed by the fact
that the input is open; that is, with no circuitry. In practical applications, however, a true
open circuit condition never exists at the input, although the difference between the real
output resistance and the open-circuit output resistance is generally negligibly low.
The npn current mirror corresponds to the small-signal equivalent circuit shown in
Fig. 4.22; for the transistors, the small-signal equivalent circuit of Fig. 2.12 on page 44
is used. The left section with transistor T1 and resistor R1 can be combined to give one
resistance Rg :9
1 1
Rg = R1 + ≈ R1 +
1 1 gm1
gm1 + +
rBE 1 rCE1
This results in almost the same small-signal equivalent circuit as in the case of the common-
emitter circuit with current feedback, as a comparison with Fig. 2.65 on page 105 reveals;
io
T1 T2
R1 R2
Rg
only resistance RC and source vg are omitted. Therefore, the output resistance of the current
mirror can be derived from the short-circuit output resistance of the common-emitter circuit
with current feedback:
⎛ ⎞
rBE 2 + Rg rCE2 >rBE 2 +Rg
β +
⎜ r ⎟ β1 βR2
⎜
ro = rCE2 ⎝1 + CE2 ⎟ ≈ rCE2 1 +
rBE 2 + Rg ⎠ R2 + rBE 2 + Rg
1+
R2
If rBE 2 1/gm1 , replacing Rg leads to:
vo
βR2
ro = ≈ rCE2 1 + (4.19)
io
ii =0 R1 + R2 + rBE 2
10 The use of amplifiers or positive feedback can achieve even higher output resistances, but only
with very accurate matching.
292 4 Amplifiers
io
T2
vBS2
R2
Fig. 4.23. A small-signal equivalent circuit for calculating the output resistance of an n-channel
current mirror
this enables us to derive the output resistance; for gm2 1/rDS2 , we obtain the following:
vo
% % & &
ro =
≈ rDS2 1 + gm2 + gm,B2 R2 (4.20)
io ii =0
rCE gm R2
rDS gm R2
npn
n-channel
gm
Fig. 4.24. Output resistances of an npn current mirror and an n-channel current mirror with the
current ratio kI = 1, Ii = Io = 100 mA and R1 = R2
This suggests that the maximum output resistance can be achieved with an ohmic feed-
back resistance if an output voltage limit of Vo,min ≈ VR2 +VCE,sat ≈ 2.8 V is accepted.
With lower current gains, the voltage limit is reduced accordingly.
– Due to the significantly lower transconductance of the MOSFETs in n-channel current
mirrors, higher feedback resistances must be used in order to achieve output resistances
that are as high as those in npn current mirrors; in this case, R2 must be replaced by
a current source, which means that the simple current mirror is converted to a cascode
current mirror.
Io Io
Io
B
Ii T3 Ii T3
I'o I'o
VB
VB
T1 T2 T1 T2
npn current mirror with cascode: The current ratio kI of the npn current mirror with
cascode shown in Fig. 4.25a can be calculated with the help of the current ratio of the
simple current mirror; for the current mirror that consists of T1 and T2 we obtain, from
(4.6):
Io
1
=
Ii IS1 1 1
1+ +
IS2 B B
The Early effect has no influence, since T2 is driven with the almost constant collector-
emitter voltage VCE2 = VB − VBE 3 ≈ VB − 0.7 V. Inserting
Io
Io
= Io +
B
leads to:
B1
Io 1 IS2
kI = = 2 ≈ (4.21)
Ii IS1 1 1 1 IS1
1+ + + 2
IS2 B B B
n-channel current mirror with cascode: In the n-channel current mirror with cascode
shown in Fig. 4.25b, Io = Io
; together with (4.18), this leads to
Io K2
kI = = (4.22)
Ii K1
4.1 Circuits 295
Io
µA Operating range of the npn current mirror with cascode
npn
100
80
n-channel
Operating range of the n-channel current
60
mirror with cascode
40
20
Vo,min,npn Vo,min,nc
0
0 1 2 3 4 Vo
V
Fig. 4.26. Output characteristics of an npn and an n-channel current mirror with cascode
Here too, the current ratio only depends on the dimensional ratios of MOSFETs T1 and T2 .
Output characteristics: Figure 4.26 shows the output characteristics of an npn and an
n-channel current mirror with cascode. The characteristic of the npn current mirror with
cascode is practically horizontal for Vo > Vo,min,npn , indicating that the output resistance
is very high. With VCE,sat ≈ 0.2 V and VBE ≈ 0.7 V, the output voltage limit is:
VCE2 > VCE2,sat must be the case to ensure that T2 operates in normal mode; consequently,
For the borderline case VB = 0.9 V, the output voltage limit is Vo,min,npn = 2VCE,sat ≈
0.4 V. The characteristic is inflected below the voltage limit.
For the n-channel cascode current mirror with Vo > Vo,min,nc , the characteristic is also
horizontal; here, the output voltage limit is:
Here, VDS3,po = VGS3 − Vth3 . To ensure that T2 works in the pinch-off region, VDS2 >
VDS2,po must be the case; thus:
Here, VDS2,po = VGS2 − Vth2 . Typical values are Vth ≈ 1 V and VGS ≈ 1.5 . . . 2 V; this
results in VB ≈ 2 . . . 3 V and Vo,min,nc ≈ 1 . . . 2 V. For ID2 = ID3 = Io and
2ID
VGS ≈ Vth +
K
296 4 Amplifiers
the dependence of the output voltage limit on the output current and the sizes of the
MOSFETs is:
1 1
Vo,min,nc = VGS2 − Vth2 + VGS3 − Vth3 = 2Io +
K2 K3
This means that the voltage limit can be reduced by increasing the size of the MOSFETs;
but the result is only influenced by the square root of the size.
Below the output voltage limit, T3 enters the ohmic region. The current, however, is
determined by T2 and thus remains approximately constant; the output resistance, on the
other hand, is markedly reduced. A further reduction in the output voltage also drives T2
into the ohmic region and the characteristic becomes identical to the output characteristic
of T2 .
Output resistance: The output resistance of the npn current mirror with cascode is
calculated by inserting the small-signal parameters of T3 and rCE2 in place of RE in (4.1):
β rCE2
ro = rCE3 1 +
rCE2 + rBE 3
If rCE2 ≈ rCE3 = VA /Io , rCE2 rBE 3 and β 1, then:
vo
ro = ≈ β rCE3 (4.23)
io
ii =0
According to (4.2), for the n-channel current mirror with cascode we obtain:
% % & &
ro = rDS3 1 + gm3 + gm,B3 rDS2
If rDS2 = rDS3 = VA /Io and gm3 rDS2 1, then
vo
% & 2
ro =
≈ gm3 + gm,B3 rDS3 ≈ µ3 rDS3 (4.24)
io ii =0
npn cascode current mirror: The current ratio of the npn cascode current mirror shown
in Fig. 4.27a can be calculated using the current ratio of the simple current mirror; according
to (4.6), the ratio for the current mirror consisting of T1 and T2 is:
Io
1
= I
Ii S1 1 1
1+ +
IS2 B B
4.1 Circuits 297
Io Io
Ii Ii
Io
B T3 T4
T3 T4
Vo Vo
Ii ' Ii ' Io'
Io'
T1 T2
T1 T2
The Early effect has no noticeable influence here, since T2 is driven by the almost constant
collector-emitter voltage VCE2 = VBE 1 + VBE 3 − VBE 4 ≈ 0.7 V.
Io Io
Ii = Ii
+ , Io
= Io +
B B
leads to:
B1
Io 1 IS2
kI = = 2 ≈ (4.25)
Ii IS1 1 2 1 IS1
1+ + + 2
IS2 B B B
The current ratio only depends on the ratio of the sizes of transistors T1 and T2 ; T3 and T4
have no influence. As kI is independent of the output voltage Vo , the first approximation
of the output resistance is infinite.
Io K2
kI = = (4.26)
Ii K1
Here too, the current ratio depends solely on the dimensional ratio of MOSFETs T1 and T2 .
298 4 Amplifiers
Io
µA
Operating range of the npn cascode current mirror
npn
100
80 n-channel
Operating range of the n-channel
cascode current mirror
60
40
20
Vo,min,npn Vo,min,nc
0
0 1 2 3 4 Vo
V
Fig. 4.28. Output characteristics of an npn and an n-channel cascode current mirror
Output characteristics: Figure 4.28 shows the output characteristics of an npn and an
n-channel current mirror with cascode. The characteristic of the npn current mirror with
cascode is practically horizontal for Vo > Vo,min,npn , indicating that the output resistance
is very high. With VCE,sat ≈ 0.2 V and VBE ≈ 0.7 V, the output voltage limit is:
Vo,min,npn = VBE 1 + VBE 3 − VBE 4 + VCE4,sat ≈ 0.9 V
It is higher than that of the current mirror with cascode, which reaches Vo,min,npn ≈ 0.4 V
at the minimum voltage VB .
For Vo > Vo,min,nc , the characteristic of the n-channel cascode current mirror is also
horizontal; the output voltage limit is:
Vo,min,nc = VGS1 + VGS3 − VGS4 + VDS4,po = VGS1 + VGS3 − Vth4
Here, VDS4,po = VGS4 − Vth4 . Typical values are Vth ≈ 1 V and VGS ≈ 1.5 . . . 2 V; this
results in Vo,min,nc ≈ 2 . . . 3 V. If we assume that all MOSFETs have the same threshold
voltage Vth , that is, if we neglect the substrate effect, then for ID1 = ID3 = Ii and
2ID
VGS ≈ Vth +
K
the dependence of the output voltage limit on the input current and the sizes of the MOS-
FETs is:
1 1
Vo,min,nc ≈ Vth + 2Ii +
K1 K3
This suggests that the voltage limit can be reduced by increasing the sizes of the MOSFETs;
but only the square root of the size influences the result. The lower limit is given by
Vo,min,nc = Vth and can be approximately reached only with very large MOSFETs. Below
4.1 Circuits 299
T3 T4
io
1 gm4vBE4 rCE4
v BE4 rBE4
gm3
T1 T2
vo
1 gm2vBE2 rCE2
v BE2 rBE2
gm1
the output voltage limit, transistor T4 enters the ohmic region. The current, however, is
influenced by T2 and thus remains approximately constant, while the output resistance
is markedly reduced. If the output voltage is further increased, T2 also enters the ohmic
region and the characteristic becomes identical to the output characteristic of T2 .
Output resistance: The small-signal equivalent circuit shown in Fig. 4.29 is used for
calculating the output resistance of the npn cascode current mirror. The following relation-
ships exist:
VA Io
rCE2 ≈ rCE4 = , gm2 ≈ gm4 =
Io VT
βVT Ii Io
rBE 2 ≈ rBE 4 = , gm1 ≈ gm3 ≈ =
Io VT k I VT
Here, VA is the Early voltage, VT is the temperature voltage, β is the small-signal current
gain of the transistors and kI is the current ratio of the current mirror. Calculating the
output resistance with kI β leads to:
vo
β β rCE4
ro =
≈ rCE4 1 + ≈ (4.27)
io ii =0 1 + kI 1 + kI
The output resistance of the cascode current mirror is higher than that of the simple current
mirror by the factor β/(1 + kI ). The maximum possible output resistance β rCE cannot
be reached, since T4 has an influence on the reference path and the voltage vBE 2 via the
base-emitter junction (see Fig. 4.29); therefore, current gm2 vBE 2 depends on the output
voltage and the output resistance of T2 is smaller than rCE2 .
In n-channel cascode current mirrors, there is no such influence on the reference path.
Thus it is possible to calculate the output resistance using (4.20), by replacing R2 by rDS2 :
% % & &
ro = rDS4 1 + gm4 + gm,B4 rDS2
300 4 Amplifiers
Vb = 5 V Vb = 5 V
RV I a = 100 µA RV I a = 100 µA
35kΩ 15kΩ
1.32 V 3.5 V
T3 T4
T3 T4
1 1
Vo 10 10 Vo
0.66 V 1.37 V
T1 T2
T1 T2
1 1
50 50
vo
% & 2
ro =
≈ gm4 + gm,B4 rDS4 ≈ µ4 rDS4 (4.28)
io ii =0
Example: An npn and an n-channel current source with an output current Io = 100 mA
is to be dimensioned for the highest possible output resistance and the lowest possible
output capacitance. The demand for a high output resistance ro requires the use of a
cascode current mirror, while the demand for a low output capacitance necessitates the use
of very small output transistors. With regard to the current ratio, there are contradictory
requirements: on the one hand, it should be as large as possible so that only a low input
current Ii = Io /kI is necessary; on the other hand, it should be as small as possible to
make the output resistance of the npn cascode current mirror as high as possible. kI ≈ 1
is selected for both current mirrors.
The circuit for the resulting npn cascode current mirror is shown in Fig. 4.30a. Here,
transistors of size 1 are used, which are rated for a collector current of 100 mA according
to Fig. 4.5; the other parameters are IS = 1 fA, B = β = 100 and VA = 100 V. According
to (4.25), the current ratio for IS1 = IS2 = IS3 = IS4 = IS is
1 1
kI ≈ = ≈ 0.96
4 1.04
1+
B
and the input current is Ii = Io /kI ≈ 104 mA. Since the collector currents of the transistors
are almost identical, the same base-emitter voltage VBE can be calculated:
Io 100 mA
VBE ≈ VT ln = 26 mV · ln ≈ 660 mV
IS 1 fA
The series resistor RV is:
Vb − VBE 1 − VBE 3 Vb − 2VBE 3.68 V
RV = ≈ = ≈ 35 k
Ii Ii 104 mA
4.1 Circuits 301
For rCE4 = VA /Io = 100 V/100 mA = 1 M, the output resistance is:
β rCE4 β rCE4
ro ≈ ≈ ≈ 50 M
1 + kI 2
The output voltage limit is Vo,min = VBE + VCE,sat ≈ 0.9 V.
The resulting circuit for the n-channel cascode current mirror is shown in Fig. 4.30b.
MOSFETs of size 10 according to Fig. 4.6 are used for T3 and T4 , since size 1 is rated for
a drain current of 10 mA and in this case 100 mA is required. MOSFETs of size 10 could
also be used for T1 and T4 ; but here MOSFETs of size 50 are used to achieve a reduction
√ on T4 , the size
of the voltage limit Vo,min . As the output capacitance is mainly dependent
of T1 and T2 has almost no bearing. K = 30 mA/V2 for size 1, γ = 0.5 V, Vinv = 0.6 V
and VA = 50 V can be taken from Fig. 4.6. The current ratio is kI = 1; consequently,
Ii = Io = 100 mA. For the MOSFETs:
mA uA
K1 = K2 = 50 K = 1.5 , K3 = K4 = 10 K = 300
V2 V2
In T1 and T2 the substrate effect has no influence, since VBS1 = VBS2 = 0; thus Vth1 =
Vth2 = Vth,0 and:
2Ii 200 mA
VGS1 = VGS2 = Vth,0 + = 1V + ≈ 1.37 V
K1 1.5 mA/V2
In contrast, the voltages in T3 and T4 are
# $
Vth3 = Vth4 = Vth,0 + γ Vinv − VBS3 − Vinv
VBS3 =VGS1 √ #√ √ $
= 1 V + 0.5 V· 1.97 V − 0.6 V ≈ 1.31 V
and:
2Ii 200 mA
VGS3 = VGS4 = Vth3 + ≈ 1.31 V + ≈ 2.13 V
K3 300 mA/V2
Consequently, the series resistor is:
Vb − VGS1 − VGS3 5 V − 1.37 V − 2.13 V
RV = ≈ ≈ 15 k
Ii 100 mA
For rDS2 = rDS4 = VA /Io = 500 k and
'
mA
gm4 = 2K4 Io = 2 · 300 mA/V2 · 100 mA ≈ 245
V
√
γ gm4 VBS4 =−VGS2 0.5 V · gm4 mA
gm,B4 = = √ ≈ 44 2
2 Vinv − VBS4 2 1.97 V V
the output resistance is:
% & 2 mA
ro ≈ gm4 + gm,B4 rDS4 ≈ 289 · (500 k)2 ≈ 72 M
V
The output voltage limit is:
Vo,min = VGS1 + VGS3 − Vth4 ≈ 1.37 V + 2.13 V − 1.31 V ≈ 2.2 V
302 4 Amplifiers
This means that with an operating voltage of 5 V, almost half of the operating voltage is
lost.
The n-channel cascode current source has a higher output resistance which, however,
comes with a disproportionately high output voltage limit, despite the fact that a reduction
has already been made by increasing T1 and T2 . If the same voltage limit as that for the npn
cascode current source is required, one can only use a simple n-channel current source,
which has a significantly lower output resistance of ro = rDS2 = 500 k; the npn cascode
current source is thus superior by a factor of 100.
Furthermore, a comparison of the cascode current mirror with the simple current mirror
with feedback is very interesting, provided that the output voltage limit is identical. The
npn cascode current mirror features a voltage limit of Vo,min = VBE + VCE,sat , which is
VBE ≈ 0.7 V higher than that of the simple npn current mirror without feedback; therefore,
feedback can be complemented with R2 = VBE /Io ≈ 7 k in order to achieve the same
voltage limit. The output resistance of the simple npn current mirror is thus:
VA Io VBE VA VBE
ro ≈ rCE2 (1 + gm R2 ) = 1+ ≈ ≈ 27 M < 50 M
Io V T Io VT Io
In this way, the output resistance of the simple npn current mirror is smaller than that of
the npn cascode current mirror, but only by a factor of 2; thus output resistances of the
same magnitude are achieved in practice with both variants. In simple n-channel current
mirrors, the voltage VGS2 ≈ 1.37 V of the n-channel cascode current mirror is available
for the feedback resistances if the same output voltage limits are also to be achieved here;
consequently, R2 ≈ 13.7 k and:
% % & & % &
ro = rDS2 1 + gm + gm,B R2 ≈ gm + gm,B R2 rDS2
mA
≈ 289 · 13.7 k · 500 k ≈ 2 M 72 M
V
The output resistance of the simple n-channel current mirror with feedback is thus signif-
icantly smaller than that of the n-channel cascode current mirror.
npn Wilson current mirror: In our calculations, we take advantage of the fact that the
Wilson current mirror contains a simple npn current mirror. With the currents Ii
and Io
,
the following applies:
Io
1
= I
Ii S2 1 1
1+ +
IS1 B B
4.1 Circuits 303
kI
Io
B
Ii Io
B
T3 1
Vo
I o' I 'i
1
B
T1 T2
1 1 B I S2
B I S1
Using
Io Io
Ii = Io
+ , Ii
= Io +
B B
the current ratio is calculated:
IS2 1 IS2
B + B1 B +1
Io IS1 B +1 IS1
kI = = ≈ (4.29)
Ii IS2 1 IS2
+B + +B
IS1 B +1 IS1
The size of transistor T3 has no influence on kI . Figure 4.31b shows the curve of kI versus
the size ratio IS2 /IS1 .
For IS1 = IS2 , we obtain:
B1
1 1
kI = ≈
2 2
1+ 1+
B + 2B
2
B2
Here, the error amounts only to 2/B 2 , compared to 2/B for the simple current mirror or
4/B for the cascode current mirror. The error of the three-transistor current mirror is also
2/B 2 , but only with the provision that all three transistors have the same current gain; but
since T3 in Fig. 4.20a carries a much lower current, in practice its current gain is lower
than that of the other transistors. This is different in the Wilson current mirror, where
IS1 = IS2 ; that is, all transistors carry approximately the same current and all transistors
have the maximum current gain, provided that they are correctly dimensioned. The fact
that the error is smallest in the Wilson current mirror with IS2 /IS1 = 1 can be seen from
the symmetry of the curve in Fig. 4.31b.
Output characteristic: The output characteristic of the Wilson current mirror corre-
sponds to that of the cascode current mirror (see Fig. 4.28 on page 298); the output voltage
304 4 Amplifiers
T3
io
T1 T2
vo
1
rCE1 gm1vBE1 r vBE1
BE1 gm2
Output resistance: The small-signal equivalent circuit shown in Fig. 4.32 is used to
calculate the output resistance of the Wilson current mirror. The following equations apply:
VA VA kI VA
rCE3 = , rCE1 ≈ = = kI rCE3
Io Ii Io
Io Ii Io gm3
gm2 ≈ gm3 = , gm1 ≈ = =
VT VT k I VT kI
βVT β βVT kI βVT kI β
rBE 3 = = , rBE 1 ≈ ≈ =
Io gm3 Ii Io gm3
Here, VA is the Early voltage, VT is the temperature voltage, β is the small-signal current
gain of the transistors and kI is the current ratio of the current mirror. For β 1, the
output resistance can be calculated as:
vo
β β rCE3 kI =1 β rCE3
ro =
≈ rCE3 1 + ≈ = (4.30)
io ii =0 1 + kI 1 + kI 2
A comparison with (4.27) shows that the output resistance of the Wilson current mirror is
the same as that of the npn cascode current mirror.
Dynamic behavior
When a current mirror is used for signal transmission, not only the output resistance is of
interest. Indeed, the frequency response of the current ratio and the step response at large
signals are also interesting. However, a general calculation of the frequency responses
is very complex, and the results are difficult to interpret because of the large number of
parameters. Therefore, the basic dynamic response of current mirrors is described on the
4.1 Circuits 305
kI
2
dB
4
0
3 1
–3
3
1 simple
–6
2 three-transistor 1,4
3 cascode 2
4 Wilson
–10
10 M 30 M 100 M 300 M 1G f
Hz
Fig. 4.33. Frequency responses of npn current mirrors with kI = 1, with a small-signal
short-circuit at the output
basis of simulation results. Four npn current mirrors are compared: the simple, the three-
transistor, the cascode and the Wilson current mirrors, each with kI = 1 and Io = 100 mA.
Figure 4.33 shows the frequency responses for a small-signal short circuit at the output
(Vo,A = 5 V or vo = 0), while Fig. 4.34 shows the responses to a jump from Io = 10 mA
to Io = 100 mA.
It can be seen that the simple current mirror features the best dynamic characteristics,
because it performs like a first-order lowpass filter. The Wilson current mirror has a some-
what higher cutoff frequency due to its complex conjugate poles, but at the cost of a step
response, which has an overshoot of about 15%. The cutoff frequency of the cascode cur-
rent mirror is lower than that of the simple current mirror by a factor of 2.5; consequently,
the settling time is prolonged accordingly. The three-transistor current mirror shows the
poorest response; it has the lowest cutoff frequency and an overshoot of more than 20%.
Io 2
4
µA
100
1,3,4
1
3
1,4
50 1 simple
2 three-transistor
2 3 cascode
4 Wilson
10
0 2 4 6 8 10 12 14 t
ns
This is caused by the low quiescent current of transistor T3 in Fig. 4.20a, which results in
a correspondingly low transit frequency.
The numeric values of the cutoff frequency, the settling time and the overshoot natu-
rally depend on the parameters of the transistors used. Other parameters would, of course,
produce other values, but the relations among the various current mirrors are almost iden-
tical.
Cascode current mirror with bias voltage: The current mirror with bias voltage shown
in Fig. 4.35a is achieved by replacing transistor T3 in the cascode current mirror of Fig. 4.27a
on page 297 by a voltage source with VCE,sat . From the equation VCE,sat + VBE 1 =
VCE2,sat + VBE 4 and VBE 1 ≈ VBE 4 , it follows that VCE2,sat ≈ VCE,sat and thus:
Vo,min = VCE2,sat + VCE4,sat = 2 VCE,sat ≈ 0.4 V
With a constant input current – that is, operating the circuit as a current source – the
bias voltage can be generated with a resistor (see Fig. 4.35b); if the base current of T4 is
neglected, then:
VCE2,sat
R1 ≈
Ii
The current ratio and the output resistance remain almost unaltered (see (4.25) and (4.27)).
Since the collector-emitter voltages of T1 and T2 are no longer practically identical as in
the case of the cascode current mirror, the current ratio shows a slight dependence on the
Early voltage of the transistors.
Ii Io Ii Io
T4 VCE4,sat T4
VCE,sat VBE4 R1
VBE1 T1 T2 VCE2,sat T1 T2
Vb Vb
Ii Io Ii Io
I0
T3 I0
T4
T3 T4
R1
( R1 )
VV
VV
T5 T1 T2
T5 T1 T2
The same procedure can be used in n-channel cascode current mirrors, according to
Fig. 4.27b. The following thus apply:
1 1
Vo,min = VDS2,po + VDS4,po = 2Io +
K2 K4
and:
VDS2,po
R1 =
Ii
The bias voltage may also be generated in a separate bias voltage path (see Fig. 4.36);
in Fig. 4.36a the following must be the case:
VV ≈ VBE 5 + I0 R1 > VCE2,sat + VBE 4
and in Fig. 4.36b
VV = VGS5 + I0 R1 > VDS2,po + VGS4
As the bias voltage is generated separately, the circuits, unlike that shown in 4.35b, may
be operated with variable input currents – that is, as current mirrors – provided that they
are dimensioned such that the conditions mentioned above are also met for the maximum
current; in other words, with a maximum VBE 4 or VGS4 . The circuits also function without
transistor T3 ; but then the collector-emitter or the drain-source voltages of T1 and T2 are
no longer identical, and the current ratio shows a slight dependence on the Early voltage
of the transistors. With the use of MOSFETs, R1 can be omitted if I0 is selected to be high
enough and the size of T5 is small enough to make VGS5 > VDS2,po + VGS4 .
Double cascode current mirror: Fig. 4.37a shows the npn double cascode current
mirror: compared to the cascode current mirror, the collector of T4 is connected to the bias
voltage Vb and a second cascode with T5 and T6 is added. When T5 and T6 are operated
with VCE > VCE,sat , the current ratio is
Io IS5
kI = ≈
Ii IS1
308 4 Amplifiers
Ii Vb Ii Vb
Io Io
T3 T4 T3 T4
1 10 1 10
T6 T6
10 1
T1 T2 T5 T1 T2 T5
1 1 1
1 1 1
R2
vo
β VA
ro =
≈ β rCE6 =
io ii =0 Io
Contrary to the cascode current mirror, there is no factor (1 + kI ) in the denominator, since
any influence on the reference path by T6 is prevented by T4 .
The sizes of the transistors can be selected such that T5 is operated with VCE5 ≈ VCE,sat
and the following output voltage limit is reached:
Vo,min = VCE5,sat + VCE6,sat = 2 VCE,sat ≈ 0.4 V
Using the equation
VBE 1 + VBE 3 = VBE 4 + VCE5 + VBE 6
under the conditions
IC1 ≈ IC3 ≈ Ii
IS2
IC4 ≈ IC2 ≈ Ii
IS1
IC5 ≈ IC6 = Io = kI Ii
and VBE ≈ VT ln(IC /IS ), we obtain:
IS4 IS6
VCE5 ≈ VT ln
kI IS2 IS3
For the dimensional ratios in Fig. 4.37a, it follows:
10 · 10
VCE5 ≈ VT ln = VT ln 100 ≈ 26 mV · 4.6 ≈ 120 mV
1·1·1
Although the voltage is below the saturation voltage VCE,sat ≈ 0.2 V that has been assumed
so far, it is sufficient in most practical applications. This becomes obvious if we consider
the output resistance and the current ratio as a function of VCE5 (see Fig. 4.38): for
VCE ≈ 120 mV the current ratio is almost one and the output resistance ro ≈ 30 M
4.1 Circuits 309
kI ro
MΩ
1.0
100
50
20
0.5
10
0
0 50 100 150 200 VCE5
mV
Fig. 4.38. Dependence of the current ratio kI and the output resistance ro on VCE5 in the npn
double cascode current mirror
amounts to one third of the maximum possible value. Making VCE = 200 mV will produce
better values, but requires size 50 for T4 and T6 :
50 · 50
VCE5 ≈ VT ln = VT ln 2500 ≈ 200 mV
1·1·1
On account of their high spatial requirements, transistors of this size are only used in
integrated circuits if it is absolutely essential for the function of the circuit. Generally, T4
and T5 are both of the same size, as this reduces the spatial requirements for the required
value of VCE5 to a minimum.
A drawback of the circuit shown in Fig. 4.37a is the high output capacitance caused by
the size of T6 . In order to reduce the size of T6 by a factor of 10 to size 1, it is necessary to
either increase the size of T4 by a factor of 10 to size 100 or to reduce current IC4 ≈ IC2
by a factor of 10. The latter option is achieved by reducing T2 by a factor of 10 or, if this
is not feasible because T2 is already at the minimum size, by increasing the size of all of
the other transistors. If the current mirror is to be used as a current source, current IC2
may also be reduced by providing T2 with a feedback resistance; this leads to the double
cascode current mirror with Widlar stage shown in Fig. 4.37b.
In Fig. 4.37a, the collector of T4 can also be used as an additional output; IC4 is thus
the output current of a cascode current mirror with kI ≈ IS2 /IS1 , and IC6 is the output
current of the double cascode current mirror with kI ≈ IS5 /IS1 .
Fig. 4.39 shows the n-channel double cascode current mirror. When T5 and T6 are
driven with VDS > VDS,po , the current ratio is
Io K5
kI = ≈
Ii K1
and the output resistance is:
vo
% & 2
ro =
≈ gm6 + gm,B6 rDS6
io ii =0
310 4 Amplifiers
Vb
Ii
Io
T3 T4
10 50 T6
50
T1 T2 T5
10 1 10
√
Neglecting the substrate transconductance gm,B6 and using gm6 = 2K6 Io and rDS6 =
VA /Io leads to:
gm,B6 gm6
2K6
ro ≈ VA2
Io3
For the circuit in Fig. 4.39, where K6 = 50K = 1.5 mA/V2 , VA = 50 V and Io = 100 mA,
we obtain an output resistance of ro ≈ 140 M.
The output voltage limit is minimum when T5 is driven with VDS5 = VDS5,po :
Vo,min = VDS5,po + VDS6,po
From the equation
VGS1 + VGS3 = VGS4 + VDS5 + VGS6
under the conditions
VGS = Vth + 2ID /K
and ID1 = ID3 = Ii , ID2 = ID4 = Ii K2 /K1 and ID5 = ID6 = Io = Ii K5 /K1 , we obtain:
VDS5 = Vth1 + Vth3 − Vth4 − Vth6
2Io K1 K6 K6 K2 K6
+ + − −1
K6 K3 K5 K5 K4 K5
For the circuit in Fig. 4.39, where Vth = Vth1 + Vth3 − Vth4 − Vth6 , we obtain:
K6 =1.5 mA/V2
2Io #√ √ √ $ Io =100mA
VDS5 ≈ Vth + 5 + 5 − 0.1 − 1 ≈ Vth + 1.15 V
K6
Voltage Vth combines the differences in the threshold voltages caused by the substrate
effect; this voltage is always negative and cannot be calculated directly. Simulation with
PSpice yields Vth ≈ − 0.3 V and VDS5 = 0.85 V. It thus follows that:
2ID5 2Io
VDS5 > VDS5,po = = ≈ 0.82 V
K5 K5
4.1 Circuits 311
Vb
Io Io
I0
T4 T4
A
Ii Ii
T3
T1 T2 V DS2 Vset T1 T2
√
For VDS6,po = VGS6 − Vth6 = 2Io /K6 ≈ 0.37 V, the output voltage limit is Vo,min =
VDS5,po + VDS6,po ≈ 1.2 V. A further reduction of Vo,min is achieved by increasing the
sizes of MOSFETs T1 , T2 and T5 proportionally; this decreases VDS5,po in accordance with
the increase in K5 .
Controlled cascode current mirror: Removing MOSFET T3 from the cascode current
mirror in Fig. 4.27b and adjusting the gate voltage of T4 by means of a control amplifier
leads to the controlled cascode current mirror shown in Fig. 4.40a; with a sufficiently high
gain A for the control amplifier, the gate voltage of T4 is adjusted such that VDS2 ≈ Vset .
For Vset ≈ VDS2,po , the circuit represents a current mirror with minimum outut voltage
limit Vo,min .
If a simple common-source circuit is used as the control amplifier, we obtain the circuit
shown in Fig. 4.40b; voltage Vset occurs as the gate-source voltage of T3 at the operating
point:
2I0
Vset = VGS3 = Vth3 +
K3
In general, all MOSFETs are operated with VGS < 2Vth and VDS,po = VGS − Vth < Vth ; in
this case, Vset = VGS3 > VDS2,po ; that is, T2 operates in the pinch-off region. If Vset is to be
kept small in order to achieve the lowest possible output voltage limit, then current I0 must
be small and MOSFET T3 large; however, this makes the bandwidth of the control amplifier
very narrow. In practice, a suitable compromise between voltage range and bandwidth must
be found for each application.
The output resistance is calculated according to the small-signal equivalent circuit in
Fig. 4.41:
rDS2 =rDS4
vo
io
–A
Fig. 4.41. Small-signal equivalent circuit of the controlled n-channel cascode current mirror
√
control amplifier, then A = gm3 rDS3 = 2K3 /I0 VA ; for I0 = 10 mA, K3 = 30 mA/V2
(T3 of size 1) and VA = 50 V, we obtain A ≈ 120. This leads to output resistances in the
G range.
In principle, the controlled cascode current mirror can also be designed with npn
transistors, but in this case it is not possible to use a simple common-emitter circuit as
the control amplifier. This is due to the fact that correct functioning requires the input
resistance ri,amp of the control amplifier to be larger than the output resistance of T2 (rDS2
for MOSFET or rCE2 for bipolar transistor). This condition is automatically met in the
case of MOSFETs, while bipolar transistors require complicated circuitry to achieve an
input resistance ri,amp that is sufficiently high. Similar conditions exist at the output: with
MOSFETs, transistor T4 puts no load on the control amplifier, which can therefore have
a high-resistance output, while with bipolar transistors the input resistance of T4 requires
a low-resistance amplifier output. A bipolar control amplifier must therefore consist of
several stages. With an ideal amplifier (ri,amp = ∞ and ro,amp = 0), the same output
resistance can be realized as in the controlled n-channel cascode current mirror: ro ≈
2 .
Agm4 rCE4
4.1.2
Cascode circuit
When calculating the cutoff frequencies of the common-emitter and common-source circuit
according to Sects. 2.4.1 and 3.4.1, the Miller effect is a limiting factor. This effect is caused
11 In the computer-aided design of discrete circuits, it is necessary to take into account the fact that
in simulation all transistors of the same type have the same data, as the simulation always uses
the same model. Therefore, the insensitivity toward parameter variations must be demonstrated
by selective parameter variations in individual transistors. This can be done using, for example,
the Monte Carlo analysis, which stochastically varies certain parameters.
4.1 Circuits 313
Vb Vb
CM RC RC
Rg Rg
T1 Vo T1 Vo
CM
Vi Vi CM
Vg Vg
(1+ A )
by the voltage drop across the Miller capacitance CM connected between the base and the
collector or the gate and the drain:
A<0
|A|1
1
vi − vo = vi − Avi = vi (1 + |A|) = − vo 1 + ≈ − vo
|A|
Here, A < 0 is the gain of the common-emitter or common-source circuit. Therefore,
the Miller capacitance affects the input side with factor (1 + |A|) and the output side
with factor (1 + 1/|A|) ≈ 1; Fig. 4.42 illustrates this, using a common-emitter circuit as
an example.12 The equivalent input capacitance CM (1 + |A|) and the internal resistance
Rg of the signal source form a lowpass filter with a relatively low cutoff frequency; this
significantly reduces the cutoff frequency of the circuit for medium and, in particular, high
internal resistances. In bipolar transistors the collector capacitance CC acts as the Miller
capacitance, compared with the gate-drain capacitance CGD in FETs.
This problem is overcome by the cascode circuit, in which a common-emitter (CE)
and a common-base (CB) or a common-source (CS) and a common-gate (CG) circuit are
connected in series; Fig. 4.43 shows the resulting circuits. At the operating point, the same
current flows through both transistors if the base current of T2 is ignored in the npn cascode
circuit: IC1,A ≈ IC2,A ≈ I0 or ID1,A = ID2,A = I0 . In the npn cascode circuit with
vo ri,CB
A = = ACE ACB
vi ro,CE + ri,CB
rCE1 1/gm2
1/gm2
= − gm1 rCE1 gm2 RC ≈ − gm1 RC
rCE1 + 1/gm2
the same gain is obtained as in the simple common-emitter circuit. On the other hand, the
operational gain of the common-emitter circuit in the cascode only amounts to:
AB,CE ≈ − gm1 ri,CB = − gm1 /gm2 ≈ − 1
Therefore, the equivalent input capacitance is CM (1 + |A|) ≈ 2CM ; in other words, the
Miller effect is avoided. No Miller effect occurs in the common-base circuit of the cascode,
because the base of T2 is at a constant voltage; the collector capacitance of T2 thus only
influences the output. These characteristics apply equally to the n-channel cascode circuit.
However, the transconductances gm1 and gm2 are only identical if the MOSFETs are of
the same size: K1 = K2 .
12 Please note that the voltages in Fig. 4.42 are large-signal voltages, but only the small-signal
portion influences the calculated result.
314 4 Amplifiers
Vb Vb
RC RD
T2 T2
VB
VB
Vo Vo
T1
T1 VCE1 VDS1
Vi Vi
A voltage source VB is required to set the operating point (see Fig. 4.43). Voltage VB
must be selected such that:
VCE1 = VB − VBE 2 > VCE1,sat or VDS1 = VB − VGS2 > VDS1,po
T1 thus operates in the normal mode or pinch-off region. Consequently13 :
,
VCE1,sat + VBE 2 ≈ 0.8 . . . 1 V
VB >
VDS1,po + VGS2 = VGS1 − Vth1 + VGS2 ≈ 2 . . . 3 V
VB is selected as close to the lower limit as possible, in order to have a maximum output
voltage range. In the npn cascode circuit, the voltage drop across two diodes is often used,
that is, VB ≈ 1.4 V – if the resultant lower voltage range is acceptable.
13 The values for the npn and the n-channel cascode are given in one equation above one another,
following after a brace.
4.1 Circuits 315
Vb Vb Vb Vb
T3 T3
I0 roS I0 roS
I0 I0
roC roC
T2 T2
VB
Vo VB Vo
T1
T1
Vi Vi
The output resistances for the simple current source are roS = rCE3 or roS = rDS3 . This
leads to the following equations for the cascode circuit with a simple current source:
cascode circuit with a simple current source
roS roC
vo
− gm1 rCE3
A = = − gm1 (roC || roS ) ≈ (4.31)
vi
io =0 − gm1 rDS3
vi rBE 1
ri = = (4.32)
ii ∞
roS roC
vo
rCE3
ro = = roS || roC ≈ (4.33)
io
v =0 i
rDS3
For the npn cascode with gm1 ≈ I0 /VT and rCE3 ≈ VA,pnp /I0 , it follows:
VA,pnp
A ≈ − (4.34)
VT
Here, VA,pnp is the Early voltage of the pnp
√ transistor T3 and VT is the temperature voltage.
For the n-channel cascode with gm1 = 2K1 I0 and rDS3 = VA,pC /I0 , this leads to:
2K1 2VA,pC
A ≈ − VA,pC = − (4.35)
I0 VGS1 − Vth,nC
Here, VA,pC is the Early voltage of the p-channel MOSFETs and Vth,nC is the threshold
voltage of the n-channel MOSFETs. If npn and pnp transistors or n-channel and p-channel
316 4 Amplifiers
MOSFETs have the same Early voltage, the magnitude of the gain corresponds to the
maximum gain µ of the common-emitter or common-source circuit:
⎧
⎪ VA
⎪
⎨ gm rCE = V ≈ 1000 . . . 6000
T
|A| ≈ µ =
⎪
⎪ 2VA
⎩ gm rDS = ≈ 40 . . . 200
VGS − Vth
Here again, the low transconductance of the MOSFET compared to the bipolar transistor
shows its negative effect.
Cascode circuit with a cascode current source: The gain increases further if the output
resistance roS is enhanced by the use of a current source with cascode to:
,
β3 rCE3 gm3 gm,B3
roS ≈ % & 2
gm3 + gm,B3 rDS3 ≈ 2
gm3 rDS3
This leads to the following equations for the cascode circuit with a cascode current source
shown in Fig. 4.45:
cascode circuit with a cascode current source
vo
vo
β2 rCE2 || β3 rCE3
ro = = r || r ≈ (4.37)
io
v =0
oS oC 2
gm2 rDS2 || gm3 rDS3
2
i
Vb Vb Vb Vb
T4 T4
I0 I0
T3 T3
VB,S
VB,S
T2 T2
VB,C
VB,C
Vo Vo
T1 T1
Vi Vi
Strictly speaking, the name cascode circuit with a cascode current source is not correct,
since in Fig. 4.45 a current mirror with a cascode is used as a current source rather than
a cascode current source; the correct term cascode circuit with a current source with a
cascode is, however, too long-winded. If a true cascode current mirror is used for the
current source, then the gain of the npn cascode is reduced by a factor of 2/3 since,
according to (4.27), for a current ratio kI = 1 the cascode current mirror has an output
resistance of only roS = β3 rCE3 /2 instead of roS = β3 rCE3 in the current mirror with a
cascode. For the n-channel cascode, the two versions are equivalent.
In cascode circuits with bipolar transistors, replacing the small-signal parameters
leads to
1
A ≈ − (4.38)
1 1
VT +
βnpn VA,npn βpnp VA,pnp
and for the cascode circuit with MOSFETs of the same size (K1 = K2 = K3 = K):
2K 4
A ≈ −
= −
(4.39)
1 1 1 1
ID 2
+ 2 (VGS − Vth )2
2
+ 2
VA,nC VA,pC VA,nC VA,pC
If the Early voltages and current gains of the npn and pnp transistors and the Early voltages
of the n- and p-channel MOSFETs are the same, then:
⎧
⎪
⎪ β gm rCE β VA β≈100
⎪
⎨ = ≈ 50.000 . . . 300.000
2 2VT
|A| ≈ 2
⎪
⎪ g2 r 2
⎪
⎩ m DS = 2
VA
≈ 800 . . . 20.000
2 VGS − Vth
It is therefore possible to achieve a gain in the region of 105 = 100 dB with one npn cascode
circuit; in contrast, the n-channel cascode circuit yields a maximum of about 104 = 80 dB.
Operational gain: The high gain of the cascode circuit is a result of the high output
resistance of the cascode and the current source:
ro = roC || roS
For β = 100, VA = 100 V and IC = 100 mA, the output resistance of the npn cascode
circuit with a cascode current source is ro = β rCE /2 = 50 M, and for K = 300 mA/V2 ,
VA = 50 V and ID = 100 mA, the output resistance of the n-channel cascode circuit with a
cascode current source is ro = gm rDS2 /2 = 31 M; the same values are assumed for the
npn and pnp transistors, as well as for the n-channel and p-channel transistors.
With load RL , the operational gain
RL
AB = A = − gm (ro || RL )
ro + R L
only reaches a value close to A if the value of RL is comparable to that of ro . In most
cases, an additional amplifier stage with input resistance ri,n is connected to the output of
the cascode circuit. If, in a CMOS circuit, the subsequent stage is a common-source or
common-drain configuration, then the cascode circuit achieves the maximum operational
gain AB = A without any special measures, since RL = ri,n = ∞. Bipolar circuits, on the
other hand, require one or several common-collector circuits for impedance conversion;
318 4 Amplifiers
Current cource
roS CoS
T2
RB2 CC2
vo
T1
Rg RB1 CC1
Fig. 4.46. Full small-signal equivalent circuit diagram of an npn cascode circuit
but the output resistance of each common-collector circuit is ro ≈ Rg /β, which means
that each collector circuit reduces the output resistance by the factor β. For β = 100 and
ro = 50 M, the output resistance is reduced to ro ≈ 500 k with one common-collector
circuit and ro ≈ 5 k with two common-collector circuits. In many operational amplifiers,
a cascode circuit with a cascode current source and three subsequent complementary
common-collector circuits is used; this results in A ≈ 2 · 105 and ro ≈ 50 .
This results in the simplified small-signal equivalent circuit shown in the upper part of
Fig. 4.47. Modification of the drawing leads to the equivalent circuit shown in the lower
4.1 Circuits 319
CC2
gm2vBE2 ro C'o
gm2vBE2
vo
vo
gm2vBE2 ro Co
Rg' CC1
Fig. 4.47. A simplified small-signal equivalent circuit diagram of the npn cascode circuit
gmvBE2 vo
ro Co = RL CL
2CC + 2CS
Rg Cc
Fig. 4.48. A simplified small-signal equivalent circuit of the npn cascode circuit, with identical
small-signal parameters for all transistors and an ohmic-capacitive load
C1 E2 BE 1
+ CES rE2
Rg + rBE 1
# $
c2 = (CE1 CC1 + CE1 CES + CC1 CES ) Rg
|| rBE 1 rE2
where the polarity of vBE 2 is taken into account.
In this case gm1 ≈ gm2 ≈ 1/rE2 , since both transistors are operating with almost
identical currents; consequently, gm1 rE2 ≈ 1. By neglecting the zero, the s 2 term in the
denominator and the middle term in c1 , one arrives at an approximation by a first-order
lowpass filter:
rBE 1 1
A1 (s) ≈
# $ C
Rg + rBE 1
1 + s (CE1 + 2CC1 ) Rg
|| rBE 1 +
ES
gm1
The small-signal equivalent circuit shown in Fig. 4.48 is achieved with Rg
= Rg +
RB1 ≈ Rg , an ohmic-capacitive load and on the assumption of identical small-signal
parameters for all transistors. By combining A1 (s) and A2 (s) according to (4.40), again
ignoring the s 2 term and replacing ro by ro || RL and Co by Co + CL , an approximation
for the frequency response of the cascode circuit is achieved:
A0
AB (s) ≈
CE + CS
1 + s (CE + 2CC ) R1 + + (2CC + 2CS + CL ) R2
gm
A0
≈ (4.41)
1 + s ((CE + 2CC ) R1 + (2CC + 2CS + CL ) R2 )
βR2
A0 = AB (0) = − (4.42)
Rg + rBE
4.1 Circuits 321
R1 = Rg || rBE
R2 = ro || RL
1
ω-3dB = 2πf-3dB ≈ % &
(CE + 2CC ) Rg || rBE + (2CC + 2CS + CL ) (ro || RL )
(4.43)
The cutoff frequency depends on the low-frequency gain A0 . On the assumption that
a change in A0 is caused by a change in R2 = ro || RL and that all other values remain
constant, a description with two time constants that are independent of A0 is achieved by
solving (4.42) for R2 and inserting it in (4.43):
1
ω-3dB (A0 ) = (4.44)
T1 + T2 |A0 |
% &
T1 = (CE + 2CC ) Rg || rBE (4.45)
Rg 1
T2 = (2CC + 2CS + CL ) + (4.46)
β gm
Due to the high gain, |A0 | T1 /T2 is generally the case; consequently:
1
ω-3dB ≈
T2 |A0 |
Thus, the cutoff frequency is inversely proportional to the gain, which leads to a constant
gain-bandwidth product (GBW ):
1
GBW = f-3dB |A0 | ≈ (4.47)
2π T2
on the other hand, is independent of the gain. In what follows, the time constant T2 will
be compared instead of GBW, since this is easier to illustrate (see (4.47)): a smaller time
constant T2 gives a higher GBW and thus a higher cutoff frequency at a given gain.
In discrete circuits with a collector resistance, the time constant T 2 of the common-
emitter (CE) circuit is calculated using (2.96) on page 125:14
CL CC + CL CL =0 1
T2,CE = CC + Rg + = C C Rg +
β gm gm
In the cascode circuit (CC), T 2 is calculated using (4.46) with CS = 0; that is, eliminating
the substrate capacitance, which does not exist in discrete transistors:
C =0
Rg 1 L Rg 1
T2,CC = (CC + CL ) + = CC +
β gm β gm
It can be seen that, especially in the case of a high generator resistance Rg and a low
load capacitance CL , the cascode circuit yields a significantly smaller time constant and
thus a higher GBW than the common-emitter circuit. If the generator resistance is very
low (Rg < 1/gm ) or the load capacitance very high (CL > β CC ), the cascode is not
advantageous.
In an integrated circuit with current sources, it is necessary to modify the time constant
of the common-emitter circuit by taking into consideration the substrate capacitance CS
of the transistor and the capacitance CoS = CC + CS of the current source. These act like
an additional load capacitance and can therefore be taken into account by replacing CL
with CC + 2CS + CL :
CC + 2CS + CL 2CC + 2CS + CL
T2,CE = CC + Rg +
β gm
Equation (4.46) is used for the cascode circuit:
Rg 1
T2,CC = (2CC + 2CS + CL ) +
β gm
For β 1, this leads to
T2,CE ≈ T2,CC + CC Rg (4.48)
Here too, the cascode circuit offers a lower time constant and thus a higher GBW. But since
CS CC almost always applies to integrated circuits, the gain in GBW achieved with the
use of a cascode circuit instead of a common-emitter circuit is clearly less than in discrete
circuits, even with a high generator resistance Rg and without a load capacitance CL ; a
typical factor is between 2 and 3. Therefore, in practice it is very often the higher gain of
the cascode circuit – especially in combination with a current source with cascode – and
not the higher cutoff frequency that is decisive.
Finally, the circuits shown in Fig. 4.49 will be compared. For very high frequencies
the corresponding responses are not shown, since they deviate from the asymptote due to
the neglected zero and pole, so that it is not possible to use the GBW to calculate the cutoff
frequency. The calculation of the low-frequency gain is based on parameters β = 100 and
VA = 100 V for npn and pnp transistors and on Rg = 0 and RL → ∞. The gain of the
cascode circuit with a simple current source is |A| = VA /VT = 4000 = 72 dB and the gain
of the cascode circuit with a cascode current source is |A| = β VA /(2VT ) = 200000 =
14 Based on R
= R + R ≈ R .
g g B g
4.1 Circuits 323
RC T2 T3 T4
T3
T2 T2
T1 T1 T1 T1
1 2 3 4
A
dB
common-emitter cascode-
circuit circuit
106
4
3
72
66
2
1
40
GBW
0
~
~ f
15 With an ideal current source, the common-emitter circuit provides its maximum gain |A| = µ =
VA /VT . The use of a simple current source with a transistor with the same parameters reduces
the output resistance from rCE to rCE || rCE = rCE /2; this cuts the gain in half. In a common-
emitter circuit with a cascode current source (not shown in Fig.. 4.49), the output resistance of
the current source is negligible; therefore, the gain |A| = VA /VT has the same value as in the
cascode circuit with a simple current source.
324 4 Amplifiers
Vb Vb Vb Vb Vb
T10 T9
T11 T2 T5
V2
T8
T12
V1
T7
T13 T4
T14 T1 T3 T6
I0
Fig. 4.50. An example of the common-emitter and cascode circuits (all transistors of size 1)
shows that the characteristics that continuously improve from circuit 1 to circuit 4 are
achieved by means of additional transistors.
Example: Circuits 2, 3 and 4 in Fig. 4.49 are operated with a quiescent current I0 =
100 mA and an operating voltage Vb = 5 V; Fig. 4.50 shows the circuits with the additions
required for setting the operating point:
– The common-emitter circuit with a simple current source (T1 and T2 ).
– The cascode circuit with a simple current source (T3 . . . T5 ).
– The cascode circuit with a cascode current source (T6 . . . T9 ).
The quiescent current is adjusted with a three-transistor current mirror (T10 . . . T12 ) which,
together with transistors T2 , T5 and T9 , forms a current source bank that mirrors the
reference current I0 to four outputs. The current through transistor T11 is fed through
transistors T13 and T14 , which operate as diodes, and generates the bias voltage V1 =
2VBE ≈ 1.4 V for transistors T4 and T7 . The bias voltage for transistor T8 can be taken
from the three-transistor current mirror: V2 = Vb − 2VBE ≈ Vb − 1.4 V = 3.6 V. In
the most simple case, the current source with reference current I0 can be realized with a
resistance R = V2 /I0 ≈ 3.6 V/100 mA = 36 k.
If we neglect the base currents, then IC,A ≈ I0 = 100 mA for transistors T1 . . . T9 ; thus
gm = IC,A /VT ≈ 3.85 mS. With the parameters shown in Fig. 4.5 on page 274 for the
npn transistors, it follows that rBE,npn = βnpn /gm ≈ 26 k and rCE,npn = VA,npn /IC,A ≈
1 M; for the pnp transistors it follows that rCE,pnp = VA,pnp /IC,A ≈ 500 k. For
junction capacitances, the approximation
,
CS0 in the reverse region
CS (U ) ≈
2CS0 in the forward region
is used instead of (2.37) on page 69; this means that the voltages at the junction capacitances
no longer have to be determined, which is otherwise the case for evaluating (2.37). The
collector and substrate diodes operate in the reverse region; therefore:
4.1 Circuits 325
Current source
roS CoS
T2
RG2 CGD2 CBD2
vo
T1
Rg RG1 CGD1 CBD1
n-channel cascode circuit: Figure 4.51 shows the full small-signal equivalent circuit of
an n-channel cascode circuit with MOSFETs T1 and T2 and a current source. The small-
signal model according to Fig. 3.48 on page 218 is used for the MOSFETs; this model
does not contain the controlled sources with substrate transconductances gm,B1 and gm,B2 ,
since:
– For T1 , the source gm,B1 vBS1 is ineffective because vBS1 = 0.
– For T2 , the controlled sources gm2 vGS2 and gm,B2 vBS2 can be combined to give one
=g
m2 + gm,B2 .
source with gm2 16
The current source is described by the output resistance roS and the output capacitance
CoS . A comparison with the small-signal equivalent circuit for the npn cascode circuit in
Fig. 4.46 yields, in addition to the usual correlations of parameters (RB = RG , rBE → ∞,
CE = CGS , etc.), the following correlations:
CS1 = CBD1 + CBS2 , CS2 = CBD2
This enables the results of the npn cascode circuit to be transferred to the n-channel cascode
circuit; from (4.43) with Rg , RL 1/gm , it follows
1
ω-3dB = 2πf-3dB ≈ (4.51)
(CGS + 2CGD ) Rg + (2CGD + 2CBD + CL ) (ro || RL )
Vb Vb Vb Vb
T1 T2
T1 T2
Vi1 Vi2 Vi1 Vi2
2I 0 2I 0
– Vb – Vb
2CGD + 2CBD + CL
T2 = (4.54)
gm1
with the low-frequency gain:
A0 = AB (0) = − gm1 (ro || RL ) (4.55)
In the n-channel cascode circuit, the low-frequency gain and the time constant T2 are not
influenced by the internal resistance Rg of the signal source due to the infinitely high input
resistance (ri = ∞).
4.1.3
Differential Amplifier
The differential amplifier is a symmetrical amplifier with two inputs and two outputs.
It consists of two common-emitter or two common-source circuits, where the emitter or
source connections are connected to a common current source; Fig. 4.52 shows the basic
circuit. In general, the differential amplifier is operated with a positive and a negative
supply voltage that are often – but not necessarily – symmetrical, as shown in Fig. 4.52.
If only one positive or negative supply voltage is available, ground can be used as the
second supply voltage; this will be described in more detail below. In integrated differential
amplifiers with MOSFETs, the bulk connections of the n-channel MOSFETs are connected
to the negative supply voltage and those of the p-channel MOSFETs to the positive supply
voltage, while in discrete MOSFETs the bulk connections are connected to the source of
the given MOSFET.
Due to the current source, the sum of the currents remains constant:17
,
IC1 + IB1 + IC2 + IB2 ≈ IC1 + IC2 with B = IC /IB 1
2I0 =
ID1 + ID2
17 The upper and lower lines behind the brace apply to the npn and the n-channel differential
amplifiers, respectively.
328 4 Amplifiers
VD VD
2 2
V i1 V i2 VCM
Fig. 4.53. Replacement of the input voltages Vi1 and Vi2 by the common-mode voltage VCM and
the differential voltage VD
For the explanations set out below, the following assumption is made: RC1 = RC2 = RC
and RD1 = RD2 = RD . Furthermore, the input voltages Vi1 and Vi2 are replaced by the
symmetric common-mode voltage VCM and the skew-symmetric differential voltage VD :
Vi1 + Vi2
VCM = , VD = Vi1 − Vi2 (4.56)
2
Thus:
VD VD
Vi1 = VCM + , Vi2 = VCM − (4.57)
2 2
Figure 4.53 shows the replacement of Vi1 and Vi2 by the symmetric voltage VCM and
the skew-symmetric voltage VD ; according to (4.57), the latter leads to two sources with
voltage VD /2.
Common-mode and differential gain: For the same input voltages (Vi1 = Vi2 = VCM ,
VD = 0), the operation is symmetrical and the current of the current source is split up into
equal portions through both transistors:
B1
IC1 = IC2 ≈ I0 or ID1 = ID2 = I0
The output voltages are thus:
Vo1 = Vo2 ≈ Vb − I0 RC or Vo1 = Vo2 = Vb − I0 RD
The variations of the common-mode voltage VCM are called common-mode voltage range
and do not change the current distribution as long as the transistors and the current source
are not overloaded; this means that in common-mode operation the output voltages remain
constant. The common-mode gain
dVo1
dVo2
ACM = = (4.58)
dVCM
VD =0 dVCM
VD =0
4.1 Circuits 329
is zero under ideal conditions. In practice, it is slightly negative: ACM ranges between
approximately ACM ≈ −10−4 …−1. This is caused by the finite internal resistance of
real current sources; the explanation of the small-signal response will go into this in more
detail.
For skew-symmetric signals with a differential voltage VD , the current distribution
changes; this also alters the output voltages. This kind of signal is called differential signal
and the related gain is called the differential gain:
dVo1
dVo2
AD = = − (4.59)
dVD
VCM =const. dVD
VCM =const.
AD
G = (4.60)
ACM
Vo1 Vo2 Vo Vo
Vi Vi1 Vi2 Vi
dVo1
dVo1 dVD
dVo1 dVCM
A1 = = +
dVi1
Vi2 =const. dVD dVi1
Vi2 =const. dVCM dVi1
Vi2 =const.
G1
1
= AD + ACM = AD 1 + ≈ AD
G
dVo2
dVo2 dVD
dVo2 dVCM
A2 = = +
dVi1
Vi2 =const. dVD dVi1
Vi2 =const. dVCM dVi1
Vi2 =const.
G1
1
= − AD + ACM = − AD 1 − ≈ − AD
G
A sufficiently high common-mode rejection produces output signals of the same amplitude
in phase opposition; this circuit is therefore used for converting a single-ended signal into
a differential signal.
In Fig. 4.54b, only output 2 is used; output 1 could also be used as an alternative. The
common-mode gain and the differential gain can be calculated from (4.58) and (4.59) by
setting Vo = Vo2 or Vo = Vo1 , depending on the output used. Since AD < 0, the version
shown in Fig. 4.54b is noninverting with Vo = Vo2 , but inverting with Vo = Vo1 . This
circuit is used for converting a differential signal into a single-ended signal.
In Fig. 4.54c, only input 1 and output 2 are used; with respect to the gain A2 , which
has already been calculated, this leads to:
dVo dVo2
G1
A = = = A2 = − AD + ACM ≈ − AD
dVi dVi1
Vi2 =const.
4.1 Circuits 331
Vb Vb
RC RC
I C1 I C2
I B1 ≈ 0 I B2 ≈ 0
T1 T2
Vo1 Vo2
VBE1 VBE2
VD VD
2 2I 0 2
– Vb
Vo1 Vo 2
,
V V
Vo1 4 Vo2
–125 – 50 – 25 0 25 50 125 VD
mV
– 5VT – 2VT – VT VT 2VT 5VT
T1 Active region Operation Active region T2
non- as an non-
conduction amplifier conduction
Fig. 4.56. Transfer characteristics of the npn differential amplifier shown in Fig 4.55, where
Vb = 5 V, RC = 20 k and I0 = 100 mA
dVo1
dVo2
I 0 RC 2V
= −
= − ≈ − ≈ − 38
dV D VD =0 dV D VD =0 2V 52 mV
T
This corresponds to the differential gain at the operating point (VD = 0, VCM = 0).
The active portion of the characteristic is in the region |VD | < 5VT ≈ 125 mV. For
|VD | > 5VT , the differential amplifier is overloaded; in this case, the current of the current
source flows almost entirely (over 99%) through one of the two transistors, while the
other is in reverse mode. For VD < − 5VT , T1 is nonconductive and output 1 reaches
its maximum output voltage Vo,max = Vb , while output 2 provides the minimum output
voltage Vo,min = Vb − 2I0 RC . For VD > 5VT , T2 is nonconductive.
Common-mode signal range: Using the transistor equations for normal mode in the
calculations presented above implied that none of the transistors enters the saturation
region. Furthermore, an ideal current source without saturation was assumed. In this case,
the characteristics are practically independent of the common-mode voltage VCM ; a minor
common-mode gain induced by the internal resistance of the current source only causes
changes in the mV range. The permissible input voltage range is determined with the help
of Fig. 4.57. Two conditions must be met:
– The collector–emitter voltages VCE1 and VCE2 must be higher than the saturation voltage
VCE,sat . From Fig. 4.57, it follows
VCE1 = Vo1 + VBE 1 − Vi1 , VCE2 = Vo2 + VBE 2 − Vi2
For VCE > VCE,sat ≈ 0.2 V, VBE ≈ 0.7 V and the minimum output voltage Vo,min =
Vb − 2I0 RC , it follows:
max{Vi1 , Vi2 } < Vb − 2I0 RC − VCE,sat + VBE ≈ Vb − 2I0 RC + 0.5 V
– V0 must not drop below the lower voltage limit V0,min of the current source; that is,
V0 > V0,min is essential. From Fig. 4.57, it follows
V0 = Vi1 − VBE 1 − (−Vb ) = Vi2 − VBE 2 − (−Vb )
Since in normal mode at least one of the transistors is conductive and operated with
VBE ≈ 0.7 V, we obtain:
min{Vi1 , Vi2 } > V0,min + (−Vb ) + VBE ≈ V0,min + (−Vb ) + 0.7 V
When using a simple npn current mirror as the current source, V0,min = VCE,sat ≈ 0.2 V
and min{Vi1 , Vi2 } > (−Vb ) + 0.9 V.
Vb Vb
RC RC
T1 T2
VCE1 VCE2
Vo1 Vo2
VBE1 VBE2
Vi1 2I0 V0 Vi2
– Vb
Fig. 4.57. A circuit for calculating the permissible input voltage range of an npn differential
amplifier
334 4 Amplifiers
Vo1 Vo2
,
V V
4
Vo1 Vo2
3
T2 enters the saturation region T1 enters the saturation region
2
–100 – 25 0 25 100 VD
mV
Fig. 4.58. Transfer characteristics of the npn differential amplifier shown in Fig. 4.55, where
Vb = 5 V, RC = 20 k and I0 = 100 mA, and where the transistors enter the saturation region
(VCM = 2.5 V)
The permissible input voltage range is usually quoted for pure common-mode operation;
that is, Vi1 = Vi2 = VCM and VD = 0. In this case, the minimum and maximum operators
are eliminated and the common-mode voltage range is:19
V0,min + (−Vb ) + VBE < VCM < Vb − 2I0 RC − VCE,sat + VBE (4.63)
For the circuit in Fig. 4.55 with Vb = 5 V, (−Vb ) = − Vb = − 5 V, RC = 20 k,
I0 = 100 mA and when a simple npn current mirror is used with V0,min = VCE,sat ,
we obtain a common-mode voltage range of − 4.1 V < VCM < 1.5 V. If this range is
exceeded, the characteristic changes; Fig. 4.58 illustrates this for VCM = 2.5 V. Since the
saturation of one transistor changes the current distribution, it also has an effect on the
characteristic of the other branch.
Within the range |VD | < 25 mV, the characteristic remains the same; use as an amplifier
is still possible even though the common-mode range is exceeded. The reason for this
apparent contradiction is that the common-mode range was defined to be the region in which
full output signal without saturation is possible. If we restrict ourselves to just a part of the
characteristic, the common-mode range is larger. In the borderline case of an infinitesimal
small differential voltage, it is sufficient that no saturation occurs with VD = 0. The
minimum output voltage is thus Vo,min ≈ Vb − I0 RC instead of Vo,min = Vb − 2I0 RC .
This leads to the common-mode voltage range in small-signal operation:
V0,min + (−Vb ) + VBE < VCM < Vb − I0 RC − VCE,sat + VBE (4.64)
For the values already mentioned, the voltage range of the circuit in Fig. 4.55 is − 4.1 V <
VCM < 3.5 V. This means that the situation shown in Fig. 4.58, with VCM = 2.5 V, is still
within the small-signal common-mode range.
19 This causes an error, because a differential voltage of at least 5V is required to reach the mini-
T
mum output voltage; therefore, one should actually use max{Vi1 , Vi2 } = VCM + VD,max /2 and
min{Vi1 , Vi2 } = VCM − VD,max /2. As the maximum differential voltage VD,max is application-
specific but very low (VD,max < VT ) in amplifiers, it is ignored in this case.
4.1 Circuits 335
Vb Vb Vb Vb
RC RC RC RC
2I0 I0 I0
Vb Vb Vb
With two resistances and one With one resistance and two
current source current sources
npn differential amplifier with current feedback: The differential amplifier can be
provided with current feedback in order to improve its linearity; Fig. 4.59 shows two options
that are equivalent in terms of the transfer characteristics. Figure 4.59a uses two resistances
RE and one current source. Without differential signal there is a voltage drop I0 RE across
both resistors; thus, the lower limit of the common-mode voltage range is increased by
this value. In Fig. 4.59b only one resistance is required, which without differential signal
carries no current. This means that the common-mode range is not reduced, but two current
sources are required.
Figure 4.60 shows the characteristics for Vb = 5 V, RC = 20 k, I0 = 100 mA and
several values of RE ; the latter are related to the transconductance of the transistors at the
operating point VD = 0:
I0 1
gm = ≈ , gm RE = 0 / 2 / 5 ⇒ RE = 0 / 520 / 1300
VT 260
With increasing feedback, the characteristics flatten out and become almost linear over a
longer portion. This means that the differential gain is reduced but remains approximately
constant over a larger region. The distortions, expressed by the distortion factor, decline
as feedback increases.
A direct calculation of the characteristics is not possible. An approximation is possi-
ble for high feedback by assuming that the base–emitter voltages remain approximately
constant. If the base currents are neglected, the differential voltage for both circuits in
Fig. 4.59 is:
Vo1 Vo2
,
V V
5
Vo1 Vo2
4
gmRE = 5
3
gmRE = 2
RE = 0
2
VD
– 0.5 – 0.1 0.1 0.5
V
–10 gmRE = 5
gmRE = 2
– 20
RE = 0
– 30
– 40
Fig. 4.60. Characteristics and differential gain of an npn differential amplifier with current
feedback (Vb = 5 V, RC = 20 k, I0 = 100 mA)
Inserting IC1 + IC2 = 2I0 and solving for IC1 and IC2 in the range 0 ≤ IC1 , IC2 ≤ 2I0
leads to
VD VD
IC1 ≈ I0 + , IC2 ≈ I0 − for |VD | < 2I0 RE
2RE 2RE
Further calculation is more complex than for the npn differential amplifier. From the
equation
2ID1 2ID2
VD = VGS1 − VGS2 = −
K K
Vb Vb
RD RD
I D1 I D2
T1 T2
Vo1 Vo2
VD VGS1 VGS2 VD
2 2I 0 2
– Vb
⎫
2 ⎪
V D RD K VD ⎪
⎪
Vo1 = Vb − I0 RD − 2K I0 − ⎪
⎪
2 2 ⎬ I0
für |VD | < 2 (4.67)
⎪
K VD 2 ⎪
K
V D RD ⎪
⎪
Vo2 = Vb − I0 RD + 2K I0 − ⎪
⎭
2 2
Outside the range of validity of (4.67), one output has the maximum output voltage
Vo,max = Vb while the other output has the minimum output voltage Vo,min = Vb −2I0 RD .
A comparison of (4.67) and the corresponding (4.62) for the npn differential amplifier
shows that the characteristics of the n-channel differential amplifier also depend on the
size of the MOSFETs, expressed by the transconductance coefficient K; on the other
hand, the size of the bipolar transistors, expressed by the reverse saturation current IS ,
has no influence on the characteristic of the npn differential amplifier. Therefore, it is
possible to selectively adjust the characteristic of the n-channel differential amplifier by
scaling the MOSFETs while keeping the external circuitry the same; in the npn differential
amplifier, this is only possible with current feedback. The typical parameter for adjusting
the characteristic according to (4.67) is the voltage:
-
I0
VDM = 2 (4.68)
K
This indicates the active segment of the characteristic according to the condition
|VD | < VDM . Since VD = 0 at the operating point, the current distribution is ID1 =
ID2 = I0 and at the same time VGS1 = VGS2 = VGS,A , insertion into the characteristic of
the MOSFET leads to the alternative expression:
√ % &
VDM = 2 VGS,A − Vth
Vo1 Vo2
,
V V
5
Vo1 Vo2
4
3 VDM = 1 V
VDM = 0.5 V
2 VDM = 0.25 V
Fig. 4.62. Transfer characteristic of the n-channel differential amplifier shown in Fig. 4.61, where
Vb = 5 V, RD = 20 k and I0 = 100 mA.
4.1 Circuits 339
Vb Vb = 5 V Vb Vb = 5 V
RD RD RD RD
20 kΩ 20 kΩ 20 kΩ 20 kΩ
T1 T2 T1 T2
15 15 150 150
V i1 Vi2 V i1 RS RS Vi2
2 kΩ 2 kΩ
2I 0 2I 0
200 µA 200 µA
– Vb = – 5 V – Vb = – 5 V
a Without current feedback and with b With current feedback and with
small MOSFETs large MOSFETs
Fig. 4.63. A comparison of n-channel differential amplifiers of the same differential gain with and
without current feedback
Figure 4.62 shows the characteristic for Vb = 5 V, RD = 20 k, and I0 = 100 mA, and
for K = 0.4 / 1.6 / 6.4 mA/V2 or VDM = 1 / 0.5 / 0.25 V. A comparison with Fig. 4.60
shows that a variation in the size of the MOSFETs in the n-channel differential amplifier
achieves an effect similar to that in the npn differential amplifier with current feedback;
the characteristics become less steep if the size of the MOSFETs in the n-channel differ-
ential amplifier decreases or if the feedback in the npn differential amplifier increases (RE
is larger). Consequently, the n-channel differential amplifier yields better linearity with
smaller MOSFETs, while larger MOSFETs lead to an increase in the differential gain.
n-channel differential amplifier with current feedback: Current feedback can also be
used in the n-channel differential amplifier to improve the linearity. However, the question
here is whether this produces a better result with the same gain than does a reduction in the
size of the MOSFETs, as discussed in the previous paragraph. For this purpose, the circuits
340 4 Amplifiers
Vo1 Vo2
,
V V
5 b
Vo1 Vo2
a
4
2
a
1 b
shown in Fig. 4.63 will be compared; their characteristics, and thus their differential gains,
are identical in the region of the operating point VD = 0; Fig. 4.64 presents the correspond-
ing characteristics. It is clear that the circuit with current feedback and larger MOSFETs
offers better linearity; however, the spatial requirements are significantly greater due to the
MOSFETs – which are ten times as large – and the necessary feedback resistances, and
the bandwidth is significantly narrower than in the circuit without current feedback, due
to the larger capacitances of the MOSFETs.
Differential amplifier with symmetric output: The circuit in Fig. 4.65a uses two cur-
rent sources with current I0 instead of the collector resistances; with regard to (4.61), the
output currents are:20
VD VD
Io1 = IC1 − I0 = I0 tanh , Io2 = IC2 − I0 = − I0 tanh
2VT 2VT
At the operating point VD = 0, there is no current at either output. The outputs must be
connected to a load so that output currents can actually flow without the transistors and
current sources entering the saturation region. The output voltages are not defined without
external circuitry.
20 As the differential amplifier as a whole represents a current node, the conditions of Kirchhoff’s
law must be met. In the following equations and in Fig. 4.65, this is only the case if the base
currents are ignored.
4.1 Circuits 341
Vb Vb Vb Vb
I0 I0 I0 I0
I o1 I o2 ID ID
I C1 I C2 I0 + ID I 0 – ID
T1 T2 T1 T2
VD VD
Vi1 Vi2 ID
2 2
2I 0 I0 I0
– Vb – Vb – Vb
a With absolute values b With differential values
Differential amplifier with asymmetric output: The circuit of Fig. 4.65a can also be
used where an asymmetric output is required, if the output that is not required is connected
to the operating voltage Vb and the related current source is removed. A better alternative,
which prevails in practical applications, is shown in Fig. 4.66a. In this case, the current
sources are replaced by a current mirror, so that the current of the output that has been
removed is mirrored to the remaining output:
IC4 ≈IC1
VD
Io = IC2 − IC4 ≈ IC2 − IC1 = − 2I0 tanh
2VT
At the operating point VD = 0, there is no current at the output. Here again, the output
must be connected to a load such that the output current can flow without driving T2 or T4
into the saturation region. Figure 4.66b shows the circuit with the differential current ID .
The current of the negative bias voltage source remains constant, while that of the positive
source changes by 2ID .
Current sources and current mirrors: In essence, all of the circuits described in
Sect. 4.1.1 can be used to obtain the current sources in Figs. 4.65 and 4.66; in prac-
tice, simple current mirrors or cascode current mirrors are mostly used as current sources.
The current mirror of Fig. 4.66 may also be made up differently; since the current ratio
342 4 Amplifiers
Vb Vb Vb Vb
T3 T4 T3 T4
I C4 I0 + ID
Io 2I D
I C1 I C2 I0 + ID I0 – ID
T1 T2 T1 T2
VD VD
Vi1 Vi2 ID
2 2
2I 0 I0 I0
– Vb – Vb – Vb
The relevant current distribution can be symmetrical, although this must not necessarily
be the case. In the transfer characteristics, the offset voltage causes a shift in the direction
of VD ; Fig. 4.67 illustrates this for Voff > 0.
As already mentioned, the offset voltage consists of a systematic portion caused by
circuit asymmetries and a stochastic portion caused by tolerances. Therefore, in practice
a likely range (e.g. 99% probability) is often quoted for the offset voltage.
21 The offset voltage is often called V (index O). As this symbol can easily be confused with V
O 0
(index zero), it is called Voff here, for simplicity.
4.1 Circuits 343
Vo1 , Vo2
Vo1 Vo2
0 Voff VD
The offset voltage can be calculated if very acurate equations are used for the transistors
and upper and lower limits are inserted for all parameters; however, the actual calculation
is very complex. It is easier to measure the offset voltage or to determine it by circuit
simulation; this is done on the basis of the circuits shown in Fig. 4.68. By feeding the output
differential voltage Vo1 − Vo2 back to input 1, the output voltages are made approximately
equal and the voltage at the input is Vi1 ≈ Voff . On the one hand, the circuit provides no true
differential signal, but, on the other hand, the actual common-mode voltage VCM ≈ Voff /2
has almost no influence on the result, due to the high common-mode rejection.
Vb Vb Vb
I0 I0 I0
Vo1 Vo2 Vo
2I 0 2I 0
– Vb – Vb
A =1 A =1
Vo,set
When measuring the offset voltage, a normal operational amplifier should not be used
as a control amplifier, since the differential amplifier causes an additional loop gain, which
results in instability in the circuit even with compensated operational amplifiers. The most
suitable type is an instrumentation amplifier with a gain of A = 1 and a cutoff frequency
fg,amp that remains below the cutoff frequency fg of the differential amplifier by a value at
least equal to the differential gain AD : fg,amp < fg /AD ; this guarantees stable operation.
In circuit simulation, a voltage-controlled voltage source with A = 1 may be used as a
control amplifier; should stability problems occur, A has to be reduced.
It is assumed that the common-mode voltage VCM,A remains within the common-mode
voltage range and has no influence on the current distribution.
– With a sufficiently low amplitude, the skew-symmetric differential signal causes a skew-
symmetric change of all currents and voltages. Consequently, all voltages in the plane
of symmetry remain constant; in Fig. 4.69a this applies to the voltage V0 at the emitter
connections of the transistors. Since a constant voltage can be replaced by a voltage
source, the resulting equivalent circuit is as shown in Fig. 4.69a (bottom): the differen-
tial amplifier is split up into two common-emitter circuits and the current sources are
eliminated. The voltage sources V0 are ideal and are short-circuited in the conversion to
the small-signal equivalent circuit. Therefore, the emitter connections of the transistors
are connected to the small-signal ground in the small-signal equivalent circuit.
– The symmetrical common-mode signal produces a symmetrical change in all currents
and voltages. This means that all currents flowing through the plane of symmetry are
zero; in Fig. 4.69b this applies to current I in the emitter cross-connection. The fact that
a dead wire can be removed leads to the equivalent circuit shown in the lower part of
Fig. 4.69b: here too, the differential amplifier consists of two common-emitter circuits.
The current sources I0 each represent one half of the original current source; Fig. 4.70
explains the conversion from an ideal to a real current source and its splitting into two
current sources. In the small-signal equivalent circuit, the current sources are omitted
and the negative supply voltage coincides with the small-signal ground.
4.1 Circuits 345
Vb Vb Vb Vb
RC RC RC RC
T1 T2 T1 T2
VD VD
V0 = const VCM I =0 VCM
2 2
I0 I0 I0 I0
– Vb – Vb – Vb – Vb
Vb Vb Vb Vb
RC RC RC RC
T1 T2 T1 T2
VD VD
VCM VCM
2 2
V0 V0 I0 I0
– Vb – Vb – Vb – Vb
Fig. 4.69. The operation of an npn differential amplifier at the operating point
2I 0 2I 0 r0 2r0 I0 I0 2r0
Fig. 4.70. Conversion from an ideal to a real current source and the split into two equivalent
current sources
This reduces the npn differential amplifier to a common-emitter circuit, for which the results
from Sect. 2.4.1 can be used. The same applies to the n-channel differential amplifier; it is
split into equivalent common-source circuits and the results from Sect. 3.4.1 can be used.
The division into separate equivalent circuits for differential and common-mode oper-
ation is an application of the Bartlett symmetry theorem – which, however, applies to linear
circuits only. Strictly speaking, the small-signal equivalent circuit should be used for the
differential amplifier in order to allow this theorem to be applied. However, the restriction
346 4 Amplifiers
to linear circuits is only required for differential operation, as here the characteristics of the
components are modulated skew-symmetrically, starting from the operating point, which
only results in skew-symmetric changes in the case of linear characteristics. In contrast, in
common-mode operation the characteristics are modulated symmetrically, which leads to
symmetrical changes even in the case of nonlinear characteristics. Therefore, the theorem
may also be applied to nonlinear circuits if the differential operation is limited to the range
in which the characteristics are almost linear; in the npn differential amplifier, this is the
range of |VD | < VT . This approach was selected because the breakdown of a differential
amplifier into two portions is easier to understand from the original circuit than from the
small-signal equivalent circuit.
Differential amplifier with resistances: Figure 4.71 shows the circuit of an npn differen-
tial amplifier together with the small-signal equivalents for the equivalent common-emitter
circuits with differential and common-mode operation which are obtained by linearization
of the subcircuits in Fig. 4.69 and by inserting the current source according to Fig. 4.70.
For the small-signal values where VD,A = 0, it follows:
It is clear that the small-signal equivalent circuit for differential operation corresponds to
that of a common-emitter circuit without feedback, and the small-signal equivalent circuit
for common-mode operation to that of a common-emitter circuit with current feedback.
With common-mode operation the output resistance 2r0 of the split current source acts as a
feedback resistance. Figure 4.72 shows the corresponding small-signal equivalent circuits
of an n-channel differential amplifier.
From the small-signal equivalent circuit for differential operation, the following pa-
rameters can be calculated: the differential gain AD , the differential output resistance ro,D
and the differential input resistance ri,D :
vo1
vo
1
AD =
=
= ACE/CS (4.73)
vD io1 =io2 =0 2vi io =0 2
vCM =0
vo1
vo
vD
2vi
ri,D = = = 2 ri,CE/CS (4.75)
ii1
vCM =0 ii
Here, it is important that the input voltage in the small-signal equivalent circuit for differen-
tial operation is vD /2 and not vD ; this is why the differential amplifier has only half the gain
but twice the input resistance of the equivalent common-emitter (CE) or common-source
(CS) circuit.
The following parameters can be calculated from the small-signal equivalent circuit
for common-mode operation: the common-mode gain ACM , the common-mode output
resistance ro,CM and the common-mode input resistance ri,CM :
4.1 Circuits 347
Vb Vb
RC RC
Vo1 T1 T2 Vo2
Vi1 Vi2
2I 0
– Vb
ii io
ii io
v i = v CM vo
RC
2r0
Fig. 4.71. npn differential amplifier with collector resistances: the circuit (above) and the small-
signal equivalent common-emitter circuits for differential operation (centre) and common-mode
operation (below)
vo1
vo
vo1
vo
vCM vi
ri,CM = = = ri,CE/CS (4.78)
ii1 ii
Here, the differential amplifier provides the same values as the equivalent common-
emitter or common-source circuits. It should be noted that the small-signal parameters
348 4 Amplifiers
Vb Vb
RD RD
T1 T2 Vo2
Vo1
Vi1 Vi2
2I 0
– Vb
ii = 0 io
v BS = 0
ii = 0 io
2 r0
v BS
Fig. 4.72. n-channel differential amplifier with drain resistances: the circuit (above) and the
small-signal equivalent common-source circuits for differential operation (centre) and
common-mode operation (below)
in (4.76)–(4.78) are from a different small-signal equivalent circuit than those in (4.73)–
(4.75); for example, AD = ACM /2 is not a consequence of (4.73) and (4.76).
Pure differential or common-mode operation must exist when measuring or simulating
these parameters. This not only applies to the input, where this is expressed by parameters
vD and vCM , but also to the output. As no specific differential and common-mode param-
eters are defined for the output, the secondary conditions vo1 = − vo2 and vo1 = vo2 must
be used to characterize the differential and common-mode operation. The consequence is
that the definitions of the differential and the common-mode output resistances are only
distinguished by the secondary conditions and not by the small-signal parameters. vo1 / io1
is formed for both output resistances; the difference is produced by the different operation
of the second output.
4.1 Circuits 349
In the npn differential amplifier, the output resistances depend on the internal resistance
Rg of the signal source, as is the case in the common-emitter circuit. As the latter is
generally smaller than the input resistances, calculations can be restricted to the short-
circuit output resistances without causing major errors; therefore, ro,D and ro,CM are
given for the secondary condition vD = vCM = 0. This dependence does not exist in the
n-channel differential amplifier due to the isolated gate connections of the MOSFETs;
here, Rg is not visible at the output.
The results for the common-emitter circuit in Sect. 2.4.1 and for the common-source
circuit in Sect. 3.4.1 lead to the following equations for the differential amplifier with
resistances:22
differential amplifier with resistance
⎧
⎪ rCE RC
⎪
⎪ − gm (RC || rCE ) 1
vo1
⎨ ≈ − gm RC
AD = = 2 2
(4.79)
vD
io1 =io2 =0 ⎪
⎪
rDS RD 1
⎪ − gm (RD || rDS )
⎩ ≈ − gm RD
2 2
⎧
⎪ rCE RC
⎪
⎨ R || r
vo1
C CE ≈ RC
ro,D = = (4.80)
io1
vo1 =−vo2 ⎪
⎪ r R
⎩ DS D
RD || rDS ≈ RD
,
vD 2rBE
ri,D = = (4.81)
ii1 ∞
⎧
⎪ RC
⎪
⎨ − 2r
vo1
0
ACM = ≈ gm gm,B (4.82)
vCM
io =0 ⎪
⎪ gm RD RD
⎩ − % & ≈ −
2 gm + gm,B r0 2r0
vo1
RC || β rCE ≈ RC
ro,CM = = (4.83)
io1
vo1 =vo2 RD || 2gm rDS r0 ≈ RD
vCM 2β r0 + rBE ≈ 2β r0
ri,CM = = (4.84)
ii1 ∞
,
AD gm r0
G = ≈ % & gm gm,B (4.85)
ACM gm + gm,B r0 ≈ gm r0
Here, the following equations were used: (2.61)–(2.63) on page 100, (2.70)–(2.72) on
page 106, (3.50)–(3.52) on page 232 and (3.56)–(3.58) on page 237; RE = 2r0 is inserted
into (2.70) and RS = 2r0 and 2gm r0 1 are inserted into (3.56).
The common-mode gain of the n-channel differential amplifier with integrated MOS-
FETs depends on the common-mode voltage VCM,A at the operating point, because the
bulk–source voltage VBS and the substrate transconductance gm,B depend on VCM,A . But
since VBS > 0 according to Fig. 4.73, the substrate transconductance is lower in the n-
22 The results for the npn and the n-channel differential amplifiers are presented following braces.
The upper values apply to the npn differential amplifier and the lower values to the n-channel
differential amplifier.
350 4 Amplifiers
VBS = 0
VBS < 0
– Vb
Fig. 4.73. Bulk–source voltage VBS in the common-source circuit and in the n-channel differential
amplifier
channel differential amplifier than in the common-source circuit and may therefore be
ignored in most practical applications. In the differential amplifier with discrete MOS-
FETs, however, VBS = 0, so that in the relevant equations the substrate transconductance
can be set to gm,B = 0. In the following considerations, the substrate transconductance is
generally ignored.
Basically, any of the circuits described in Sect. 4.1.1 can be used to realize the current
source; the output resistance r0 has a significant influence on the common-mode gain and
the common-mode rejection ratio. In practice, a simple current mirror is usually used.
vo1
1
AD = = − gm ro,D (4.86)
vD
io1 =io2 =0 2
vo1
ro,CM
ACM =
≈ − (4.87)
vCM io1 =io2 =0 2r0
ro,D ≈ro,CM
AD ro,D
G = ≈ gm r0 ≈ gm r0 (4.88)
ACM ro,CM
If the output resistances ro,D and ro,CM are almost identical, as is the case in the differ-
ential amplifier with resistances, then the common-mode rejection only depends on the
transconductance of the transistors and the output resistance r0 of the current source.
Current feedback, as shown in Fig. 4.59 on page 335 or in Fig. 4.63b on page 339,
can be taken into account simply by replacing the transconductance gm with the reduced
transconductance:
4.1 Circuits 351
⎧
⎪ gm
⎪
⎪
⎨ 1 + g m RE
The differential gain thus drops accordingly. The common-mode gain remains the same
because the feedback resistance in the small-signal equivalent circuit for common-mode
operation is connected in series to the output resistance r0 of the current source, and can
be ignored since r0 RE , RS . Thus the common-mode rejection G = AD /ACM is lower
with current feedback.
Differential amplifiers with simple current sources: Figure 4.74 shows an npn and an
n-channel differential amplifier, both with simple current sources instead of resistances. In
the small-signal equivalent circuit and in the equations, the resistances are replaced by the
output resistance of the simple current source: RC → rCE3 for the npn differential amplifier
and RD → rDS3 for the n-channel differential amplifier. For differential amplifiers with
simple current sources, this leads to:
differential amplifier with simple current source
vo1
1
AD =
= − gm1 ro,D
vD io1 =io2 =0 2
⎧ rCE1 ≈rCE3
⎪
⎨ rCE1 || rCE3 rCE3
vo1
≈ 2 (4.90)
ro,D = ≈
io1
vo1 =−vo2 ⎪
⎩
rDS1 ≈rDS3
rDS3
rDS1 || rDS3 ≈ 2
vo1
ro,CM
ACM = ≈ −
vCM
io1 =io2 =0 2r0
,
vo1
rCE1 ≈rCE3
AD ro,D rDS1 ≈rDS3 gm r0
G = ≈ gm1 r0 ≈ (4.92)
ACM ro,CM 2
The input resistances ri,D and ri,CM remain unaltered; in other words, (4.81) and (4.84)
also apply to the differential amplifier with current sources.
After inserting gm1 = I0 /VT , rCE1 = VA,npn /I0 and rCE3 = VA,pnp /I0 , the differen-
tial gain for npn differential amplifiers with simple current sources is:
1
AD = − (4.93)
1 1
2VT +
VA,npn VA,pnp
Here, VA,npn and VA,pnp are the Early voltages of the transistors; the temperature voltage
at T = 300 K is VT ≈ 26 mV. The transistor parameters and the quiescent current I0
352 4 Amplifiers
Vb Vb Vb Vb Vb Vb
T4 T3 T4
T3
I0 I0 I0 I0
T1 T2
T1 T2
Vi1 Vi2 Vi1 Vi2
2I 0 2I 0
– Vb – Vb
have no influence on the differential gain. From Fig. 4.5, the values for the transistor are
VA,npn = 100 V and VA,pnp = 50 V; consequently, AD = − 640.
√ For n-channel differential amplifiers with simple current sources, where gm1 =
2K1 I0 , rDS1 = VA,nC /I0 and rDS3 = VA,pC /I0 , the differential gain is as follows:
K1 1 1
AD = − = − (4.94)
2I0 1 1 1 1
+ (VGS1 − Vth1 ) +
VA,nC VA,pC VA,nC VA,pC
Here, VA,nC and VA,pC are the Early voltages of the MOSFETs. In this case, the differential
gain also depends on the sizes of the MOSFETs T1 and T2 , expressed by the transcon-
ductance coefficient K1 ; this gain increases when the size of the MOSFETs is increased.
From Fig. 4.6, the values for the MOSFETs are VA,nC = 50 V and VA,pC = 33 V; for the
typical value VGS1 − Vth1 = 1 V, it follows that AD = − 20.
Differential amplifiers with cascode current sources: The differential gain can be
increased by using current sources with cascode or cascode current sources23 instead
of the simple current sources; Fig. 4.75 shows the resulting circuits when using current
sources with cascode. Strictly speaking, the name differential amplifier with cascode cur-
rent sources is not quite correct, but it is preferred to the longer designation differential
amplifier with current sources with cascode.
When using the current sources with a cascode, the output resistance of the current
source increases from rCE3 or rDS3 to:
,
β3 rCE3
roS ≈ % & 2 gm3 gm,B3
gm3 + gm,B3 rDS3 ≈ 2
gm3 rDS3
23 The difference is shown in Fig. 4.25 on page 294 and Fig. 4.27 on page 297.
4.1 Circuits 353
Vb Vb Vb Vb Vb Vb
T6 T5 T6
T5
T4
T3
V1 V1 T3 T4
I0 I0 I0 I0
T1 T2
T1 T2
Vi1 Vi2 Vi1 Vi2
2I 0 2I 0
– Vb – Vb
This leads to the following equations for the differential amplifier with cascode current
sources:
vo1
1
AD =
= − gm1 ro,D
vD io1 =io2 =0 2
,
vo1
vo1
ro,CM
ACM =
≈ −
vCM io1 =io2 =0 2r0
,
vo1
β1 rCE1 || β3 rCE3
ro,CM =
≈ (4.96)
io1 vo1 =vo2 2gm1 rDS1 r0 || gm3 rDS3
2
AD ro,D
G = ≈ gm1 r0 (4.97)
ACM ro,CM
354 4 Amplifiers
Here, the common-mode output resistance ro,CM is typically larger than the differential
output resistance ro,D by a factor of 20 . . . 200; as compared to the differential amplifier
with resistances, this reduces the common-mode rejection accordingly:
gm1 r0
G ≈
20 . . . 200
For npn differential amplifiers with cascode current sources, the differential gain is
obtained by inserting gm1 = I0 /VT and rCE1 = VA,npn /I0 :
VA,npn µ
AD = − = − (4.98)
2VT 2
K1 VA,nC µ
AD = − VA,nC = − = − (4.99)
2I0 VGS1 − Vth1 2
Here, µ represents the maximum gain of the MOSFETs introduced in connection with the
common-source circuit. For VA,nC = 50 V and VGS1 − Vth1 = 1 V, the differential gain
is AD = − 50, as compared to AD = − 20 in the n-channel differential amplifier with
simple current sources.
The differential amplifier with cascode current sources is used whenever the pnp or p-
channel transistors have a clearly lower Early voltage than the npn or n-channel transistors.
In this case, the simple current sources only yield an insufficient gain.
Vb Vb Vb Vb Vb Vb
T8 T7 T8
T7
T6
T5
V2 V2 T5 T6
I0 I0 I0 I0
T3 T4
T4
T3
V1 V1
T1 T2
T1 T2
Vi1 Vi2 Vi1 Vi2
2I 0 2I 0
– Vb – Vb
For the cascode differential amplifier, it follows from (4.36) and (4.37) that:
Cascode differential amplifier
vo1
1
AD = = − gm1 ro,D
vD
io1 =io2 =0 2
,
vo1
β3 rCE3 || β5 rCE5
ro,D =
≈ (4.100)
io1 vo1 =−vo2 =0 2
gm3 rDS3 || gm5 rDS5
2
vo1
ro,CM
ACM =
≈ −
vCM io1 =io2 =0 2r0
,
vo1
β3 rCE3 || β5 rCE5
ro,CM =
≈ 2 (4.101)
io1 vo1 =vo2 =0 gm5 rDS5
AD ro,D
G = ≈ gm1 r0
ACM ro,CM
356 4 Amplifiers
With common-mode operation of the n-channel cascode differential amplifier, the output
resistance at the drain connection of T3 increases to 2gm1 gm3 rDS32 r and can be ignored.
0
In the npn cascode differential amplifier, the maximum output resistance β3 rCE3 at the
collector of T3 is already achieved with differential operation; it is not possible to increase
it any further.
For the npn cascode differential amplifier, the insertion of the small-signal parameters
leads to
1
AD ≈ − (4.102)
1 1
2VT +
βnpn VA,npn βpnp VA,pnp
and for the n-channel cascode differential amplifier with MOSFETs of the same size – that
is, the same transconductance coefficients K – we obtain:
K 2
AD ≈ −
= −
(4.103)
1 1 1 1
ID 2
+ 2
(VGS − Vth )
2
2
+ 2
VA,nC VA,pC VA,nC VA,pC
With the bipolar transistors from Fig. 4.5, the differential gain is AD ≈ − 38500 and with
the MOSFETs from Fig. 4.6 it is AD ≈ − 1500.
If the Early voltage and the current gain of the npn and pnp transistors and the Early
voltages of the n-channel and p-channel MOSFETs have the same magnitude, then:
⎧
⎪
⎪ β gm rCE β VA β≈100
⎪
⎨ = ≈ 25.000 . . . 150.000
4 4VT
|AD | ≈ 2
⎪
⎪ g2 r 2
⎪
⎩ m DS =
VA
≈ 400 . . . 10.000
4 VGS − Vth
Therefore, it is possible to achieve a differential gain in the region of 105 = 100 dB with
one npn cascode differential amplifier; in contrast, the maximum differential gain with an
n-channel cascode differential amplifier is approximately 104 = 80 dB.
Differential amplifier with a current mirror: The use of a current mirror produces a
differential amplifier with an asymmetric output; Fig. 4.77a shows a simple circuit design
that has already been presented in Fig. 4.66 on page 342 and studied with regard to the
large-signal performance. The use of the cascode current mirror in the cascode differential
amplifier results in the circuit shown in Fig. 4.77b. The current ratio of the current mirror
must be kI = 1 (in practice, kI ≈ 1).
The small-signal parameters can be derived easily if the following properties are con-
sidered:
– The current mirror doubles the output current in the case of differential operation (see
Fig. 4.66); this increases the differential gain by a factor of 2.
– In common-mode operation the currents change proportionally and are subtracted at
the output by the current mirror. With ideal subtraction by an ideal current mirror, the
output voltage remains constant; consequently, ACM = 0. Real current mirrors generate
a small common-mode gain.
– The output resistance ro corresponds to the differential output resistance ro,D of the
corresponding symmetric circuit.
4.1 Circuits 357
Vb Vb Vb Vb
T3 T4 T7 T8
T5 T6
Vo Vo
T4
T3
V1
T1 T2 T1 T2
Vi1 Vi2 Vi1 Vi2
2I0 2I0
– Vb – Vb
a Differential amplifier with b Cascode differential amplifier with
a simple current mirror a cascode current mirror
This leads to the basic equations for an asymmetric differential amplifier with a current
mirror:
vo1
AD = = − g m ro (4.104)
vD
io =0
vo1
ACM = ≈ 0 (4.105)
vCM
io =0
AD
G = → ∞ (4.106)
ACM
The output resistance of the differential amplifier with a simple current mirror is
,
vo1
rCE2 || rCE4
ro =
≈ (4.107)
io1 vD =0 rDS2 || rDS4
and for the cascode differential amplifier with a cascode current mirror:
358 4 Amplifiers
ro,D ro,D
ro,CM – ro,D
vD = vi1 – vi2 r=
2
vi1 + vi2
vCM =
2
2ri,D – ri,CM
' =
ri,D ≈ ri,D
2ri,CM – ri,D
vi1 vi1
ri,CM ri,CM
⎧ β6 rCE6
⎪
⎨ β4 rCE4 ||
vo1
2
ro = = (4.108)
i
o1 vD =0 ⎪
⎩ gm4 r 2 || gm6 r 2
DS4 DS6
In the npn cascode differential amplifier with a cascode current mirror, it should be noted
that the output resistance of the cascode current mirror with kI = 1 is only half of the
output resistance of a current source with a cascode (see (4.23) and (4.27)).
Equivalent circuit: With the help of the small-signal parameters of a differential ampli-
fier, we obtain the equivalent circuit shown in Fig. 4.78. On the input side, it comprises a π
network with three resistances, to simulate the input resistances ri,D and ri,CM of the npn
differential amplifier; the resistances are omitted in the n-channel differential amplifier.
Since the resistances ri,CM also affect the differential operation, the cross-resistance must
have the value
2ri,D ri,CM
ri,D =
2ri,CM − ri,D
This ensures that the effective differential input resistance is ri,D . In practice, ri,CM ri,D
and thus ri,D ≈ ri,D . On the output side, a T network consisting of three resistances sim-
ulates the actual output resistances. The T network has the advantage that the differential
output resistance has a direct influence, which is important in practical applications, and
that for ro,D = ro,CM the resistance r enters the short-circuited state. Two current sources
controlled by the differential voltage vD and the common-mode voltage vCM are con-
nected to each output; the corresponding transconductance values are gm /2 for differential
operation and 1/(2r0 ) for common-mode operation.
! 3 "
vD vD 1 vD
vo1 = − I0 RC tanh = − I 0 RC − + ···
2VT 2VT 3 2VT
Inserting vD = v̂D cos ωt leads to:
!
"
3 3
vD vD vD
vo1 = − I0 RC − + · · · cos ωt − − · · · cos 3ωt + · · ·
2VT 32VT3 96VT3
For small amplitudes (vD < 2VT ), an approximation of the distortion factor of the npn
differential amplifier without current feedback results from the ratio of the amplitudes at
3ωt and ωt:
1 v̂D 2
k ≈ (4.109)
48 VT
For VT = 26 mV and where the maximum distortion factor is given, we obtain:
√ √
v̂D < VT 48k = 180 mV · k
The following must be the case: v̂D < 18 mV for k < 1%. This makes the npn differential
amplifier considerably more linear than the common-emitter circuit, for which an amplitude
of only v̂i < 1 mV is permissible for k < 1%. Furthermore, in order to reduce the distortion
factor, the amplitude must be reduced proportionally to the root of the distortion factor,
and not linearly as is the case in the common-emitter circuit.
The calculation is only valid as long as no overload occurs at the output; this was implied
by the assumption of an ideal tanh characteristic. However, the gain of most differential
amplifiers with current sources is so high that a differential operation of as little as a
few millivolts causes the output to be overloaded; this is particularly the case in cascode
differential amplifiers. In this case, the differential amplifier works virtually linearly up
to the point of overload of the output and the distortion factor is correspondingly low. As
soon as overload at the output sets in, however, the distortion factor increases rapidly.
The differential voltage of the npn differential amplifier with current feedback is:
VD = VBE 1 + IC1 RE − VBE 2 − IC2 RE = VBE 1 − VBE 2 + (IC1 − IC2 ) RE
If VD is replaced by VD
= VBE 1 − VBE 2 it follows from (4.61) that
VD
vD
vD = vD + 2I0 RE tanh
2VT
It follows from (4.62) that:
vD
vo1 = −I0 RC tanh
2VT
results in
Series expansion and the elimination of vD
3
I0 RC VT vD
vo1 = − vD − + ···
I0 RE + V T 12 (I0 RE + VT )3
which allows us to calculate the distortion factor of an npn differential amplifier with
current feedback:
360 4 Amplifiers
2 gm =I0 /VT
2
VT vD 1 v̂D
k ≈ = (4.110)
48 (I0 RE + VT )3 48 (1 + gm RE )3 VT
As the feedback factor (1 + gm RE ) has a cubic influence on the distortion factor but only a
linear influence on the differential gain, the distortions drop quadratically with the feedback
factor when the output amplitude is kept constant. Thus, the linearizing effect of the current
feedback is much stronger in the differential amplifier than in the common-emitter circuit,
in which the distortions at the output only drop linearly with the feedback factor when the
output amplitude is constant.
With the same approach in the n-channel differential amplifier, we obtain the distortion
factor of an n-channel differential amplifier:
√
2
K v̂D gm = 2KI0 2
K v̂D RS =0 2
K v̂D
k ≈ # $3 = = (4.111)
64I0 1 + 2KI0 RS 64I0 (1 + gm RS )3 64I0
Here too, the feedback factor (1+gm RS ) has a cubic effect. In contrast to the npn differential
amplifier, the size of the MOSFETs influences the results in the form of the transconduc-
tance coefficient K. Without feedback (RS = 0) the distortion factor increases linearly
size of the MOSFETs (k ∼ K), but it reduces when the negative feedback is high
with the √
(k ∼ 1/ K for gm RS 1). The equations are only valid under the condition that the
output is not overloaded.
Relating the distortion factor to the amplitude v̂o at the output of a differential amplifier
with resistances and demanding a certain differential gain leads to an expression that is
helpful for dimensioning the circuit. We shall therefore look at the differential amplifiers
with current feedback shown in Fig. 4.79, which enter the state of a corresponding differ-
ential amplifier without current feedback when RE = 0 or RS = 0. For the npn differential
amplifier, we obtain:
Vb Vb Vb Vb
RC RC I 0 RC RD RD I 0 RD
I0 I0
RE RE RS RS
2I 0 2I 0
– Vb – Vb
Fig. 4.79. Circuits for comparing the distortion factors of npn and n-channel differential amplifiers
4.1 Circuits 361
2 ⎫
1 v̂D ⎪
⎪
knpn ≈ ⎪
⎬
48 (1 + gm RE )3 VT |AD |VT v̂o2
⇒ knpn ≈
⎪
⎪ 6 (I0 RC )3
|AD | ≈
v̂o
=
g m RC ⎪
⎭
v̂D 1 + g m RE
Here, I0 RC is the voltage drop across the collector resistance (see Fig. 4.79a). For the
n-channel differential amplifier:
2 ⎫
KvD ⎪
⎪
knC ≈ ⎪
3 ⎬
64I0 (1 + gm RS ) |AD | (VGS − Vth ) v̂o2
⇒ knC ≈
v̂o g m RD ⎪ ⎪
⎪ 32 (I0 RD )3
|AD | ≈ = ⎭
v̂D 1 + g m RS
Here, I0 RD is the voltage drop across the drain resistance (see Fig. 4.79b). In both dif-
ferential amplifiers, the distortion factor is proportional to the third power of the voltage
drop across resistances RC and RD . As this voltage drop must be selected as a function of
the supply voltage Vb , the distortion factor increases almost cubically when Vb decreases:
halving the supply voltage thus results in an eight-fold distortion factor. The feedback
resistances RE and RS are not shown explicitly, as their values are in a fixed relation to
RC and RD , respectively, due to the presupposed constant differential gain. From the ratio
knC 3 VGS − Vth VT =26 mV VGS − Vth
≈ =
knpn 16 VT 140 mV
it follows that the distortion factor of an npn differential amplifier is usually lower than
that of an n-channel differential amplifier with the same differential gain.
Example: In the description of the n-channel differential amplifier with current feedback,
the characteristics of the circuits shown in Fig. 4.63 on page 339 were compared (see
Fig. 4.64). The outcome was that the characteristics of the differential amplifier without
current feedback are less linear than that of the differential amplifier with current feedback.
This result can be verified by means of the distortion factor approximations. Both circuits
have the same quiescent current and the same differential gain; that is, the same output
amplitude and the same input amplitude v̂D . For the differential amplifier without feedback,
where I0 = 100 mA, K = 15 · 30 mA/V2 = 0.45 mA/V2 (size 15) and v̂D = 0.5 V,
the distortion factor is k ≈ 1.76%; for the differential amplifier with feedback, where
K = 150 · 30 mA/V2 = 4.5 mA/V2 (size 150) and RS = 2 k, and with the same values
for all other parameters, the distortion factor is k ≈ 0.72%. This confirms the previous
result.
Vb+ Vb Vb
Vb+ Vb
Vb V > 0
Vb– Vb
Vb– – Vb
Fig. 4.80. Supply voltages for differential amplifiers: general, symmetrical and unipolar
Supply voltages: In general, a differential amplifier is provided with the two supply
voltages Vb+ and Vb− ; where Vb+ > Vb− . The voltage difference Vb+ − Vb− must be at
least large enough to keep all transistors in the normal or pinch-off region, and small
enough to ensure that the maximum permissible voltages are not exceeded in any transistor.
Theoretically, all combinations that meet these conditions are possible, but in practice two
cases are particularly common:
– A symmetrical voltage supply, with Vb+ > 0 and Vb− = − Vb+ . In this case, the supply
voltage terminals are usually called Vb and −Vb . Examples: ±12 V, ±5 V.
– A unipolar voltage supply, with Vb+ > 0 and Vb− = 0. Here, the terminal Vb− is
connected to ground. The terminal Vb+ is usually called Vb . Examples: 12 V; 5 V; 3.3 V.
Figure 4.80 shows a comparison of the general case and the two practical solutions. For
unipolar voltage supply only one supply voltage source is required.
Differential amplifier with resistances: Fig. 4.81a shows the common method of setting
the operating point in a differential amplifier with resistances, taking an npn differential
amplifier as an example. Current 2I0 is derived from the reference current I1 by an npn
current mirror, the current ratio being kI = 2I0 /I1 . Current I1 can be most easily adjusted
with a resistance R1 . Voltage V0 at the output of the current mirror must not drop below
the lower limit V0,min – in the simple current mirror VCE,sat or VDS,po – this represents a
lower limit of the common-mode voltage range.
4.1 Circuits 363
R1
I1 VBE
2I 0 2I 0
V0 R0 V0
Fig. 4.81. Common methods for setting the operating point in npn differential amplifiers with
resistances
For only slight variations in the common-mode voltage, the current source can be
replaced by a resistance (see Fig. 4.81b):
V0 VCM − VBE − Vb−
R0 = =
2I0 2I0
In this case, the common-mode rejection is comparatively low because resistance R0 is
usually significantly lower than the output resistance r0 of a real current source.
Differential amplifier with current sources: Fig. 4.82 uses an npn differential amplifier
to show the common method of setting the operating point in differential amplifiers with
simple or cascode current mirrors. As in the differential amplifier with resistances, the
current mirror 2I0 is realized by an npn current mirror with the current ratio kI = 2I0 /I1 .
A pnp current mirror with two outputs is used for the current sources at the output; here,
the same reference current I1 is used, which leads to a current ratio of kI = I0 /I1 . Here
again, a resistance R1 is the simplest method of adjusting current I1 . Voltage V1 for the
cascode stage is adjusted to Vb+ − 2VEB ≈ Vb+ − 1.4 V by the two pnp transistor diodes.
Cascode differential amplifier: The cascode differential amplifier with cascode current
sources requires two auxiliary voltages; Fig. 4.83 shows an npn cascode differential am-
plifier as an example of a typical circuit. The current is adjusted in the same way as in the
differential amplifier with current sources. Voltage V2 for the pnp cascode stage is again
set to Vb+ − 2VEB ≈ Vb+ − 1.4 V by two pnp transistor diodes. Voltage V1 for the npn
cascode stage is provided via a voltage divider that consists of resistances R1 and R2 and
a collector circuit for impedance conversion; the current of the collector circuit is set by
an additional current source. The value of voltage V1 affects the voltage range at the input
and at the output: a relatively high voltage V1 provides a larger common-mode voltage
range at the input and a lower voltage range at the output; a lower voltage has the opposite
effect.
364 4 Amplifiers
V1
I0 I0
I0 I0
R1
R1
I1 I1
2I 0 2I 0
V b– V b– V b– V b–
Fig. 4.82. A common method for setting the operating point in npn differential amplifiers with
current sources
Differential amplifier with folded cascode: In the ideal case, the input and output volt-
age ranges should cover the entire range of the supply voltages. The differential amplifier
with folded cascode shown in Fig. 4.84 is very close to this ideal. It is evolved from the
normal cascode differential amplifier by folding the cascode stage and the output current
sources downward and adding two more current sources. The input and the output can
now be modulated almost over the entire range of the supply voltages; a particular conse-
quence of this is that the output voltages may be lower than the input voltages. However,
the small-signal response remains the same. In practice, an asymmetrical output is usually
used, by replacing the output current sources with a cascode current mirror; this results in
the circuit shown in Fig. 4.85, which is predominantly used as an input stage in operational
amplifiers because of its wide voltage range and its high differential gain and common-
mode rejection. Resistance R1 is often replaced by the reference current source described
in Sect. 4.1.5, to make the quiescent currents independent of the supply voltages.
Output voltage control: In all symmetrical differential amplifiers with current sources,
the output voltages at the operating point are undefined without external circuits. The
reason for this is the small difference between the currents of npn and pnp transistors or
n-channel and p-channel transistors, which cause the outputs to reach either the upper
or lower voltage limits. With low-resistance loads at the outputs, the operating point is
defined by the loads; they absorb the transistor differential currents. If, however, high-
4.1 Circuits 365
V2
I0 I0
R1
V1
R2
2I 0
Fig. 4.83. A common method for setting the operating point in an npn cascode differential
amplifier with cascode current sources
Vb+ Vb+
2I 0 2I 0
I0 I0
V1
2I 0 I0 I0
V b– V b– V b–
2I 0 2I 0
I0 I0
I0 I0
R1
2I 0
Fig. 4.85. A common configuration of a differential amplifier with a folded cascode and an
asymmetrical output
resistive loads are connected, the output voltages must be controlled in order to prevent
overload in the circuit; this is done by suitably controlling either current source 2I0 or the
two output current sources I0 .
If common-collector or common-drain stages are connected to the outputs for
impedance conversion, the current source 2I0 can be controlled by adjusting the qui-
escent currents of these circuits via resistances that are connected to the reference path of
the current source; Fig. 4.86 illustrates this method using an npn differential amplifier with
npn common-collector circuits as an example. For R2 = R3 , the outputs at the operating
point are:
R2
Vo,A = Vb− + VBE 7 + I1 R2 = Vb− + VBE 7 1 + with VBE 7 ≈ 0.7 V
2R4
This presupposes that the current mirror T7 ,T8 has the current ratio 2, as in the noncontrolled
mode. Alternatively, resistance R4 can be omitted and the operating point can be adjusted
with the current ratio kI of the current mirror T7 ,T8 ; it thus follows:
1 1
kI (I0 + 2I1 ) ≡ 2I0 ⇒ I1 = I0 −
kI 2
The output voltages are related to the supply voltage Vb− , which is particularly disadvan-
tageous in circuits with variable supply voltages. This is remedied in the version shown
in Fig. 4.87, in which the output voltages relate to supply voltage Vb+ and where the pnp
current sources are controlled. Here, the following applies:
4.1 Circuits 367
T3
T9 T4
I0 I0
T5 T6
R1
V o,A Vo,A
I0 T1 T2
R2 R3
I1 2I 0 I1
2I 1
R4 T7 T8
V b– V b– V b–
Fig. 4.86. Control of the output voltages in a differential amplifier with common-collector circuits
(in relation to the supply voltage Vb− )
R4
T12 T4
2I 1 T3
R2 R3 T5 T6
I0 I1 I1 I0 I0
Vo,A Vo,A
T1 T2
R1
T7
2I 0
T8 T11
T9 T10
Fig. 4.87. Output voltage control in differential amplifiers with common-collector circuits (in
relation to the supply voltage Vb+
368 4 Amplifiers
T3
T11 T4 T7 T8
I0 I0
R1
I1 I1
Vo,A
T5 T6
Vo,A
R2
T1 T2
2I 1
2I 0
T10 T9
Fig. 4.88. Output voltage control with a subsequent npn differential amplifier
R2
Vo,A = Vb+ − VEB12 − I1 R2 = Vb+ − VEB12 1 + with VEB12 ≈ 0.7 V
2R4
Again, resistance R4 can be omitted and the operating point can be adjusted using the
current ratio kI of current mirrors T12 ,T3 and T12 ,T4 :
I0 1
I1 = −1
2 kI
Here, condition kI < 1 must be met; that is, T12 must be larger than T3 and T4 .
In both versions resistances, R2 and R3 must not be too small, as they present a load
to the output and thus reduce the differential gain. Therefore, in differential amplifiers
with very high output resistances it is often necessary to connect two common-collector
circuits in series before the resistances can be connected. In the corresponding circuits with
MOSFETs, however, a single common-drain circuit prevents any effect on the differential
gain as a result of the resistances.
The same method can be used when another npn differential amplifier is connected
instead of the common-collector circuits; Fig. 4.88 shows the corresponding circuit. With
the current ratio kI of the current mirror T9 ,T10 , the parameters are:
I0
I1 = , Vo,A = Vb− + VBE 9 + 2I1 R2 + VBE 5
kI
4.1 Circuits 369
R1 Vb+ – 0.7 V
T3 T9
T4
I0 I0 2I 1
Vb+ – 1.4 V
Vb+ – 1.4 V
T5 T6
I1 I1
T1 T2
2I 0
T10 T11 T7 T8
Fig. 4.89. Output voltage control with a subsequent pnp differential amplifier
If a pnp differential amplifier is connected, the circuit shown in Fig. 4.89 can be used,
in which the pnp current sources are controlled without additional resistances; here, the
voltage is
Vo,A = Vb+ − VEB9 − VEB5 ≈ Vb+ − 1.4 V
and with the current ratio kI of the current mirrors T9 ,T3 and T9 ,T4 the current is:
I0
I1 =
2kI
In this version the gain of the control loop is very high and may have to be limited by current
feedback resistances in the current mirrors, which means that resistances corresponding
to the current ratio have to be inserted on the emitter side of T3 , T4 and T9 . Such circuits
are used especially in precision operational amplifiers
All methods of controlling the output voltages increase the common-mode rejection,
because they compensate the unidirectional change of the output voltages caused by the
common-mode voltage. Therefore, operational amplifiers provided with the circuit shown
in Fig. 4.89 have a particularly high common-mode rejection and, due to the two differential
amplifiers, a particularly high differential gain.
CL RL vo1 vo2 RL CL
Differential amplifier
Rg Rg
vi1 vi2
vg1 vg2
v o1 (s)
AB,D (s) =
(4.113)
v g,D (s)
vg,CM =0
v o1 (s)
AB,CM (s) =
(4.114)
v g,CM (s)
vg,D =0
AB,D (s)
GB (s) = (4.115)
AB,CM (s)
In what follows, the prefix operational is omitted for the sake of simplicity.
The symmetry characteristics are utilized again to calculate the frequency responses. In
this way, we can convert the symmetrical differential amplifier back to the corresponding
common-emitter, common-source or cascode circuits. This is not possible with asymmetric
differential amplifiers with current mirrors because of the asymmetry; in addition, the
frequency response of the current mirror has to be taken into account. For the calculation
of static parameters, an ideal current mirror was assumed; this meant that the results of the
symmetric differential amplifier could simply be transferred to the asymmetric differential
amplifier. This procedure may be used here as well, since current mirrors generally have
a very high cutoff frequency; an ideal frequency response is thus assumed for the current
4.1 Circuits 371
Frequency response and the cutoff frequency of the differential gain: The frequency
response of the differential gain can be approximately described by a first-order lowpass
filter:
A0
AB,D (s) ≈ (4.116)
s
1+
ωg
If the internal resistance Rg of the signal source and the load resistance RL are taken into
account, then the operational gain A0 at low frequencies is:
ri,D RL
A0 = AB,D (0) = AB = AD (4.117)
ri,D + 2Rg ro,D + RL
The -3 dB cutoff frequency f-3dB , at which the magnitude of the gain is 3 dB lower, is
derived from (4.116): ω-3dB ≈ ωg . This can be described by means of the low-frequency
gain A0 and two time constants:
For |A0 | T1 /T2 , the cutoff frequency is inversely proportional to the magnitude of the
gain A0 and the result is a constant gain–bandwidth product (GBW ):
1
GBW = f-3dB |A0 | ≈ (4.119)
2π T2
The time constants T1 and T2 for the different types of differential amplifiers are
described in the following sections:
2.4.1 Common-emitter circuit: (2.92), (2.96), (2.99)–(2.101) pp. 124ff.
3.4.1 Common-source circuit: (3.77), (3.80), (3.83) pp. 246ff.
4.1.2 Cascode circuit (4.45), (4.46), (4.53), (4.54) pp. 321 and 326
Figure 4.91 contains an overview of the case in which the capacitances of the npn and pnp
transistors and the n-channel and p-channel MOSFETs are identical. In order to differen-
tiate, for time constant T2 all capacitances with factor 2 must be replaced by the sum of
the corresponding values:
All other capacitances relate to the npn transistors in the npn differential amplifier or to
the n-channel MOSFETs in the n-channel differential amplifier; this also applies to the
capacitances shown with factor 2 in the time constant T1 .
Some of the equations in Fig. 4.91 have been modified, as a comparison with the
original calculated forms will show:
372 4 Amplifiers
&
T1 = CE + CC Rg || rBE
With resistances C S + CL C C + CS + CL
T2 = CC + Rg +
gm
≈ g /(1 + g R ) and C
≈ C /(1 + g R )
with gm m m S GS GS m S
T1 = (CGS + CGD ) Rg
With current sources 2CGD + 2CBD + CL
T2 = CGD Rg +
gm
T1 = (CGS + 2CGD ) Rg
With a cascode 2CGD + 2CBD + CL
T2 =
gm
Fig. 4.91. Time constants for the cutoff frequency of the differential gain
– The extrinsic base resistance and the gate resistance have been neglected; that is, Rg
=
Rg + RB and Rg
= Rg + RG have been replaced by Rg .
– For npn differential amplifiers, the basic equations for the common-emitter circuit have
been expanded by adding the substrate capacitance CS ; furthermore, CL has been re-
placed by CL + CS , as the substrate capacitance has the effect of a load capacitance.
– In the basic equations for the common-source circuit for n-channel differential ampli-
fiers, the drain–source capacitance CDS , which only occurs in discrete MOSFETs, has
been replaced by the bulk–drain capacitance CBD .
4.1 Circuits 373
CGD = CGD,ov W
CS ≈ CJ 0,S
With current feedback, some parameters are transformed by the feedback factor;
Fig. 4.91 only reflects this for the differential amplifier with resistances, but the same
can be done for other types.
The small-signal parameters of integrated bipolar transistors and MOSFETs, which
are required to evaluate the time constants, are listed in Fig. 4.92; these are taken from
Fig. 2.45 on page 83 (without CE and CC ), (4.49) and (4.50) on page 325 and Fig. 3.52
on page 223. For the junction capacitances CC , CS and CBD , the corresponding zero
capacitance C(U = 0) is used without regard for the actual reverse voltage; the true
capacitance is lower.
The frequency responses of the differential gain are illustrated in Fig. 4.93. The values
of the low-frequency gain apply to npn differential amplifiers; the values for correspond-
ing n-channel differential amplifiers are about ten times lower. Differential amplifiers with
simple and cascode current sources achieve a higher differential gain than the differen-
tial amplifier with resistances, but feature a lower gain–bandwidth product due to the
additional capacitances of the current source transistors. The cascode differential ampli-
fier with cascode current sources offers both the highest differential gain and the highest
gain–bandwidth product.
The differential gain and the gain–bandwidth product of the differential amplifier with a
simple current mirror are approximately twice as large as those of corresponding symmetric
differential amplifiers; therefore, both circuits have the same cutoff frequency. This also
applies to the n-channel cascode amplifier with a cascode current mirror. In the npn cascode
differential amplifier with a cascode current mirror, the gain–bandwidth product is also
about twice as high as that of the npn cascode differential amplifier with a cascode current
mirror, but the differential gain is only slightly higher, due to the low output resistance of
the cascode current mirror compared to the cascode current source; the cutoff frequency
is therefore higher. The frequency responses of differential amplifiers with current mirrors
are not shown in Fig. 4.93 for the sake of clarity.
AD
4
10000
1000
2
100
1
4
1
10
2, 3
1
~
~ f [log]
f – 3 dB GBW
Fig. 4.93. Frequency responses of the differential gain (the numeric values apply to npn
differential amplifiers)
gain only by the impedance of the current source, which causes frequency-dependent
current feedback; the approximate frequency response of the common-mode gain can
thus be calculated from the frequency response of the differential gain by replacing the
transconductance gm by the reduced transconductance:
gm r0 1
gm 1 + sC0 r0
gm,red (s) = ≈
2 C0
1 + gm 2 r0 || 2r0 1 + s
sC0 2gm
RB CC
Fig. 4.94. Dynamic small-signal equivalent circuit of an npn differential amplifier with resistances,
with common-mode input signal
4.1 Circuits 375
As the common-mode signal provides the full common-mode voltage at each input, the
result must be multiplied by two. Taking the output resistances into account, it follows
from (4.116):
gm,red (s)ro,CM A0 ro,CM 1 + sC0 r0
AB,CM (s) ≈ 2AB,D (s) ≈
gm ro,D gm r0 ro,D C0 s
1+s 1+
2gm ωg
If one inserts the common-mode rejection
gm r0 ro,D
G =
ro,CM
and replaces the time constant C0 r0 by the cutoff frequency of the common-mode rejection
1
ωg,G = 2πfg,G = (4.120)
C 0 r0
then:
s
1+
A0 ωg,G
AB,CM (s) ≈ (4.121)
G s s
1+ 1+
2Gωg,G ωg
s
1+
2Gωg,G
GB (s) ≈ G (4.122)
s
1+
ωg,G
Figure 4.95 shows the frequency responses |AB,D |, |AB,CM | and |GB | for the cases fg,G <
fg and fg,G > fg .
The case fg,G < fg is typical for differential amplifiers with resistances or with simple
current sources. The magnitude of the common-mode gain increases in the region between
the common-mode cutoff frequency fg,G and the cutoff frequency fg , remains constant
above fg and at high frequencies is twice as high as the magnitude of the differential gain.
The common-mode rejection decreases from the common-mode cutoff frequencyfg,G to
higher frequencies at the rate of 20 dB per decade, and approaches 1/2 at high frequencies.
The situation fg,G > fg mostly occurs in cascode differential amplifiers that have a
relatively low cutoff frequency fg because of their very high low-frequency gain; fg is still
very low even if the gain–bandwidth product is high. The common-mode gain decreases
between the cutoff frequency fg and the common-mode cutoff frequency fg,G , is constant
above fg,G and at high frequencies is twice as high as the magnitude of the differential
gain. The common-mode rejection is the same as for fg,G < fg .
The simplified derivation of the frequency response of the common-mode gain makes
understanding easier, but leads to inaccuracies:
– Due to the frequency-dependent feedback in common-mode operation, the cutoff fre-
quency fg is different than in differential operation. In most circuits this effect is in-
significant, but in some it is quite considerable; it then creates an additional pole and
an additional zero in the common-mode rejection. In the differential amplifier with re-
sistances, this creates a region in which the common-mode rejection decreases at a rate
of 40 dB per decade, and in the differential amplifier with cascode current mirrors it
376 4 Amplifiers
A B , GB
10000
GB
1000
A B,D
100
10
2Gfg,G
f g,G
1
~
fg f [log]
0.1
A B,CM
A B , GB A B,D
10000
1000 GB
100
10 A B,CM
2Gfg,G
f g,G
1
~
fg f [log]
0.1
Fig. 4.95. Frequency responses |AB,D |, |AB,CM | and |GB | for the cases fg,G < fg (above) and
fg,G > fg (below)
creates a region in which the common-mode rejection increases; Fig. 4.96 illustrates
these special cases.
– In the npn differential amplifier, the differential and the common-mode portions of
the input signal are attenuated differently because of the different input resistances
in differential and in common-mode operation. Thus, especially in the case of high-
resistance sources, the low-frequency value of the operational common-mode rejection
GB (s) does not correspond to the common-mode rejection G, but is reduced by a value
equal to the ratio of the voltage divider factors:
ri,CM
Rg ri,CM
ri,CM + 2Rg 2Rg
≈ 1+
ri,D ri,D
ri,D + 2Rg
This effect is not seen with low-resistance sources where Rg ri,D .
4.1 Circuits 377
AB AB
GB GB
[log] [log] A B,D
GB
GB
A B,D
A B,CM
1 1
~
~
f [log] f [log]
A B,CM
Example: The various npn and n-channel differential amplifiers are compared below.
All circuits are designed for a unipolar supply voltage Vb = 5 V and an output voltage
Vo,A = 2.5 V. The parameters for bipolar transistors are taken from Fig. 4.5 on page 274
and those for MOSFETs from Fig. 4.6 on page 274. The quiescent current is I0 = 100 mA
in npn differential amplifiers and I0 = 10 mA in n-channel differential amplifiers. Bipolar
transistors of size 1 are generally used per 100 mA quiescent current; this corresponds to the
typical value quoted in Fig. 4.5. According to Fig. 4.6, MOSFETs of size 1 would also be
sufficient, but the resulting gate–source voltage of |VGS | ≈ 1.8 . . . 2 V (|VBS | = 0 . . . 1 V)
is too high for the given supply voltage of 5 V; therefore n-channel MOSFETs of size 5
(VGS ≈ 1.4 . . . 1.6 V) and p-channel MOSFETs of size 2 (VGS ≈ − 1.6 . . . − 1.8 V) are
used per 10 mA quiescent current. Since the dimensional ratio of n-channel and p-channel
MOSFETs of size 1 is exactly 2/5, all MOSFETs, with the exception of the MOSFET in
the current source, have the same geometric size:
W = 15 mm , L = 3 mm
The common-mode voltage at the input is VCM,A = 1 V in npn differential amplifiers
and VCM,A = 2 V in n-channel differential amplifiers; thus the current sources in the
common-emitter or common-source branch are operated just above their signal limits.
Figure 4.97 shows the differential amplifiers with resistances; the collector and drain
resistances are chosen such that the desired output voltage Vo,A = 2.5 V is reached:
.
RC Vb − Vo,A 25 k
= =
RD I0 250 k
In contrast, the operating point is not automatically adjusted in the differential am-
plifiers with simple current sources and simple current mirrors shown in Fig. 4.98. As,
in general, the collector and drain currents of transistors T1 and T3 or T2 and T4 are not
exactly the same at the desired operating point, the transistor with the larger current enters
the saturation or the pinch-off region; this situation overloads the outputs. In an integrated
circuit the actual operating point depends on the external components at the outputs and
an operating point control circuit, should this exist; the latter is described in more detail
378 4 Amplifiers
5V 5V 5V 5V 5V 5V
R3 R1 R2 R3 R1 R2
41.5 kΩ 25 kΩ 25 kΩ 358 kΩ 250 kΩ 250 kΩ
T1 T2 T1 T2
1V 1 1 1V 5 5
2V 2V
0.3 V 0.5 V
202 µA 20 µA
T4 0.7 V T3
T4 T3
1 2 1.4 V
5 10
in the section on setting the operating point in differential amplifiers. During the circuit
simulation the desired operating point can be set by, for example, connecting the outputs
to a voltage source with Vo,A via very high inductances (e.g. L = 109 H); this keeps the
DC output potential at Vo,A , while for AC voltage we have an open circuit due to the in-
ductances that present a very high impedance even at low frequencies. This method must
be used for all differential amplifiers with current sources and current mirrors. For the
differential amplifiers in this example, an operating point of Vo,A = 2.5 V is assumed, but
the necessary external circuitry or operating point control circuit is not illustrated.
Auxiliary voltages are required to set the operating point of the cascode transistors
in the differential amplifiers with cascode current sources in Fig. 4.99 and the cascode
differential amplifiers with cascode current sources and cascode current mirrors as shown
in Figs. 4.100 and 4.101, respectively; the generation of these voltages is detailed in
Sect. 4.1.5.
The small-signal parameters of the transistors can be determined with the help of the
information in Fig. 4.92 on page 373 and the parameters in Figs. 4.5 on page 274 and 4.6
on page 274, and on the basis of the quiescent currents and the transistor sizes. By means of
the following equations, it is possible to calculate the gain, the output and input resistances
of differential amplifiers for differential and common-mode operation:
The operational differential gain A0 is determined from (4.117), the time constants T1
and T2 from Fig. 4.91, the gain–bandwidth product (GBW ) from (4.119), the -3 dB cutoff
4.1 Circuits 379
5V 5V 5V 5V 5V 5V
3.4 V
T7 4.3 V T3 T4 T7 T3 T4
1 1 1 2 2 2
100 µA 100 µA 10 µA 10 µA
R1 R1
T1 T2 T1 T2
35 kΩ 200 kΩ
1V 1 1 1V 5 5
2V 2V
104 µA 10 µ A
0.3 V 0.5 V
202 µA 20 µA
T6 0.7 V T5 T6 T5
1 2 5 1.4 V
10
5V 5V 5V 5V 5V 5V
R1 R1
41.5 kΩ T3 T4 358 kΩ
T3 T4
1 1 2 2
100 mA 100 µA 10 µA 10 µA
T1 T2 T1 T2
1V 1 1 1V 5 5
2V 2V
0.3 V 0.5 V
202 µA 20 µA
T6 0.7 V T5 T6 T5
1 2 5 1.4 V
10
Fig. 4.98. Example: differential amplifiers with simple current sources and simple current mirrors
frequency f-3dB from (4.118) and the cutoff frequency fg,G of the common-mode rejection
from (4.120).
When calculating the small-signal parameters of the npn transistors, the slight dif-
ferences in the quiescent currents of the individual transistors are neglected; that is,
|IC,A | ≈ I0 ≈ 100 mA is used. Consequently,
npn: gm = 3.85 mS , β = 100 , rBE = 26 k , rCE = 1 M ,
CE = 0.6 pF , CC = 0.2 pF , CS = 1 pF
pnp: β = 50 , rCE = 500 k , CC = 0.5 pF , CS = 2 pF
380 4 Amplifiers
5V 5V 5V 5V 5V 5V
3.4 V
T9 4.3 V T5 T6 T9 T5 T6
1 1 1 2 2 2
2.5 V 4.3 V
4.3 V 4.3 V
3.6 V T3 T4 T3 T4
1 1 2 2
R1 100 µA 100 µA R1 10 µA 10 µA
35 kΩ 200 kΩ
T1 T2 T1 T2
1V 1 1 1V
5 5
2V 2V
104 µA 10 µA
0.3 V 0.5 V
202 µA 20 µA
T8 0.7 V T7 T8 T7
1 2 5 1.4 V
10
For the current source, r0 = VA,npn /(2I0 ) = 500 k. The output capacitance C0 of the
current source is the sum of the substrate capacitance and the collector capacitance of the
current source transistors. Since the transistors are of size 2, both partial capacitances are
twice as high as in the other npn transistors; therefore C0 = 2(CS + CC ) = 2.4 pF. From
(4.120), the cutoff frequency of the common-mode rejection is thus fg,G = 133 kHz. The
resulting values for the npn differential amplifiers are listed in Fig. 4.102. In differential
amplifiers with a current mirror, the values for common-mode operation are determined
with the help of circuit simulation; they are quoted in parentheses.
With I0 = 10 mA, the values for MOSFETs are:
n-channel: K = 150 mA/V2 , gm = 54.8 µS , rDS = 5 M ,
CGS = 18 fF , CGD = 7.5 fF , CBD = 17 fF
p-channel: K = 60 mA/V2 , gm = 34.6 µS , rDS = 3.3 M ,
CGD = 7.5 fF , CBD = 17 fF
This assumes that the drain areas are 5 mm long and 2 mm wider than the channel width W .
Consequently,
AD = (15 + 2) · 5 µm2 = 85 µm2 ⇒ CBD = CJ
AD = (0.2 · 85) fF = 17 fF
For the current source, r0 = VA,nC /(2I0 ) = 2.5 M. The output capacitance C0 of the
current source consists of the bulk–drain and the gate–drain capacitances of the current
source MOSFET and the bulk–source capacitances of MOSFETs T1 and T2 ; the latter have
the same size as the bulk–drain capacitances because of the symmetrical configuration.
4.1 Circuits 381
5V 5V 5V 5V 5V 5V
3.4 V
3.6 V T5 T5 T6
T6
1 2 2
1
100 µA 100 µA 10 µA 10 µA
1.4 V T3 T4 T3 T4
1 1 5 5
2.5 V
R1 T1 T2 R1 T1 T2
34.5 kΩ 1V 1 1 1V 200 kΩ 5 5
2V 2V
106 µA 0.3 V 10 µA 0.5 V
204 µA 20 µA
Fig. 4.100. Example: cascode differential amplifiers with cascode current sources
With the drain area of the current source MOSFET AD = (32 · 5) µm2 = 160 µm2 , the
output capacitance is:
C0 = CJ
AD + 2CGD + 2CBD = (0.2 · 160 + 2 · 7.5 + 2 · 17) fF = 83 fF
The cutoff frequency of the common-mode rejection is thus fg,G = 767 kHz. The resulting
values for the n-channel differential amplifiers are listed in Fig. 4.103. Again, the values
for common-mode operation in differential amplifiers with current mirror are determined
by circuit simulation.
A comparison of the values for npn and n-channel differential amplifiers shows that the
differential gain in npn differential amplifiers is approximately ten times higher than that
of the corresponding n-channel differential amplifiers; only in cascode differential ampli-
fiers is the difference less. It must be taken into consideration that n-channel MOSFETs
were chosen that are five times as large as the √ size that the quiescent current calls for;
this increases the differential gain by a factor of 5. The reason for the lower differential
gain in the n-channel differential amplifiers is the lower maximum gain of the MOSFETs.
The situation improves with the MOSFETs of the cascode differential amplifiers, as their
output resistance rises without any limit with increasing current feedback, while in bipolar
transistors it is limited to β rCE . This means that with additional cascode stages the differ-
382 4 Amplifiers
5V 5V 5V 5V 5V 5V
R1 R1 T7 T8
41 kΩ
T7 T8 358 kΩ
2 2
1 1
4.3 V 3.4 V
T5 T6
T5 T6
2 2
1 1
3.6 V 1.6 V
2.5 V 2.5 V
100 µA 100 µA 10 µA 10 µA
1.4 V T3 T4 T3 T4
1 1 5 5
2.5 V
1V
T1 T2 T1 T2
1V 1 1 1V 5 5
2V 2V
204 µA 20 µA
ential gain of an n-channel cascode differential amplifier may be increased almost without
any limitation.
Usually, additional amplifier stages are connected to the outputs of a differential am-
plifier. To ensure that the differential gain is fully maintained, the input resistances of these
stages must be higher than the output resistances of the differential amplifier. In CMOS
circuits this condition is met automatically, because of the isolated gate connections of
the MOSFETs, which means that the maximum operational gain AB,D = AD is achieved
without additional measures. In bipolar circuits, on the other hand, each output must be pro-
vided with an impedance converter with one or more common-collector circuits, to reduce
the output resistances to a value below the input resistance of the next stage. Impedance
converters are explained in more detail in Sect. 4.1.4.
Due to the significant variations in gain, a meaningful comparison of the cutoff fre-
quencies of the differential amplifiers described here is only possible on the basis of the
gain–bandwidth product. Due to the very low capacitances of the integrated MOSFETs,
and despite their low quiescent currents, n-channel differential amplifiers achieve higher
values than npn-differential amplifiers. Since the input capacitances of the subsequent
amplifier stages are also very low, this advantage of integrated circuits is maintained to
its full extent. However, if higher load capacitances are connected or exist outside an in-
tegrated circuit, npn differential amplifiers reach a higher gain–bandwidth product, due
4.1 Circuits 383
to the higher transconductance of the bipolar transistors. This can be seen from the time
constant T2 in Fig. 4.91, for the borderline case of high load capacitances CL :
⎧
⎪ Rg 1
⎪
⎨ CL + npn differential amplifier
β gm
lim T2 =
CL →∞ ⎪
⎪ C
⎩ L n-channel differential amplifier
gm
If we assume load capacitances of CL = 100 pF for the npn differential amplifiers and
CL = 10 pF for the n-channel differential amplifiers, which makes the ratio of the qui-
escent current to the load capacitance equal in both cases, the gain–bandwidth product is
GBW ≈ 4.4 MHz for the npn differential amplifier and GBW ≈ 870 kHz for the n-channel
differential amplifier. Here too, it must be taken into account that the chosen n-channel
MOSFETs are already five times larger than would have been necessary due to the quies-
cent current;
√ this increases the transconductance and thus the gain–bandwidth product by
a factor of 5 for a capacitive load.
384 4 Amplifiers
Summary
Due to its particular characteristics, the differential amplifier is one of the most important
circuits in integrated circuit engineering. It is found not only in amplifiers but also in
comparators, ECL logic circuits, voltage regulators, active mixers and numerous other
circuit designs. It has a special ranking among amplifier circuits, primarily on account of
the almost unlimited choice of common-mode voltage levels at the input, which means
that any signal source with a DC voltage within the common-mode signal range can be
connected; voltage dividers for setting the operating point or coupling capacitances are not
required. The differential amplifier is therefore a true DC voltage amplifier. As it virtually
amplifies the differential signal alone, it is the choice for a feedback control circuit, since
by subtraction it determines the deviation and subsequently amplifies it; in other words, it
combines the operational blocks subtractor and control amplifier of an automatic control
circuit. It thus forms the basis of the operational amplifier. In this respect, the differential
amplifier is the simplest operational amplifier, and the operational amplifier is the better
differential amplifier.
4.1 Circuits 385
4.1.4
Impedance Converters
The output resistance of an amplifier stage with a high voltage gain is generally very high
and must be reduced by means of an impedance converter before other amplifier stages
or load resistances can be connected without loss of gain. Single-stage or multi-stage
common-collector and common-drain circuits are used as impedance converters.
⎧
⎪
⎪ Rg 1 gm Rg β Rg
⎪
⎨ + ≈ Common-collector circuit
β gm β
ro = (4.123)
⎪
⎪ 1 gm gm,B
1
⎪
⎩ ≈ Common-drain circuit
gm + gm,B gm
Vb Vb Vb Vb
Rg Rg T1
R1 R1
T1
vg vg
I0 ro I0 ro
T2 T3 T2 T3
Output voltage: At the operating point of both circuits, the output voltage is lower than
the input voltage by a value equal to a base–emitter or gate–source voltage. A pnp common-
collector circuit or a p-channel common-drain circuit could be used as an alternative; in this
case, the output voltage at the operating point is higher than the input voltage. However,
pnp transistors generally have a lower current gain than npn transistors, and p-channel
MOSFETs are larger in size than n-channel MOSFETs with the same transconductance
coefficient; that is, they have higher capacitances.
Vb Vb Vb Vb Vb Vb
T1
R1 T1 R1
T2 T2
I1 I2 I1 I2
T4
T3 T5 T3 T4 T5
current I2 of the second stage; the effective source resistance at this point is Rg /β. The
quiescent currents of both stages are thus:
10β 2 VT β≈100
2600 V 10I2 B≈β≈100
260 V
I2 ≈ ≈ , I1 ≈ ≈ (4.126)
Rg Rg B Rg
Vb Vb
T1
G1 T2
G2
Rg Ci1
Cg ro1 Ci2
I1 I2 ro2
vg Cg v1 v1 v2 v2
Fig. 4.106. Two-stage common-drain circuit: circuit (above) and the small-signal equivalent
circuit (below)
tances and input capacitances depend on the sizes24 G1 and G2 of the MOSFETs T1 and
T2 :
ro
ro
24 In this context, size means the electrical and not the geometric size; that is, G ∼ K.
4.1 Circuits 389
2Cox WL
Ci
≈ 0.2 ·
Output voltage: In a two-stage npn common-collector circuit, the output voltage at the
operating point is lower than the input voltage by a value equal to 2 VBE ≈ 1.4 V. In a
two-stage common-drain circuit, the voltage offset of 2 VGS ≈ 3 . . . 4 V is so high that
an input voltage of at least 4 . . . 5 V is required when utilizing the operation limit of the
current source of about 1 V. With impedance converters that consist of more than two
stages, the voltage offset increases further. As an alternative, one may design one or more
stages as pnp common-collector or p-channel common-drain circuits; this compensates
the base–emitter or gate–source voltage fully or partially. Figure 4.107 shows an example
of a two-stage impedance converter with Vi,A ≈ Vo,A .
I2 T1 I2
T1
V i,A Vo,A Vi,A Vo,A
V BE2 VGS2
VBE1 T2 VGS1
T2
I1 I1
a npn-pnp b n-channel–p-channel
( VBE1 ≈ – VBE2 ) ( VGS1 ≈ – VGS2 )
Figure 4.108 shows the basic circuit of a single-stage complementary impedance con-
verter with bipolar transistors and with MOSFETs. The quiescent currents must be adjusted
using bias voltage sources, the practical aspects of which will be described later. At the
operating point, the input and output voltages are identical; that is, there is no offset volt-
age. For reasons of symmetry, the circuits are shown with a symmetrical voltage supply;
however, a unipolar voltage supply is also possible.
Complementary impedance converters offer the advantage of providing a high output
current in both directions; Fig. 4.109 illustrates this with a sudden change of the input
voltage by comparing a complementary and a simple common-collector circuit. In the
complementary common-collector circuit, the output current is supplied via an active
common-collector circuit in both directions and can therefore become very high; in this
case, the other common-collector circuit is nonconductive. In the case of a sudden step-
Vb Vb
T1
T1
VBE1 VGS1
V= 0 V= 0
Vi VBE2 Vo Vi VGS2 Vo
T2
T2
– Vb – Vb
a With bipolar transistors b With MOSFETs
Vb Vb
T1 T1
≈ I C1 ≈ I C2
T2 T2
– Vb – Vb
Vb Vb
T1 T1
≈ I C1 – I 0 I0
I0 I0 I0
– Vb – Vb – Vb
Fig. 4.109. Comparison of a complementary and a simple common-collector circuit with regards
to their response to a sudden change in the input voltage
down in the input voltage in the simple common-collector circuit, the output current is
supplied by the current source; in other words, it is limited to the quiescent current. For
this reason, complementary impedance converters are used whenever simple impedance
converters would require a disproportionately high quiescent current.
I1 = kI I0
The circuit shown in Fig. 4.110a may be regarded as one npn and one pnp common-
collector circuit connected in parallel; the output resistance is thus:
392 4 Amplifiers
Vb Vb Vb Vb
I0 I0
T1
T1
T3
T3
I1 I1
I1 I1
T4 T4
T2
T2
I0 I0
– Vb – Vb – Vb – Vb
Vb Vb Vb
T5 T7
I1
VBE3 T1
VBE1
T3
I2
– Vb
R1
Vi Vb Vo
I2
T4 VBE2
VBE4 T2
I1
T6 T8
– Vb – Vb – Vb
emitter voltages of npn and pnp transistors of the same size are different for the same
current. The offset voltage can be minimized by scaling the transistors accordingly.
The general calculation of the quiescent currents and the offset voltage is based on the
following equation:
VBE 3 + VBE 1 − VBE 2 − VBE 4 = 0
The base–emitter voltages are:
⎧
⎪ IC
⎪
⎨ VT ln I npn transistor
S
VBE =
⎪
⎪ − IC
⎩ − VT ln pnp transistor
IS
Inserting and dividing by VT leads to:
− IC3 IC1 − IC2 IC4
− ln + ln + ln − ln = 0
IS3 IS1 IS2 IS4
Neglecting the base currents allows us to insert − IC2 = IC1 ≈ I2 and − IC3 = IC4 ≈ I1 ;
it thus follows that
IS3 IS4 I22
ln ≈ 0
IS1 IS2 I12
394 4 Amplifiers
and, consequently:
I2 IS1 IS2 √ IS1 IS2
kI = ≈ = gnpn gpnp with gnpn = , gpnp = (4.132)
I1 IS3 IS4 IS4 IS3
Here, gnpn is the size ratio of the npn transistors T1 and T4 ; gpnp is the size ratio of the
pnp transistors T2 and T3 .
In general, the same size ratios and the same sizes for T1 and T2 are chosen – for
example, size 10 for T1 and T2 and size 1 for T3 and T4 – then kI ≈ gnpn = gpnp = 10
and I2 ≈ 10 I1 . Factor 10 is typical for practical applications because here, as is the case
in simple multi-stage common-collector circuits, a quiescent current ratio of about B/10
is used and B ≈ β ≈ 100 is a typical value for integrated transistors.
The offset voltage is Voff = Vi,A − Vo,A ; from Fig. 4.111, it follows:
I2 I1 IS3 I2
Voff = VBE 1 + VBE 3 ≈ VT ln − VT ln = VT ln
IS1 IS3 IS1 I1
When equal size ratios and the same sizes for T1 and T2 are selected, then kI = I2 /I1 ≈
gnpn = gpnp ; consequently,
IS2 IS3 IS,pnp
Voff ≈ VT ln = VT ln = VT ln
IS1 IS4 IS,npn
Here, IS,npn and IS,pnp are the reverse saturation currents of npn and pnp transistors of
the same size; for example, size 1. The value for the transistors in Fig. 4.5 on page 274 is
IS,npn = 2 IS,pnp ; consequently, Voff = VT · ln 0.5 ≈ − 18 mV.
The offset voltage becomes zero when the reverse saturation currents of transistors T1
and T2 reach the same level. For the transistors in Fig. 4.5, this means that T2 must be twice
as large as T1 and – in order to maintain the same size ratios – must be twice the size of T4 .
In practice, this causes a significant reduction in the offset voltage; typical values are in the
range of a few millivolts. The reason for the remaining offset voltage is the asymmetrical
current distribution caused by the differing current gains of npn and pnp transistors. This
could also be eliminated by:
– Slightly adapting the size of T1 or T2 .
– Slightly increasing T8 until the absolute values of the collector currents in T3 and T4 are
the same; then the relatively high base current of T2 , which is due to the lower current
gain of the pnp transistors, is also provided by the lower current mirror T6 ,T8 .
Despite these measures, the offset voltage of the circuit cannot be reduced as much as in
the circuit shown in Fig. 4.110a, because here the offset voltage depends on the ratio of
the reverse saturation currents of the pnp and npn transistors, which varies in practice due
to production tolerances.
The two-stage complementary common-collector circuit can be regarded as a series
connection of two single-stage complementary common-collector circuits, which means
that the output resistance can be calculated by applying (4.130) on page 392 twice.
Vb Vb Vb Vb Vb Vb
T7
T5 – 5V
2.2 V
V BS1
T1
V BS3 – 7.2 V
2.8 V – Vb
T3
5V
– Vb
0V 0V
R1
Vb
– 5V
T4
V BS4 Vb
– 2.8 V V BS2
T2 7.2 V
– 2.2 V
T6 5V
T8
– Vb – Vb – Vb – Vb – Vb – Vb
to the optimum for the two-stage common-drain circuit (see (4.128) on page 388). The
quiescent current and the size of the MOSFETs in the second stage can be calculated using
(4.131) on page 392, by giving the value desired for the output resistance.
Since the bulk–source voltages change during operation, the quiescent current of the
second stage also varies. Here too, it is necessary to ensure, by circuit simulation, that the
circuit meets the given requirements in the desired operation range. The quiescent current
is usually largest when the input voltage is approximately in the middle of the supply
voltage range, and it decreases when it approaches one of the supply voltages. In contrast,
the quiescent currents of the first stage remain constant, as they are determined by the
current mirror.
4.1.5
Circuits for Setting the Operating Point
In integrated circuits, the operating point is set in most cases by injecting the quiescent
currents by means of current sources or current mirrors. Therefore, the setting of a stable
operating point requires, first and foremost, reference current sources that are insensitive to
temperature variations and independent of the supply voltage. On the other hand, reference
voltage sources are seldom required; for example, the auxiliary voltages necessary to adjust
the operating point of cascode stages can be generated without particularly complex circuit
designs and without stringent demands on stability. The most important reference current
396 4 Amplifiers
sources are described below, followed by an examination of the circuits used for current
distribution.
Characteristics: Figure 4.114 shows the characteristic of the VBE reference current
source with R1 = 6.6 k and R2 = 36 k. For Vb > 1.4 V, the current is approximately
constant; only in this region does the circuit act as a current source.
When calculating the characteristic, the dependence of the base–emitter voltage VBE 2
on the current IC2 ≈ I2 must be taken into account:
⎛ ⎞
VBE 2
⎝ ⎠ I2
I2 ≈ IC2 = IS2 e VT −1 ⇒ VBE 2 ≈ VT ln +1
IS2
Here, IS2 is the reverse saturation current of T2 and VT is the temperature voltage; at room
temperature, VT ≈ 26 mV. The reference current is:
I2 IS2
VT I2 VT I2
Iref ≈ ln +1 ≈ ln (4.133)
R1 IS2 R1 IS2
For
Vb − VBE 1 − VBE 2 Vb − 1.4 V
I2 = ≈
R2 R2
Vb Vb
R2
I2 I ref = I C1
I B1
T1
I C2 VBE1
I B2
T2
R1
VBE2
Fig. 4.113. Basic circuit of a VBE reference current source
4.1 Circuits 397
I ref
µA
100
80 Eq. (4.135)
Eq. (4.134)
60
40
20
0
0.7 1.4
0 1 2 3 4 5 Vb
V
Fig. 4.114. Characteristic of a VBE reference current source with R1 = 6.6 k and R2 = 36 k
VBE reference current source with a current mirror: A significant improvement of the
circuit behavior is achieved when current feedback is introduced by means of a current
mirror; Fig. 4.115 shows the circuit with a simple current mirror. Current I2 is no longer
adjusted by a resistor, but is derived from the reference current. In standard circuits, all
transistors are of the same size; in this case, the current mirror has a current ratio kI ≈ 1,
so that I2 ≈ Iref . Inserting this into (4.133) leads to the transcendental equation:
VT Iref
Iref ≈ ln +1
R1 IS2
The solution to this equation depends solely on VT , R1 and IS2 , and no longer on the supply
voltage Vb . In practice, a very low dependence remains due to the Early voltage of the
transistors, which is ignored in this context.25 As the current I2 is now also stabilized, the
base–emitter voltage VBE 2 can be regarded as constant and the following approximation
25 When the Early effect is taken into account, the calculation shows that the Early factor 1 + V /V
A
only enters the argument of the logarithm and its effect is thus attenuated by a factor of 20 . . . 30;
this leads to an output resistance which is the same as that reached in a cascode circuit.
398 4 Amplifiers
Vb Vb Vb Vb Vb
T3
T4 T3 T4 T6
T5 T7
I2 I ref
I ref
100 µ A
T1 T1
T2 T2
UBE2 R1
R1
6.2 k Ω
can be used:
VBE 2
Iref ≈ (4.136)
R1
The practical configuration of the VBE reference current source with a current mirror
is shown in Fig. 4.115b. Transistor T5 expands the current mirror T3 ,T4 to form a three-
transistor current mirror, in which T6 is an additional output for the reference current. The
additional output must be provided with a cascode stage T7 to ensure that the independence
of the supply voltage is not impaired by the Early effect of T6 . To make certain that the
desired reference current is obtained at the output, R1 must be somewhat smaller than
calculated in (4.136), to compensate the current losses caused by the various base currents.
Figure 4.116 shows the resulting characteristics for R1 = 6.2 k at room temperature (T =
27 ◦ C) and at the limits of the temperature range for general applications (T = 0 . . . 70 ◦ C).
Temperature sensitivity: One disadvantage of the VBE reference current source is the
relatively high temperature sensitivity, caused by the temperature dependence of the base–
emitter voltage. From (2.21) on page 54, it follows that dVBE /dT ≈ − 1.7 mV/K; this
causes a current change of
dIref 1 dVBE 2 1.7 mV/K
= ≈ − (4.137)
dT R1 dT R1
and a temperature coefficient of:
VBE 2 ≈0.7 V
1 dIref 1 dVBE 2
= ≈ − 2.5 · 10−3 K −1
Iref dT VBE 2 dT
This means that a temperature increase of 4 K causes a reference current reduction of 1%.
4.1 Circuits 399
I ref
µA
107
0
100 27 T
°C
88
70
80
60
40
20
0
0 1 2 3 4 5 Vb
V
Fig. 4.116. Characteristics of the VBE reference current source in Fig. 4.115
Starting circuit: In addition to the desired operating point, the VBE reference current
source has another operating point at which all transistors are at zero current. Whether
this second operating point is stable or instable depends on the leakage currents of the
transistors; these depend on the production process used and are disregarded in most
simulation models. If lateral pnp transistors are used to make up current mirror T3 . . . T5 ,
the relatively high leakage current of T4 , which is caused by its large surface, is high
enough to provide a sufficiently high starting current for T1 . In this case, there is no stable
operating point with zero current. Otherwise, a separate starting circuit is required, to
provide the starting current that is then cut off when the desired operating point has almost
been reached.
Figure 4.117 shows a simple and commonly used starting circuit [4.1, 4.2]. It consists
of diodes D1 . . . D4 designed as transistor diodes and the resistances R3 . Diodes D1 . . . D3
Vb Vb Vb Vb
R3 T3
100 kΩ T4 T6
D4
V1 V2 T5 T7
R2
6.2 kΩ I ref
D1 100 µA
T1
D2
T2
D3
R1
6.2 kΩ Fig. 4.117. VBE reference current
source with a starting circuit
400 4 Amplifiers
and resistance R3 form a simple reference voltage source of V1 = 3 VBE ≈ 2.1 V, which
provides T1 with a starting current by means of D4 and resistance R2 . Resistance R2 is
rated such that the starting collector current of T4 causes voltage V2 to increase until D4
enters the blocking state at the desired operating point. When selecting
VBE
R2 ≈ ≈ R1
Iref
the voltage at the desired operating point is V1 = V2 ; D4 is thus in the blocking state.
Resistance R3 must be selected small enough to make the starting current sufficiently high
even with the supply voltage at a minimum; on the other hand, it must not be too small to
prevent the cross-current through diodes D1 . . . D3 from becoming too high at a maximum
supply voltage.
Example: The VBE reference current source in Fig. 4.117 is to be rated for a ref-
erence current of Iref = 100 mA. For the npn transistors in Fig. 4.5 on page 274,
VBE ≈ VT ln Iref /IS ≈ 0.66 V; (4.136) thus leads to R1 ≈ 6.6 k. Circuit simu-
lation is used for fine-tuning to R1 = 6.2 k. For the starting circuit, this leads to
R2 = R1 = 6.2 k. The value for resistance R3 can be in a wide range; here, it is
selected such that at a maximum supply voltage of Vb = 12 V the current in the starting
circuit is lower than the reference current: R3 ≈ (Vb − 3VBE )/Iref ≈ 100 k.
VT IS1
Iref ≈ ln for IS1 > IS2 und kI ≈ 1 (4.138)
R1 IS2
Since Iref must be positive, the limitation IS1 > IS2 is necessary in (4.138); this means that
T1 must be larger than T2 . Usually, Iref and IS1 /IS2 ≈ 4 . . . 10 are specified to calculate
R1 .
The PTAT reference current source also has a second operating point at zero current,
which has to be eliminated by a starting circuit. Figure 4.118b shows one possible circuit,
which has already been used and described in connection with the VBE reference current
source. However, here, resistance R2 must be higher than in the VBE reference current
4.1 Circuits 401
Vb Vb Vb Vb Vb
R3
T4 T3 T4 T3
D4
V1 V2
I2 I ref = I C1 R2 I ref
D1
I B1
T2 T1 D2 T2 T1
VBE1 D3
VBE2
R1 R1
source, in order to render voltage V2 sufficiently high at the desired operating point; the
guidance value here is Iref R2 ≈ 2 VBE ≈ 1.4 V.
To ensure that the PTAT reference current source provides a current independent of
the supply voltage, it is necessary to add cascode stages to eliminate the Early effect of
transistors T1 and T4 and an output. Figure 4.119 shows a practical solution, which has the
following additional features compared to Fig. 4.118b:
– T5 is added to current mirror T3 ,T4 to form a three-transistor current mirror and the
cascode stage T6 is added at the output.
Vb Vb Vb Vb
R3 T4 T3 T8
100 kΩ
1 1 1
T6 T5 T9
1 1 1
D4
V2
T7 I ref
V1
1 100 µA
R2
14 kΩ
D1
D2 T2 T1
1 4
D3
R1
320 Ω Fig. 4.119. Practical configuration
of a PTAT reference current source
402 4 Amplifiers
I ref
µA 70
T
100 27 °C
0
80
60
40
20
0
0 1 2 3 4 5 Vb
V
Fig. 4.120. Characteristics of the PTAT reference current source in Fig. 4.119
– Transistor T1 is provided with cascode stage T7 , which uses voltage V2 of the starting
circuit as the base bias voltage.
– Transistor T8 and the corresponding cascode stage T9 are used to provide the reference
current output node.
Figure 4.120 shows the resulting characteristics for various temperatures.
Controlled PTAT reference current source: Figure 4.121 shows the principle of a
controlled PTAT reference current source; here, the PTAT current according to (4.138) is
adjusted not with a current mirror, but with two control amplifiers A1 and A2.
If both control amplifiers have high-resistance inputs and a gain A, then:
V1 = A (V0 − V4 )
V2 = V3 − IC2 R2
V3 = A (V0 − V2 )
V4 = V3 − IC1 R2
V3
R2 R2
V4 V2
–
I C1 I C2 A2
+
– V1 T1
A1 T2
+
V0 R1 V0
Fig. 4.121. Controlled PTAT
reference current source
4.1 Circuits 403
v4 v3 v4
– v3 –
A 1
A1 –A gm1R2 – A A2 A1 – A gm1R2 gm2R2
v1 vx v2 v1
gm2R2
–
Fig. 4.122. Equivalent control circuit of the controlled PTAT reference current source
If the gain A is sufficiently high and the circuit is stable, then the parameters at the operating
point are V2 = V4 = V0 and IC1 = IC2 = Iref ; the latter only applies to the PTAT current
according to (4.138), because of the common-base voltage V1 . The stability is checked
using the small-signal equivalent circuit; this leads to
v1 = − Av4
v2 = v3 − iC2 R2 = v3 − gm2 R2 v1
v3 = − Av2
v4 = v3 − iC1 R2 = v3 − gm1 R2 v1
with the transconductances:
Iref Iref
gm1 = , gm2 = > gm1
VT + Iref R1 VT
Figure 4.122 shows the equivalent control circuit for the static situation; due to
A1
v3 A
= ≈ 1
vx 1+A
the circuit with control amplifier A2 can be replaced by a direct connection. This means
that transistor T1 with gm1 R2 provides positive feedback and transistor T2 with gm2 R2
negative feedback for control amplifier A1; since gm2 > gm1 , the circuit is statically
stable. Dynamic stability must be ensured by way of frequency response compensation for
the two control amplifiers; this will be outlined in more detail below.
Figure 4.123 shows a practical configuration for the controlled PTAT reference current
source. One common-emitter circuit (T3 ,T5 ) with a subsequent common-collector circuit
(T4 ,T6 ) is used for each of the control amplifiers; an additional nonlinear level converter
(R6 ,T7 ), which linearizes the circuit with regard to large signals, is contained in control
amplifier A1. The voltages V0 correspond to the base–emitter voltages of transistors T3
and T5 at the operating point: V2 ≈ V4 ≈ V0 ≈ 0.7 V. This means that transistors T1 and
T2 are operated with a constant collector voltage; the Early effect thus has no influence on
the reference current. Decoupling is done by connecting additional transistors to voltage
V1 ; this is indicated on the far left of Fig. 4.123. Here too, the decoupling transistors may
need a cascode stage to eliminate their Early effect. Figure 4.124 shows the characteristics
for various temperatures.
Frequency response compensation is required for both control amplifiers to ensure that
the circuit is dynamically stable; this is done by capacitances C1 and C2 . The component
dimensions are determined by circuit simulation. For this purpose, a transient analysis
is performed in which a current source injects a short current pulse into node V1 ; this
404 4 Amplifiers
Vb Vb Vb > 2 V Vb Vb
I ref = 100 µA
R5 2 I ref R4
20 kΩ 20 kΩ
T6 T4
1 2
U3
C2 C1
R6 4 pF R2 R3 1 pF
1 kΩ 1 kΩ 1 kΩ
V4 V2
I ref T5 T3
I ref 1 I ref I ref 1
T7 V1 T1 T2
1 4 1
R1
370 Ω
A1 A2
allows us to evaluate the pulse response at the different nodes and select the capacitances
accordingly. Without compensation the circuit is usually unstable; in such cases, the circuit
simulation produces undamped oscillation.
Temperature sensitivity: Since the current of the PTAT reference current source is
proportional to the temperature voltage VT , the influence of the temperature sensitivity is:
kT dVT k
VT = ⇒ = ≈ 86 µV/K
q dT q
I ref
A
70
T
100 27 °C
0
80
60
40
20
0
0 1 2 3 4 5 Vb
V
Fig. 4.124. Characteristics of the controlled PTAT reference current source in Fig. 4.123
4.1 Circuits 405
Use in bipolar amplifiers: Despite its high temperature sensitivity, the PTAT reference
current source is used as the reference source for quiescent currents in bipolar amplifiers.
In such cases the temperature sensitivity may even be an advantage, as the gain of bipolar
amplifier stages without current feedback is proportional to the transconductance gm =
IC,A /VT of the transistors; with IC,A ∼ Iref ∼ VT the transconductance, and thus the
gain, remains constant.
dIref
dIref
+ = 0
dT
dT
Vb Vb Vb Vb Vb Vb Vb
T3
T4 T11 T16 T15
T8
T10
T6 T12
T9
R3
T5
D4 I ref
T7
R2
D1
D2 T14
T2 T1
T13
D3
R1 R4
The transistor diode T10 has been added to allow the bases of the pnp cascode transistors
to be connected to the emitter of T5 ; this reduces the errors caused by the base currents.
The VBE reference current source T13 ,T14 is connected to the original output T8 ,T9 ; in this
case, it is already supplied with a stabilized current and feedback via a current mirror is
not required. With T11 ,T12 , the PTAT reference current source has an additional output
at which the current of the VBE reference current source is added via the current mirror
T15 ,T16 . Equations (4.136)–(4.139) determine the ratio of the currents
dVT
Iref ,VBE dVBE Iref ,PTAT dVT Iref ,VBE VBE dT
+ = 0 ⇒ = − ≈ 1.3
VBE dT VT dT Iref ,PTAT VT dVBE
dT
and the reference current:
Iref = Iref ,VBE + Iref ,PTAT ≈ 2.3 · Iref ,PTAT ≈ 1.77 · Iref ,VBE
A reference current of Iref = 100 mA results in Iref ,PTAT ≈ Iref /2.3 ≈ 43 mA and
Iref ,VBE ≈ Iref /1.77 ≈ 57 mA.
Vb Vb Vb Vb
of a depletion FET in the pinch-off region as the reference current; We can thus operate
with VGS = 0, or in the case of current feedback with a resistance with VGS < 0.
The pinch-off current sources with junction FETs in Fig. 4.126b are realized in inte-
grated circuits by means of a pinch resistor. This is an integrated high-resistance resistor,
which is pinched off as the voltage increases. Since its basic construction corresponds to
that of a junction FET, its characteristic behavior is practically identical. The disadvantages
are the high production tolerances, which are typically in the range of ±30% [4.1].
Bipolar circuits: Figure 4.127 shows a typical circuit for setting the operating point in
bipolar amplifier circuits. It comprises of a PTAT reference current source (T1 . . . T8 ) with
a starting circuit (D1 . . . D5 ) as well as an npn (T9 ) and a pnp collector circuit (T11 ) with
the relevant current sources (T10 ,T12 ) to provide the auxiliary voltages V1 and V2 for the
cascode stages; the transistor diode D6 represents a simple method of producing additional
auxiliary voltages. As the PTAT reference current source offers not only decoupling at the
current mirror T3 . . . T6 but also decoupling at the Widlar current mirror T1 ,T2 , the Widlar
current mirror is, in contrast to Fig. 4.119, expanded by T8 to form a three-transistor current
mirror to reduce the error caused by the base currents; this, however, makes an additional
transistor diode necessary in the starting circuit, to raise the starting voltage accordingly.
Resistance R3 represents a p-channel pinch resistance. This is nothing unusual, as in most
cases resistances rated around 100 k can only be produced in this form. The fact that
with large voltages pinch resistances act as constant current sources (see the section on
pinch-off current sources) is of advantage here, since this limits the current in the starting
circuit; likewise, the production tolerances have no negative effect, since the current in the
starting circuit may vary by almost one order of magnitude without restricting the function.
Simple current sources or current sources with a cascode of any current ratio can be
connected to the decoupled signals and the auxiliary voltages; Fig. 4.127 exemplifies this
with each current source with cascode. Other auxiliary voltages, such as voltage V3 , can
408 4 Amplifiers
Vb Vb Vb Vb Vb Vb Vb
T4 T3 T9 T12 Vb – 0.7 V
1 1 1 1
( R3 )
T6 T5 V1 Vb – 1.4 V
1 1
D6 I1
D5 I ref V3 Vb – 2.1 V
100 µA
R2 Vb
14 kΩ I2
D1
T8 T7 V2 1.4 V
D2 1 1
Fig. 4.127. A typical circuit with a PTAT reference current source to set the operating point in
bipolar amplifier circuits (numeric example for Iref = 100 mA at Vb > 3.5 V, using the bipolar
transistors in Fig. 4.5 on page 274)
MOS circuits: Figure 4.128 shows a typical circuit for setting the operating point in
MOS amplifier circuits. It comprises a VGS reference current source (T1 ,T2 ) with current
mirror (T3 ,T4 ) and starting circuit (T5 ,T6 ), as well as a decoupling with an auxiliary voltage
generator (T8 . . . T12 ). The starting circuits supplies the starting current via T5 , which is
turned off by T6 after the reference current becomes active. The depletion MOSFET T7
serves as the quiescent current source (the pinch-off current source) for T6 ; its current must
be smaller than the reference current to be able to turn the starting circuit off via T6 . The
size of T7 depends on the threshold voltage of the depletion MOSFET, which is determined
by the given manufacturing process.
A circuit with this configuration only makes sense if the production tolerances of
resistance R1 and the threshold voltage of T2 are lower than the tolerance of the threshold
voltage of T7 ; otherwise, it is better to use the current of the pinch-off current source T7
as the reference current.
4.2
Properties and Parameters
The properties of an amplifier are described by the characteristic parameters. These are
based on the characteristic curves of the amplifier. The small-signal parameters (e.g. the
4.2 Properties and Parameters 409
Vb Vb Vb Vb Vb Vb
Vb 1.8
V1 Vb
V2
2 1.8
Fig. 4.128. A typical circuit using a VGS reference current source to set the operating point in
MOS amplifier circuits (numeric example for Iref = 10 mA at Vb > 7 V, using the MOSFETs in
Fig. 4.6 on page 274)
gain) are determined by linearization at the operating point, and the nonlinear parameters
(e.g. the distortion factor) are determined by series expansion. As direct calculation of the
characteristics is often not possible, it may be necessary to rely on measurements or circuit
simulations.
4.2.1
Characteristics
An amplifier with one input and one output is usually described by two families of char-
acteristics; using the parameters of Fig. 4.129, the currents are:
Ii = fI (Vi , Vo )
Io = fO (Vi , Vo )
Vb
Rg Ii Io
Vg Vi Vo
RL
– Vb
Fig. 4.129. Voltages and currents in an amplifier with one input and one output
410 4 Amplifiers
In most amplifiers, the effect that the output has on the input is negligibly small in the
interesting region; that is, the input characteristic is almost independent of the output
voltage. This leads to:
Ii = fI (Vi ) (4.140)
Io = fO (Vi , Vo ) (4.141)
With an open output, this leads to the open-circuit transfer characteristic:
Io = fO (Vi , Vo ) = 0 ⇒ Vo = fT r (Vi ) (4.142)
This is often simply referred to as the transfer characteristic.
If the amplifier is operated with a signal source with an internal resistance Rg and a
load RL , then according to Fig. 4.129:
Vg − Vi Vo
Ii = , Io = − (4.143)
Rg RL
The straight lines described by these equations are known as the source and load charac-
teristics. Insertion into (4.140) and (4.141) leads to the nonlinear set of equations
Vg = Vi + Rg fI (Vi )
(4.144)
0 = Vo + RL fO (Vi , Vo )
and thus the operational transfer characteristic:
Vo = fT r,B (Vg ) (4.145)
It is only possible in exceptional cases to solve (4.142) and the set given in (4.144), as
well as to determine the operational transfer characteristic. In practice, circuit simulation
programs are used, which solve the equations in the course of a DC analysis for several
points and plot the characteristics graphically. If the characteristics of an amplifier are
available in a graphic plot, the set given in (4.144) can also be solved graphically by drawing
the straight lines according to (4.143) into the family of input or output characteristics and
determining the intersections.
Example: Applying the transport model in Fig. 2.26 on page 61 to the common-emitter
circuit shown in Fig. 4.130 leads to:
Vb Vb
RC RC
Io Io
I B,I
Rg Ii Ii BN I B,N
Vo RL Vo
– BI I B,I
Vg Vi Vi
I B,N
Vo − Vb
Io = fO (Vi , Vo ) = + BN IB,N − (1 + BI ) IB,I
RC
⎛ ⎞ ⎛ ⎞
Vi Vi −Vo
Vo − V b 1 + B
+ IS ⎝e VT − 1⎠ − IS ⎝e VT − 1⎠
I
=
RC BI
For practical applications, only the region of normal transistor operation, Vo > VCE,sat ≈
0.2 V, is of interest; in this region, the output voltage has no effect on the input characteristic.
If the reverse currents are neglected, it follows that:
V
IS V i
Ii = fI (Vi ) = e T
BN
V
Vo − V b i
Io = fO (Vi , Vo ) = + IS e VT
RC
The characteristics are shown in Fig. 4.131. Here, the open-circuit transfer characteristic
can be calculated straight forward:
Vi
fO (Vi , Vo ) = 0 ⇒ Vo = fT r (Vi ) = Vb − IS RC e VT
The straight-line source characteristic in Fig. 4.131a and the straight-line load characteristic
in Fig. 4.131b are determined for Vg = 1 V, Rg = 100 k and RL = 10 k. The
intersections provide Vi (Vg = 1 V) ≈ 0.69 V and Vo (Vi = 0.69 V) ≈ 1 V. Thus, one
Ii Io Vi
Straight-line 0.7 0.69 0.68 0.67
source V 0.65
characteristic 1 0
10 Vg 1V
5 Vo
Rg 100
0.1 V
5
Straight-line
3.2 load
characteristic
RL 10
1
0.69
0.5
0.1 0.5 1.0 V
Fig. 4.131. Characteristics of the common-emitter circuit in Fig. 4.130, with Vb = 5 V and
RC = 10 k
412 4 Amplifiers
Vo
V
Fig. 4.132. Operational transfer characteristic of the common-emitter circuit in Fig. 4.130, with
Vb = 5 V, RC = 10 k, Rg = 100 k and RL = 10 k
4.2.2
Small-Signal Characteristics
The small-signal characteristics describe the quasi-linear behavior of an amplifier when
driven with low amplitudes around one operating point; this mode of operation is called
small-signal operation.
Operating Point
The operating point A is characterized by the voltages Vi,A and Vo,A and by the currents
Ii,A and Io,A :
Ii,A = fI (Vi,A ) , Io,A = fO (Vi,A , Vo,A )
In general, the operating point depends on the signal source and the load. One exception
are amplifiers with AC coupling via coupling capacitances or transformers, which allow
the operating point to be set independent of the signal source and the load. However, for
calculating the small-signal parameters, it is of no importance how the operating point is
adjusted.
Small-Signal Values
In the analysis of the small-signal characteristics, only the deviations from the operating
point are considered. These are described by the small-signal values:
vi = Vi − Vi,A , ii = Ii − Ii,A
vo = Vo − Vo,A , io = Io − Io,A
As the operating point values Vi,A , Ii,A , Vo,A and Io,A normally correspond to the DC
components of Vi , Ii , Vo and Io , the small-signal values have no DC component; that is,
they have zero mean value.
4.2 Properties and Parameters 413
Example:
Vi,A = V0
Vi = V0 + v1 cos ω1 t + v2 cos ω2 t ⇒
vi = v1 cos ω1 t + v2 cos ω2 t
Linearization
Insertion of the small-signal values into the characteristics given in (4.140) and (4.141)
and series expansion at the operating point lead to26 :
Ii = Ii,A + ii = fI (Vi,A + vi )
∂fI
1 ∂ 2 fI
2 1 ∂ 3 fI
3
= fI (Vi,A ) + vi +
v +
v + ···
∂Vi
A 2 ∂Vi2
i 6 ∂Vi3
i
A A
Io = Io,A + io = fO (Vi,A + vi , Vo,A + vo )
∂fO
∂fO
= fO (Vi,A , Vo,A ) + vi + vo
∂Vi
A ∂Vo
A
1 ∂ 2 fO
2 1 ∂ 2 fO
1 ∂ 2 fO
2
+
v +
vi vo +
v + ···
2 ∂Vi2
i 2 ∂Vi ∂Vo
2 ∂Vo2
o
A A A
With sufficiently small input signals, the series expansion can be aborted after the linear
term; this results in linear relations between the small-signal values:
∂fI
ii = vi
∂Vi
A
∂fO
∂fO
io = v + vo
∂Vi
A ∂Vo
A
i
The transition to these linear equations is called linearization at the operating point.
Small-Signal Parameters
The partial derivations resulting from the linearization, which are evaluated at operating
point A, are called the small-signal parameters.
– The small-signal input resistance ri :
vi ∂fI
−1
ri = = (4.146)
ii ∂Vi
A
– The small-signal output resistance ro :
vo
∂fO
−1
ro = = (4.147)
io
vi =0 ∂Vo
A
This is also called the short-circuit output resistance, since the input is short-circuited
with regard to small signals (vi = 0). In practice, this means that a voltage source with
sufficiently low internal resistance, which keeps the input voltage constant at Vi,A , is
connected to the input.
26 A partial differentiation of the input characteristic f is also used in what follows; this indicates
I
that fI is generally dependent on a second variable (Vo ).
414 4 Amplifiers
∂fO
∂fO
−1
A = = − (4.148)
vi
io =0 ∂Vi
A ∂Vo
A
This is also called the open-circuit gain, as there is no load connected to the output; that
is, it is open with regard to small signals (io = 0). The gain can also be calculated from
the open-circuit transfer characteristic given in (4.142):
dfT r
A =
dVi
A
– The transconductance gm :
io
∂fO
gm = = (4.149)
vi
vo =0 ∂Vi
A
This is of minor importance in amplifiers that have a low-resistance output (small ro )
and thus provide an output voltage, but it plays an important role in transistors and
amplifiers with high-resistance outputs (high ro ). From a comparison of (4.147) and
(4.148), it follows:
A
gm = − of A = − gm ro (4.150)
ro
This means that one of the parameters A, ro and gm is redundant.
dfT r,B
AB =
dV
g A
4.2 Properties and Parameters 415
Vb
I i = I i, A + i i I o = I o, A + i o
V i = V i, A + v i V o = V o, A + v o
– Vb
ii ro io ii io
vi ri Av i vo vi ri gm v i ro vo
Rg ii ro io
vi vo
vg ri Av i RL
Fig. 4.134. Small-signal equivalent circuit for an amplifier with a signal source and a load
Example: For the common-emitter circuit in Fig. 4.130a on page 410, the following
characteristics were determined:
V V
IS V i Vo − V b i
Ii = fI (Vi ) = e T , Io = fO (Vi , Vo ) = + IS e VT
BN RC
For Vg = 1 V, Rg = 100 k and RL = RC = 10 k, we obtained Vi ≈ 0.69 V and
Vo ≈ 1 V. This point is now used as the operating point; for IS = 1 fA, BN = 100 and
VT = 26 mV, it follows:
Vi,A ≈ 0.69 V , Ii,A = fI (Vi,A ) ≈ 3 mA
Vo,A
Vo,A ≈ 1 V , Io,A = − ≈ − 100 mA
RL
The upper portion of Fig. 4.135 shows the circuit with the operating point values.
From (4.146) for
Vi
∂fI
IS
Ii
Ii,A 3 mA
= e T
=
V
= ≈ ≈ 0.115 mS
∂Vi A VT B N
VT A VT 26 mV
A
we obtain an input resistance of ri ≈ 8.7 k; similarly, for
416 4 Amplifiers
Vb = 5 V
RC
10 k Ω I o,A ≈ – 100 µA
Rg
100 k Ω I i,A ≈ 3 µA RL
Vo,A ≈ 1 V 10 k Ω
Rg ro
100 kΩ 10 k Ω
vg vi ri vo RL
8.7 kΩ
–115 v i
10 k Ω
Fig. 4.135. Example: the common-emitter circuit with an operating point (above) and the resulting
small-signal equivalent circuit (below)
∂fO
1
= = 0.1 mS
∂Vo A RC
it follows from (4.147) that the output resistance is ro = RC = 10 k, and from (4.149)
for
Vi
∂fO
IS V
300 mA
= e T
≈ ≈ 11.5 mS
∂Vi A VT
26 mV
A
we obtain the transconductance of gm ≈ 11.5 mS. The gain A can be determined from
gm and ro using (4.150): A = − gm ro ≈ − 115. The lower portion of Fig. 4.135 shows
the resulting small-signal equivalent circuit. From (4.153), the operational gain is thus
AB ≈ − 4.6; this corresponds to the slope of the operational transfer characteristic in
Fig. 4.132 on page 412 at the operating point shown.
∂fI
∂fI
−1
Ar = = − (4.155)
vo
ii =0 ∂Vo
A ∂Vi
A
27 Here, only the static reactive effect will be looked at; but due to parasitic capacitances many
amplifiers have dynamic reactive effects that become evident at higher frequencies.
4.2 Properties and Parameters 417
ii ri ro io ii gm,r vo gm vi io
vi vo vi vo
A r vo A vi ri ro
Fig. 4.136. Small-signal equivalent circuits for an amplifier with reverse gain
ii
∂fI
gm,r = = (4.156)
v
∂V
o vi =0 o A
From a comparison of (4.146) and (4.155), it follows:
Ar
gm,r = − or Ar = − gm,r ri (4.157)
ri
This means that one of the parameters Ar , ri and gm,r is redundant.
Figure 4.136 shows the small-signal equivalent circuits for amplifiers with reverse gain;
the parameters are:
vi
vi = Ar vo + ii ri or ii = gm,r vo + (4.158)
ri
vo
vo = A vi + io ro or io = gm vi + (4.159)
ro
In this case, input resistance ri is also known as the short-circuit input resistance, as it is
determined with the output short-circuited for small signals (vo = 0). Furthermore, gain
A is also called the forward gain and transconductance S the forward transconductance if
the difference with regard to the related reverse parameter is to be emphasized.
In addition to the two small-signal equivalent circuits shown in Fig. 4.136, there are a
further two circuits, since the input or output can be shown in terms of the given gain or the
given transconductance. The two mixed forms are, however, very seldom used. These four
possible presentations must not be confused with the four four-pole representations using
the Y, Z, H or P matrices, since here the controlled sources are always voltage-controlled;
therefore, the four small-signal equivalent circuits are variations of the Y representation
with:
1 1
y 11 = , y 12 = gm,r , y 21 = gm , y 22 =
ri ro
The small-signal equivalent circuit in Fig. 4.136b corresponds to the common Y repre-
sentation. The three other small-signal equivalent circuits are obtained by converting the
current source into an equivalent voltage source in either the input circuit or the output
circuit, or both; this causes the transconductances gm and gm,r to change into the gains A
and Ar , respectively.
For operation with a signal source with internal resistance Rg and a load RL , the
operational gain AB can be calculated directly with the help of the small-signal equivalent
circuit in Fig. 4.137; the result is a complex expression that allows no insight into the
relationships. The procedure is therefore broken down into three steps:
418 4 Amplifiers
Rg ii ri ro io
Fig. 4.137. Small-signal
vg vi
Ar vo A vi
vo
RL
equivalent circuit for
calculating the operational
gain of an amplifier with
reverse gain
– First, the operational gain is calculated for operation with an ideal signal voltage source;
that is, Rg = 0:
vo
vo RL
AB,0 =
= = A (4.160)
vg Rg =0 vi ro + R L
This comprises the open-circuit gain A and the voltage divider factor at the output and
is independent of the reverse gain Ar . Subscript 0 of AB,0 indicates that Rg = 0.
– Then, the operational input resistance ri,B is calculated:
vi ri ri
ri,B = = = (4.161)
ii RL 1 − Ar AB,0
1 − Ar A
ro + R L
In amplifiers with reverse gain (Ar = 0), this value depends on the load RL ; in amplifiers
without reverse gain (Ar = 0), we obtain ri,B = ri .
– With the help of the operational input resistance, the voltage divider factor at the input,
and thus the operational gain, can be calculated:
ri,B ri,B RL
AB = AB,0 = A (4.162)
Rg + ri,B Rg + ri,B ro + R L
Inserting ri,B = ri leads to the operational gain for amplifiers without reverse gain
according to (4.153).
Therefore, it is possible to treat an amplifier with reverse gain like an amplifier without
reverse gain provided that the operational input resistance ri,B is used instead of the input
resistance ri . This is the reason why Sects. 2.4 and 3.4, on the calculation of basic transistor
circuits, show how the input resistance can be calculated for a given load RL provided
that such a dependence – that is, reverse gain – exists; this replaces the calculation of
the reverse gain Ar or the reverse transconductance gm,r . The operational gain of basic
transistor circuits can thus be calculated using (4.160)–(4.162) by replacing ro with the
short-circuit output resistance ro,s and ri,B with the input resistance ri for operation with
a load RL :
ro = ro,s , ri,B = ri (RL )
While the gain A presents no interpretation problems, specifications for the input and
output resistances generally require the operating conditions to be taken into consideration;
these relationships are listed in Fig. 4.138.
Fig. 4.138. Input and output resistances for the amplifier in Fig. 4.137, for different operating
conditions. Note that ri and ro are short-circuit resistances by definition
Example: Fig. 4.140 again shows the common-emitter circuit from Fig. 4.130a; the only
nonlinear component is the transistor. The small-signal equivalent circuit of the circuit
is derived by inserting the small-signal equivalent circuit of the transistor. The transistor
parameters β and VA and the collector current IC,A at the operating point are required to
calculate the parameters gm , rBE and rCE ; for β = 100, VA = 100 V and IC,A = 300 mA,
we obtain:
IC,A 300 mA β 100
gm = = ≈ 11.5 mS , rBE = = ≈ 8.7 k
VT 26 mV gm 11.5 mS
VA 100 V
rCE = = ≈ 333 k
IC,A 300 mA
From a comparison with Fig. 4.133b on page 415, we obtain ri = rBE ≈ 8.7 k,
ro = rCE || RC ≈ 9.7 k, gm ≈ 11.5 mS (here the transconductance of the amplifier
corresponds with the transconductance of the transistor) and A = − gm ro ≈ − 112.
420 4 Amplifiers
Components
Circuit
Linearisation of the
characteristics at the
operating point: rD
∂ fI ∂ fI
ii = vo + vo
∂Vi A ∂ Vo A
gmvBE
∂f ∂f rBE rCE
io = O vi + O vo
∂Vi A ∂Vo A
Small-signal parameters: ro , ri , A , gm
Vb
RC
I o,A
I C,A
Rg I i,A = I B,A Vo,A RL
Vg,A Vi,A
Rg
vi vo
vg rBE vBE gmvBE rCE RC RL
Fig. 4.140. Example: common-emitter circuit with an operating point (top) and the resulting
small-signal equivalent circuit when using the small-signal equivalent circuit of the transistor
(bottom)
The values for A and ro deviate slightly from the values in Fig. 4.135, as the small-
signal equivalent circuit of the transistor also takes into account the Early effect, which is
represented by resistance rCE and which was ignored when calculating the values from
the characteristics.
Series connection of amplifiers without reverse gain: Several amplifiers without re-
verse gain connected in series can be combined into one single amplifier. The following
applies to n amplifiers:
– The input resistance corresponds to the input resistance of the first amplifier: ri = ri1 .
– The output resistance corresponds to the output resistance of the last amplifier: ro =
ro(n) .
– The gain corresponds to the product of the individual gains and the voltage divider
factors between each pair of subsequent amplifiers:
422 4 Amplifiers
Amplifier 1 Amplifier 2
Rg i i1 ro1 i i2 = – i o1 ro2 io2
Fig. 4.141. Example: series connection of two amplifiers without reverse gain
/
n /
n−1
ri(i+1)
A = A(i) · (4.163)
ro(i) + ri(i+1)
i=1 i=1
Here, the voltage divider factors at the input (i = 0) and output (i = n) must be added.
Example: The small-signal parameters of the two amplifiers without reverse gain con-
nected in series in Fig. 4.141 are
ri2
ri = ri1 , ro = ro2 , A = A1 A2
ro1 + ri2
and the operating gain is:
ri RL ri1 ri2 RL
AB = A = A1 A2
Rg + r i ro + R L Rg + ri1 ro1 + ri2 ro2 + RL
Series connection of amplifiers with reverse gain: It is a very complex task to determine
the small-signal parameters of amplifiers with reverse gain connected in series. In contrast,
the calculation of the operating gain AB is simple: if the operating input resistances ri,B(i)
are used instead of the input resistances ri(i) , it is possible to proceed as with amplifiers
without reverse gain connected in series and use (4.164). The operating input resistances
can be determined backwards: The operating input resistance of the last amplifier depends
on the load RL ; this in turn forms the load of the second-last amplifier and so on. For n
amplifiers, the following applies:
In general, this backwards calculation can only be performed with numeric values, as
inserting several formulas into one another soon leads to extremely comprehensive ex-
pressions. In relation to this, it should be noted that the dependence on RL means that the
gain A cannot be calculated using (4.163); the operating input resistances are not defined
in this case.
4.2 Properties and Parameters 423
v1 v2 ri2 v3 v4
A r1 v2 A1 v1 A2 v2 A r3 v4 A3 v3
v1 ~ v4
ri,B1 A1 v1 v2 ri2 A2 v2 v3 ri3 A3 v3
Fig. 4.142. Conversion of amplifiers with reverse gain connected in series into one amplifier
without reverse gain
Series connection of at least one amplifier without reverse gain: As mentioned above,
a chain of amplifiers in series has no reverse gain if at least one amplifier in the series has
no reverse gain. In such cases, it is possible to determine the small-signal parameters A, ri
and ro by successively converting the amplifiers with reverse gain into amplifiers without
reverse gain; Fig. 4.142 shows an example of this. The procedure is based on the fact that
an amplifier with reverse gain arranged in front of or behind an amplifier without reverse
gain can be converted into an amplifier without reverse gain; the successive application of
this rule effectively removes the reverse gain from all amplifiers.
First, we will look at amplifier 1 in Fig. 4.142. It is arranged in front of the amplifier 2
without reverse gain and is therefore operated with a defined load; that is, ri2 . This allows
the operating input resistance ri,B1 = ri,B1 (ri2 ) to be calculated and the conversion to be
performed.
Amplifier 3 in Fig. 4.142 is arranged behind the amplifier 2 without reverse gain and is
therefore operated with a defined internal resistance of the signal source; that is, ro2 . This
allows the operating output resistance ro,B3 = ro,B3 (ro2 ) to be calculated. In addition, it
is necessary to change the gain of the voltage-controlled voltage source from A3 to:28
ro,B3
Ã3 = A3
ro3
Example: Figure 4.143 shows a three-stage amplifier with one common-emitter circuit
with voltage feedback at the input and one at the output (T1 and T3 ) and a common-emitter
circuit with current feedback in between (T2 ). The common-emitter circuits with voltage
feedback have a reverse gain that cannot be neglected and which is mainly caused by the
resistances R21 and R23 ; the reverse gain of common-emitter circuit with current feedback,
28 This formula results when converting the values, but can also be derived by observing the short-
circuit current at the output: with a short at the output (v4 = 0) the current A3 v3 /ro3 flows in the
original amplifier and current Ã3 v3 /ro,B3 flows in the converted amplifier; since in both cases
v3 is the same due to Ar3 v4 = 0, the equality of the currents leads to the formula for the gain
factors.
424 4 Amplifiers
Vb Vb Vb
RC1 RC2 RC3
Ck1 R21 1 kΩ 1 kΩ R23 1 kΩ
v1 ~ v4
ri,B1 A1 v1 v2 ri2 A2 v2 v3 ri3 A3 v3
ii ro io
vi vo
ri Avi
on the other hand, is practically zero. The small-signal values A, ri and ro are determined
below.
The supply voltage is Vb = 1.7 V; this causes a quiescent collector current of 1 mA
in all three transistors. If β = 100 and VA = 100 V, then gm = IC /VT = 38 mS and
rBE = βVT /IC = 2.6 k; compared to the resistances in the circuit, the collector-emitter
resistance rCE = VA /IC = 100 k can be neglected.
First, the parameters of the common-emitter circuit with current feedback are deter-
mined:
– From (2.70), we obtain the gain:
gm RC2
A2 = − = − 30
1 + gm RE2
– From (2.71), we obtain the input resistance:
ri2 = rBE + βRE2 = 3.3 k
– From (2.72), we obtain the output resistance:
ro2 = RC2 = 1 k
4.2 Properties and Parameters 425
The common-emitter circuits with voltage feedback lack the resistance R1 that exists
in the basic circuit of Fig. 2.66; therefore, it is necessary to provide the formulas for this
situation first:
– From the deduction for gain A, the formulas can be used with the conditions that rCE
RC , β 1 and R1 = 0:
− gm R2 + 1
A =
R2
1+
RC
– The short-circuit input resistance for R1 = 0 is taken from the deduction for the input
resistance:
ri = ri,s = rBE || R2
– The operating input resistance ri,B corresponds to the open-circuit input resistance in
the deduction if RC is replaced by RC and RL connected in parallel (see Fig. 2.69); if
rCE RC , β 1, βRC rBE , βRC R2 , and R1 = 0, then:
1 R2
ri,B = 1+
gm RC || RL
– From the deduction for the short-circuit output resistance, the formula can be used with
the conditions that rCE RC , β 1 and R1 = 0:
ro = RC || R2
– The same formula is used with R1 = Rg to calculate the operating output resistance,
since in this case the internal resistance Rg replaces the missing resistance R1 :
% &
rBE Rg + R2 + Rg R2
ro,B = RC ||
rBE + βRg
These formulas yield the following values for the first common-emitter circuit with voltage
feedback with R2 = R21 = 700 and RC = RC1 = 1 k:
A1 = − 15 , ri,B1 (RL = ri2 ) = 50 , ro1 = 412
For the second common-emitter circuit with voltage feedback with R2 = R23 = 740
and RC = RC3 = 1 k, the values are:
A3 = − 15.6 , ri3 = 576 , ro3 = 425
ro,B3 (Rg = ro2 ) = 49 , Ã3 = − 1.8
All elements of the small-signal equivalent circuit shown in the centre of Fig. 4.143
are now determined and the series connection can be combined:
ri2 ri3
A = A1 A2 Ã3 = − 263
ro1 + ri2 ro2 + ri3
ri = ri,B1 = 50
ro = ro,B3 = 49
This represents an amplifier that is matched to 50 on both sides. If we use a 50 signal
source and a 50 load, then the voltage dividers at the input and output have the factor 1/2;
426 4 Amplifiers
consequently, the operating gain is AB = A/4 = − 66. Circuit simulation with PSpice
yields ri = ro = 50 and AB = − 61.
It should be noted that the gain is achieved by the first two stages; while the third stage,
together with the voltage divider factor between the second and the third stage, provides
attenuation. The third stage only serves as an impedance converter from ro2 = 1 k to
ro,B3 = 50 ; it is necessary to use a common-emitter circuit with voltage feedback, as a
directly coupled npn common-collector circuit cannot be used due to the low DC output
voltage of the second stage (V3,A ≈ 0.7 V). Also, the circuit is to be produced in HF
semiconductor technology, in which no pnp transistors are available that are fast enough.
This detailed example shows that a multi-stage amplifier can be calculated exactly using
the method described here.29 However, it also shows that the parameters in the small-signal
equivalent circuit must be calculated very carefully, and that it may be necessary to revert
to the full equations of the basic transistor circuits.
4.2.3
Nonlinear Parameters
In relation to the small-signal parameters, the question arises as to how high the maximum
signals around the operating point may become before leaving the small-signal mode.
From a mathematical point of view, the small-signal equivalent circuit is only valid for
infinitesimal – that is, arbitrarily small – signals. In practice, the nonlinear distortions that
increase disproportionately with an increasing amplitude are crucial and should not exceed
an application-specific limiting value.
The nonlinear response of an amplifier is described by the distortion factor, the com-
pression point and the intercept points. These can be calculated from the coefficients of
the series expansion for the transfer characteristic. If it is not possible to calculate a closed
representation of the transfer characteristic, the transfer characteristic must be measured
or determined by means of circuit simulation.
dfT r,B
1 d 2 fT r,B
2
= fT r,B (Vg,A ) + vg +
v
dVg
A 2 dVg2
g
A
1 d 3 fT r,B
3 1 d 4 fT r,B
4
+
v +
v + ···
6 dVg3
g
24 dVg4
g
A A
This leads to the small-signal parameters:
Rg
Vg RL Vo = fTr,B (Vg )
vg = Vg – Vg,A , vo = Vo – Vo,A
Rg
vg RL vo = a1 vg + a 2 v 2g + a 3 v 3g + ...
Fig. 4.144. Nonlinear amplifier (top) and series expansion at the operating point (bottom)
dfT r,B
1 d 2 fT r,B
2 1 d 3 fT r,B
3 1 d 4 fT r,B
4
vo = vg +
v +
v +
v + ···
dVg
A 2 dVg2
g 6 dVg3
g 24 dVg4
g
A A A
0 n
1 d fT r,B
= an vgn with an =
(4.165)
n! dVgn
n=1...∞ A
The coefficients a1 , a2 , . . . are known as the coefficients of the Taylor series. Coefficient
a1 corresponds to the small-signal operating gain AB and is nondimensional; all of the
other coefficients are dimensional parameters:
1
[an ] = n−1 for n = 2 . . . ∞
V
Example: For the common-emitter circuit in Fig. 4.135 on page 416, it is comparatively
simple to calculate the series expansion of the operating transfer characteristic; we start
with the series expansion of the input equation
IC Rg IC
Vg = Ii Rg + Vi = IB Rg + VBE = + VT ln
B IS
at the operating point:
iC R g iC
vg = + VT ln 1 +
B IC,A
IC,A Rg iC VT i C 2 VT iC 3
= + VT − + − ···
B IC,A 2 IC,A 3 IC,A
The use of
vo
iC = −
RC || RL
and Vk = IC,A (RC || RL ) leads to
IC,A Rg vo VT vo 2 VT vo 3
vg = − + VT − − − ···
B Vk 2 Vk 3 Vk
428 4 Amplifiers
If we use the values RC = RL = 10 k, Rg = 100 k, IC,A = 300 mA, B = 100 and
VT = 26 mV, then
5.78 vo2 2.57 vo3 1.28 vo4 0.685 vo5
vg = − 0.2173 vo − − − −
103 V 103 V2 103 V3 103 V4
and, after inversion:
0.563 vg2 vg3 2 vg4 4 vg5
vo = − 4.6 vg − + − + − ···
V V2 V3 V4
Consequently:
0.563 1 2 4
a1 = − 4.6 , a2 = − , a3 = , a4 = − , a5 =
V V2 V3 V4
The bn coefficients are obtained by converting the terms cosn ωt into the form cos nωt and
by sorting them according to frequencies. We can see that the even coefficients a2 , a4 , . . .
generate a DC component b0 – that is, a shift in the operating point; with the amplitudes
used in practical applications the shift is minor and can be neglected. Furthermore, at
even-numbered multiples of the frequency ω the even coefficients produce additional por-
tions. Similarly, at odd-numbered multiples of frequency ω the odd coefficients a3 , a5 , . . .
produce additional portions. The odd coefficients have an effect on the amplitude of the
useful signal; therefore, with higher amplitudes the operating gain is no longer constant.
The portion at the frequency ω is called the fundamental wave. The other portions are
known as harmonic waves and are numbered according to their order: the first harmonic
wave at 2ω, the second harmonic wave at 3ω and so on. Alternatively, the portions can
also be called harmonics: the first harmonic at ω, the second harmonic at 2ω and so on.
In practice, amplitudes are used at which the harmonic waves are significantly smaller
than the fundamental. In this case, only the first term of the expressions in parentheses in
(4.166) must be taken into account; that is, the coefficients bn are approximately constant
and no longer depend on the input amplitude v̂g , but only on the coefficient an of the
characteristic:
an
bn ≈ n−1 for n = 1 . . . ∞ (4.167)
2
For the amplitudes of the fundamental and the harmonic waves, it follows:
v̂o(F W ) = |b1 |v̂g ≈ |a1 |v̂g
a2
v̂g2
2
(4.168)
2
a3
v̂g3
4
..
.
Thus the amplitude of the fundamental shows a linear increase with the input amplitude,
while the amplitudes of the harmonic waves increase disproportionately. A requirement
for the approximation is the condition:
v̂o(F W ) v̂o(1.H W ) , v̂o(2.H W ) , . . .
Insertion of the coefficients leads to
|b1 |v̂g |b2 |v̂g2 , |b3 |v̂g3 , |b4 |v̂g4 , |b5 |v̂g5 , . . .
and, solving this for v̂g ,
b1
b1
b1
b1
v̂g
, 3
, 4
, . . .
b2 b3 b4 b5
n − 1
b1
n − 1
a1
(4.167)
v̂g minn
b
= 2 minn
a
(4.169)
n n
Example: With (4.167) and the coefficients a1 , . . . , a5 on page 428, we obtain for the
common-emitter circuit in Fig. 4.135:
430 4 Amplifiers
a2 0.282 a3 0.25
b1 ≈ a1 = − 4.6 , b2 ≈ =− , b3 ≈ = 2
2 V 4 V
a4 0.25 a5 0.25
b4 ≈ =− 3 , = 4
b5 ≈
8 V 16 V
All other coefficients also have the magnitude 0.25. From (4.169) we obtain, for the am-
plitude:
v̂g min (16.3 V ; 4.3 V ; 2.6 V ; 2 V ; . . .) = 1 V
The minimum is reached for n → ∞. For v̂g = 100 mV, (4.168) yields the fundamental
of v̂o(F W ) ≈ 460 mV, the first harmonic wave of v̂o(1.H W ) ≈ 2.82 mV and the second
harmonic wave of v̂o(2.H W ) ≈ 0.25 mV.
Distortion Factor
For sinusoidal signals, the distortion factor k is used as a measure of the nonlinear distor-
tions:
The distortion factor k describes the ratio of the effective value of all harmonic
waves in one signal to the effective value of the total signal.
For a sinusoidal signal without harmonic waves, k = 0.
Described by a polynomial
Vo,max
Vo,A
Vo,min
Vg,A Vg
Fig. 4.145. Validity range of the series expansion for the operating transfer characteristic
4.2 Properties and Parameters 431
The DC component b0 is neglected. For small signals with a low distortion factor, the
harmonic waves can be neglected when calculating the effective value of the total signal.
Consequently:
0
bn2 v̂g2n
n=2...∞
k ≈
b1 v̂g
Often, not all of the harmonic waves are transferred in systems with filters; partial
distortion factors
bn v̂ n
bn
kn =
=
v̂gn−1 for n = 2 . . . ∞
b1 v̂g
b1
that describe the ratio of the effective values of individual harmonic waves to the funda-
mental wave are thus quoted. This allows the distortion factor k to be calculated from the
partial distortion factors:
0
kn2
kn 1 0
n=2...∞
k = 0 ≈ kn2 (4.171)
1 + kn2 n=2...∞
n=2...∞
a2 a4 v̂g2 15a6 v̂g4
+ + + · · ·
b2
a2
2 2 32
v̂g
k2 =
v̂g =
v̂g ≈
b1
3a v̂ 2
5a v̂ 4
2a
a + 3 g
+
6 g
+ · · ·
1
1
4 8
a3 5a5 v̂g2 21a7 v̂g7
+ + + · · ·
b3
2
a3
2
4 16 64
2
v̂
k3 =
v̂g =
v̂g ≈
g
b1
3a v̂ 2
5a v̂ 4
4a
a + 3 g
+
6 g
+ · · ·
1
1
4 8
b4
a4
3
k4 =
v̂g3 ≈
v̂
b1 8a1
g
..
.
bn
an
kn =
v̂gn−1 ≈
n−1
It can be seen that for small amplitudes the nth partial distortion factor only depends on
the coefficients a1 and an , and increases with the power (n − 1) of the input amplitude. At
medium amplitudes other components become influential and cause a deviating response.
At very high amplitudes, the amplifier becomes fully saturated; in this case the output
provides a square-wave signal with:
⎧
⎨ 0 for n = 2, 4, 6, . . .
kn = 1
⎩ for n = 3, 5, 7, . . .
n
The distortion factor is k ≈ 0.48. In practice, saturation is usually not exactly symmetrical,
so that the even-numbered partial distortion factors do not approach zero.
Example: With the coefficients an from page 428, we obtain the following partial distor-
tion factors for the common-emitter circuit in Fig. 4.135:
0.061 v̂g 0.054 v̂g2 0.054 v̂g3 0.054 v̂g4
k2 ≈ , k3 ≈ , k4 ≈ , k5 ≈
V V2 V3 V4
Figure 4.146 shows the curves of k2 . . . k5 . In the quasi-linear region (I), the partial dis-
tortion factors are in accordance with (4.172); in the graph with two logarithmic scales,
the powers of v̂g become straight lines with the respective slopes shown. In the region
of weak saturation (II), the partial distortion factors increase markedly. With a further in-
crease in saturation, the output signal shows segments where some partial distortion factors
approach zero; in Fig. 4.146 this is the case for v̂g ≈ 0.2 V and v̂g ≈ 0.5 V. In the region
of strong saturation (III), the output signal is almost rectangular; the distortion factors are
k3 ≈ 1/3, k5 ≈ 1/5 and k2 , k4 → 0.
It follows from Fig. 4.146 that the distortion factor k in the quasi-linear region roughly
corresponds to the partial distortion factor k2 :
a2
k ≈ k2 ≈
v̂g
2a
1
All other partial distortion factors are clearly lower. In circuits with a symmetric charac-
teristic (a2 = 0) k2 = 0; here, in the quasi-linear region the following applies:
a3
2
k ≈ k3 ≈
v̂
4a
g
1
An example of this is the differential amplifier.
Compression Point
The odd coefficients of the series expansion also affect the amplitude of the fundamental
wave (see (4.166)); this makes the effective operating gain of the circuit dependent on the
amplitude:
3a3 2 5a5 4 35a7 6
A
B (v̂g ) = b1 = a1 + v̂ + v̂ + v̂ + · · ·
4 g 8 g 64 g
With an increasing amplitude, the absolute value of the operating gain may initially increase
(a3 /a1 > 0, gain expansion) or decrease (a3 /a1 < 0, gain compression), starting from
|AB | = |a1 |. When saturation sets in, the value always declines and approaches zero as
saturation increases. This region is not covered by the series expansion.
4.2 Properties and Parameters 433
kn
I II III
1.0
I : quasi-linear region k3
II : weak saturation 1/3
III : strong saturation 1/5
k5
0.1
k2
0.01
k3
k4
k2 , k4 , k 5 k2
–3
10
k4 , k5
–4
10
–5
10
k2 k3 k4 k5
–6
10
~
~
–4 –3
10 10 0.01 0.1 1.0 10.0 vg
V
Fig. 4.146. A plot of the partial distortion factors k2 . . . k5 for the common-emitter circuit in
Fig. 4.135
For amplifiers, the 1 dB compression point is quoted as a characteristic figure for the
onset of saturation:
The 1 dB compression point describes the amplitude at which the operating gain
drops to 1 dB below the small-signal operating gain due to the onset of saturation.
We can distinguish between the input compression point v̂g,comp with
A (v̂g,comp )
= 10−1/20 · |AB | ≈ 0.89 · |AB | (4.173)
B
AB AB
AB AB
0.89 A B 0.89 A B
~ ~
~
~
vg,comp vg vg,comp vg
Fig. 4.147. Magnitude of the operating gain, including the 1 dB compression point
Example: Circuit simulation for the common-emitter circuit in Fig. 4.135 yields
v̂g,comp ≈ 0.3 V and v̂o,comp ≈ 1.2 V.
Passband
Passband
Fig. 4.148. Portions resulting from feeding a single-tone signal (top) and a two-tone signal
(bottom) to a system with a characteristic of fifth order
The sum is of practical relevance only in so far as the portions are still within the passband.
At low amplitudes, the coefficients cn are approximately constant:
3a3 5a5
c1 ≈ a1 , c3 ≈ , c5 ≈ , ...
4 8
It follows:
2n + 1
c2n+1 ≈ a2n+1 for n = 1, . . . , ∞ (4.176)
2n+1
3a3
3
v̂o,IM3 = |c3 |v̂g3 ≈
v̂ (4.177)
4
g
5a5
5
v̂o,IM5 = |c5 |v̂g ≈
5
v̂
8
g
..
.
Intermodulation ratio: The abbreviations IM3 and IM5 are also used to refer to the
intermodulation ratio:
The ratio of the useful signal amplitude to the amplitude of a certain intermodu-
lation product is known as the intermodulation ratio.
Use of the amplitudes from (4.177) leads to:
v̂o,us
c1
1
4a1
1
IM3 =
=
2 ≈
(4.178)
v̂ o,IM3 c v̂ 3 3a
v̂ 2
g 3 g
v̂o,us
c1
1
8a1
1
IM5 = =
4 ≈
(4.179)
v̂o,IM5 c5 v̂g 5a5
v̂g4
In practice, the intermodulation ratios are usually quoted in decibels (dB):
IM3dB = 20 dB · log IM3 , IM5dB = 20 dB · log IM5
4.2 Properties and Parameters 437
4a1
(4.178)
v̂g,IP3 = v̂g
=
3a
(4.180)
IM3=1 3
8a1
(4.179)
v̂g,IP5 = v̂g
=
5a
(4.181)
IM5=1 5
and output intercept points (output IP, OIP):
v̂o,IP3 = |a1 |v̂g,IP3 , v̂o,IP5 = |a1 |v̂g,IP5 (4.182)
The latter exceed the input intercept points by a value equal to the small-signal operating
gain (|a1 | = |AB |) and are often simply called intercept points IP3 and IP5, without
explicit reference to the output.
Figure 4.149 shows the amplitude curve of the useful signal v̂o,us = |c1 |v̂g and the
intermodulation products v̂o,IM3 = |c3 |v̂g3 and v̂o,IM5 = |c5 |v̂g5 as a function of the input
amplitude v̂g , in the form of a diagram with two logarithmic axes. With small amplitudes,
this results in straight lines with slopes equal to 1 for v̂o,us , 3 for v̂o,IM3 and 5 for v̂o,IM5 .
The intercept points IP3 and IP5 are determined by extrapolation, as the points at which
the straight lines are intercepted. Furthermore, examples of the intermodulation ratios IM3
and IM5 and the compression point are indicated.30
With the help of the intercept points, we can calculate the amplitudes of the intermod-
ulation products and the intermodulation ratios for any given input and output amplitudes
in the quasi-linear region. According to the approximations in (4.177), the amplitudes of
the intermodulation products, when referenced to the input intercept points, are:
3a3
3
(4.180) |a1 |v̂g3
v̂o,IM3 ≈
v̂ =
3a3
|a1 |v̂ 3 =
4
g
4a
1
g 2
v̂g,IP3
30 Due to b = c , two-tone operation of the amplifier results in a different compression point than
1 1
one-tone operation [see (4.166) and (4.175)]; c1 ≈ b1 ≈ a1 is true for small amplitudes only.
Therefore, diagrams such as that in Fig. 4.149 usually present the curve of the intermodulation
products in two-tone operation and the curve of the useful portion in one-tone operation. This has
no bearing on the intercept points, as they are determined by means of extrapolated values.
438 4 Amplifiers
vo
[log]
v o,IP5
v o,IP3
vo,comp
vo,us
IM 3 IM 5
vo,IM3 vo,IM5
~
~
Fig. 4.149. Intercept points at the input (v̂g,IP3 ,v̂g,IP5 ) and at the output (v̂o,IP3 ,v̂o,IP5 ) and the
intermodulation ratios IM3 and IM5
5a5
5
5a5
(4.181) |a1 |v̂g5
v̂o,IM5
v̂ =
|a1 |v̂ 5 =
8
g 8a1
g 4
v̂g,IP5
With reference to the output intercept points, and taking v̂o,us = |a1 |v̂g and (4.182) into
consideration, we obtain:
% &3
|a1 |v̂g3 |a1 |v̂g 3
v̂o,us
v̂o,IM3 ≈ 2 = % &2 = 2
v̂g,IP3 |a1 |v̂g,IP3 v̂o,IP3
% &5
|a1 |v̂g5 |a1 |v̂g 5
v̂o,us
v̂o,IM5 ≈ 4 = % &4 = 4
v̂g,IP5 |a1 |v̂g,IP5 v̂o,IP5
From the approximations in (4.178) and (4.179), and taking v̂o,us = |a1 |v̂g and (4.182)
into consideration, we obtain the intermodulation ratios:
% &2
4a1
1 (4.180) v̂g,IP3
2
|a1 |v̂g,IP3 2
v̂o,IP3
IM3 ≈
= = % &2 =
3a3
v̂g2 v̂g2 |a1 |v̂g
2
v̂o,us
4.2 Properties and Parameters 439
% &4
8a1
1 (4.181) 4
v̂g,IP5 |a1 |v̂g,IP5 4
v̂o,IP5
IM5 ≈
= = % &4 = 4
5a
v̂ 4
5 g v̂g4 |a1 |v̂g v̂o,us
In general, the following applies:
v̂g,IPn n−1 v̂o,IPn n−1
IMn ≈ = (4.184)
v̂g v̂o,us
Example: Using (4.180)–(4.182) and the coefficients an on page 428, the intercept points
for the common-emitter circuit in Fig. 4.135 are:
v̂g,IP3 = 2.5 V ⇒ v̂o,IP3 = 11.4 V , v̂g,IP5 = 1.2 V ⇒ v̂o,IP5 = 5.4 V
These are always clearly higher than the actual amplitudes. For a two-tone signal with
v̂g = 100 mV, (4.177) yields v̂o,us = 460 mV, v̂o,IM3 ≈ 0.7 mV and v̂o,IM5 ≈ 0.024 mV;
with (4.178), IM3 ≈ 610; and with (4.179), IM5 ≈ 19000.
Rg
1 2
vg vg1 vo1 = vg2
RL vo2 = vo
a1,1 = A B1 a1,2 = A B2
a2,1 a2,2
a3,1 a3,2
...
...
By way of contrast, amplifier 2 is operated with an ideal signal voltage source, since the
voltage vg2 is fed directly to the amplifier input (see Fig. 4.150); therefore, Rg = 0 and:
RL
a1,2 = AB2 = A2
ro2 + RL
Distortion factor in series connections: For series connections, the partial distortion
factors are obtained using (4.172):
a2
k2 ≈
v̂g , k3 ≈
a3
v̂ 2 , . . .
2a
4a
g
1 1
If we assume that all harmonic distortions add up – in other words, that all terms in
parentheses in (4.185) have the same sign – and we take into account the fact that v̂g2 ≈
|a1,1 |v̂g1 , then the partial distortion factors of the series connection can be expressed by
the partial distortion factors of amplifier 1
a2,1
a3,1
2
k2,1 ≈
v̂g1 , k3,1 ≈
v̂ , ...
2a1,1
4a1,1
g1
and by the partial distortion factors of amplifier 2:
a2,2
a1,1 a2,2
k2,2 ≈
v̂g2 ≈
v̂g1
2a1,2
2a1,2
a3,2
2
a2 a
3,2
2
k3,2 ≈
v̂ ≈
1,1
v̂ , ...
4a1,2
g2
4a1,2
g1
k2 ≈ k2,1 + k2,2
k3 ≈ k3,1 + k3,2 + 2k2,1 k2,2
k4 ≈ k4,1 + k4,2 + 2k3,1 k2,2 + 3k2,1 k3,2 + k2,1
2
k2,2
..
.
If all of the partial distortion factors are much smaller than one, then the products of the
partial distortion factors are negligible:
k2 ≈ k2,1 + k2,2 , k3 ≈ k3,1 + k3,2 , k4 ≈ k4,1 + k4,2 , ...
Thus, the partial distortion factors in series connections can be obtained from the sum of
the partial distortion factors of both amplifiers. This result can be extended to cover the
series connection of any number of amplifiers:
The partial distortion factors of a series connection of several amplifiers corre-
spond approximately to the sum of the corresponding partial distortion factors of
the individual amplifiers.
The equation for the series connection of M amplifiers is:
0
kn ≈ kn,m (4.186)
m=1...M
If a compensation of harmonic waves occurs in a series connection, then the partial dis-
tortion factors of the series are smaller than the sum; therefore, the sum can be considered
to be an estimate of the upper limit (the worst case).
4.2 Properties and Parameters 441
For the general case, it is not possible to describe a simple relationship between the
total distortion factor k of a series connection as calculated using (4.171) on page 431
from the partial distortion factors and the distortion factors of the individual amplifiers.
However, in practice, one partial distortion factor is usually dominant, so that k ≈ k2 or –
in the case of symmetrical characteristics – k ≈ k3 ; in this case, (4.186) can be used and
the distortion factor of the series connection can be evaluated by adding up the distortion
factors of the individual amplifiers.
Intercept points of series connections: For the input intercept point IIP3 of the series
connection, it follows from (4.180) and (4.185) that:
3a3
3a 2
=
3,1 3a1,1 a3,2 3a 2,1 a 2,2
=
+ +
2
v̂g,IP3 4a 1
4a 1,1 4a 1,2 2a 1,2
If we assume that the first two terms have the same sign and the third term is negligible,
because the denominator is the product of two comparably small values a2,1 and a2,2 , this
expression can be presented with the help of the intercept point of amplifier 1
4a1,1
v̂g1,IP3 =
=
4AB1
3a3,1
3a3,1
4a1,2
4AB2
v̂g2,IP3 =
3a3,2
3a3,2
as:
1 1 |AB1 |2
IIP3 : 2
≈ 2
+ 2
v̂g,IP3 v̂g1,IP3 v̂g2,IP3
With
v̂o1,IP3 = |AB1 | v̂g1,IP3 , v̂o2,IP3 = |AB2 | v̂g2,IP3
this leads to the output intercept point OIP3:
1 1 1
OIP3 : ≈ +
2
v̂o,IP3 |AB2 |2 v̂o1,IP3
2 2
v̂o2,IP3
The intercept point IP5 is obtained in the same way:
1 1 |AB1 |4
IIP5 : 4
≈ 4
+ 4
v̂g,IP5 v̂g1,IP5 v̂g2,IP5
1 1 1
OIP5 : ≈ +
4
v̂o,IP5 |AB2 |4 v̂o1,IP5
4 4
v̂o2,IP5
Using the equation for parallel circuits,
1 1 1
= + ⇒ c = a || b
c a b
we obtain:
442 4 Amplifiers
v̂g2,IP3 2
IIP3 : 2
v̂g,IP3 ≈ v̂g1,IP3
2
||
|AB1 |
% &2
OIP3 : 2
v̂o,IP3 ≈ |AB2 |v̂o1,IP3 || v̂o2,IP3
2
v̂g2,IP5 4
IIP5 : 4
v̂g,IP5 ≈ v̂g1,IP5
4
||
|AB1 |
% &4
OIP5 : 4
v̂o,IP5 ≈ |AB2 |v̂o1,IP5 || v̂o2,IP5
4
It can be seen that with the help of the operating gains AB1 and AB2 the intercept points
of the amplifiers can be converted to the input or output of the series connection and can
be connected in parallel to the second and fourth power.
This result can be extended to apply to the series connections of any number of ampli-
fiers:
The input intercept point IIPn of amplifiers in series connection can be determined
by converting the intercept points of the individual amplifiers to the input by means
of the operating gains and by connecting them in parallel with the power (n − 1).
In the same way, the output intercept point OIPn can be determined by conversion
to the output.
4.2.4
Noise
The basics for discussing the noise and calculating the noise figure are described in
Sect. 2.3.4, using a bipolar transistor as an example. Here, the results are converted to
amplifiers in general. For an explanation of the term noise density, we recommend that
you read Sect. 2.3.4 on page 82.
vi vo vi i r,0 vo
v r,0 i r,0
nV/ Hz pA/ Hz
100
50
20 i r,0
10 With noise region
5
v r,0
2
1
~
~
Fig. 4.152. Typical noise density curves of an amplifier with bipolar transistors
Signal generator Noise sources of the amplifier Signal generator with equivalent
with noise source noise source
v r,g vr,0 vr
Rg Rg
vg i r,0 vg
a With the noise source of the signal generator and b With an equivalent noise source
the equivalent noise sources of the amplifier
source of the signal generator and the noise sources of the amplifier can be combined into
one equivalent noise source (see Fig. 4.153b). Consequently:
|v r (f )|2 = |v r,g (f )|2 + |v r,0 (f )|2 + Rg2 |i r,0 (f )|2 (4.187)
It is assumed that the amplifier noise originates from the signal generator, and the ratio of
the noise density of the equivalent noise source to the noise density of the signal generator
is known as the spectral noise figure [4.6]:
In other words:
The noise density of the equivalent noise source representing the noise from the
signal generator and from the amplifier is higher than the noise density of the
signal generator by a value equal to the spectral noise figure F (f ). Thus the
4.2 Properties and Parameters 445
spectral noise figure describes the factor by which the noise already existing in
the signal generator is increased as a result of the amplifier noise. This means
that the noise density at the amplifer output is higher than the noise density at the
output of a noiseless amplifier with the same gain by a value equal to the spectral
noise figure. Therefore, a noiseless amplifier has a spectral noise figure of one.
In order to prevent the noise figure from depending on the properties of one particular
signal generator, an ideal signal generator with a noise density that corresponds to the
thermal noise density of the internal resistance Rg is used for calculations [4.7]:
nV Rg
|v r,g | = 4kT Rg = 0.13 √ · (4.189)
Hz
The noise density in real signal generators is usually much higher.
When quoting the noise figure F in practice, we are referring to the spectral noise figure
for the frequency range that is of interest for the intended application. If the noise figure in
this region is not constant, then, strictly speaking, the mean noise figure must be calculated
using the integral given in (4.233). In this book, we follow the general convention and use
the term F (f ) only in cases in which the frequency dependence of the noise figure is
expressly being referred to; noise densities are treated accordingly.
The relationship between the noise densities, the internal resistance Rg of the signal
generator and the noise figure is illustrated in Fig. 4.154; the graph shows separate curves
for the portions |v r,g |, |v r,0 | and Rg |i r,0 | that originate in the white noise region of the
equivalent noise source for the amplifier shown in Fig. 4.152. On the two logarithmic axes,
these portions have the slopes 0, 1/2 and 1, respectively:
1/2
|v r,0 | = const. ∼ Rg0 , |v r,g | ∼ Rg , Rg |i r,0 | ∼ Rg
In this plot, the noise figure F corresponds to the ratio of the noise density |v r | of the
equivalent noise source and the noise density |v r,g | of the signal generator; this is shown
|v|
nV/ Hz
100 F
Optimum
operating point
10
| vr |
| v r,0 |
1
F
| v r,g |
Rg |–i r,0 |
0.1 Rg,opt
~
~
Fig. 4.154. Noise density of the equivalent noise source for the amplifier in Fig. 4.152 within the
white noise region
446 4 Amplifiers
100
50
20
10
Optimum
5 operating point
Fopt
2
1 Rg,opt
~
~
Fig. 4.155. Noise figure of the amplifier in Fig. 4.152 in the white noise region
on a separate diagram in Fig. 4.155. Due to the different slopes, there is always one
point at which the noise figure reaches a minimum; this point is indicated on the curves
of Figs. 4.154 and 4.155 as the optimum operating point. The corresponding internal
resistance is known as the optimum source resistance Rgopt .
A fundamental property can be derived from the curves in Fig. 4.154:
If operated with an internal resistance that is clearly below the optimum source
resistance, the noise density of the equivalent noise source depends primarily on
the noise voltage density of the amplifier; similarly, if operated with an internal
resistance that is clearly above the optimum source resistance, then the noise
density depends primarily on the noise current density of the amplifier. This is
also true for the noise figure.
Therefore:
No general statement can be made for operation with an internal resistance in the region of
Rgopt , since in this case the ratio of the noise densities of the amplifier to the noise density
of the signal generator is decisive.
An amplifier has low noise if there is a region in which the portions |v r,0 | and Rg |i r,0 |
caused by the amplifier are clearly below the noise density |v r,g | of the signal generator.
The borderline case is where the noise density of the signal generator is equal to the sum
of the noise densities of the amplifier:
!
|v r,g |2 = |v r,0 |2 + Rg2 |i r,0 |2
The resulting noise figure is F = 2; therefore F < 2 is often defined as a condition for a
low-noise amplifier.
4.2 Properties and Parameters 447
The white noise region: In the white noise region, the frequency dependence can be
neglected; it thus follows:
|v r,0 |
Rgopt = (4.192)
|i r,0 |
|v r,0 | |i r,0 |
Fopt = 1 + (4.193)
2kT
Here, |v r,0 | and |i r,0 | are the noise densities in the white noise region. The quantity equation
for Fopt is:
T =300 K |v r,0 | |i r,0 |
Fopt = 1 + 0.12 · √ · √
nV/ Hz pA/ Hz
√
For the √amplifier in Fig. 4.152, the values are |v r,0 | = 1.1 nV/ Hz and |i r,0 | =
1.8 pA/ Hz; it follows that Rgopt = 610 and Fopt = 1.24.
At the optimum operating point, the absolute noise source quantities of the amplifier
are of the same value: |v r,0 | = Rgopt |i r,0 |. This relationship is shown in Fig. 4.154: the
corresponding straight lines intersect at the optimum operating point.
100
3
50
2
20 1
10
2
Fmin = 1
1
~
~
Fig. 4.156. A plot of the noise figures of three amplifiers with different optimum noise figure
that |v r,0 | and |i r,0 | are known. It is also possible to start with Fopt and Rgopt ; from (4.192)
and (4.193), it follows:
% & 2kT % &
|v r,0 |2 = 2kT Rgopt Fopt − 1 , |i r,0 |2 = Fopt − 1
Rgopt
and by inserting this into (4.188), we obtain:
1% & Rg Rgopt
F = 1+ Fopt − 1 + (4.194)
2 Rgopt Rg
Rg = Rgopt leads by definition to F = Fopt ; if Rg = Rgopt , then F > Fopt .
It should be noted that the increase in the noise figure is not only due to the ratio of the
resistances, but also to the optimum noise figure:
An amplifier with lower noise not only has a lower optimum noise figure but also
a wider minimum. In the borderline case of a noiseless amplifier, the minimum
becomes infinitely wide; that is, F = 1 for all values of Rg .
Figure 4.156 illustrates this for three amplifiers with different optimum noise figures:
amplifier 1 with Fopt = 1.12, amplifier 2 with Fopt = 2.2 and amplifier 3 with Fopt = 13.32
For Rg Rgopt , the noise figure is almost only dependent on |v r,0 |, while for Rg
Rgopt it is almost only dependent on |i r,0 |:
⎧
⎪
⎪ |v r,0 |2 1% & Rgopt
⎪
⎨ 4kT R = Fopt − 1 für Rg Rgopt
g 2 Rg
F ≈
⎪
⎪ Rg |i r,0 |2 1% & Rg
⎪
⎩ = Fopt − 1 für Rg Rgopt
4kT 2 Rgopt
This relationship has already been pointed out in one of the emphasized notes. The corre-
sponding asymptotes are shown in Fig. 4.156.
32 For amplifier 1 it was assumed that |v
√ √
r,0 | = 1 nV/
√
Hz and |i r,0 | = 1 pA/ Hz. The values for
amplifiers 2 and 3 are higher by the factors 10 and 10, respectively.
4.2 Properties and Parameters 449
A note on the selection or dimensioning of amplifiers: The parameters Fopt and Rgopt
play an important role in the optimum operation of a given amplifier. On the other hand,
making a choice between several standard amplifiers must always be done on the basis of
the noise figure F for the given source resistance Rg : the optimum amplifier is the one that
has the lowest noise figure when operated with the given source resistance. The operating
noise figure is calculated using (4.188) and (4.189):
|v r,0 (f )|2 + Rg2 |i r,0 (f )|2 T =300 K |v r,0 (f )|2 + Rg2 |i r,0 (f )|2
F = 1+ = 1+
4kT Rg 1.656 · 10−20 V2/ · Rg
Here, the parameters Fopt and Rgopt are only relevant because, according to (4.194), they
can be used to calculate the operating noise figure. Therefore, the best choice is not the
amplifier with the lowest noise figure Fopt or the amplifier whose optimum source resis-
tance Rgopt best corresponds to the given source resistance Rg ; usually, neither produces
an optimum result. Accordingly, when dimensioning an integrated amplifier it is the op-
erating noise figure with the given source resistance that should be used as a criterion for
optimization; the optimizations Fopt or Rgopt as individual parameters are useless. There
is only one exception: if an impedance transformation of Rg to Rgopt is made by means
of a transformer, then F = Fopt can be achieved; this allows the amplifier with the lowest
noise figure Fopt to be selected provided that the required transformation ratio Rg /Rgopt
can be realized.
Calculating the noise figure: As the noise densities of the equivalent noise sources are
related to the source voltages and not to the input voltages of the amplifiers, it is not possible
to use the operating gain values AB1 = v2 /vg and AB2 = vo /v2 in this case. Instead, we
must use the noise operating gains
Fig. 4.157. Equivalent circuit for calculating the noise figure of two amplifiers connected in series
450 4 Amplifiers
A 1 v1 ri1
AB,r1 = = A1
vg Rg + ri1
A 2 v2 ri2
AB,r2 = = A2
A1 v1 ro1 + ri2
and the load factor:
RL
kL =
ro2 + RL
It thus follows:
vo
AB = = AB,r1 AB,r2 kL
vg
The noise operating gains are made up of the voltage divider factor at the input and the
open-circuit gain, and indicate the gain from one source voltage to the next.
At first, all noise sources are converted to the output of the series connection by means
of the relevant gain; without a signal voltage (vg = 0) it follows:
% & % &
vo = vr,g + vr,01 + Rg ir,01 AB,r1 AB,r2 kL + vr,02 + ro1 ir,02 AB,r2 kL
The voltage of the equivalent noise source for the series connection is obtained by conver-
sion to the signal generator:
vo vr,02 + ro1 ir,02
vr = = vr,g + vr,01 + Rg ir,01 + (4.195)
AB AB,r1
Since all noise sources are independent, the noise density of the equivalent noise source
is:
|v r,02 |2 + ro1
2
|i r,02 |2
|v r |2 = |v r,g |2 + |v r,01 |2 + Rg2 |i r,01 |2 +
A2B,r1
For the noise figure of the series connection, it follows:
|v r |2 |v r,01 |2 + Rg2 |i r,01 |2 |v r,02 |2 + ro1
2
|i r,02 |2
F = = 1+ + (4.196)
|v r,g |2 |v r,g |2 A2B,r1 |v r,g |2
The noise figure of the first amplifier is:
|v r,01 |2 + Rg2 |i r,01 |2 |v r,01 |2 + Rg2 |i r,01 |2
F1 = 1 + = 1+
|v r,g |2 4kT Rg
For the second amplifier, Rg is replaced by ro1 . To calculate the noise figure, we assume an
ideal signal generator with thermal noise from the internal resistance. The thermal noise
|v r,o1 |2 = 4kT ro1
must therefore be used as a reference value. This does not mean that resistance ro1 in
Fig. 4.157 produces thermal noise, but only that the noise figure of the second amplifier is
determined for an internal signal generator resistance of the value ro1 . Therefore:
|v r,02 |2 + ro1
2
|i r,02 |2 |v r,02 |2 + ro1
2
|i r,02 |2
F2 = 1 + = 1+
|v r,o1 |2 4kT ro1
By inserting the noise figures F1 and F2 into (4.196), we obtain:
4.2 Properties and Parameters 451
F2 − 1 ro1
F = F1 +
A2B,r1 Rg
which by generalization leads to the noise figure of a series connection of n amplifiers:
noise figure of a series connection of n amplifiers
F2 − 1 ro1 F3 − 1 ro2
F = F1 + 2
+ 2 + ···
AB,r1 R g AB,r1 A2B,r2 Rg
⎛ ⎞
n ⎜
⎜ ⎟
0 ⎜ F(i) − 1 ro(i−1) ⎟
⎟ (4.197)
= F1 + ⎜ i−1 ⎟
⎜/ Rg ⎟
i=2 ⎝ ⎠
A2 B,r(k)
k=1
ri(k) (4.198)
AB,r(k) = A(k) with ro0 = Rg
ro(k−1) + ri(k)
The noise figure of the first amplifier directly affects the noise figure of the series
connection; for all subsequent amplifiers, the additional noise figure
FZ = F − 1 (4.199)
which is rated according to the inverse square of the previous operating noise gain and the
ratio of the source resistances, enters into the result. A minimum noise figure is achieved
mainly by optimizing the first amplifier:
– Minimizing the noise figure F1 .
– Maximizing the noise operating gain AB,r1 by maximizing A1 and ri1 .
– Minimizing the output resistance ro1 .
The last item is of particular importance for small internal resistances Rg , since the factor
1/A2B,r1 can be overcompensated by the factor ro1 /Rg ; the noise figure F2 dominates
in this case. If the second amplifier also features a high-noise operating gain, then the
contributions of the subsequent amplifiers are negligible.
33 Here, we use r.m.s. values; that is, P = u2 /R; a distinction between DC and AC voltages is
therefore not required.
452 4 Amplifiers
Consequently:
2
PA,amp ri Rg Rg
GA = = A2 = A2B,r (4.200)
PA,g Rg + r i ro ro
Insertion into (4.197) leads to:
⎛ ⎞
n ⎜
⎜ ⎟
F2 − 1 F3 − 1 0 ⎟
⎜ F(i) − 1 ⎟
F = F1 + + + · · · = F1 + ⎜ i−1 ⎟ (4.201)
GA1 GA1 GA2 ⎜/ ⎟
i=2 ⎝ ⎠
G A(k)
k=1
The following relationship should be noted:
Rg 2 ro1 ro(i−2)
GA1 GA2 · · · GA(i−1) = A2B,r1 AB,r2 · · · A2B,r(i−1)
ro1 ro2 ro(i−1)
Rg
= A2B,r1 A2B,r2 · · · A2B,r(i−1)
ro(i−1)
Equation (4.201) is often quoted in the form
F2 − 1 F3 − 1
F = F1 + + + ···
G1 G1 G2
with the power gains G(i) not specified in detail. Since a number of different power gains
are used in radio-frequency engineering, we expressly emphasize that, in general, the
available power gain GA must be used; only in cases where all amplifiers are matched
are all power gains equal and the general term “power gain” can be used without further
specification.
Equivalent noise sources: The equivalent noise sources of the series connection in
Fig. 4.157 can be determined by means of the noise voltage of the equivalent noise source;
inserting AB,r1 into (4.195) and arranging the terms into groups with and without Rg leads
to:
vr,02 + ro1 ir,02 vr,02 + ro1 ir,02
vr = vr,g + vr,01 + + Rg ir,01 +
A A r
3 45 1 6 3 45 1 i1 6
vr,0 ir,0
The equivalent noise sources are interdependent, as the noise sources of the second am-
plifier enter into the noise voltage source vr,0 and the noise current source ir,0 . Since
calculations with interdependent noise sources are rather complex, we will refrain from
using this model, but we nevertheless wish to emphasize an important relationship: the
equivalent noise sources of a multi-stage amplifier, which are regarded as basic transistor
circuits connected in series, are only approximately independent if the contribution of the
second and any subsequent stage can be neglected in at least one of the two equivalent
noise sources.
Ohmic resistance: For ohmic resistances, we prefer to use the model of a noise current
source with the noise density:
454 4 Amplifiers
C D
v r,T
v r,F
B R
G
R i r,R
i r,T
i r,F v r,R
E S
|v r,T | = 2 kT + 4 kT RB 8 kT
2 2 2
|v r,F | = |v r,R | = 4 kTR
gm 3 gm
2
2 kTgm 8 kTg m f
| i r,R | = 4 kT
2 2 2
| i r,T | = | i r,F | =
ˇ 3 f T2 R
Fig. 4.158. Noise sources of a bipolar transistor, a MOSFET and an ohmic resistance
4kT
|i r,R |2 =
R
However, this method only determines the thermal noise of an ideal resistance; real resis-
tances in integrated circuits may have a clearly higher noise density, depending on their
construction.
Bipolar transistors: According to (2.49) and (2.50), the following √ applies to a bipolar
transistor in the white noise region; that is, for fg(1/f ) < f < fC / β ≈ fC /10:
2kT VT 2kT
|v r,T |2 = + 4kT RB = + 4kT RB
IC,A gm
2qIC,A 2kT gm
|i r,T |2 = =
β β
Here, gm = IC,A /VT and VT = kT /q. We shall limit our analysis to small and medium
currents; this eliminates the third term of (2.49). It is obvious that a low-noise bipolar
transistor must have a low base spreading resistance RB and a high current gain β. While
the current gain is determined by the technology, the base spreading resistance can be
influenced by scaling: in general, RB is inversely proportional to the size of the transistor.
It is therefore possible to reduce the noise voltage density in the medium current range by
enlarging the transistor. But this is at the cost of the bandwidth, since the capacitances of
the transistor increase while the transconductance remains unchanged.
Figure 4.159 shows the noise figure of a bipolar transistor with β = 100 and RB = 10
in the IC,A –Rg plane. Inserting the noise densities into (4.192) and (4.193) leads to:34
β RB →0
β β≈100 10 0.26 V
Rgopt,T = 1 + 2gm RB ≈ ≈ =
gm gm gm IC,A
1 RB →0
1 β≈100
Fopt,T = 1 + 1 + 2gm RB ≈ 1+ ≈ 1.1
β β
34 The relationships |v 2 2 2 2
r,T | = |v r,0 (f )| and |i r,T | = |i r,0 (f )| apply since vr,T and ir,T are
the equivalent noise sources of the bipolar transistor.
4.2 Properties and Parameters 455
Rg
Ω
100M
20 100 1000 10000
10M
5
2
1M
1.2 100
100k
1.2 20
10k
2 1.2 2 5
1k
5 1.2
20
100
100 2
10
1000 100 20 5
1
10000
~
~
However, this optimum only applies if gm and/or IC,A are regarded as given variables
while Rg varies. The relationship is different if Rg is given and the optimum quiescent
current IC,A is to be determined; in this case:
VT β
IC,A opt (Rg ) = '
Rg2 + RB2
RB 1 RB 2
Fopt,T (Rg ) = 1 + + 1+
Rg β Rg
This optimum has already been determined in Sect. 2.3.4 [see (2.54) and (2.57)]. In practice,
it is more significant, because Rg is usually given and can only be matched by means of a
transformer or a resonance transformer. By contrast, it is easy to alter IC,A .
The difference between Fopt,T and Fopt,T (Rg ) is caused by the base spreading resis-
tance; for RB = 0, the values are equal. To illustrate this, Fig. 4.160 shows the noise figure
of a bipolar transistor with β = 100 and RB = 100 , together with the curves for Fopt,T
and Fopt,T (Rg ). It is clear that Fopt,T represents the optimum in the Rg direction and that
Fopt,T (Rg ) is the optimum in the IC,A direction. If Rg < RB , then Fopt,T (Rg ) increases
rapidly; for Rg = 1 , the noise figure is 100. Figure 4.160 also shows that there is an
upper limit for the quiescent current:
VT β β≈100 0.26 V
IC,A opt (Rg → 0) = ≈
RB RB
A higher quiescent current is pointless.
For currents in the range of IC,A = 10 mA . . . 1 mA, Rgopt,T ≈ 26 k . . . 260 . For
higher√quiescent currents, the base spreading resistance has to be taken into account. The
value βRB ≈ 10RB can be regarded as the lower limit for Fopt,T (Rg ); the portions
in Fopt,T (Rg ) that are caused by β and Rg are then of approximately the same size and
456 4 Amplifiers
Rg
Ω
2 5 20
100
100k
1.2
20
10k
1.2 2 5
1k
Fopt,T
5 2
100
20 5
10
20
100
1
~ Fopt,T ( Rg )
Fig. 4.160. Optimum noise figures
~
Fopt,T (Rg ) ≈ Fopt,T still applies. The upper limit for Rgopt,T depends on the required
bandwidth; IC,A cannot be reduced as one wishes, since in the range of very small currents
the transit frequency fT decreases proportionally to IC,A (see Fig. 2.44 on page 80). The
limiting current, below which the transit frequency declines, is:
VT % & VT % &
IC,A = CJ,E + CJ,C ≈ 2CJ 0,E + CJ 0,C
τ0,N τ0,N
For an npn transistor of size 1 with the parameters given in Fig. 4.5, the limit current
is 100 mA; consequently, Rgopt = 5.7 k and Fopt = 1.22. In pnp transistors the base
spreading resistance is usually lower than in npn transistors; with low-resistance sources
in particular, this produces a lower noise figure. However, even in complementary tech-
nology, the transit frequency of a pnp transistor is lower than that of an npn transistor. In
technologies where only lateral pnp transistors are available, the transit frequency of the
pnp transistors is lower than that of npn transistors by up to three orders of magnitude.
Here, gm = 2KID,A . This description applies up to the transconductance cutoff fre-
quency fY 21s , which is usually higher than the transit frequency fT . The noise current
density depends on the frequency; in other words, there is no region of white current noise.
But when we examine the basic circuits, we will note that additional noise sources that
cover the current noise of the MOSFET in a more or less wide range are caused by the
external circuitry; this makes the equivalent noise current density of the circuit independent
of the frequency in this range.
Inserting the noise densities into (4.192) and (4.193) leads to:35
fT
Rgopt,F (f ) =
gm f
4 f
Fopt,F (f ) = 1 +
3 fT
For f fT we have Fopt,F → 1, which means that at the optimum operating point
a MOSFET is practically free of noise. In narrow-band applications, the frequency de-
pendency can be neglected and the centre frequency can be used for f . In contrast, for
wide-band applications the mean noise current density
fU
1
i [fL , fU ]
2 = |i r,F (f )|2 df
r,F
fU − fL fL
2
8kT gm fU3 − fL3 fU fL fU
= ≈ 8kT gm
fT2 f U − fL fT
must be used in the range between the lower cutoff frequency fL and the upper cutoff
frequency fU ; for fU fL , this leads to:
fT
√
Thus, we only have to insert f = 3fU in order to achieve optimum values for a wide-band
application. Here too, Fopt,F → 1, since in many applications the upper cutoff frequency
fU is at least 100 times lower than the transit frequency. From the equations for Rgopt,F ,
one must not conclude that the matching to a source can be optimized by the proper choice
of the transit frequency; the noise figure has a minimum value for fT → ∞ not only
in optimum operation but, indeed, for any given operation. Figure 4.161 shows the noise
figure of a MOSFET for
2ID,A 2ID,A
VGS − Vth = 1 V ⇒ gm = =
VGS − Vth 1V
and fU /fT = 100 in the ID,A –Rg plane. Due to the diagonal plots across the entire plane,
it is not necessary to distinguish between optimization in the Rg and ID,A directions, which
means that the equation for Rgopt,F can be solved for gm and can be used to calculate the
optimum transconductance.
35 The relationships |v 2 2 2 2
r,F | = |v r,0 (f )| and |i r,F (f )| = |i r,0 (f )| apply since vr,F and ir,F
are the equivalent noise sources of the MOSFET.
458 4 Amplifiers
Rg
Ω
1.2 2 5 20 1000
100M
100
20
10M
5
1M
1.2 1.2 2
2
100k
5
10k
20
1k
100 1.2
100
2
1000 5
10
20
10000 100
1
~
~
Unlike the bipolar transistor, the MOSFET offers an additional degree of freedom,
since the transconductance not only depends on the current ID,A , but also on the size of
the MOSFET, which is expressed by the transconductance coefficient K: gm = 2KID,A .
It is only the transconductance that enters the noise density; in other words, the choice
of ID,A and K does not affect Rgopt,F and Fopt,F in the white noise region. However, the
width of this range depends on this choice, since it influences the 1/f cutoff frequency.
With the given area A = W L, we prefer to increase L at the expense of W , since:
- -
ID,A 1 ID,A L
= ID,A W −1/2 L−3/2
1/2
fg(1/f ) ∼ k(1/f ) ∼ 2
K L W
This causes K to decline. The minimum 1/f noise is thus achieved with MOSFETs that are
geometrically large but electrically small, and that are operated using high currents. Here,
as in the case for the bipolar transistor, there is a conflict with the bandwidth:
-
W
gm 2KID,A ID,A
fT ∼ ∼ ∼ L = ID,A W −1/2 L−3/2
1/2
C A WL
It follows that fg(1/f ) ∼ fT , which means that a reduction in the 1/f cutoff frequency will
also reduce the transit frequency.
It follows:
3
gm,T > gm,F
4
Otherwise, the MOSFET is generally superior in the white noise region.
In practice, the limit Rg,T ↔F is seldom of interest, since general limitations are caused
by the technology employed, which cannot be chosen with regard to the noise figure of
one amplifier alone; it is of greater interest to determine the region in which the required
noise figure is achieved. Solving (4.194) for Rg results in a quadratic equation with the
solution:
⎛ ⎞
2
F − 1 F − 1
Rg,l/u = Rgopt ⎝ ± − 1⎠
Fopt − 1 Fopt − 1
For F > Fopt , the lower limit is achieved with Rg,l < Rgopt and the upper limit with
Rg,u > Rgopt ; for F = Fopt , we have Rg,l = Rg,u = Rgopt and no solution exists for
F < Fopt . Furthermore, Rg,l Rg,u = Rgopt
2 . For
% &
F − 1 > 2 Fopt − 1
it is possible to perform a series expansion of the square root and interrupt the series after
the linear term;36
Rgopt Fopt − 1 F −1
Rg,l ≈ , Rg,u ≈ 2Rgopt
2 F −1 Fopt − 1
The results for a bipolar transistor are:
1 1 2β
Rg,lT ≈ + RB , Rg,uT ≈ (F − 1)
2gm F −1 gm
The base spreading resistance RB only influences the lower limit. The values for a MOSFET
in wide-band applications are:
2 F − 1 fT 2
Rg,lF ≈ , Rg,uF ≈
3gm (F − 1) 2gm fU
Only the upper solution Rg,uF depends on the frequency.
Optimum operating point: For both the bipolar transistor and the MOSFET, the opti-
mum source resistance in the white noise region depends primarily on the transconduc-
tance; in principle, this is also the case for the 1/f and the high-frequency noise, as is shown
in the relevant equations in Sects. 2.3.4 and 3.3.4. The operating point plays a predominant
role in the optimization of the noise figure. Bipolar transistors allow no margin due to the
relationship gm = IC,A /VT ; that is, an optimum collector current IC,A opt (Rg ) exists for
every source resistance. MOSFETs, on the other hand, allow variations in the ratio of the
transconductance coefficient K to the drain current ID,A , since gm = 2KID,A .
In practice, it is not usually possible to select the collector or drain current at the oper-
ating point solely on the basis of noise considerations, since conflicting demands exist with
regard to the bandwidth, the impedance level and – due to the increasing miniaturization
and portability of modern systems – the power consumption. The tendency is favorable:
with increasing frequencies, the resistances must be made lower due to the unavoidable ca-
pacitances, which leads to a reduction of the source resistances in the circuits; furthermore,
36 For a > 2, the following applies: a 2 − 1 ≈ a − 1/(2a).
4.2 Properties and Parameters 461
1:n
2
Rg Fopt n Rg LTr Fopt
R gopt R gopt
n Rg ≈ R gopt F ≈ Fopt
2
Noise matching with a transformer: If no DC voltage gain is needed and the demands
with respect to noise are particularly high, then a transformer can be used for noise match-
ing; it will transform the value of the internal resistance Rg to a value in the region of the
optimum source resistance Rgopt . This method is used with very small internal resistances
(Rg < 50 ) in particular, since amplifiers with a correspondingly low optimum source
resistance are not available. Figure 4.162 shows the transformation of Rg to n2 Rg by means
of a 1:n transformer; a numeric example is presented at the end of Sect. 2.3.4.
The lower cutoff frequency is determined by the inductance of the transformer:
n2 Rg
fL =
2π LTr
Therefore, in LF applications it is necessary to use transformers with a high inductance
and correspondingly large dimensions, which is impractical in most applications; how-
ever, transformers for the frequency range from 1 MHz to 1 GHz are available as SMD
components, with a volume of 0.1 . . . 0.5 cm3 .
Noise matching with a resonance transformer: For high frequency and a low band-
width, a special resonance transformer can be used instead of the standard transformer.
Very often, a π element with two capacitances and one inductance, known as a Collins
filter or a Collins transformer, is used. An RF amplifier usually contains two resonance
transformers; one at the output for power matching and one at the input for power or noise
matching. Figure 4.163 shows one version with discrete components and one with strip
lines. Dimensioning is discussed in more detail in Sect. 27.2.7.
L2
L1
C2o C2b
C1o C1b
Fig. 4.163. Noise matching at the input and power matching at the output using Collins filters
Procedure for calculating the equivalent noise sources: Every noise source of an
amplifier can be converted into an equivalent noise voltage source and an equivalent noise
current source at the amplifier input. This is done in four steps:
– Calculation of gain A = vo /vi using a driving voltage from an ideal voltage source
(vi = vg ), and calculation of the transimpedance RT = vo / ii using a drive signal
from an ideal current source (ii = ig ). Since vi = ii ri , we have RT = A ri ; this is of
importance since in the basic circuits we have only calculated A and ri , and not RT .
– Calculation of the short-circuit output voltage
vo,S = AS,x vr,x or vo,S = RS,x ir,x for vi = 0
and the open-circuit output voltage
vo,O = AO,x vr,x or vo,O = RO,x ir,x for ii = 0
for each noise source vr,x and/or ir,x .
– Calculation of the equivalent noise voltage
AS,x vr,x RS,x ir,x
vr,0x = or vr,0x =
A A
and the equivalent noise current
AO,x vr,x RO,x ir,x
ir,0x = or ir,0x =
RT RT
for each noise source vr,x and/or ir,x .
– Calculation of the noise densities of the equivalent noise sources:
0 0
vr,0 = vr,0x ⇒ |v r,0 |2 = |v r,0x |2
x x
0 0
ir,0 = ir,0x ⇒ |i r,0 | = 2
|i r,0x |2
x x
This assumes that the noise sources vr,x and/or ir,x are independent; thus the equivalent
noise sources vr,0x and/or ir,0x are also independent and the noise densities can be added.
Figure 4.164 shows the first three steps of this procedure, using the example of a noise
current source ir,x .
4.2 Properties and Parameters 463
vg vo vo
ig
vo = Avg vo = R T ig
v o,S vo,O
i r,x i r,x
vo,S vo,O
vr,0x i r,0x
Fig. 4.164. Method for calculating the equivalent noise sources for a noise current source ir,x
In general, every noise source contributes to the equivalent noise voltage source and the
equivalent noise current source; therefore, strictly speaking, the equivalent noise sources
are always dependent. However, the ratio of the values is mostly such that each noise
source contributes significantly to only one equivalent noise source, while its contribution
to the other equivalent noise source is negligibly low; thus the equivalent noise sources are
practically independent.
Vb Vb
i r,R1 R1 i r,RC RC
R1 RC
v r,T
vo
vi
v BE i r,T rBE g m vBE
Vo
Vi
R2 RE
i r,R2 R2 i r,RE RE
The current feedback primarily affects the equivalent noise voltage density. By way
of contrast, the internal resistance Rb of the base voltage divider only influences the
noise current density; this influence is negligible for Rb 2rBE . It should be noted
that (4.204) uses the sum of the feedback resistance RE and the base spreading resistance
RB ; therefore, all noise equations for bipolar transistors can be used if we replace RB
with RB + RE . However, RE is not usually an independent parameter, but is correlated
to the transconductance via the loop gain kE = gm RE ; in this case, the calculation of the
optimum quiescent current leads to:
β≈100
VT β Rg >RB 0.26 V
IC,A opt (Rg ) = ' 1 + 2kE ≈ 1 + 2kE
Rg2 + RB2 Rg
Due to the current feedback, the optimum noise figure increases to:
RB 1 RB 2
Fopt,T (Rg ) = 1 + + 1+ 1 + 2kE
Rg β Rg
Therefore, no current feedback is used if optimization is based on noise alone. In practice,
it is often the amplitude limit, given by an permissible distortion factor or intermodulation
ratio, that must be optimized. The potential gain in amplitude due to the linearizing effect
of the current feedback may be higher than the loss caused by the increased noise. An
example of this kind of application is the receiving amplifier of a mobile telephone, which
may receive extremely varying input signals depending on the distance to the transmitter;
here, it is necessary to sacrifice sensitivity in order to enable high input signals to be
processed with low intermodulation.
Common-source circuit with current feedback: The equivalent noise densities of the
common-source circuit with current feedback in Fig. 4.166 correspond to those of the
common-emitter circuit with current feedback; but here the equivalent noise current density
is determined by the gate voltage divider, since the noise current density of the MOSFET
is so low that it can be ignored. If Rb = R1 || R2 and the substrate transconductance is
disregarded, then:
kS =gm RS 8kT
3
|v r,0 |2 ≈ |v r,F |2 + 4kT RS = 1 + kS (4.206)
3gm 2
Vb Vb
R1 RD
Vo
Vi
R2 RS
4kT 4kT
|i r,0 |2 ≈ |i r,F (f )|2 + ≈ (4.207)
Rb Rb
Here, kS = gm RS is the loop gain. It is necessary to select Rb as high as possible to prevent
the noise figure from declining considerably in the region of high source resistances.
In the common-source circuit direct optimization is not possible, as there are no opposed
values; here, it is necessary to make gm and Rb as high as possible and the loop gain kS as low
as possible. In LF applications the minimum for the noise figure is not very pronounced.
If kS is given by the permissible distortion factor, then the transconductance should be
increased until the noise figure drops below the desired limit value. This may be achieved
by using the lower limit Rg,lF for the source resistance and taking into consideration the
additional factor
3
1 + kS
2
from (4.206); if Rg = Rg,lF , then:
2 3
gm = 1 + kS
3Rg (F − 1) 2
With increasing frequencies the minimum becomes more pronounced; in this case, we
use the equation for the optimum source resistance in wide-band applications – that is,
Rgopt,F [fL , fU ] – and, likewise, take into consideration the additional factor:
fT 3
gm = √ 1 + kS
3Rg fU 2
Common-emitter circuit with voltage feedback: Figure 4.167 shows the common-
emitter circuit with voltage feedback and the corresponding small-signal equivalent circuit
with all noise sources. In this circuit the resistance R1 shown in Fig. 2.66 on page 108
is omitted, since the source resistance Rg takes over the function of R1 ; however, Rg
belongs to the signal source and therefore does not contribute to the noise of the circuit.
An additional resistance R1 is undesirable, as it reduces the gain and increases the noise
figure.
i r,R2
Vb
i r,RC RC
RC R2
R2
vr,T
Vo vo
Vi vi
i r,T vBE rBE g m vBE
Vb Vb Vb Vb Vb
R2 R2
Vo
Vo
Vi Vi
Fig. 4.168. A practical design for a low-noise common-emitter circuit with voltage feedback
The common-source circuit with voltage feedback: The equivalent noise densities of
the common-source circuit with voltage feedback correspond to those of the common-
emitter circuit with voltage feedback; but here the equivalent noise current density is
determined by the feedback resistance, since the noise current density of the MOSFET is
negligibly low:
8kT
|v r,0 |2 ≈ |v r,F |2 = (4.210)
3gm
4kT 4kT
|i r,0 |2 ≈ |i r,F (f )|2 + ≈ (4.211)
R2 R2
The practical design is shown in Fig. 4.168; here, however, it is not essential that an
impedance converter in the form of a common-drain circuit follows, as a common-source
circuit also features a high input resistance. This circuit is ideal for high-resistance sources
provided that the upper cutoff frequency remains far below the transit frequency. This
circuit is preferred in optical receivers for frequencies around 10 MHz; Fig. 4.169 shows
the corresponding circuit. The feedback resistance R2 , together with the capacitance CD
of the photodiode and the gate–source capacitance CGS , form a lowpass filter that limits
the bandwidth; therefore, in practical applications R2 must be chosen according to the
4.2 Properties and Parameters 469
Vb Vb Vb
R2
Vo
CD CGS
Fig. 4.169. A practical design for an optical
detector circuit with a photodiode
required bandwidth. Since photodiodes are high-resistance components, only the noise
current density has an influence; by means of the sensitivity of the diode, the noise current
density is converted to a corresponding illuminance, known as the noise equivalent power
(NEP) [4.7].
Common-collector and common-drain circuits: Figure 4.170 shows the basic design
of both a common-collector and a common-drain circuit used as impedance converters.
The noise sources of the common-collector circuit are:
% &
gm RE vr,T + RE ir,T + ir,RE gm RE 1 ir,T + ir,RE
vr,0 = ≈ vr,T +
gm R E + 1 gm
ir,RE
ir,0 = ir,T +
β
If gm RE 2 and β 1, then the equivalent noise densities are:
2kT
|v r,0 |2 ≈ |v r,T |2 = + 4kT RB (4.212)
gm
2kT gm
|i r,0 |2 ≈ |i r,T |2 = (4.213)
β
Vb Vb
Vi
Vi
Vo Vo
RE RS
a Common-collector b Common-drain
circuit circuit Fig. 4.170. Impedance converters
470 4 Amplifiers
Vb Vb
RC RD
Vo Vo
Vi RE Vi RS
Vb Vb
Fig. 4.171. Common-base and
a Common-base circuit b Common-gate circuit common-gate circuits
4.2 Properties and Parameters 471
additional base spreading resistance, so that the equivalent noise voltage density increases
accordingly; the same applies to the resistance RBV in Sect. 2.4.3.
The corresponding values for the common-gate circuit are:
8kT
|v r,0 |2 ≈ |v r,F |2 = (4.218)
3gm
4kT 4kT 1 1
|i r,0 | ≈ |i r,F (f )| +
2 2
+ ≈ 4kT + (4.219)
RS RD RS RD
The noise current density of the MOSFET is negligible.
Current source: For a current source, the noise current density at the output is of interest;
it should be so low that the noise figure of the circuit to which the current source is
introduced shows no or very little increase. Usually, a current source is used instead of a
high-value resistor; for example, a collector or drain resistor. In the small-signal equivalent
circuit of a circuit, this is shown by the output resistance ro and a noise current source;
the noise current density |i o,r |2 is significantly greater than that of a corresponding ohmic
resistor:
4kT
|i o,r |2
ro
Therefore, a current source may drastically increase the noise figure of a circuit, even if
an ohmic resistance at the same location had no notable influence.
Figure 4.172 shows the circuit and the small-signal equivalent circuit of a current
source based on a simple current mirror with current feedback. A strict analysis yields the
result
!
β vr,T 1 + ir,R1 R1
io = ir,RV ri + gm1 ri + vr,T 2
rBE 2 + βR2 + ri 1 + gm1 R1
"
+ ir,T 2 (ri + R2 ) + ir,R2 R2
Vb
RV i r,RV RV v r,T2 vo
Io
io
gm1vr,T1 1 i r,T2 rBE2 gm2 vBE2
v
T1 T2 gm1 BE2
Vo
R1 R2 i r,R1 R1 i r,R2 R2
where
RV (1 + gm1 R1 )
ri =
1 + gm1 (RV + R1 )
is the internal resistance of the left branch. Here, our observations are restricted to the case
of cross-symmetrical transistors and resistors with the current ratio:
gm2 R1
kI = =
gm1 R2
Furthermore, we assume that RV 1/gm1 + R1 ; thus ri ≈ 1/gm1 + R1 and:
β gm2 * % & +
io ≈ vr,T 1 + vr,T 2 + kI ir,R1 + ir,R2 + (1 + kI ) ir,T 2 R2
β + kI 1 + gm2 R2
In general, the current ratio is much smaller than the current gain: kI β. Consequently:
2 1 # $ 2
gm2
|i o,r | ≈
2
(1 + kI ) |v r,T 2 |2 + 4kT R2 + (1 + kI )2 |i r,T 2 |2 R22
1 + gm2 R2
An approximation for this expression can be reached by examining the extreme situations
without feedback (R2 = 0) and with high feedback (gm2 R2 1); this provides accurate
results for the extreme situations and only deviates very slightly from the exact value in
the region gm2 R2 ≈ 1:
1 + kI
|i o,r |2 ≈ + (1 + kI )2 |i r,T 2 |2
1 1
+
2
gm2 |v r,T 2 |2 |i r,R2 |2
4kT (1 + kI ) 2kT gm2
≈ + (1 + kI )2 (4.220)
2 β
+ R2
gm2 (1 + 2gm2 RB2 )
The noise current density declines with increasing feedback, since the denominator of the
first term increases with R2 ; the lower limit is determined by the second term. For practical
applications, a current ratio of kI = 1 is selected in order to prevent an unnecessary increase
in the noise current density. A ratio of kI < 1 may be selected for highly demanding
applications.
For kI ≤ 1, both terms in (4.220) have approximately the same size for gm2 R2 ≈ β,
which means that a further increase in R2 has very little effect; the voltage drop across R2 is:
gm2 R2 ≈β β≈100
IC2 R2 = gm2 R2 VT ≈ βVT ≈ 2.6 V
In practice, it is seldom possible to select the voltage drop across R2 to be so high that the
lower limit is reached; therefore, especially in circuits with very low supply voltages, no
low-noise current sources can be realized.
Without feedback or with low feedback, the noise voltage densities of the transistors
are of dominating influence. In this case, large transistors can be used to keep the base
spreading resistance, and thus the noise voltage densities, low; however, this causes an
increase in the output capacitance. With medium and high feedback, the influence of the
noise voltage densities of the transistors is low; the transistor size can then be chosen
according to the normal scaling.
4.2 Properties and Parameters 473
|i o,r | ro
pA/ Hz MΩ
Bipolar: ro
20
10
MOS : ro
2
Bipolar: |i o,r |
1
MOS : |i o,r |
0.5
~
~
Fig. 4.173. Noise current densities and output resistances of a bipolar and a MOS current
source with Io = 100 mA as a function of the output voltage limit Vo,min (VCE,sat = 0.2 V,
VDS,po = 0.5 V)
Vb Vb Vb Vb
RV RV RV
Io Io
Io
T3 T4 T3
T3 1 1 1
1
Vo Vo Vo
T1 T2 T1 T2 T1 T2
10 10 10 10 10 10
R R R R R R
source; this is the case even after the MOSFETs have been changed in size, which means
that in terms of noise the MOS current source is always slightly better. The bipolar current
source, however, has a significantly higher output resistance; for this reason it is always to
be preferred in situations that require a high output resistance.
Figure 4.174 shows other current sources with bipolar transistors, on the basis of three-
transistor, cascode and Wilson current mirrors. The noise current densities differ only in
terms of their lower limit at high feedback: for the three-transistor current mirror this limit
is lower than in the current source with a simple current mirror, while in the other two
current mirrors the limit is higher than in the current source with a simple current mirror.
The differences are small and of no significance in practice, as such high feedbacks can
only seldom be used in practice. Without feedback or with low feedback, the noise current
density again depends on the size of transistors T1 and T2 , while the sizes of transistors T3
and T4 have no noticeable influence. For this reason, one can optimize the current sources
with cascode or Wilson current mirrors without feedback by selecting large transistors for
T1 and T2 to minimize the noise current density and keeping the other transistors small to
minimize the output capacitance (see Fig. 4.174b,c). With the same output voltage limit,
the output resistance of the current source with cascode or Wilson current mirrors is higher
than for the current source with a simple current mirror; however, the noise current density
is also higher. Therefore, one must check which parameter is more important for the given
application.
Vb Vb
R1 R2
T1 T2 i o,r ro
v r,T3
RV
vo
vi = vBE3
Vo i r,T3 rBE3 gm3 vBE3
T3
Vi
follows:
|i o,r |2 gm3 R2 2+2kI
|v r,0 |2 = |v r,T 3 |2 + 2
≈ |v r,T 3 |2 (4.222)
gm3
|i o,r |2
|i r,0 |2 = |i r,T 3 |2 + ≈ |i r,T 3 |2 (4.223)
β32
The equivalent noise current density is hardly increased by the current source; this is also
the case without current feedback (R1 = R2 = 0). In contrast, a current feedback with
gm3 R2 2 + 2kI is required in order to prevent the equivalent noise voltage density
from increasing significantly; without current feedback (R1 = R2 = 0) and under the
assumption that the base spreading resistances RB2 and RB3 are equal, it follows:
kI =1
|v r,0 |2 ≈ (2 + kI ) |v r,T 3 |2 = 3 |v r,T 3 |2
This also applies to the common-source circuit.
If the common-emitter or common-source circuit already includes current feedback
through resistance RE or RS , then the reduced transconductance
gm gm
gm,red = or gm,red =
1 + g m RE 1 + g m RS
must be used to convert the noise current density of the current source; this increases
its influence. However, the noise voltage density is also increased by resistances RE and
RS ; here, the noise of the current source for R2 (1 + kI )RE or R2 (1 + kI )RS is
negligible.
Common-collector and common-drain circuits with a current source: Here, the same
conditions apply as for the common-emitter and common-source circuits; that is, the equiv-
alent noise current density is hardly increased. However, current feedback of the current
source is required in order to maintain the equivalent noise voltage density. In practice, how-
ever, current sources without current feedback are usually used, since common-collector
476 4 Amplifiers
Vb
RC v r,T2 i r,RC RC
T1
Vi v i = vBE1
i r,T1 rBE1 gm1 vBE1 rCE1
and common-drain circuits are generally used as impedance converters if high source re-
sistances are involved; in this case, the noise figure depends primarily on the equivalent
noise current density, so that an increase in the equivalent noise voltage density by a factor
of three has almost no effect on the noise figure.
Cascode circuit: Figure 4.176 shows the circuit and the small-signal equivalent circuit
of a cascode circuit with bipolar transistors. Since the small-signal characteristics of cas-
code and common-emitter circuits are essentially the same, one can use the equations for
the common-emitter circuit; in particular, the noise of the collector resistance RC can be
disregarded. The noise sources of transistor T2 are also negligible:
– The noise voltage source vr,T 2 has virtually no effect, as the current is impressed by
T1 and there is very little change in the current distribution due to the collector–emitter
resistances rCE1 and rCE2 .
– The noise current source ir,T 2 affects the output in the same way as source gm1 vBE 1
and is therefore converted to the input by means of current gain β; this can be neglected
compared to ir,T 1 .
For the cascode circuit, it follows:
|v r,0 |2 ≈ |v r,T 1 |2 (4.224)
|i r,0 |2 ≈ |i r,T 1 |2 (4.225)
For versions with current feedback or a base voltage divider, the equations of the corre-
sponding common-emitter circuit can be used. Similarly, the equations for the common-
source circuit apply to the cascode circuit with MOSFETs.
Differential amplifier: Figure 4.177 shows the circuit design and the small-signal equiv-
alent circuit of a differential amplifier with bipolar transistors and a resistor for setting the
quiescent current. For the sake of clarity, the small-signal equivalent circuit of the transis-
tors is shown in the form of a block diagram.
4.2 Properties and Parameters 477
Vb Vb
RC RC i r,RC RC RC i r,RC
T1 T2 T1 T2
Vi1 Vi2 vi1 i r,T1 i r,T2 v i2
R0 R0
i r,R0
–Vb
The differential amplifier can be reduced to a common-emitter circuit; this method has
already been utilized in Sect. 4.1.3. It follows that the noise of the collector resistances can
also be neglected in the case of the differential amplifier. The two noise voltage sources
vr,T 1 and vr,T 2 are combined to form one equivalent noise voltage source
|v r,0 |2 = |v r,T 1 |2 + |v r,T 2 |2 (4.226)
which is arranged in front of one of the two inputs. This is possible regardless of the
external circuitry, as the noise voltage sources enter directly into the equivalent noise
source. By way of contrast, the contribution of the noise current sources depends on the
source resistances of the two inputs; this is the reason why – provided that the transistors
are equal – an equivalent noise current source with
|i r,01 |2 = |i r,02 |2 = |i r,T 1 |2 = |i r,T 2 |2 (4.227)
is connected to both inputs. The influence of the noise current source ir,R0 depends on the
circuitry at the output and is discussed separately; in most cases it can be neglected.
The noise equivalent circuit shown in Fig. 4.178 is obtained with two signal sources;
for the equivalent noise source, it thus follows:
vo1 vo2
Signal source 1 Signal source 2
vr,g1 v r,0 Differential vr,g2
Rg1 amplifier Rg2
Signal-to-noise ratio: The SNR is the ratio of the useful signal power to the noise power:
Pus
SNR = (4.229)
Pn
Pn is the noise power in the frequency interval fL < f < fU . Since the power of a
signal is proportional to the square of its effective value (the root-mean-square value), the
signal-to-noise ratio of the signal generator is:
2 2
vgeff vgeff
SNRg = 2
= fU
(4.230)
vr,geff
|v r,g (f )|2 df
fL
The amplifier increases the noise density by the spectral noise figure F (f ); the signal-to-
noise ratio at the amplifier input is thus
2 2 2
vgeff vgeff vgeff
SNRi = 2
= fU
= fU
(4.231)
vreff
|v r (f )|2 df F (f ) |v r,g (f )|2 df
fL fL
which is lower than the signal-to-noise ratio of the signal generator.
Mean noise figure: The mean noise figure corresponds to the relation of the signal-to-
noise ratios [2.9]:
fU
F (f ) |v r,g (f )|2 df
SNRg fL
F = = fU (4.232)
SNRi
|v r,g (f )|2 df
fL
Usually, it is quoted in decibels (dB):
FdB = 10 log F
If SNRi,dB = 10 log SNRi and SNRg,dB = 10 log SNRg , then:
FdB = SNRg,dB − SNRi,dB
When operated with an ideal signal generator with the frequency-independent noise
density |v r,g (f )|2 = 4kT Rg , this expression can be moved in front of the integral, so that:
fU
1
F = F (f )df (4.233)
fU − fL fL
In this case the mean noise figure F is obtained by averaging across the spectral noise
figure F (f ). Often, F (f ) is constant in the given frequency interval; thus F = F (f ),
which is generally referred to as the noise figure F .
480 4 Amplifiers
Use of weighting filters: In some applications a weighting filter with the transfer function
H B (s) is used for determining the noise power; the signal-to-noise ratios are then calculated
with the weighted noise density:
|v r(B),g (f )|2 = |H B (j 2πf )|2 |v r,g (f )|2
Thus:
2
vgeff
SNRB,g = fU
|H B (j 2πf )|2 |v r,g (f )|2 df
fL
2
vgeff
SNRB,i = fU
F (f ) |H B (j 2πf )|2 |v r,g (f )|2 df
fL
This is used if the noise causes stronger interference in certain ranges of the given frequency
interval than in other ranges. The weighting filter, whose absolute frequency response is
proportional to the disturbing effect of the noise, provides more meaningful values for the
SNRs. A weighted noise figure can also be introduced, but this is not common.
A typical use for a weighting filter is to determine the signal-to-noise ratio of an audio
amplifier. Such amplifiers usually operate in the frequency range 20 Hz < f < 20 kHz.
Since the human ear is particularly sensitive to noise in the range of 1 kHz < f < 4 kHz,
a weighting filter is used that enhances this range and suppresses other frequency ranges.
This filter is known as an A filter; the SNR is thus quoted in decibel A or dBA.
Bandwidth of the amplifier: The bandwidth of an amplifier must at least cover the given
frequency interval of the useful signal fL < f < fU to provide equal amplification; this
means that the noise is also equally amplified in this range. In practice, the bandwidth is
usually wider than required; that is, the amplifier also amplifies the ranges f < fL and
f > fU with its operating gain AB (s). These ranges do not contain any useful signal,
but only the noise of the signal generator and the amplifier. This means that without any
limitation of the frequency range the amplifier provides the noise power at its output:
∞
Pn,o = |AB (j 2πf )|2 |v r (f )|2 df
0
∞
= |AB (j 2πf )|2 F (f ) |v r,g (f )|2 df (4.234)
0
As the useful signal is amplified with the useful gain AB,us , which is assumed to be
constant in the range fL < f < fU , the signal-to-noise ratio at the amplifer output
without limitation of the frequency range is:
|AB,us |2 vgeff
2
|AB,us |2 vgeff
2
SNRo = = ∞ (4.235)
Pn,o
|AB (j 2πf )|2 F (f ) |v r,g (f )|2 df
0
This is lower than the SNR obtained using (4.231), since the total noise is taken into
account, rather than just the portion in the range fL < f < fU .
In practice, the noise power Pn,o plays a major role, as it may be considerably higher
than the useful power if the bandwidth of the amplifier is correspondingly large; as a
4.2 Properties and Parameters 481
2
| AB | Equivalent noise
bandwidth Br
2
| A B,us |
0
~
consequence, the following components of the signal processing chain are primarily driven,
and in some cases even overdriven, by the amplified noise.
The signal-to-noise ratio SNRo is only of importance if the noise outside the range
fL < f < fU is transmitted to the output of the signal processing chain and actually
causes interference here. In this event, the weighting filter comes into play: the signal-
to-noise ratio SNRi is obtained using (4.231) by introducing an additional ideal bandpass
filter with the lower cutoff frequency fL and the upper cutoff frequency fU into (4.235).
Equivalent noise bandwidth: If the noise densities of the signal generator and the
amplifier are almost constant within the transfer bandwidth, which means that the noise
figures are also almost constant, then the noise power at the amplifier output is:
∞
Pn,o ≈ F |v r,g | 2
|AB (j 2πf )|2 df = F |v r,g |2 |AB,us |2 Br
0
The bandwidth
∞
|AB (j 2πf )|2 df
Br = 0
(4.236)
|AB,us |2
is known as the equivalent noise bandwidth. It indicates the bandwidth of an ideal filter
with a gain of AB,us which has the same noise power Pn,o as the amplifier. This means that
the area under the plot of the squared magnitude |AB (j 2πf )|2 is replaced by a rectangle
of the same area with height |AB,us |2 and width Br (see Fig. 4.179).
In the following example, we will calculate the equivalent noise bandwidth of an
amplifier with a transfer function that corresponds to a lowpass filter of first order:
A0 A20
AB (s) = ⇒ |AB (j 2πf )|2 = 2
s f
1+ 1+
ωg fg
For |AB,us |2 = A20 , it follows:
∞
1 f ∞ π
Br = 2 df = fg arctan = fg ≈ 1.57 · fg
0 f f g 0 2
1+
fg
482 4 Amplifiers
Order Br /f-3dB
Multiple pole Butterworth
1 π/2 = 1.57 π/2 = 1.57
2 1.22 1.11
3 1.15 1.05
4 1.13 1.03
5 1.11 1.02
The equivalent noise bandwidth of a first-order lowpass filter is higher than the cutoff
frequency fg by a factor of 1.57, which in this case corresponds to the 3 dB cutoff frequency
f-3dB . Figure 4.180 includes the factors for lowpass filters of higher orders for cases
with multiple poles and for a Butterworth characteristic with a maximum flat magnitude
frequency response. In this and most other cases found in practice, the equivalent noise
bandwidth is larger than the 3 dB bandwidth; however, it may also be smaller.
The equivalent noise bandwidth is not often used in practical calculations; instead,
the 3 dB bandwidth is used and the slight error that occurs especially with higher orders
is accepted. For measuring noise densities, however, the equivalent noise bandwidth is
of high importance; the frequency range of interest is scanned with a bandpass filter of
narrow bandwidth and, with the help of the equivalent noise bandwidth, the noise power
at the output is converted into the noise density to be determined.
Chapter 5:
Operational Amplifiers
5.1
General
Operational amplifiers are available in a wide variety of monolithic integrated circuits and
differ little from discrete transistors in terms of size and price. As their characteristics are
ideal in many respects, they are much easier to use than discrete transistors. The advantage
of the classical operational amplifier is its high accuracy at low frequencies. But for many
applications it is too slow. For this reason, versions with a modified architecture were de-
veloped that feature desirable high-frequency characteristics. Today, there are virtually no
fields of application in which discrete transistors offer advantages. The detailed descrip-
tion of the internal design of operational amplifiers provided in this chapter is intended
only to explain certain properties and characteristics of integrated circuits. Nowadays, cir-
cuit design on a transistor level is only needed for the development of specific integrated
circuits.
Today, a vast selection of operational amplifiers is available; they differ not only in their
data but also in their basic construction. Four families can be distinguished, whose internal
design and effect on the parameters will be described in the sections below. Section 5.6
compares the four different versions in order to establish their common features and their
differences.
Circuit calculations are performed on models that are explained in respect to the internal
design. It is, of course, not possible to explain every single transistor, since this would render
the circuit analysis too complicated. Instead, macromodels that allow an easy interpretation
of the entire circuit are used. Depending on the effect to be analyzed, only the relevant
section of the circuit is modeled in more detail. In many cases, the calculation of operational
amplifier circuits is so simple that the quickest way to do it is by hand. With the aid of
macromodels, the behavior of a circuit can be studied in more detail using simulation
programs such as PSpice. This approach provides information on the capability of the
484 5 Operational Amplifiers
Io
VD
Vo
VP VN
O 1 14 O
1 8 O 1 8 + N 2 – – 13 N
V
+ +
2 7 V
+
2 7 O P 3 12 P
N – N –
+ –
P 3 +
6 O P 3 +
6 N V 4 11 V
–
5 + 10 P
V
–
4 5 V
–
4 +
5 P P +
– –
N 6 9 N
O 7 8 O
Fig. 5.2. Typical terminal configuration of operational amplifiers in dual inline cases, seen from top
circuit as early as in the design phase. The hardware construction is not carried out until
the simulation results are satisfactory.
Figure 5.1 shows the graphic symbol for the operational amplifier. It has two inputs –
one inverting and the other noninverting – and one output.
The ideal operational amplifier only amplifies the differential voltage VD = VP − VN
between the input terminals. The noninverting input is called the P-input and is identified
by the plus (+) sign in the graphic symbol; the inverting input is called the N-input and is
identified by the minus (−) sign. For the power supply of the operational amplifier, there
are two supply voltage terminals which provide operating voltages that are positive and
negative in respect to ground in order to allow quiescent input and output potentials of 0 V.
The operational amplifier has no ground terminal, even though the input and output voltages
are related to ground. Typical supply voltages are ±15 V for general applications; but
today’s tendency is to use voltages of ±5 V and there is a trend to reduce the voltages further.
Typical pin-outs for operational amplifiers are illustrated in Fig. 5.2. As many circuits
require more than one operational amplifier, there are dual and quadruple operational
amplifiers on the market to save space and money.
5.1.1
Types of Operational Amplifier
As shown in Fig. 5.3, there are four different types of operational amplifier. The differences
are the high-resistive or low-resistance inputs and outputs.All four types have high-resistive
noninverting inputs.
In the standard operational amplifier (the voltage feedback operational amplifier) the
inverting input also has high resistance, which means that it is voltage controlled. Its output
5.1 General 485
Vo Io
VD VD
V o = ADVD I o = gm,DVD
Vo Io
VD VD
IN IN
Vo = IN Z = ADVD I o = kI IN = gm,DVD
Fig. 5.3. Graphic symbols and transfer equations for the four types of operational amplifier
acts like a voltage source with a low internal resistance; that is, a low-resistance output. For
this reason, standard operational amplifiers are also known as VV operational amplifiers.
The first V (voltage) indicates the voltage control at the (inverting) input, while the second
V indicates the voltage source at the output. At one time, only this type existed; even today,
it is still the most important type and it holds the highest market share. The output voltage
Vo = AD VD = AD (VP − NN ) (5.1)
is identical to the amplified input voltage difference; AD is the differential gain. Values of
AD = 104 ...106 are necessary to create circuits with high negative feedback. The transfer
characteristics of an ideal VV operational amplifier are shown in Fig. 5.4a. The differential
gain
dVo
AD = (5.2)
dVD
b
corresponds to the slope of the curve in the operating or bias point. It is obvious that
fractions of 1 mV are sufficient to obtain the maximum output swing. The linear operating
range Vo, min < Vo < Vo, max is called the output voltage swing. After this limit is reached,
an increase in VD causes no further increase in Vo ; that is, the amplifier is overdriven. The
literature often specifies an ideal operational amplifier that features a differential gain of
AD = ∞; here, we do not wish to share this point of view, since it makes a thorough
understanding more difficult.
Unlike the normal operational amplifier, however, the operational transconductance
amplifier also has high-resistance output. This acts like a current source, whereby the
current is controlled by the input voltage difference VD . Therefore, its graphic symbol
shows a current source symbol at the output (see Fig. 5.3). This is an operational amplifier
486 5 Operational Amplifiers
with a voltage-controlled inverting input and an output that acts like a current source. The
transconductance amplifier is thus also known as a VC-OPA. The output current
Io = gm,D VD = gm,D (VP − VN ) (5.3)
is proportional to the input voltage difference. The differential transconductance
dIo
gm,D = (5.4)
dV
D b
indicates the degree of output current increase with a rising input voltage. The differential
transconductance generally corresponds to the transconductance of a transistor, and here
too a transistor is used to determine the transconductance. The term “transconductance
amplifier” is based on the fact that the transconductance gm,D governs the behavior of
the amplifier. Figure 5.4b shows the typical transfer characteristic of a VC operational
amplifier. It can be seen that small voltage differences are sufficient to achieve the maximum
output current swing.
The two operational amplifiers with current input in Fig. 5.3 have a low-resistance
inverting input; that is, they are current controlled. At first glance this seems to be a
drawback, but for high frequencies it provides major advantages, as we will see below,
since:
– The internal signal path is shortened and the oscillation tendency reduced.
– The gain of the OPA can be adapted to the specific demand.
The transimpedance amplifier (current feedback amplifier) shown in Fig. 5.3 has a
current-controlled inverting input and a voltage source at the output; it is thus a CV-OPA.
The output voltage
Vo = AD VD = IN Z (5.5)
can be calculated either from the differential gain – as with the standard OPA – or from
the input current IN and an internal impedance Z in the megaohm range. Due to this
characteristic impedance Z, the CV-OPA is also known as the transimpedance amplifier.
The current amplifier (diamond transistor, drive-R amplifier) has a current-controlled
input like the CV-OPA and a current-controlled output like the VC-OPA. Therefore, it is
known as a CC-OPA. The transfer response
Io = gm,D VD = kI IN (5.6)
5.1 General 487
dIo
kI = (5.7)
dIN
b
for calculations, because it is in the range kI = 1 . . . 10 depending on the type. The current
amplifier is also known as a diamond transistor (with the trade name of Burr Brown) as in
many respects it behaves like an ideal transistor (as will be shown in Sect. 5.5).
5.1.2
Principle of Negative Feedback
The VV operational amplifier is used as an example to explain the principle of negative
feedback. An operational amplifier with negative feedback can be regarded as a closed-
control circuit for which the rules of automatic control engineering apply. Figure 5.5 shows
the design of a general closed-control circuit. The target value is derived from the reference
variable using the reference variable shaper, given here as multiplication by kF . The actual
value is derived from the output variable, using the controller given here, as multiplication
by kR . In the controlled system, the difference between the target and the actual value is
multiplied by AD . The equation for the system deviation,
VD = kF Ve − kR Vi
VD
VD
kF = and kR = − (5.8)
Vi
Vo =0 Va
Vi =0
The gain of the control loop in Fig. 5.5 can be calculated from Vo = AD VD and (5.1.2):
kR AD 1
Vo kF AD kF
A = = ≈ (5.9)
Vi 1 + k R AD kR
In an operational amplifier circuit, the operational amplifier represents the controlled sys-
tem. The reference variable shaper and the controller are formed by the operational ampli-
fier’s external circuitry. The subtraction is achieved either by the differential input of the
differential amplifier or by the external circuitry.
operational amplifier
reference target output
variable reference value + Deviation Controlled variable
variable shaper +
kF Vi VD = kF Vi – kRVo system
Vi – Vo
AD
actual value
Controller
kR Vo
kR
Noninverting Amplifier
In the general control loop shown in Fig. 5.5, if the setpoint value is identical to the
reference variable and the controller is a voltage divider, then the circuit is a noninverting
amplifier, as shown in Fig. 5.6. For a qualitative investigation of the transient response,
the input voltage is changed from zero to a positive value Vi . In the first instant the output
voltage, and hence the feedback voltage, is still zero. Therefore, the voltage at the amplifier
input is VD = Vi . Since this voltage is amplified with the high differential gain AD , Vo and
thus the feedback voltage kR Vo rapidly assume positive values; this leads to a reduction in
VD . The fact that a change in the output voltage counteracts a change in the input voltage
is typical of the negative feedback principle. This suggests that a stable condition will
ultimately be reached.
In order to calculate the steady state values, it is assumed that the output voltage rises
until it reaches the same level as the amplified input voltage difference:
Vo = AD VD = AD (VP − kR Vo )
This leads to the voltage gain
⎧
Vo AD ⎨ 1 for kR AD 1
A = = = kR (5.10)
Vi 1 + k R AD ⎩
AD for kR AD 1
where
g = k R AD (5.11)
represents the loop gain. If the loop gain g 1, then the 1 in the denominator of (5.10)
can be disregarded, thus giving the gain of the feedback circuit:
Vo 1 RN
A = = = 1+ (5.12)
Vi kR R1
In this case it is determined by the external circuitry only and not by the amplifier. This
approximation can be derived directly from the circuit because a large loop gain results in
VD = 0; that is, VN = Vi . For the negative feedback voltage divider, it follows:
R1 Vo RN
Vi = Vo ⇒ A = = 1+
R1 + R N Vi R1
VD = VI – kR Vo
Vi Vo Vi VD
Vo
VN
Fig. 5.6. Control engineering model of the noninverting amplifier, using the VV-OPA as an example
5.1 General 489
This leads to the most important rule for calculating operational amplifier circuits:
The output voltage of an operational amplifier takes on a value at which the input
voltage difference is zero.
A condition for this is that the loop gain is high and that true negative feedback takes
place, with no positive feedback; otherwise a Schmitt trigger occurs, as described in
Sect. 6.5.2 on page 601. If the loop gain g 1, then A = AD as seen in (5.10); in
this case, the gain is not affected by the negative feedback.
A useful method for calculating the loop gain results from (5.11) and (5.12) if g 1:
AD
g = kR AD = (5.13)
A
In order to prevent the error caused by the approximation in (5.12) from exceeding 10/00 ,
a loop gain of g = 1000 is required. If the gain of the negative feedback circuit is to
be A = 100, the required differential gain can be calculated from (5.13): AD = gA =
1000 · 100 = 105 . This explains why the differential gain of the operational amplifier must
be as high as possible. Four different gains must be distinguished:
AD the open-loop gain of the amplifier
A the closed-loop gain of the amplifier
g the loop gain, g = AD /A
kR the feedback factor β
The literature sometimes specifies an additional gain – noise gain, which is the inverse
value of the feedback factor; this represents the gain determined by the external circuitry.
The loop gain may also be exemplified: we make Vi = 0 and interrupt the loop at the input
of the external circuitry as shown in Fig. 5.7a; then we introduce a test signal VS at this
point and measure the signal at the other side of the interruption; in other words, at the
amplifier output. As can be seen in Fig. 5.6,
Vo = − kR AD VS = − gVS (5.14)
The test signal is amplified with gain g = kR AD while passing through the open loop.
The loop can also be interrupted at the inverting input and a test signal introduced there
(see Fig. 5.7b). In this case, the gain is initially AD , followed by kR ; the loop gain is once
again g = kR AD .
AD AD
A D VD = – kR A D VS A D V D = – A D VS
VD VD
Vo Vo
RN ~ VS RN
kR VS
R1 VS ~ R1 kR Vo = – kR AD VS
The loop gain can also be measured in a closed loop. To this end, a voltage Vi is fed
into the input of the circuit and measured as VN and VD according to Fig. 5.6b. The ratio
of these two voltages represents the loop gain:
VN kR Vo k R Vo
= = = kR AD = g (5.15)
VD VD Vo /AD
Inverting Amplifier
In addition to the circuitry shown in Fig. 5.6, there is a second fundamental method of
making an operational amplifier to an amplifier with negative feedback. Here, the feedback
must, of course, come from the output toward the inverting input too, in order to achieve
negative and no positive feedback. However, it is also possible to supply the input voltage
to the base point of the feedback voltage divider instead of the noninverting input. This
leads to the circuit shown in Fig. 5.8. Inserting kF and kR into (5.9) results in
RN
− AD kR AD 1
Vo − k F AD R1 + R N RN
A = = = ≈ − (5.16)
Vi 1 + k R AD R1 R1
1+ AD
R1 + R N
This means that this is an inverting amplifier. This can also be seen in the circuit itself when
a positive input voltage is applied. Since this voltage reaches the inverting input via R1 ,
the output voltage becomes negative. In the ideal operational amplifier with AD = ∞, the
output voltage becomes negative until VD = 0; thus, this is also called a virtual ground.
In order to obtain the output voltage, the nodal equation is applied to the inverting input,
leading to
0 Vi Vo
= + = 0
R1 RN
I =0
This equation can directly be solved for Vo :
RN Vo RN
Vo = − Vi ⇒ A = = −
R1 Vi R1
Compared to the noninverting amplifier in Fig. 5.5, the voltage gain here is negative and its
value is reduced by 1. Of course, it is also possible to calculate the gain of the circuit shown
VD = kFVi – kRVo RN
R1
Vi Vo Vi VD
Vo
Fig. 5.8. Circuitry of an operational amplifier as an inverting amplifier, using the VV-OPA as an
example. The values shown for kF and kR are in accordance with the definitions in (5.8)
5.2 Normal Operational Amplifier (VV-OPA) 491
in Fig. 5.8, for a finite differential gain AD . In this case, one must take into consideration
the fact that VD = 0.
From
VE + VD Vo + VD
+ = 0
R1 RN
and Vo = AD VD , it follows:
Vo RN AD AD
A = = − = kF (5.17)
Vi R1 AD + R N + R 1 1 + k R AD
The loop gain determines the deviation from the ideal performance. For g = kR AD 1
it follows:
kF AD kF −RN /(R1 + RN ) RN
A = = = = − (5.18)
kR A D kR R1 /(R1 + RN ) R1
In the simplest configuration, the external circuitry consists of a voltage divider only,
as shown in Figs. 5.6 and 5.8. The use of an RC network produces an integrator, a dif-
ferentiator, or an active filter. Nonlinear components such as diodes can also be used in
the external circuitry to form exponential functions or logarithms. These applications are
described in Sect. 11.7 on page 739. The explanations here are limited to the simplest
ohmic feedback.
5.2
Normal Operational Amplifier (VV-OPA)
This section explains the various ways to design an operational amplifier so that the user
understands the practical consequences that result from the internal design. The purpose is
not to recommend building operational amplifiers from discrete transistors and resistors.
This would not only result in greater costs, but also clearly poorer data.
An amplifier that is to be used as an operational amplifier must meet several require-
ments, which determine its internal design. Nowadays, as shown in Fig. 5.3, there are four
different types of operational amplifier on the market, each of which comes in several
variations depending on the given field of application. All types comply with the following
general requirements:
– Direct voltage coupling.
– Differential input.
– Zero input and output voltage operating point
Operational amplifiers can be built with bipolar or field effect transistors, or with a
combination of both. The following description is generally based on bipolar transistors.
Differential amplifiers are mostly used for the input stage, because this compensates for
both the base–emitter voltages and the temperature sensitivity.
The use of npn transistors for the amplification renders the output potential of the
amplifier stage positive to the input potential. To reduce the output operating point to zero,
it is essential to shift the potential downwards in a stage of the amplifier. This can be
achieved by two fundamentally different methods, which are shown in Fig. 5.9.
– Zener diodes, as shown in Fig. 5.9a, cause virtually no attenuation of the desired sig-
nal, due to their low dynamic internal resistance. However, sufficient current must flow
492 5 Operational Amplifiers
+ +
– –
a Zener diode b Complementary
transistors Fig. 5.9. Methods for potential shift
through the Zener diode to insure that its noise does not reach too high a level. Conse-
quently, these devices should only be used after emitter followers. Another disadvantage
is their fixed potential shift, which does not adapt itself to the operating voltage. How-
ever, it is advantageous that only npn transistors are required, which makes this method
particularly suitable for high-frequency amplifiers.
– Complementary transistors, as shown in Fig. 5.9b, represent the easiest and most ele-
gant method of compensating the potential shift of the amplifier stage with that of the
subsequent stage. In most cases pnp transistors are used in the current mirror configu-
ration, as shown in Fig. 4.66 on page 342. A disadvantage here, however, is the fact that
the transit frequencies of pnp transistors in integrated circuits are often clearly poorer.
Only the more modern, more complex production processes enable the manufacture of
equivalent pnp transistors.
5.2.1
Principle
VV operational amplifiers have a voltage-controlled (in other words, high-resistance) input
and a low-resistance output. Therefore, it is convenient to use a differential amplifier at
the input and an emitter follower at the output. This creates the simplest VV operational
amplifier, as shown in Fig. 5.10. The circuit was only extended by an additional Zener diode
at the output to bring the quiescent output potential down to 0 V. Operational amplifiers
should meet the following three requirements:
– Common-mode voltage: almost up to the operating voltages.
– Output voltage swing: almost up to the operating voltages.
– Differential gain: as large as possible: AD = 104 . . . 106 .
In Fig. 5.10, the positive limit of the common-mode voltage (see Sect. 4.1.3 on
page 333) is reached at VN = VP = VGl = 7.5 V, because otherwise the collector–
base diode of T2 becomes conductive. The negative limit is given by the current source I0 .
If one assumes a minimum voltage drop of 1 V, the emitter potential of the differential am-
plifier may drop to −14 V. This results in a minimum common-mode voltage of −13.4 V.
Consequently, the limits of the common-mode voltage can be expressed in the inequality:
−13.4 V < VGl < +7.5 V.
The positive limit of the output voltage is reached when the transistor T2 becomes
nonconductive; the base potential of the emitter follower then rises to 15 V and the output
5.2 Normal Operational Amplifier (VV-OPA) 493
+ + 15 V +
RC
7.5 kΩ
T3 T1 rm T3
I0 + Iq I0 – Iq VP 1 1
0V 6.9 V 26 Ω
0V
T1 T2 0V VD RC Vo
rm Iq
VP VN 7.5 kΩ
Iq 3 mA
Vo VN 1
RC 26 Ω
I0 T2
1mA 5 kΩ
– 15 V – –
Fig. 5.10. Simple operational amplifier with dimensioning examples, and the resulting quiescent
voltages and quiescent currents. The emitter current is divided into two parts to allow the current
difference Iq to be indicated. In this and the following circuits, it is just as good to use a common
emitter current of Ik = 2I0
voltage to +7.5 V. The lower control limit is determined by T2 , because its collector poten-
tial cannot fall below 0 V, since otherwise the collector–base diode becomes conductive.
At this point the related output voltage is −7.5 V. For the output swing, this leads to the
inequality −7.5 V < Vo < +7.5 V. This range becomes even more unfavorable when a
positive common-mode voltage is applied. For VGI = 5 V, the negative output swing is
even limited to −2.5 V.
The model for the differential gain of the operational amplifier also shown in Fig. 5.10
is equal to that of the differential amplifier if one takes into consideration the fact that
the emitter follower features a voltage gain of almost 1 and the Zener diode causes no
attenuation. Therefore, we can use the model shown in Fig. 5.10 for calculating the volt-
age gain. If a differential input voltage VD is applied, this voltage also lies between the
transconductance resistances of the input transistors, producing a current
VD 1 IC 1 1 mA mA
Iq = = VD = VD = 19 VD
2 rm 2 VT 2 26 mV V
At the collector resistance RC this current causes a voltage change that is identical to the
output voltage, due to the almost constant potential shift between collector of T2 and output:
1 IC R C VRC 7.5 V
Vo = Iq RC = VD = VD = VD = 144 · VD
2 VT 2 VT 2 · 26 mV
One can see that the common-mode and the output voltage range, as well as the differ-
ential gain of the circuit, are not even close to reaching the desired values; the operational
amplifier shown in Fig. 5.10 therefore needs to be improved in all respects. A marked im-
provement is achieved by replacing the Zener diode used for potential shift with a current
mirror made of pnp transistors. Figure 5.11 illustrates this variation. Here, the common-
mode control is clearly improved, as the collector potential of T2 is close to the positive
operating voltage: −13.4 V < VGl < +14.4 V. At the same time the output swing is
improved, since the emitter follower at the output allows a swing close to the positive and
negative operating voltages. If a minimum voltage drop of 1 V is required for the current
sources, then we achieve −14 V < Vo < +13.8 V.
The differential gain of the circuit shown in Fig. 5.11 can easily be calculated if one
takes into account the fact that the current mirror T2 , T3 only reverses the direction of the
494 5 Operational Amplifiers
Fig. 5.11. Operational amplifier with a current mirror for potential shift
T1 T5
rS T3 T4
VP 1
26 Ω
VD r3 rCE4 Vo
rS Iq V1 V1 S4
26 Ω 100 kΩ
VN 1
26 Ω
T2
Fig. 5.12. Model of the amplifier shown in Fig. 5.11, with two amplifier stages
collector current of T2 . The necessary load resistance consists of the parallel arrangement
of all resistances r4,ges connected to the collector of T4 . In the event of ideal current sources
only rCE4 needs to be considered, because the input resistance of the emitter follower is
infinitely high in the no-load condition:
rCE4 1 IC2 VA 1 1 100 V
AD = = = µ = = 1923 (5.19)
2 rS 2 VT IC2 2 2 26 mV
The differential gain can, however, also be calculated on the basis of two amplifier
stages, as illustrated in Fig. 5.12. The first stage is formed by the differential amplifier,
with the transdiode T3 as collector resistance. It has the gain
r3 1 IC2 VT 1
A2 = = = (5.20)
2 rS 2 VT IC3 2
since both collector currents are equal. The second amplifier stage is formed by transistor
T4 , operated in common-emitter configuration with the voltage gain
IC4 VA 100 V
A4 = S4 rCE4 = = µ = = 3846 (5.21)
VT IC4 26 mV
assuming that the internal resistances of the current sources are infinite. Thus, in accordance
with (5.19), the differential gain of the entire operational amplifier is AD = 1923.
5.2.2
Multipurpose Amplifiers
The operational amplifier shown in Fig. 5.11 is far from being an ideal multi-purpose
operational amplifier because of its gain of only AD ≈ 2000. There are two methods for
substantially increasing the voltage gain:
5.2 Normal Operational Amplifier (VV-OPA) 495
+15 V + + +
14.4 V
T3 T4
13.8 V
T5
2I q
I0 – Iq I0 + Iq T6
T1 T2
Iq
VN VP 1 mA 3 mA Vo
I0 I0 I1 I2
1mA 1mA
–15 V – – –
Fig. 5.13. Operational amplifier with two stages for voltage amplification. The values shown are
examples of quiescent potentials and quiescent currents
– By increasing the internal resistance at the collector of T4 . This node is the high-
impedance point of the circuit, which means that it is the point of the highest ohmic
resistance in the signal path. Its impedance determines the voltage gain and the cutoff
frequency of the circuit. The internal resistance of this high-impedance point can be
increased by using cascode circuits; this method is used in the wide-band amplifiers
described in Sect. 5.2.6.
– By amplifying the voltage in two stages. This method is used in the multipurpose am-
plifiers that are outlined below.
As we have seen, the circuit in Fig. 5.11 can be regarded as an amplifier with two amplifier
stages. Due to its low collector resistance, the differential amplifier with the transdiode
load T3 only provides a gain of 1/2. This means one must increase the collector resistance
of T2 in order to enhance the voltage gain. In Fig. 5.13, the current source T4 has been
introduced for this purpose. It is very useful to complement this current source with T3
to make it a current mirror; in this way the current changes in T1 are used to double the
current changes at the output of the differential amplifier and thus the differential gain. An
even more important advantage of the current mirror is the fact that the quiescent current of
T4 always has the appropriate value, regardless of the quiescent current I0 . The tolerances
of I0 thus have no influence on the offset voltage of the differential amplifier. This makes
it easier to integrate the circuit. To prevent the second amplifier stage from compromising
the high internal resistance at the collector of T2 , a Darlington configuration with its high
input resistance must be used for T5 , as described in Sect. 2.4.4 on page 159.
Most integrated multipurpose amplifiers are based on the principle shown in Fig. 5.13.
However, the input differential amplifier is realized by means of several npn and pnp
transistors, which act together like a pnp differential amplifier. In this case, an npn transistor
must be used in the second stage for potential shift. In Fig. 5.14, one can see that this leads
to a circuit that is exactly complementary to that in Fig. 5.13. In integrated circuits much
smaller quiescent currents are used. The collector currents of the differential amplifier are
only 10 mA. In integrated operational amplifiers the final stage is always designed as a
496 5 Operational Amplifiers
+15 V + + +
I0 I0 I1
10 µA 10 µ A 300 µ A
Iq
T6
T1 T2
VN VP Cc
30 pF
T7 Vo
I0 + Iq I0 – Iq
–13.8 V
T5
2I q
T3 T4
–14.4 V
–15 V – – –
Fig. 5.14. Operational amplifier of the 741 class. This circuit illustrates the principle only; due to
technological restrictions, the differential amplifier is made up of an array of transistors.
Capacitance Cc serves to correct the frequency response; its effect is described in Sect. 5.2.7.
Current 2Iq is not the base current of T5 , but the signal current that determines the voltage gain at
this point
rm1
VP
gm2V1
VD V1 V2 Vo
rm1
VN
complementary emitter follower, with the aim of maintaining positive and negative output
currents that are high compared to the quiescent current.
The model shown in Fig. 5.15 can be used to calculate the differential gain of the
operational amplifier. The transistors T1 and T2 of the input differential amplifier are
represented by voltage followers. The emitters are connected via the transconductance
resistances rm1 = 1/gm1 . The current Iq indicates how much the current through one
transistor is increased and the current through the other transistor is decreased when an
input signal is applied: Iq = VD /2rm1 . This current reaches the output of the differential
amplifier through the current mirror and produces a voltage across the existing internal
resistance. This voltage can be determined:
VD 1 M
V1 = −2Iq R1 = −2R1 = − VD = −200 · VD
2rm1 5 k
For the parameters shown in the model, the differential amplifier has a voltage gain of
A1 = V1 /VD = −200 and the transconductance
2Io 1 mA
gm1 = = = 0.2 .
VD rm1 V
5.2 Normal Operational Amplifier (VV-OPA) 497
The Darlington circuit T5 amplifies the voltage V1 and provides the output current gm2 V1
that produces a voltage drop across the internal resistance R2 . This voltage can be deter-
mined as:
mA
V2 = −gm2 V1 R2 = −5 · 100k · V1 = −500 · V1
V
For the parameters shown in the model, the second amplifier stage has a gain of
A5 = −500. Under the assumption that the emitter follower has a voltage gain of 1 at
its output, we obtain an overall gain for the model of
AD = A1 A2 = (−200) · (−500) = 105
5.2.3
Operating Voltages
So far, we have used a symmetrical operating voltage of ±15 V. Normal operational
amplifiers such as those described here then have a common-mode control range and an
output swing of approximately ±13 V. This is shown in Fig. 5.16a. The limitation results
from the internal construction with a minimum voltage drop of 2 V. Of course, 15 V can
be added to both operating voltages without the operational amplifier noticing this, since
there is no ground connection. This case is illustrated in Fig. 5.16b. Here, the operational
amplifier can be operated with one single supply voltage. This, however, causes a 15 V
shift in positive direction in the common-mode control range and the output swing, so that
the quiescent input and output potentials of 0 V can no longer be reached; this leads to
2 V < VCM , Vo < 28 V. This causes the operational amplifier to lose an important property
that renders its application so simple: zero quiescent input and output potential. This can be
remedied by generating an additional positive auxiliary potential of +15 V as a reference
for all other voltages;1 but this still makes a second voltage source necessary, resulting
in a number of disadvantages. If, however, it is known from the start that no negative
common mode and output voltages will occur, one can drive the operational amplifier
with an asymmetrical operating voltage in order to increase the positive control range.
For the example shown in Fig. 5.16c, this results in a control voltage of −1 V < VCM ,
Vo < +25 V.
+15 V + 30 V + 27 V
+13 V + 28 V + 25 V
control range
control range
control range
0V + 15 V 0V
–13 V + 2V – 1V
–15 V 0V –3V
a Normal operation b Operation with a single c Operation for positive
operating voltage voltage
Fig. 5.16. Influence of the operating voltages on the common-mode control range and the output
swing.
1 A circuit designed for this purpose is the rail-splitter TLE2426, from Texas Instruments.
498 5 Operational Amplifiers
+
– operating voltage Only one single positive operating voltage of +5 V
+
–
Normal operation Normal OPA Single-supply OPA Rail-to-rail OPA
Fig. 5.17. Control range of operational amplifiers when driven with low operating voltages
Operational amplifiers intended for a nominal operating voltage of ±15 V can also
usually be driven with ±5 V, but then the control range is reduced to ±3 V – as shown
in Fig. 5.17a – if one again assumes a minimum voltage drop of 2 V. There is a growing
desire to supply operational amplifiers with a single operating voltage of not more than
+5 V, or even +3.3 V, since in most cases this voltage is already available for the supply
of digital circuits. Most multipurpose amplifiers are not specified to operate at such low
voltages. Even if they were operated at +5 V, they would be of little use because the control
range would be reduced to 2 V < VCM , Vo < 3 V, as shown in Fig. 5.17b. For this reason,
single-supply amplifiers have been developed, whose common-mode control range and
output swing include the negative operating voltage, as shown in Fig. 5.17c. Even with a
negative operating voltage of 0 V, they allow quiescent input and output potentials as low
as 0 V. Indeed, there are operational amplifiers whose common-mode control range and
output swing reach to both the negative and the positive operating voltage. Such amplifiers
are known as rail-to-rail amplifiers; their control range is shown in Fig. 5.17d.
5.2.4
Single-Supply Amplifiers
The classic single-supply amplifier is model LM324, the basic design of which is shown
in Fig. 5.18. This circuit is related to the multipurpose amplifier shown in Fig. 5.14, but it
features some modifications to allow a control range down to negative operating voltages:
+ 15 V + +
Ik 6 µA
I1
100 µA
0.6 V
1.2 V T8
0.6 V 0.6 V
0V T1 T2 0V 0V
T5 T6
VN VP Vo
T9
0.6 V 0.6 V
T7
I2
T3 T4
Fig. 5.18. Basic circuit of the single-supply amplifier LM324. The potentials shown apply to a
control voltage equal to the negative operating voltage; in other words, zero potential in this case
5.2 Normal Operational Amplifier (VV-OPA) 499
– The emitter followers T5 and T6 have been added to shift the emitter potential of the
differential amplifier upward by 0.6 V. This keeps the collector–emitter voltage of the
differential amplifier at 0.6 V even in the most critical case of 0 V input voltage, as shown
here.
– The second amplifier stage T7 is shown here as a simple common-emitter configuration,
in order to achieve a quiescent base potential of 0.6 V. The Darlington circuit shown in
Fig. 5.14 would result in a quiescent potential of 1.2 V, so that with an input voltage of
0 V, transistor T2 would enter the saturation region.
– The current source I2 is added to allow an output swing down to almost 0 V. Of course,
for output voltages under 0.6 V transistor T9 is nonconductive, so that in this region the
output can only accept currents lower than I2 .
Phase Reversal
If a single-supply amplifier such as that shown in Fig. 5.18 is driven in a range below the
negative operating voltage, the collector–emitter voltage of the transistors in the differential
amplifier T1 , T2 is still at a level of 0.6 V. This value is clearly above the saturation voltage
of VCE,sat = 0.2 V. For this reason, the common-mode voltage may even be 0.4 V below
the negative operating voltage. For even more negative common-mode voltages, transistor
T2 enters the saturation region and its base–collector diode becomes forward biased. Then
the emitter of T6 is connected to the base of T7 and the inverting gain of T2 changes to
a noninverting signal transfer. With a further voltage drop at the P-input, transistor T7
becomes nonconductive and the output voltage increases until it reaches the positive limit
of the control range. This effect is known as phase reversal. The consequence of this effect
in practical applications is shown in Fig. 5.19 for a noninverting amplifier that amplifies a
sinusoidal alternating voltage. Due to the limitation caused by the final stage, the output
voltage cannot become negative. However, it is not clipped to 0 V as one might expect,
but due to the phase reversal jumps to the positive control limit if the input voltage drops
below the reversal voltage Vr ≈ −1 V. Phase reversal can have a very negative effect in
single-supply amplifiers if sufficiently negative input voltages are applied. This may be
prevented; for example, by using a Schottky diode at the input, which becomes conductive
at −0.5 V. The better solution, however, would be to use operational amplifiers, which,
due to their circuit design, do not cause phase reversal.
+ 15 V Vo Vo
V V
15 15
Vi Vo
10 10
Vo
5 5
0
–2 0 2 4 6 8 Vi / V Vi Vr t
+ 15 V + +
50 µA 25 µA
0V 3V 0V
T7
T1 T2 1V
VN VP
Cc
1.5 V 1.5 V 0V
Vo
T3 T4
T5 T6
Fig. 5.20. Single-supply CMOS operational amplifier of the TLC series. The substrates of the
n-channel FETs are connected to zero potential, and the substrates of the p-channel FETs to the
positive operating voltage
CMOS operational amplifiers do not suffer from phase reversal, since the isolated gate
electrodes do not allow any current flow. An effect similar to the conductive mode of the
base–collector diode in a bipolar transistor does not exist for the MOSFET. A common
circuit is illustrated in Fig. 5.20. A comparison with Fig. 5.18 shows that the design is very
similar. The p-channel FETs T1 and T2 form the differential amplifier. Both output signals
are combined by the current mirrors T3 and T4 and transferred to the second amplifier
stage T5 . The source follower T7 serves as an impedance converter. The only difference
is the operating mode of T6 . It does not act as a complementary source follower, but in
the common-source configuration it amplifies the signal, as does T5 . This enables this
transistor to reduce the output voltage to 0 V when T7 is nonconductive, as shown in the
example of Fig. 5.20. Therefore, the current source required at the output of LM324 is not
needed in this case. Here, the threshold voltage of all MOSFETs is |Vth | = 1 V. For the p-
channel MOSFETS this value is raised to 2.5 V due to the substrate effect because all bulk-
electrodes of these MOSFETs are connected to the positive supply voltage. Therefore the
bulk-source voltage of the input transistors amounts VBS = 12 V. This effect is quite useful,
because sufficient drain–source voltage remains, even with a common-mode voltage, down
to the negative operating voltage as shown in the example. As this effect does not exist
in bipolar transistors, the LM324 shown in Fig. 5.18 requires the additional transistors T5
and T6 for level shift.
5.2.5
Rail-to-Rail Amplifiers
Rail-to-rail amplifiers are special operational amplifiers that allow the common-mode
control range to extend not only to the negative operating voltage, as is the case with
single-supply amplifiers, but also to the positive operating voltage. In Fig. 5.20 the en-
hancement MOSFETs provide a potential shift that is sufficient to allow common-mode
control down to the negative operating voltage. However, common-mode control up to the
positive operating voltage is not possible, since this would require the source potential of
the differential amplifier to increase above the positive operating voltage. The rail-to-rail
5.2 Normal Operational Amplifier (VV-OPA) 501
+5V + +5 V +
Ik Ik
100 µA 100 µA
T1 T2 T1 T2
– 3.5 V + 4.7 V
VP VN VP VN
–5V –5V +5V +5 V
– 4.4 V – 4.4 V
T3 T4 T3 T4
–5V – –5V –
ID
– 1.5 V – 0.8 V + 0.3 V + 1.0 V
VGS
50 µA
OP OP
a VCM = – Vb b VCM = + Vb
Fig. 5.21. Rail-to-rail CMOS differential amplifier. The transfer characteristic of the MOSFETs is
shown for both extreme situations. OP marks the operating point. This principle is used, for
example, in the LMC6484
amplifier with this configuration becomes possible due to the use of MOSFETs which –
as previously seen – are in enhancement mode at the negative control limit, but which
change to the depletion mode at the positive control limit. The – usually rather disturbing –
substrate effect is used to shift the threshold voltage. This method is shown in Fig. 5.21.
In Fig. 5.21b, the maximum positive common-mode control voltage is VBS ≈ 0 and
the transistors are in depletion mode; this leaves a voltage drop of 0.3 V for the current
source. With a maximum negative common-mode voltage, the bulk–source voltage of
VBS = 8.5 V causes a shift in the threshold voltage, as explained in Sect.3.3.1:2
Vth = Vth,0 − γ ( Vinv + VBS − Vinv )
√ √ √
= 1 V − 0.8 V( 0.6 V + 8.5 V − 0.6 V)
= 1 V − 1.8 V = −0.8 V
At the operating point, this leads to a voltage of VGS = −1.5 V, as shown in Fig. 5.21a.
For the differential amplifier, this leaves a voltage of VDS = −0.9 V, which is sufficient to
remain above the pinch-off voltage:
VDS < VDS, po = VGS − Vth = −1.5 V + 0.8 V = −0.7 V
2 In practice, special MOSFETs with a higher substrate effect are used. The substrate control factor
√ √
is therefore assumed to be γ = 0.8 V , which is higher than the usual value of γ = 0.6 V .
502 5 Operational Amplifiers
This method is not practicable for bipolar transistors because they offer a fixed base-
emitter voltage of 0.6 V. They require the use of two complementary single-supply dif-
ferential amplifiers, one of which can be driven up to the positive and the other down to
the negative operating voltage, and the combination of their output signals. This method
is shown in Fig. 5.22. The differential amplifier consisting of pnp transistors T1 , T2 allows
input signals down to the negative operating voltage, while becoming nonconductive at
common-mode voltages close to the positive operating voltage; in this region, the npn
differential amplifier T3 , T4 connected in parallel takes over. The subsequent transistors
T5 − T8 combine the output signals of the differential amplifiers, so that all current changes
produced enhance the gain: the output current at the collectors of T6 and T8 is 4Iq . If one
of the two differential amplifiers fails because the common-mode voltages are close to the
operating voltages, the transconductance of the rail-to-rail input stage, and thus the volt-
age gain, halve in value. However, this effect is not noticeable in a circuit with negative
feedback, in which the gain is determined by the external circuitry.
In principle, it is possible to use the voltage V2 as the output voltage of the operational
amplifier. Transistors T9 − T12 then form a conventional complementary emitter follower.
But this would not provide a rail-to-rail output; the limits for the output voltages would
be approximately 1 V below the operating voltages. A rail-to-rail output stage can only
be realized with complementary transistors whose emitters are connected to the operating
voltages. In the common-emitter configuration, the resultant output voltage swing almost
reaches the operating voltages. In this case, the minimum voltage drop reaches a collector–
emitter saturation voltage VCE,sat of T15 or T16 , which for low currents amounts to only a
few millivolts.
The two output transistors are more difficult to drive in this configuration for the
following reasons:
– Throughout the entire control range, a constant quiescent current must be achieved via
the output-stage transistors.
– When controlling or loading the output, the current must be increased through one of
the transistors and decreased through the other.
– Output currents that are high compared to the quiescent current must be possible, so
that the output-stage transistors are operated in class-AB mode.
There are several different circuit designs for driving the output transistors. The circuit
shown in Fig. 5.22 is very reliable and requires no specific technological tricks; this makes
it easier to understand. In order to accurately control the output transistors T15 and T16 ,
they are expanded to current mirrors by adding transistors T13 and T14 .
In the analysis of the output stage, we start with the assumption R3 = ∞. Driving the
voltage followers T9 − T12 with a positive signal causes a current to flow through R2 that
increases the collector current of T11 and decreases that of T12 . The current difference is
identical to the current flowing through R2 . This is even the case if the current through R2
is so high that one of the transistors becomes nonconductive. For this reason, the output
current difference of T11 , T12 is proportional to the current through R2 . To allow high
output currents, the areas of the output transistors have been multiplied in size by a factor
of ten (A = 10).
+ 4.4 V
+ 4.8 V
+ 0.6 V
+ 0.6 V + 4.2 V
I2 V2 Io
V1
VP VN Vo
– 0.6 V – 4.2 V
– 0.6 V
– 4.8 V
– 4.4 V
–
5.2 Normal Operational Amplifier (VV-OPA)
Fig. 5.22. Example of an operational amplifier with rail-to-rail input and output.
Quiescent current: for example, I0 = 10 mA and I1 = 100 mA
503
504 5 Operational Amplifiers
rS V1 I 2 V2 R3 Io
1 1
VD Vo
rS Iq 4I q R1 R2 10 I2
1
Fig. 5.23. Model of a rail-to-rail operational amplifier for analyzing the output stage
Due to its common-emitter configuration, the rail-to-rail output stage features a high
output resistance and thus is a current output. In order to achieve a VV operational amplifier
with a low-resistance output, internal negative voltage feedback was realized with resistors
R2 and R3 in the output stage. The rail-to-rail output stage is a CC operational amplifier:
this circuit design is described in more detail in Sect. 5.5.
The behavior of the rail-to-rail output stage is explained most easily using the model
in Fig. 5.23. The situation may seem somewhat confusing, as the output stage represents
a current-controlled current source with negative voltage feedback. To analyze the circuit,
one can apply the nodal equations to both nodes of the output stage:
V2 Vo − V2
I2 − + = 0 (5.22)
R2 R3
V2 − Vo
+ 10 I2 − Io = 0 (5.23)
R3
For Io = 0, this leads to the open-circuit gain of the output stage:
11 R2 + 10 R3 111
Vo = V2 = V2 ≈ 10 V2 for R3 = 10 R2
11 R2 11
From (5.22) and (5.23), where V2 = 0, this leads to an output resistance:
Vo 1 1
ro = − = R3 = 1 k ≈ 100
Io 11 11
This means that the resistance is low compared to the output resistance of the output
transistors that amounts to rCE = VA /(10I1 ) = 100 V/(10 · 100mA) = 100 k.
5.2.6
Wide-Band Operational Amplifiers
In wide-band amplifiers, a single amplifier stage is preferred to bring the full voltage gain,
since this usually eliminates the need for a frequency response correction that would in turn
affect the bandwidth. However, the maximum gain achievable with one bipolar transistor
is limited:
IC VA VA 100 V
AD = µ = gm rCE = · = = ≈ 4000
VT IC VT 26 mV
This is insufficient even for a universal wide-band operational amplifier. A higher voltage
gain may be achieved by increasing the internal resistance above rCE . This is possible with
the cascode circuit shown in Fig. 5.24a. Unlike Fig. 5.11, here the additional transistor T6
together with T4 forms the cascode circuit. Its output resistance has already been determined
in Sect. 4.1.1 on page 294; according to (4.23) its value is ro = βCE . This means that in
5.2 Normal Operational Amplifier (VV-OPA) 505
+5V + + +5V +
Vo Vo
2I 0 I0 2I 0 I0
–5V – – –5V – –
a pnp cascode b Folded cascode
this case the output resistance is higher by the current gain β than that of a single transistor.
We thus obtain an open-circuit gain in the cascode circuit for β = 100:
VA 100 V
AD = βgm rCE = βµ = β = β ≈ 400.000
VT 26 mV
This method allows us to build a high-gain operational amplifier with a single amplifier
stage. The circuit shown in Fig. 5.24a may be simplified by omitting the current mirror. In
Fig. 5.24b, this eliminates the phase shift caused by the current mirror without degrading
the voltage gain. Only the polarity of the gain changes, but this can be compensated by
interchanging the inputs.
The realization of this principle in practice is shown in Fig. 5.25. In order to make
the circuit symmetrical, the two output signals of the differential amplifier are used and
combined by the current mirror T7 , T8 . To prevent a reduction of the high internal resistance
at the collector of T4 , a cascode current mirror according to Fig. 4.27 is used. Its output
resistance is βrCE /2.
As the high-impedance node (the collector of T4 ) has a very high internal resistance due
to the cascode circuit, a simple emitter follower is not sufficient to serve as an impedance
converter. The emitter followers T9 and T10 not only cause additional impedance trans-
formation, but also generate the bias voltage necessary for the complementary emitter
followers T11 and T12 .
The model shown in Fig. 5.26 can be used to calculate the voltage gain. The maxi-
mum voltage gain is achieved for RE = 0. This causes a current in the input differential
amplifier of
VD I0
Iq = = VD
2 rm 2 VT
At the high-impedance node of the circuit, which is represented here by R2 = βrCE
βrCE /2 = βrCE /3, this current causes a voltage drop:
β=100
I0 β VA β VA VA =100 V
Vo = 2 Iq R2 = VD = VD = 1.3 · 105 VD
V T 3 I0 3 VT
506 5 Operational Amplifiers
+5V + + + +
2I 0 2I 0 1.2 V I1
I0 + Iq
+4.4 V I0 – Iq
+ 0.6 V
T3 T4 T11
I0 + Iq I0 – Iq
+3.8 V T9
I0 + Iq
2I q –
T1 RE T
2 0V
VP VN – 3.8 V + Vo
Iq I0 – Iq
T5 T6 T10
– 4.4 V – 0.6 V
T12
T7 T8
I0 I0 I1
–5V – – – – –
Fig. 5.25. Operational amplifier with a complementary cascode differential amplifier (folded cas-
code). This principle is used, for example, in the AD797 from Analog Devices, the OP640 from Burr
Brown, and the LT1363 from Linear Technology
rS
VP 1 1
Iq Vo
VD 2I q R2
Fig. 5.26. Model of the
RE operational amplifier shown
rS
VN 1 in Fig. 5.25 for calculating
the differential gain
+5V + +
2I 0 1.2 V + +
I0 + Iq
+ 4.4 V
T1 T2 T5
+3.8 V
Iq R T9
E
I0 I0 T7
I0 + Iq
– – 2I q –
VP + + VN +
I0 – Iq Vo
I0 I0 T8
I q RE T10
– 3.8 V
T3 T4 T6
– 4.4 V
I0 – Iq – –
2I 0 1.2 V
–5V – –
Fig. 5.27. Push–pull operational amplifier. This principle is used, for example, in the EL2038 from
Elantec and the HFA0001 from Harris
resulting in a limited slew rate. One can naturally make the quiescent currents sufficiently
high, but this also increases the power dissipation of the operational amplifier. However,
a criterion for the quality of a given circuit design is whether a high bandwidth can be
achieved despite low quiescent currents. A satisfactory circuit is one that can provide high
charge currents for the parasitic capacitances even at low quiescent currents. In comple-
mentary emitter followers, this principle is common in class AB operation. A remarkable
development in the area of wide-band amplifiers is that of a differential amplifier suitable
for class AB operation.
An operational amplifier with a differential amplifier in AB mode at the input is shown
in Fig. 5.28. Transistors T1 − T8 form two voltage followers that are connected via the
resistance RE ; they form the differential amplifier. Here, the emitter current Iq = VD /RE
is not limited; it rises continuously with the voltage difference at the input. The output
signals are decoupled by the current mirror T9 , T11 or T10 , T12 . With Iq > I0 , the entire
current flows through the upper signal path (see Fig. 5.28). For this reason, the charge of
capacitance Cc at the high-impedance node can be reversed as quickly as desired if the
voltage difference at the input is high. Here, the slew-rate limitation that occurs in all of
the operational amplifiers discussed so far (see Sect. 5.2.7) is practically nonexistent.
The model shown in Fig. 5.29 will be used to explain the functional principle of the
amplifier. The voltage followers at the inputs are coupled via the resistance RE and the
two output resistances rm = 1/gm are in series. This results in
VD
Iq =
RE + 2rm
508 5 Operational Amplifiers
+5V + + + + + +
I0
+ 4.4 V
T9 T10
I0
T3 I0 + 1 2 Iq
T7 T15
T1 (I q ) T13
I0 – 1 2 Iq T5
– (0) I0 + 1 2 Iq Iq –
(I q ) –
+
VN + RE I q I0 – 1 2 Iq VP + Vo
Cc
I0 + 1 2 Iq (0)
T6 T14
T2 (I ) T
T4 q 8 T16
I0 – 1 2 Iq
I0
(0)
I0 T11 T12
– 4.4 V
–5V – – – – – –
Fig. 5.28. Operational amplifier with a push–pull differential amplifier in AB mode (current on
demand). The values quoted in parentheses reflect the conditions with a high current Iq > I0 . This
principle is used, for example, in the OP 467 models from Analog Devices, the LT1819 from Linear
Technology, and the LM7171 from National
rm
VP 1 1
Iq
Vo
VD Iq R2 Cc
rm RE
VN 1
Fig. 5.29. Model to explain the mode of operation of the operational amplifier shown in Fig. 5.28
This current is mirrored to the output and produces a voltage drop Vo at the resistance R2 ,
which represents the parallel arrangement of all impedances occurring at the output of the
current mirror. This voltage drop can be determined as:
VD R2
Vo = Iq R2 =
RE + 2rm
This voltage drop is transferred to the output via a voltage follower. The differential gain
of the circuit is thus
Vo R2 R2
AD = = =
VD RE + 2rm RE, ges
The highest voltage gain is reached when RE = 0. With R2 = rCE /2 we obtain:
1 R2 1 I C 1 VA 1 VA 1
AD = = = = µ ≈ 1000
2 rm 2 VT 2 I C 4 VT 4
With a simple common-emitter current mirrors (T10 , T12 ), no higher voltage gain can be
expected. But here too, it can be increased significantly by using a cascode current mirror
5.2 Normal Operational Amplifier (VV-OPA) 509
as seen in Fig. 5.25, since this causes the internal resistance at the high-impedance node
to increase by the current gain β.
It may be somewhat unusual to build a differential amplifier from two voltage followers,
and to transfer the output signal by means of two complementary current mirrors from the
supply terminals, but this arrangement can also be beneficial in other situations – for
example, in rail-to-rail output stages, as shown in Fig. 5.22.
There is also another way of interpreting the functional principle of the voltage fol-
lowers shown in Fig. 5.28: transistors T3 and T7 form an npn differential amplifier and
transistors T4 and T8 form a pnp differential amplifier, whose inputs are connected in
parallel and whose outputs are combined. This illustrates the relationship with the push–
pull operational amplifier in shown Fig. 5.27. Both circuits use complementary differ-
ential amplifiers to generate output signals in phase opposition, which are then ampli-
fied in the subsequent stage. The circuit shown in Fig. 5.27 limits the output current of
the differential amplifier to 2I0 . The current of the operational amplifier in Fig. 5.28 is,
however, not limited; for this reason, it is essential to use the current mirrors here; other-
wise, the advantage of the unlimited output currents of the differential amplifier would be
lost.
5.2.7
Frequency Compensation
Basic Principles
When using an operational amplifier as an voltage amplifier, the feedback, as shown in
Fig. 5.30, must always take place between the output and the inverting input, so that
negative feedback is achieved. Direct (positive) feedback is undesirable, because it may
produce a latching circuit with bistable behavior.
Operational amplifiers such as model 741, shown in Fig. 5.14, are multi-stage ampli-
fiers, where every stage has lowpass characteristic. The model in Fig. 5.31 shows the most
important lowpass filters of the amplifiers. The differential amplifier has the lowest cutoff
frequency of fc1 = 10 kHz, as it is operated with very low currents and the effective re-
sistance at the collector is very high. The cutoff frequency of the second amplifier stage is
clearly higher due to the higher currents, and amounts to fc2 = 100 kHz. In low-cost tech-
VD
Vi
Vo
Vi VD
Vo
Fig. 5.30. Comparison of noninverting and inverting amplifiers. For Vi = 0, both circuits are
R
identical. Then VD = R +R
1
Vo = kVo (k is used here instead of kR )
1 N
510 5 Operational Amplifiers
+ 1 1
Vo
VD gm1 VD R1 C1 V1 V1gm2 R2 C2
mA 1MΩ 16 pF mA 100 k Ω 16 pF
0.2 5
– V V
Fig. 5.31. The three most important cutoff frequencies in operational amplifiers of the 741class
nologies, the quality of pnp transistors is greatly inferior to that of the npn types; therefore,
pnp transistors cause a third lowpass filter with a cutoff frequency of fc3 = 1 MHz.
Each lowpass filter gives rise to a reduction in the gain by 20 dB per decade above
the cutoff frequency, and causes an additional phase lag which, at the cutoff frequency,
amounts to 45◦ and thereafter increases up to 90◦ , as described in Sect. 29.3.1 on page 1488.
Figure 5.32 shows the resulting Bode diagram. The gain reduction by 20 dB per decade
starts at fc1 , with a phase shift of 45◦ . From fc2 onward, this value falls by 40 dB per
decade and the phase shift is as much as 135◦ ; this value is a combination of 90◦ from
the first lowpass filter and 45◦ from the second. Due to the third lowpass filter, above fc3
the gain drops at a rate of −60 dB per decade and the phase shift rises asymptotically to
−270◦ . At a frequency of (here) f180 = 300 kHz, the phase lag passes −180◦ . At this point
the functionality of the inputs interchanges and the negative feedback turns into positive
feedback.
Whether or not this circuit oscillates at this frequency depends on whether or not the
oscillating condition is met:
,
|g| = |k||AD | ≡ 1 Amplitudecondition
g = k AD ≡ 1 ⇒ (5.24)
φ(kAD ) ≡ 0◦ , 360◦ , . . . Phasecondition
5 A D0 AD
10 4
4
g0 A 1 = 10 g
10 g =1
3
3 A 2 = 10
10 g =1
2
10 1
A0 =
10 k
f c1 f c2 f c3
1
1 10 100 1k 10 k 100 k 1M 10 M f/Hz
ϕ
– 90°
f 180
α
– 180°
– 270°
Fig. 5.32. Bode diagram of a noncompensated operational amplifier of the 741 class
5.2 Normal Operational Amplifier (VV-OPA) 511
This condition comprises two parts: the amplitude condition and the phase condition. An
oscillation with constant amplitude arises only if both conditions are met. This is the case
in Fig. 5.32 with a feedback, to achieve a gain of A2 = 1, 000. At the frequency f180 , the
loop gain is then |k|AD | = 1. The loop gain can be taken directly from the Bode diagram.
Since g = kAD = AD /A becomes lg g = lgAD − lgA in the logarithmic diagram, the
(logarithmic) loop gain is identical to the distance between the differential gain and the
negative feedback gain.3 From Fig. 5.32 one can see that this distance becomes smaller
as the frequency increases and reaches zero at the intersection with the set gain; at these
points, g = 1.
If kAD > 1 and the phase condition is met, an oscillation with an increasing amplitude
occurs. In this case, the oscillation amplitude increases until the amplifier is overdriven.
If kAD < 1, then there is a damped oscillation. For an amplifier, this is the only case
of interest. In our example, it occurs when the gain set by the negative feedback exceeds
1,000; for example, A1 = 10,000. For the frequency f180 it follows that g = kAD = 1/10;
the loop gain is thus below the oscillating point by a factor of ten. This is also called a gain
margin of ten; this means that the loop gain can be increased by a factor of ten before an
undamped oscillation occurs.
It is more common to specify the phase shift distance to −180◦ , when the amplitude
condition g = kAD = 1 is met. The value
α = 180◦ − ϕ(fk ) (5.25)
is called phase margin. It states the angle by which the phase shift may increase before an
undamped oscillation occurs at the critical frequency fk at which the amplitude condition
is met. In the Bode diagram of Fig. 5.32, it is marked by circles.
The phase margin is a particularly useful parameter for judging the damping effect
and the tendency toward oscillation of a given system. Figure 5.33 shows the transient
responses for various phase margins and the corresponding frequency responses. It is clear
that a declining phase margin causes move overshoot in the step response and an increasing
peaking of the frequency response. The aperiodic limit is reached at a 90◦ phase margin:
there is no overshoot, but there is clearly a longer rise time and a drastically reduced
Vo 30° A
Vi 30°
45°
45°
60° 60°
1.0 90°
1.0
90°
0.3
0.5
0.1
0 0.5 1.0 t/µ s 10 k 100 k 1M f/Hz
a Step response b Frequency response
Fig. 5.33. Step response and frequency response for several phase margins α. Overshoot in the
time interval corresponds to peaking in the frequency range
5
10 Uncompensated
4 Compensated
10
3
10
2 '
f c1 f c1
10
f c2
10 f c1 A3 = 1 g fk f c3
1
1 10 100 1k 10k 100k 1M 10M f/Hz
ϕ
Compensated Uncompensated
–90°
–180° = 45°
–270°
Fig. 5.34. Bode diagram of an operational amplifier of the 741 class with universal frequency
compensation
Pole Splitting
Pole splitting utilizes the Miller effect in order to reduce the compensation capacity down to
a value that can be integrated. In the common-emitter circuit, the collector–base capacitance
represents the Miller capacity. It results in a significantly narrowed bandwidth, since it has
the effect of an input capacitance that is increased by the voltage gain (refer to Sect. 2.4.1
on page 125). This otherwise disadvantageous effect can be of benefit here, as it enables the
use of a lower compensation capacitance. The Miller capacitor Cc is shown in the model
shown in Fig. 5.35. For a voltage gain of 500, a capacitance of only 160 nF/500 = 320 pF
is sufficient.
Utilization of the Miller effect has a second advantage: since the Miller capacitor
induces negative voltage feedback, the output resistance of the amplifier stage is also
reduced. In our example, this results in an increase of the cutoff frequency fc2 from 100 kHz
to 10 MHz; it thus exceeds the third cutoff frequency of 1 MHz. As the cutoff frequency
Cc
30 pF
+ 1 1
Vo
VD gm1VD R1 C1 V1 gm2V1 R2 C2
mA 1 MΩ 16 pF mA 100 k Ω 16 pF
0.2 5
– V V
5
10 Uncompensated
10
4 Compensated
10
3 f'c1 f c1 f c2 f'c2
2
10
10
A3 = 1 g f c3
1
1 10 100 1k 10k 100k 1M 10M f/Hz
ϕ
Compensated
–90°
Uncompensated
–180°
–270°
c c c
c c c
c
c c c c
Fig. 5.37. Comparison of general and adapted frequency compensation for the gains of Amin = 1,
10, and 100
amplifiers provide a good transient response for any external circuitry and remain stable
as long as the negative feedback does not produce an additional phase lag. However, using
them to produce a gain A > 1 gives away bandwidth. This disadvantage can be avoided
by adapted frequency compensation. For a gain of A0 = 10 we obtain, from (5.26):
This means the first cutoff frequency can be increased by the gain A0 , compared to the case
of general frequency compensation. This makes the cutoff frequency of the amplifier with
negative feedback constantly equal to the second cutoff frequency, as can be seen from
Fig. 5.37. The phase margin is always 45◦ . In contrast, general frequency compensation
provides a constant gain–bandwidth product; that is, the bandwidth decreases by the same
factor as the gain increases.
With operational amplifiers of the 741 class, the adapted frequency compensation can
only be used up to a cutoff frequency of fg1 = 100 Hz; this results in a compensation
capacitance of Cc = 3 pF. A higher decompensation renders the pole splitting ineffective,
and the second cutoff frequency is again reduced to 100 kHz.
The terminals of the compensation capacitor are critical. Therefore, in more recent
operational amplifiers they are no longer accessible to the user. Instead, for some types of
operational amplifier fully compensated and partially compensated terminals are available
(for example, for a minimum gain of Amin = 2, 5, or 10).
Slew Rate
Besides reducing the bandwidth and the loop gain, frequency compensation has yet another
disadvantage: the maximum speed the output voltage can change – the slew rate – is limited
to a relatively low value. The reason for this can easily be seen in the equivalent circuit
shown in Fig. 5.38.
If, in the event of overdrive, only T2 is conductive, then I1 = 2I0 . If only T1 is
conductive, then the entire current flows through the current mirror, resulting in I1 = −2I0 .
The charge current of Cc is limited to the maximum output current of the differential
516 5 Operational Amplifiers
+15 V
2I 0
20 µA
T1 T2
VN VP
Cc
I1 30 pF
T3 T4
Vo
–15 V
Fig. 5.38. Model to explain the slew rate, using a 741 class amplifier as an example. The second
amplifier stage with the Miller capacitor is shown here as integrator
amplifier: I1max = ±2I0 = ±20 mA.4 As the compensation capacity carries the full
output voltage, I = C V̇ leads to the slew rate SR
dVo
I1max 2I0 20 mA V
SR =
= = = = 0.6 (5.29)
dt max Cc Cc 30 pF ms
This means that the maximum change in the output voltage within 1 ms is 0.6 V. A
rectangular-wave signal with an output amplitude of ±20 V has a rise- or falltime of
Vo 20V
t = = = 33 ms
SR 0.6 V/ms
Likewise, a sinusoidal input voltage swing can produce no faster change at any point than
that allowed by the slew rate. If we take the output voltage to be Vo = V̂o sin ωt, then the
maximum rate of voltage change occurs at the point of zero crossing resulting in
dVo 7o ω = 2πf V 7o
SR = = V (5.30)
dt
This allows us to calculate the frequency up to which an undistorted sinusoidal signal with
full output amplitude is possible:
SR 0.6 V/ms
fp = = = 10 kHz (5.31)
7
2π Vo 2π · 10 V
This value is known as the power bandwidth, because it represents the frequency up
to which the full output power is available. In amplifiers of the 741 class it is only fp =
10 kHz, as can be seen, even though the small-signal bandwidth is fT = 1 MHz. According
to (5.30), the maximum output voltage swing decreases when the frequency f is increased:
V7o = SR (5.32)
2πf
As shown in Fig. 5.39, an amplifier of the 741 class achieves the full output swing up to
10 kHz. At 100 kHz this drops to 1 V, and at 1 MHz as low as 0.1 V.
4 The maximum current of the second amplifier stage, shown here as an integrator, is also limited,
but at 300 mA it is significantly higher and therefore has no limiting effect.
5.2 Normal Operational Amplifier (VV-OPA) 517
Vo
V
10
8
6 Partially
4 compensated
Fully compensated A min = 10 Fig. 5.39. Dependence of
2
the output swing on the
0 frequency in an amplifier of
1 10 100 1k 10k 100k 1M f/Hz the 741 class
When the output signal exceeds the slew rate limitation, the curve segments can be
replaced by straight lines that represent the slope of the slew rate. This is shown in Fig. 5.40.
When the slew rate is considerably exceeded the output signal assumes a triangular shape
and, with the exception of the frequency, has little in common with the undistorted signal.
In order to improve the slew rate, one could assume on the basis of (5.29) that it
increases as the current I0 increases. However, in order to examine this we also have to
take into account the current dependence of Cc , as shown in (5.28):
2I0 2I0
SR = = 2πfT (5.33)
Cc gm1
For the given transit frequency, the slew rate increases as the current I0 increases at a
given transconductance. For bipolar transistors, however, the ratio I0 /S1 is constant, since
the transconductance is proportional to I0 :
2I0 2I0
= = 4VT ≈ 100 mV
S1 2I0 /4VT
For the slew rate, it follows:
4 · 2I0 VT V
SR = 2πfT = 8πVT fT = 8π · 26 mV · 1 MHz = 0.6 (5.34)
2I0 ms
Therefore the slew rate does not depend on the current I0 , since the required compensation
capacity increases at the same rate as I0 . However, current feedback of the input differential
amplifier allows the current I0 to be increased with a constant transconductance; this is
often utilized in wide-band operational amplifiers. Favorable values are also achieved with
operational amplifiers by using field effect transistors at the input, as they have a much
lower transconductance as bipolar transistors, at the same current I0 .
5
R1 RN 10
4
10 kΩ 10 kΩ 10
3
Vi Rc 10
1.25 kΩ 2 f c1 f c2 = f k
10
g 1
Cc Vo 10
1.3 nF A1 k
A min = 10 1
1 10 100 1k 10k 100k f/Hz
a Input compensation b Frequency response
Fig. 5.41. Use of a partially compensated amplifier with Amin = 10 at a gain of Amin = 1
Capacitive Load
If a capacitive load CL is connected to the output of an operational amplifier, it forms,
together with the output resistance ro , an additional lowpass with a cutoff frequency fcC , as
shown in Fig. 5.42. Operational amplifiers with a simple emitter follower at the output have
output resistances (of the open amplifier) in the range of ro ≈ 1 k, while in Darlington
circuits and in RF operational amplifiers this value is mostly below 100. If the load
capacitance is low (CL < 100 pF), the additional cutoff frequency fcC is higher than the
second cutoff frequency of the amplifier; the phase margin then decreases only slightly. For
higher load capacitances, the additional cutoff frequency drops below the second cutoff
frequency; this case is shown in Fig. 5.43. As one can see, the phase shift above fcC
becomes so large that the circuit would oscillate at lower closed loop gains. In order to
achieve stable operating conditions in spite, additional frequency compensation is required.
As standard operational amplifiers usually have internal compensation, it is not possible
to subsequently reduce the minimum cutoff frequency fc1 . But with the aid of input
compensation it is possible to carry out additional compensation by external circuitry. This
method has already been demonstrated in Fig. 5.41. In the example shown in Fig. 5.43,
the gain was still 1,000 for a second cutoff frequency fcC = 1 kHz, making it necessary
to reduce the loop gain by this factor. Such a great attenuation is not practicable with input
compensation.
ro
1
VD C2 Vo CL
gm1VD R1 C1 V1 gm2V1 R2 C2
Fig. 5.42. Operational amplifier with capacitive load. The cutoff frequencies of a fully
compensated amplifier of the 741 class are fc1 = 10 Hz and fc2 = 1 MHz
AD
5
10
4 With capacitive load
10
3 General compensation
10
2
10
10 fc1 fcC fc3
1
1 10 100 1k 10 k 100 k 1M 10 M f/Hz
ϕ
–90°
–180°
–270° With capacitive load
R1 RN
Vi
ro
V2 R iso
Vo CL
Fig. 5.44. Isolation resistance for phase compensation with a capacitive load
It is better to connect an isolation resistance in series with the capacitive load, as shown
in Fig. 5.44. For high frequencies at which the load capacitor acts as a short-circuit, the
output of the amplifier is connected solely to a voltage divider formed by ro and Riso ,
which produces no phase lag. From the Bode diagram in Fig. 5.45 it can be seen that,
compared to Fig. 5.43, the phase does not change up to 1 kHz, but above this frequency
it approaches the nonloaded situation. At the critical frequency fk = fc2 = 100 kHz, the
phase margin is 90◦ ; this determines the transient response of the circuit, whereby it is
of no importance that the phase margin is smaller at lower frequencies. Here we have the
specific case of the phase margin decreasing with less feedback: for a gain of A = 10, the
critical frequency is 10 kHz; the phase margin at this point is only 45◦ .
An example will serve to explain the dimensioning of the components. An amplifier
with an open-circuit output resistance of ro = 1 k is to be loaded at the output with a
capacity of CL = 160 nF. This results in a cutoff frequency of
1 1
fcC = = = 1 kHz (5.35)
2π ro CL 2π 1 k 160 nF
5
AD
10
4
10
3
With RC load
10
2 General compensation
10
f c2
10 f c1 f cC f ck
A1 g
1
1 10 100 1k 10 k 100 k 1M 10 M f/Hz
ϕ
–90°
–180°
α = 45° α = 90°
–270°
Fig. 5.45. Reversal of the phase shift above fck by way of isolation resistance
5.2 Normal Operational Amplifier (VV-OPA) 521
R1 RN R1 RN
Vi Cc Vi Cc
– ro – ro
V2 V2 Riso
+ CL Vo + CL Vo
In order to reduce the phase shift caused by the load, when the critical frequency fc2 =
100 kHz is reached, we take fck = 10 kHz according to Fig. 5.45. It then follows from
(5.35)
1 fcC 1 kHz
Riso = = ro = 1 k = 100 (5.36)
2πfck CL fck 10 kHz
To achieve as much bandwidth as possible, one can reduce Riso slightly. Due to the
reduced phase margin this leads to a gain peaking as shown in Fig. 5.33, which can com-
pensate the gain reduction caused by the lowpass filter Riso CL within a certain frequency
range.
According to Fig. 5.44, the use of an isolation resistance is disadvantageous in many
applications, as the load is not operated from a low source resistance. The conventional
circuitry can then be expanded by the capacitance Cc shown in Fig. 5.46a. This can com-
pensate the phase lag caused by the load. Its size should be continually increased until the
desired transient response or frequency response is achieved.
In very difficult cases, one can insert an additional isolation resistance as shown in
Fig. 5.46b. In order to realize the proper output voltage Vo at the load capacitor the amplifier
produces a leading voltage V1 . Feeding this voltage back via the compensation capacitance
Cc enhances the stabilizing effect.
Cc2 Rc
ro
1
Vo CL Fig. 5.47. Internal
VD S1VD R1 Cc1
compensation for a
capacitive load
5
10
10
4 2-pole uncompensated
1-pole
3 '
fc1 fc1
10 fc2
'
fc2
2
10
10
A g
1
1 10 100 1k 10 k 100 k 1M f/Hz
ϕ
–90°
1-pole
–180°
2-pole uncompensated
–270°
Fig. 5.48. 2-pole frequency compensation of an operational amplifier. 1-pole correction without
pole splitting is also shown for comparison
1
Cc2
VD 1.6 nF Vo
VD gm1 R1 C1 Cc1 V1 V1 gm2 R2 C2
1MΩ 16 pF 144 pF 100 kΩ 16 pF
Rc2
1kΩ
1 ' = 1
f c1
' = = 1 kHz f c2 = 1 kHz
2π R1 ( C1 + Cc1 ) 2π R2 ( C2 + Cc2 )
is already −90◦ and increases to −180◦ . In order to maintain a sufficient phase margin
around the critical frequency, a zero point is introduced by Rc2 , which shifts back the phase
of one lowpass filter. This reduces the phase shift; the phase margin then reaches 45◦ and
more in the range from 100 kHz to 1 MHz. The amplifier is thus correctly compensated for
gains in the range of A = 1 − 10. For less negative feedback with a gain of A = 1,000,
the critical frequency is 10 kHz; but at this point the phase margin is almost zero and the
transient response is correspondingly poor.
In Fig. 5.48, one can see that the 2-pole compensation creates a higher loop gain
without degrading the frequency response of the feedback circuit. The nonlinear distor-
tions are also reduced to the same extent as the loop gain increases. This technology is
therefore particularly interesting in electro-acoustics to insure low nonlinear distortions.
However, bipolar compensation impairs the transient response: there is an overshooting
and undershooting tendency in the range of some percent, which decays very slowly. And
since bipolar frequency compensation must be tailored to each individual application, it is
rarely used. It is more convenient to use a faster operational amplifier with 1-pole standard
compensation. This technology is applied only in cases in which it is essential to keep the
nonlinear distortions as low as possible even up to high frequencies.
5.2.8
Parameters of Operational Amplifiers
The most important parameters of operational amplifiers are listed in Fig. 5.50. These
parameters, together with their influence on the noninverting and inverting amplifiers, are
described below.
The œA741 and TLC272 amplifiers are older standard types that feature no particularly
high-quality data compared to newly developed types. The reason for their widespread use
in large numbers even today is that they are very low-priced. The TLC272 is built ex-
clusively from n-channel and p-channel enhancement MOSFETs. Its maximum operating
voltage is therefore limited to 16 V. As its common-mode control range and output swing
reach the negative operating voltage, it is known as a single supply amplifier. Due to the
MOS input its input currents are extremely low and the input resistances are correspond-
ingly high. These values are usually not determined by the chip but by the case and the
PCB.
The OP177 is an operational amplifier that is suitable for achieving a particularly high
precision. On the one hand, its offset voltage is very low and can be totally neglected
in most applications and the user must insure that the thermal voltages at the soldering
point cause no major errors. On the other hand, this type of amplifier features extremely
high values for differential gain and common-mode rejection, which can be assumed to be
infinite.
The AD797 is√a particularly low-noise amplifier for audio applications. Its noise voltage
density of 1 nV/ Hz is at the limit of technical feasibility. Its noise current, however, is
not lower than in normal operational amplifiers. For this reason, the AD797 is particularly
advantageous with low-resistance sources (see page 537). The gain–bandwidth product
of 110 MHz seems to be unnecessarily high for audio applications. However, this is the
precondition for a high loop gain and thus low distortions. Up to 20 kHz, the distortions
remain 120 dB below the useful signal.
524 5 Operational Amplifiers
The LM7171 is a particularly fast operational amplifier that can be used up to 200 MHz.
This can be seen from the high bandwidth and slew rate. But this is at the cost of poor
DC data; the offset voltage drift and the input quiescent currents are high, while the
differential gain is low.
For the calculation and design of operational amplifier circuits, it is generally possible
to carry out a precise analysis of the circuit with all of its error sources. But it is easier to
initially assume that the operational amplifier to be ideal and then calculate the deviations
caused by the individual parameters of the real operational amplifier.
5.2 Normal Operational Amplifier (VV-OPA) 525
Vo Vo,max Vo
10 V 10 V
∆V o ∆Vo
AD = ∆Vo ACM = ∆Vo
∆V D ∆VCM
for VCM = 0 ∆VD for Vo = 0 ∆VCM
–10 V –10 V
10 10
Fig. 5.51. Differential and common-mode control. The values shown are typical of an operational
amplifier of the 741 class.
526 5 Operational Amplifiers
Since the transfer characteristics are approximately linear within the permissible control
limits, (5.40) also applies in large-signal situations:
Vo = AD VD + ACM VCM
This equation can be solved for VD ; at the same time, the common-mode gain can be
replaced by the more conventional common-mode rejection CMR = AD /ACM :
,
Vo VCM Vo /AD for VCM = 0
VD = − = (5.41)
AD CMR −VCM /CMR for Vo = 0
This represents the common definition of the differential gain
∂Vo
Vo
AD = = (5.42)
∂VD
dVCM = 0 VD
VCM = 0
and an additional definition for the common-mode rejection:5
VCM
CMR = = · = = (5.43)
ACM ∂VD ∂Vo ∂VD
dVo = 0 VD
Vo =0
The relationships between the common-mode and the differential voltages are obtained
by applying a differential voltage at a certain common-mode voltage that is so high that it
causes a zero output voltage. This is the voltage required to compensate the effect of the
common-mode control. The slope of this function is the common-mode rejection, which is
shown in Fig. 5.51b. From the abrupt drop in the common-mode rejection, one can clearly
see the limits of the common-mode control capability. The limit defined by the circuitry
in Fig. 5.14 is due to the fact that one of the transistors in the differential amplifier or the
corresponding current source enters the saturation region. A comparison of Figs. 5.51a and
5.51b shows that the differential gain and the common-mode rejection look very similar.
In (5.41) one can see that the input voltage VD consists of two parts: the first is
determined by the output swing and the second is added during the common-mode control.
As AD and CMR are generally very high, within the linear operating region voltage VD
is usually very low, with values in the millivolt range. In order to take the effects of the
finite differential gain and the common-mode rejection into account, it is easiest to use the
models shown in Fig. 5.52. From the condition of an ideal operational amplifier that the
output voltage assumes a value at which the input voltage difference becomes zero, we
can calculate the output voltage:
Vo VCM
Vi − + = kVo
AD CMR
Since, in a noninverting amplifier, the common-mode voltage
VCM = (VP + VN ) /2 ≈ VP = Vi (5.44)
is practically equal to Vi , the gain is
Vo AD 1 AD 1 RN
A = = 1+ ≈ ≈ = 1+ (5.45)
Vi 1 + kAD CMR 1 + kAD k R1
5 For the common-mode gain and common-mode rejection, only the absolute value is specified;
therefore the sign is of no importance. To avoid the impression that other effects could be com-
pensated by common-mode rejection, it is always good practice to use the less favorable sign.
5.2 Normal Operational Amplifier (VV-OPA) 527
Vo VCM
VD RN
AD CMR
R1
ideal ideal
Vi VD 0V Vi 0V
Vo Vo
RN
kVo VCM
CMR
R1
Vo
AD
Fig. 5.52. Effect of the finite differential gain and the common-mode rejection on the gain
The deviation from the ideal behavior caused by the finite differential gain is
1 AD
−
A Aid − A k 1 + kAD 1 1
= = = ≈ (5.46)
A Aid 1/k 1 + kAD g
Thus, the relative deviation from the ideal behavior is the same as the reciprocal value of
the loop gain; therefore, it is generally very low. Production tolerances and the changes in
differential gain due to temperature variations are also reduced by the same factor.
The inverting amplifier in Fig. 5.52b has a common-mode voltage VCM = VD/2 Vi .
This means that the finite common-mode rejection CMR has no influence on the voltage
gain; the gain is thus as defined in (5.17).
Offset Voltage
The transfer characteristic of a real operational amplifier does not pass through zero, but is
shifted by the input offset voltage; this is illustrated in Fig. 5.53. The offset voltage is usually
in the millivolt range; high-quality operational amplifiers even feature offset voltages in
the microvolt range, as shown in Fig. 5.50. But even a low offset voltage can cause the
amplifier to be overdriven if both inputs are connected to ground (that is, VD = 0); this
can also be seen in Fig. 5.53. The reason for this is the high differential gain that amplifies
even small offset voltages to such a degree that the output swing is exceeded.
Vo Vo,max
10 V
ideal real
Fig. 5.53. Effect of the offset voltage on the transfer characteristic of an operational amplifier
528 5 Operational Amplifiers
V1
A1 A2
Vi Vo
VO1 VO2
However, operational amplifiers are not usually operated in open-circuit mode but with
negative feedback; then, the error caused by the offset voltage is only amplified with the
closed loop gain as the input signal. The effect is similar to that achieved with the signal
voltage source connected in series. If this small error still causes disturbance, the offset
voltage can be brought down to zero. Some operational amplifiers have special terminals
for connecting a trimmpoti to adjust the offset voltage. However, it is often better to use
a type with so small an offset voltage that the problem does not arise. Type OP177 in
Fig. 5.50 illustrates how low the offset voltage may be. It is usually much cheaper for the
operational amplifier to be balanced by the manufacturer rather than by the user, since this
would require not only a trimmer but also a measuring station, a technician, and adjustment
instructions.
The offset voltage is caused by various factors. In addition to the tolerances when
matching the input transistors, the input amplifier and the subsequent circuitry also have
asymmetries and tolerances. The greatest influence, however, is the input stage. This is
exemplified by the model of a two-stage amplifier shown in Fig. 5.54, where the offset
voltage is supplied to the input of each stage. This leads to the output voltage:
Vo = (V1 + VO2 ) A2 = [(Vi + VO1 ) A1 + VO2 ]A2
= A1 A2 Vi + A1 A2 VO1 + A2 VO2
In order to determine the offset voltage of the entire circuit in relation to the input, one
sets Vo = 0 and calculates the input voltage for this situation:
1
Vi (Vo = 0) = VO = − VO1 − VO2 (5.47)
A1
The input offset voltage of the first stage affects to full amount – whereas the offset of the
second stage is reduced by the factor 1/A1 . The goal is therefore to make the gain of the
first stage as high as possible.
With the offset voltage reduced to zero, only its dependence on temperature, time, and
the operating voltage can be noticed:
∂VO ∂VO ∂VO
dVO (ϑ, t, Vb ) = dϑ + dt + dVb (5.48)
∂ϑ ∂t ∂Vb
Here, ∂VO /dϑ is the temperature drift, with typical values of 3 − 10 mV/K. The long-term
drift ∂VO /∂t is several microvolts per month. This can be regarded as a low-frequency
portion of the noise. The supply voltage rejection ratio ∂VO /∂Vb characterizes the influence
that the supply voltage variations have on the offset voltage. It has a value of 10−100 mV/ V.
To keep this contribution to the offset voltage low, the supply voltage should be regulated.
According to Fig. 5.53, the transfer characteristic of an operational amplifier with an
offset voltage within the linear control range is:
Vo = AD (VD − VO ) (5.49)
5.2 Normal Operational Amplifier (VV-OPA) 529
RN
VO
R1
ideal ideal
Vi VO 0V Vi 0V
Vo Vo
RN
kVo
VO
R1
Fig. 5.55. Influence of the offset voltage in noninverting and inverting amplifiers
In order to reduce the output quiescent potential to zero, it is necessary to either zero the
offset voltage or apply a voltage VD = VO to the input. This results in the rule:
The offset voltage is the voltage required at the input to achieve a zero output
voltage.
To examine the effect of the offset voltage in circuits with negative feedback, use of the
equivalent circuits in Fig. 5.55 is recommended. If Vi = 0, then both circuits are identical.
This leads to an offset voltage at the output:
RN
Vo (Vi = 0) = − 1 + VO (5.50)
R1
This complies with the voltage gain of the noninverting amplifier. Thus, the offset voltage of
a noninverting amplifier is amplified like the input voltage, while in an inverting amplifier
this is only an approximation.
Input Currents
The input quiescent current of an operational amplifier corresponds to the base or gate
current of the input transistors. Its value depends on the current used to drive the input
transistors. In multipurpose amplifiers with bipolar transistors at the input, operated with
collector currents of 10 mA, the input quiescent currents may be 0.1 mA. In wide-band
amplifiers with collector currents up to 1 mA, the input currents may be several microam-
peres. With Darlington circuits at the input, the input quiescent current is in the nanoampere
range. The lowest input quiescent currents are found in operational amplifiers with field
effect transistors at the input. Here, they often amount to a few picoamperes only.
Since the input transistors are operated with constant collector currents, the base cur-
rents are constant as well at common mode input signals; therefore, the inputs represent
constant current sources. In reality, the input currents are similar, but not identical. This is
the reason why data sheets specify the mean input quiescent current
1
IB = (IP + IN ) (5.51)
2
and the input offset current
IO = |IP − IN | (5.52)
530 5 Operational Amplifiers
IP IB 1 I IB
2 O
VP VP VP
VN Vp VN Vp VN Vp
IN IB – 1 IO IB IO
2
Fig. 5.56. Converting the input currents to bias and offset currents
IB IB RN
Rg VP R1 VP
ideal ideal
0V 0V
Vi Vi
Vo Vo
VN RN VN
IB I0 CB RB IB IO
R1
amplifier to amplifier, as does its polarity. In principle one could adjust the offset current in
the same way as the offset voltage, but it is better to dimension the circuitry so low-ohmic
that the resulting error remains negligible. Furthermore, as for the offset voltage, the offset
current is also temperature sensitive; the offset current drift indicates the rate at which it
changes with temperature.
For the inverting amplifier shown in Fig. 5.57b, the noninverting input usually carries
ground potential. The input current thus causes an offset of IN RN at the output. Here too,
this error can be compensated by connecting the noninverting input indirectly to ground
via resistance RB = RN R1 , so that the total resistances of both inputs are equal. The error
IO RN caused by the offset current is the only remaining error. To prevent resistance RB
from generating additional noise, the alternating voltages are shorted by capacitance CB .
We have shown that the error caused by the input currents rises proportionally to the
dimensions of the resistors connected. Therefore, these resistors should be dimensioned
as low as possible so that this error remains small. If you must use high feedback resistors,
an operational amplifier with sufficiently low input currents should be selected. There are
very significant differences, as can be seen in Fig. 5.50.
Input Resistances
As in the case of the differential amplifier, the operational amplifier has two input re-
sistances, the differential resistance and the much larger common-mode resistance. The
influence that the common-mode resistance may have on the noninverting amplifier can
be taken from the equivalent circuit in Fig. 5.58a. The resistors are connected between
input and ground; that is, they are in parallel to the inputs and therefore they are not af-
fected by the negative feedback. The common-mode resistance at the noninverting input
causes an attenuation whereas the resistance at the inverting input results in an increase in
gain. After matching the internal resistances of both inputs to Rg = RN R1 , the effect is
fully compensated. Due to their very high resistance common mode resistance has little
influence.
In order to examine the effect of the differential input resistances, it is necessary to
take a real operational amplifier with finite gain and common-mode rejection. We thus look
at Fig. 5.58b and calculate the current flowing through the differential input resistance.
According to (5.41), the current is:
rCM
VD
Vi Ii VP Vi Ii
Vo Vo
Vg Vg
Ii
rCM
Fig. 5.58. Effect of the differential and common-mode input resistances in noninverting amplifiers
532 5 Operational Amplifiers
RN
rCM
R1
Vi rD VD 0V
Vo
rCM VD Fig. 5.59. Input resistance in an
inverting amplifier
VD Vo VCM 1
Ii = = +
rD AD CMR rD
For Vo = Vi /k, VCM = Vi , and g = kAD , we have a contribution to the input resistance
due to rD :
Vi g CMR g rD for CMR g
rD = = rD = (5.56)
Ii g + CMR CMR rD for g CMR
Therefore, the differential input resistance is significantly increased by the negative feed-
back, as the differential voltage VD , which is only a fraction of the input voltage Vi .
The resulting input resistance of the noninverting amplifier is thus ri = rCM ||rD
; since
both portions are very high, we obtain input resistances in the gigaohm range, even with
operational amplifiers with bipolar transistors.
A differential voltage can, of course, also be caused by the offset voltage, so that
VD = VO . In this case, a constant current Ii = VO /rd flows through rD , causing a
constant offset at the output. The output voltage caused by Ii can be calculated according
to Fig. 5.58b:
R1 + RN VO
Vo = − RN + Rg
R1 rD
Furthermore, even if the resistances are balanced, the offset is not compensated, because
both terms have the same sign. In the balanced situation of Rg = RN ||R1 , the equation is
simplified:
2 RN
Vo = − VO (5.57)
rD
The situation is much easier for the inverting amplifier shown in Fig. 5.59. Since the
differential voltage VD is in the millivolt range, the inverting input represents a virtual
ground here. Therefore, resistance R1 acts as if connected to a real ground. The input
resistance of the circuit is thus identical to R1 . It is not influenced measurably by the
differential and the common-mode input resistances of the amplifier. However, R1 is usually
in the range of 1 − 100 k, and is thus several orders of magnitude lower than the input
resistance of a noninverting amplifier.
Output Resistance
As Fig. 5.50 indicates, in terms of output resistance, real operational amplifiers are far from
ideal. However, the output resistance is reduced by negative feedback: the output voltage
reduction induced by the load is fed back to the inverting input via the voltage divider
5.2 Normal Operational Amplifier (VV-OPA) 533
Vo' /AD
Io ro Io
VD
Vi
Vo' Vo RL
RN
k Vo
R1
RN , R1 in Fig. 5.60. This increases VD and counteracts the original decrease of the output
voltage.
The model shown in Fig. 5.60 is used for quantitative analysis. Neglecting the current
through the negative feedback voltage divider, the output voltage is calculated from
Vo
Vi − = kVo and Vo
= Vo + Io ro
AD
which leads to the output voltage:
AD Vi − Io ro Vi Io ro
Vo = ≈ −
1 + kAD k g
In addition to the normal output voltage, there is also a reduction caused by the current,
but this is reduced by the loop gain. Consequently, the output resistance is:
dVo ro
ro
= − = (5.58)
dIo g
This means that the negative feedback reduces the output voltage by the loop gain.
Multiplier
0.33 mV
o o
rCM
Vo VCM
CMR VO
Vi
o o o
.
VD Vo
Vg .
rCM
.
2ID RN .
I0 RN
Fig. 5.61. Static error of a noninverting amplifier with a closed loop gain A = 10 for an
operational amplifier of the 741 class
This error also undergoes a 10-fold gain and at the output amounts to 0.33 mV. According
to Fig. 5.55, the effect of the offset voltage can be regarded similarly: a voltage of 1 mV
at the input produces an error of 10 mV at the output.
With matched input resistances, the output quiescent current IB has no effect. The
offset current causes an error of:
Vo = IO RN = 20 nA · 100 k = 2 mV
If the source had no internal resistance, it would be possible to insert a 10 k resistance with
a capacitor in parallel to suppress unwanted noise. If the input resistances are not matched,
the input quiescent current must be taken into account, which would then produce a voltage
error of IB RN = 8 mV.
The common-mode input resistances are obviously very high compared to all other
resistances, so that they seldom produce an error. Here, their effect is cancelled out because
the input resistances are matched. But the differential input resistance produces an error
that is caused by the current:
VD 1.5 mV
ID = = = 1.5 nA (5.59)
rD 1 M
VD
Vo = 2 · RN = 2 · RN ID = 2 · 1.5 nA · 100 k = 0.3 mV
rD
The error caused by the output resistance must also be evaluated. Assuming that the output
5.2 Normal Operational Amplifier (VV-OPA) 535
is loaded with the resistance RL = 1 k, the output current is Io = 10 V/1 k = 10 mA.6
Across the output resistance that is transformed according to (5.58), there is a voltage
drop of
ro 1 k
Vo = Io = 10 mA = 1 mV
kAD 104
When calculating the error, we have not taken the sign into consideration. Due to
errors caused by the differential gain and the output resistance, the output voltage becomes
smaller. But the signs of the offset voltage, the offset current, and the common-mode
rejection are not defined; it is therefore impossible to say with which sign they influence
the output voltage. The magnitude of the individual errors is more important: in this
example, no single error exceeds 1% of the output voltage. Most disturbing is the 10 mV
error caused by the offset voltage, as it is independent of the output voltage value. With
an output voltage of 100 mV, this amounts to 10%. For this reason, the offset voltage is an
important parameter when choosing an operational amplifier.
Bandwidth
Operational amplifier as a lowpass filter: Now that we have seen that a frequency-
corrected operational amplifier behaves approximately like a first-order lowpass filter, its
frequency response can be simply described as:
AD0
AD = (5.60)
f
1+j
fg
The differential gain of the open amplifier is usually very high; it often reaches values of
AD0 = 105 and above, as shown in Fig. 5.62. On the other hand, the cutoff frequency of
the open amplifier is usually very low and often amounts to only fg = 10 Hz.
A
Integrator approximation
5
fg
10
4
A D0 AD
10
3
10 g
+ 1 2 A 100
Vo 10 fT = Af’g = const
VD VD gm1 R1 Ck A 10
10
mA 33 pF A1
0.2 500 M Ω
V 1
–
1 10 100 1k 10k 100k fT f/Hz
Fig. 5.62. Frequency-corrected operational amplifier as a first-order lowpass filter for calculating
the frequency response of a circuit with negative feedback. An amplifier of the 741 class is taken as
an example. fg
is the cutoff frequency of the amplifier with feedback.
6 For a standard operational amplifier, this is a relatively high current, which is not far from the
maximum output current of 20 mA. Such high currents should only be allowed if they are un-
avoidable, since the operational amplifier becomes heated up by the power dissipation. The offset
voltage drift and offset current drift then cause additional errors.
536 5 Operational Amplifiers
According to (5.10), the frequency response of the circuit with negative feedback is:
AD 1/k
A = = (5.61)
1 + kAD 1
1+
kAD
Inserting (5.60) leads to:
kAD0 1
AD0 1 1/k
A = ≈ (5.62)
1 + kAD0 f f
1+j 1+j
fg (1 + kAD0 ) kfT
A comparison of the right-hand sides of (5.61) and (5.62) shows that one can use a simplified
frequency response of an open amplifier:
fT
AD = −j (5.63)
f
This represents the frequency response of an integrator; for this reason it is also known as
the integrator approximation of the operational amplifier. As shown in Fig. 5.62b, only low
frequencies lead to a deviation from the exact frequency response of the open amplifier:
here, the gain approaches infinity while the actual gain approaches AD0 . Insertion of the
integrator approximation (5.63) into (5.61) results in:
,
AD A0 A0 = 1/k fürf fg
A = = =
(5.64)
1 + kAD A0 A = −jf /f fürf f
1+j f D T g
fT
where A0 = 1/k is the gain defined by the negative feedback. In this way, we obtain the
result of (5.62) with less calculation and without further approximation. Thus, up to the
cutoff frequency fg
= fT /A0 = fT k, the gain of the circuit with negative feedback has a
value determined by the feedback; above the cutoff frequency the gain is the same as in the
open amplifier. This is illustrated in Fig. 5.62b, which includes the frequency responses for
different closed loop gains. It follows that the product of closed loop gain and bandwidth
is constant as can be seen in the figure:
fT = A0 · fg
= GBW = Gain BandWidth product
A RN
A = 1+
R1
+ 10
Vi
8
– CN RN Vo
1 pF 100 k Ω 6
C1
4 A = 1+
C1 R1 CN
1 pF 11.1 k Ω 2
1
f g = 1.65 MHz f
a Circuit b Frequency response
Fig. 5.63. Cutoff frequency caused by parasitic capacitances of the feedback resistances with an
ideal amplifier
Noise
The noise of an operational amplifier can be described by the noise voltage and noise
current densities related to the input, as is the case for discrete transistors. Typical values
are listed in Fig. 5.50 on page 524. These have to be multiplied by the root of the bandwidth
in order to calculate the noise voltage and the noise current:
√ √
Vn = Vnd B or In = Ind B (5.66)
Resistances also produce noise; their noise power
Pn = 4 kT B (5.67)
is independent of the value of the resistance. The value k represents Boltzmann’s constant
and T is the absolute temperature; at room temperature we have 4kT = 1.6 × 10−20 W s.
This can be used to calculate the noise voltage:
√ √ B R
Vn = P R = 4kT BR = 0.13 nV (5.68)
Hz
√
A 10 k resistance has a noise voltage density of Vnd = 13 nV / Hz. Figure 5.64 shows all
the noise voltage sources of an operational amplifier configured as a noninverting amplifier.
It can be seen that every resistance is a noise voltage source, and that the noise voltage
of the operational amplifier acts like the offset voltage and the noise current like the input
quiescent current. In the internal resistance Rg of the signal source, the input noise current
of the amplifier generates a noise voltage In Rg which, together with the noise of Rg and the
voltage noise of the amplifier, is amplified in the same way as the useful signal. The noise
of resistance R1 is amplified with the gain of 9 of the inverting amplifier. The noise current
at the inverting input produces a voltage drop across RN . In order to calculate the resulting
noise voltage at the amplifier’s output, the individual noise components cannot be simply
added. Since these are uncorrelated noise sources, the squares of the noise components
must be added:
-
0
Vn, tot = Vn2 (5.69)
This has the consequence that smaller contributions have almost no influence √ on the
result. Thus, in this example, the resulting noise voltage density Vno,tot = 353 nV/ Hz. To
538 5 Operational Amplifiers
nd
Vnd
Vo
nd .
Vg
Multiplier
Fig. 5.64. Noise sources of a noninverting amplifier, taking the 741 class as an example. The
closed loop gain of the circuit is A = 10
calculate the noise√ voltage, the bandwidth must be taken into consideration. This requires
multiplication by B and a correction factor of π/2 = 1.57, which takes into account
the fact that the noise above the cutoff frequency will not suddenly become zero, but will
decline in the same way as a first-order lowpass filter (see Fig. 4.180 on page 482). For a
bandwidth of B = 100 kHz, the example in Fig. 5.64 gives an output noise voltage
π 353 nV √
Vno, tot = √ 100 kHz = 175 mV (5.70)
2 Hz
In order to reduce the noise, it is necessary to configure a circuit with lower resistance
values and to use an operational amplifier with less voltage and current noise. Reducing
the resistances in Fig. 5.64 by a factor of 100 causes noise voltage reduction by a factor
√ of
ten. Using the modelAD797 amplifier, which has a noise voltage density of only 1 nV/ Hz,
we obtain a noise voltage of only Vno,tot = 11 mV at the output for the same bandwidth.
Since it already exists at the amplifier input, the internal resistance of the input voltage
source Rg represents a lower noise limit. This value can be calculated using (5.68). For
the purpose of comparison, one can convert the noise voltage at the output of the amplifier
to the value at the input by dividing it by the gain. The noise of the amplifier, including
its circuitry, is obtained by taking all resistors including the generator resistor and the
amplifier in Fig. 5.64 into account. This allows a comparison to be made in Fig. 5.65a as
to how much the amplifier contributes to the total noise. The voltage noise of the amplifier
is dominant in the case of low source resistances, while at high source resistances it is
the current noise that produces a noise voltage across Rg , which dominates. Since this
voltage is proportional to Rg , it rises in the logarithmic graph with double the slope of
the resistance noise Rg which rises with the square root of Rg according to (5.68). In
order to reduce the noise at low generator resistances, an amplifier with low voltage noise
must be√ used. For the purpose of comparison, we chose the AD797 amplifier, whose value
1 nV / Hz is only 1/10 of the voltage noise. It is obvious that source resistances in the
5.2 Normal Operational Amplifier (VV-OPA) 539
Vnd F
nV/ Hz
4
1000 10
3
100 Rg 10
µA 741 µA 741
2
10 10
AD 797
1 10
AD 797
0.1 1
10 100 1k 10k 100k 1M 10M Rg 10 100 1k 10k 100k 1M 10M Rg
Ω Ω
a The noise voltage b The noise figure
Fig. 5.65. Dependence of the noise voltage and the noise figure on the source resistance, using
models mA741 and AD797 as examples
range of 500 bring us close to the theoretical limits. For high generator resistances, the
low noise voltage offers no advantage. In this case, an amplifier with low current noise is
more favorable.
It is obvious from Fig. 5.65a that it is not the absolute noise value that determines the
quality of an amplifier, but the factor by which the circuit with amplifier produces more
noise than the generator resistance itself.
The noise figure indicates by which factor the noise power with a real amplifier
is higher than that of an ideal – that is, noise-free – amplifier at constant source
resistance noise.
It is calculated most easily from the relation
2
total noise voltage
F = (5.71)
noise voltage of the source resistance
Due to the ratio it makes no matter whether the quotient is taken at the input or the output
of the amplifier. You only have to translate all noise voltages with the gain to the output or
the input. For the example in Fig. 5.64 we obtain a noise figure
√
2 √
2
(353/10) nV/ Hz 353 nV/ Hz
F = √ = √ = 7.4
13 nV/ Hz (13 · 10) nV/ Hz
The interdependence of the noise figure and source resistance is shown in Fig. 5.65b. As
can be seen, there is a distinct minimum; the optimum noise figure is achieved with an
optimum generator resistance:
,
Vnd 10 nV/2 pA = 5 k for mA741
Rg, opt = = (5.72)
Ind 1 nV/2 pA = 0.5 k for AD797
There are systematic differences in the noise characteristic between the various tech-
nologies used in the construction of an input differential amplifier. A comparison is shown
in Fig. 5.66. Operational amplifiers with bipolar transistors√at the input have the lowest
noise voltage, which in good models amounts to only 1 nV/ Hz. Junction FETs at the in-
put result in clearly higher noise voltages, even in high-quality models. CMOS operational
amplifiers have the highest noise voltage but the lowest current noise at high frequencies.
Junction FETs are superior at low frequencies.
540 5 Operational Amplifiers
Vnd I nd
nV/ Hz fA/ Hz
Bipolar
1000 1000
100
100
CMOS 10
Vnd JFET
10 JFET 1
CMOS
Bipolar 0.1
1 f cI
fcV 0.01 f min f max I nd
Fig. 5.66. Voltage and current noise of a low-noise operational amplifier with bipolar transistors,
with junction FETs and MOSFETs at the input
Below a certain frequency there is an increase in both the voltage and current noise,
as shown in Fig. 5.66. As the noise density is inversely proportional to the frequency, this
noise is called 1/f noise. The frequency at which it enters the white noise is clearly higher
in CMOS operational amplifiers than in types with bipolar transistors or junction FETs at
the input. Data sheets usually specify the noise density in the region of white noise; this is
the region in which the noise density is independent of the frequency. In order to evaluate
the contribution of the noise voltage in the 1/f region, we require integration over the
noise density; this leads to:
f
max
fcV fmax
Vn = Vnd2 + 1 df = Vnd fcV ln + (fmax − fmin ) (5.73)
f fmin
fmin
f
max
fcI fmax
In
= 2
Ind + 1 df = Ind fcI ln + (fmax − fmin ) (5.74)
f fmin
fmin
Here, fmax and fmin are the cutoff frequencies of the region of interest and fcV and fcI
are the cutoff frequencies of the 1/f noise. They are shown in Fig. 5.66 as examples of
the current noise of a CMOS operational amplifier. For the frequency range from 100 Hz
to 100 kHz, the noise current is:
fA 100 kHz
In = 0.01 √ 1 MHz ln + (100 kHz − 100 Hz) = 26fA
Hz 100 Hz
5.3
Transconductance Amplifier (VC-OPA)
5.3.1
Internal Construction
The simplest circuit for a VC operational amplifier is achieved by taking the operational
amplifier shown in Fig. 5.11 and removing the emitter follower. The result is the circuit
shown in Fig. 5.67. Here, the characteristic value is the transconductance, which can be
calculated from the model:
Iq Ios 1 1 1 I0
gD = = = = gm = (5.75)
VD VD 2 rm 2 2 VT
With no load at the output, the current Iq flows through the output resistance rCE4 and
generates the open-circuit voltage gain
Vo Iq 1 I 0 VA 1 100 V
= rCE4 = gD rCE4 = = = 1923
VD VD 2 V T I0 2 26 mV
which is identical to that of the VV operational amplifier in Fig. 5.11. The amount of the
voltage gain with a connected load depends very much on the load resistance, since here the
output resistance of 100 k is much higher than in a VV operational amplifier, where it is
1 k or less. For high-resistive loads that permit a high gain, the VC operational amplifier
behaves almost identically to the VV operational amplifier.
In a practical realization as in Fig. 5.68, both output currents of the input differential
amplifier are used to control the output. This not only leads to doubled output currents but
also to markedly improved zero stability, as the quiescent currents I0 cancel each other out
at the output.
A special feature is the fact that the user can control the current I0 . The easiest way to
do this is to use a resistor Rcont with a voltage drop that is 0.6 V lower than the negative
operating voltage. A nominal current Icont = 0.5 mA therefore requires a resistance of
14.4 V
Rcont = = 28.8 k
0.5 mA
The maximum output current is thus Io,max = Icont = 2I0 = 0.5 mA. The transconduc-
tance of the circuit according to (5.75), and thus the voltage gain, can be controlled by
adjusting the current Icont . If a load resistance RL that is small compared to the output
resistance of the circuit, the output voltage is
rS
VP
VD Ios
VN rS
VP VN
Ios
Fig. 5.67. Simple VC operational amplifier. The values shown are for I0 = 1 mA
542 5 Operational Amplifiers
. .
VN VP
Ios
Icont
Icont
Rcont
. .
Fig. 5.68. Schematic circuit for the CA3080 from Harris. It is also known as an operational
transconductance amplifier (OTA) or gm -cell
RL
Vo = gD RL VD = Icont VD
2 VT
This characteristic can be utilized to multiply two voltages by making current Icont
proportional to a second input voltage. A more comprehensive description of such circuits
is contained in Sect. 11.8.2 on page 754.
Nowadays, model CA3080 is of no practical significance because of its outdated tech-
nology and the small output currents. But the MAX436 from Maxim or the OPA615 from
Burr Brown are modern successors that operate in push–pull AB mode and thus provide
high output currents. The circuit in Fig. 5.69 is obtained by taking the VV operational
amplifier shown in Fig. 5.28 and removing the impedance converter at the output stage.
This circuit has the particular advantage of working even with currents Iq > 2 I0 when the
upper or lower half of the circuit is without current.
The newer types enable the user to reduce the transconductance by means of an external
emitter resistance RE . Figure 5.70 shows how the resistance is connected in series with
the transconductance resistances of the input transistors. If we then take into consideration
the fact that the current mirrors of the MAX436 have a transformation ratio of kI = 8, the
transconductance of the circuit becomes
Ios kI Iq kI kI
gD = = = ≈
VD VD 2 rm + R E RE
Calculation of the voltage gain of the circuit takes the load resistance into account:
ro RL
kI RL
A = gD (RL ||ro ) = (RL ||ro ) ≈ kI
RE RE
As the gain can be adjusted to any value due to the current feedback via RE , no additional
negative voltage feedback is required.
One can see that all of the VC operational amplifiers shown here use transistors in
common-emitter configuration at their output in order to achieve a high output resistance.
For this reason, no special circuitry is required to realize a rail-to-rail output, as is the case
5.3 Transconductance Amplifier (VC-OPA) 543
+5V + + + +
I0 + 4.4 V
T9 T10
I0 A=8
T3 8( I0 + 1/2 Iq )
T1 T7
I0 – 1/2 I q T5 (8 Iq )
(0) I 0 + 1/2 I q
– (I q) – 8 Iq
+ RE Iq +
VN I 0 – 1/2 I q VP I os
I 0 + 1/2 I q (0) T6
T2 ( I q ) T8
T4 8( I0 – 1/2 Iq )
I0 (0)
I0 T11 T12
– 4.4 V
A=8
–5V – – – –
Fig. 5.69. Example of a modern VC operational amplifier (e.g., the MAX436). The circuit is also
known as a wide-band transconductance amplifier (WTA). The values in parentheses represent the
case Iq > I0
rS Iq
VP 1
0.15 Ω
ro
VD RE k I Iq RL Vo
4 kΩ
rS
kI = 8
VN 1
0.15 Ω
for the VV operational amplifiers described in Sect. 5.2.5. However, some of the standard
VC operational amplifiers use Wilson current mirrors that require a minimum voltage drop
of 0.8 V, as described in Sect. 4.1.1 on page 303.
5.3.2
Typical Applications
VC operational amplifiers are particularly suitable for driving coaxial cables. It is assumed
that their output resistance is high compared to the characteristic impedance of the cable.
This allows the cable to be terminated at either end by the characteristic resistance Rt con-
nected in parallel. The tranconductance of the amplifier is controlled by current feedback
through RE . The output voltage is:
1 kI Rt
Vo = Io R t = Vi
2 2 RE
In order to make Vo = Vi , it is necessary to use a negative current feedback resistance of
RE = kI Rt /2. The advantage of the parallel termination is the fact that the voltage at the
coaxial cable is the same as the output voltage of the amplifier. With low operating voltages
544 5 Operational Amplifiers
kI = 8
Io
Vi
Rt Rt Vo = Vi
RE
75 Ω 75 Ω
300 Ω
MAX 436 A
kI = 8
10
Vi
CE Ro Co Vo 1
RE 16 nF 125 Ω 120 pF f highpass f lowpass
100 Ω 0.1
1k 10 k 100 k 1M 10 M 100 M f/Hz
1 1
fhighpass = flowpass =
2πRE CE 2π Ro Co
At medium frequencies, the gain is A = kI Ro /RE . The capacitive load at the output is
not critical, as the circuit has no voltage feedback. But even with voltage feedback, VC
operational amplifiers are not affected by capacitive loads, because the high impedance
point with the lowest cutoff frequency is at the output. The load capacitance lowers the
cutoff frequency and thus improves the circuit stability.
5.4
Transimpedance Amplifier (CV-OPA)
5.4.1
Internal Design
The most simple type of CV amplifier is shown in Fig. 5.73b. For comparison, a normal VV
amplifier is illustrated in Fig. 5.73a. Transistor T1 of the differential amplifier in the VV
operational amplifier can be regarded as an impedance converter for the inverting input.
This is omitted in the CV operational amplifier, resulting in a low-resistance inverting
input. However, the emitter–base voltage of T2 must then be compensated at some other
point. This is done by the pnp transistor T1 at the noninverting input in Fig. 5.73b. In the
CV operational amplifier, the npn emitter follower at the emitter of T2 is replaced by a pnp
emitter follower at the base of T2 . Transistors T1 and T2 form a voltage follower that leads
from the noninverting input to the inverting input. Its output resistance is rS = 1/gm . The
signal at the noninverting input is transferred to the inverting input by two-step impedance
conversion. In this way, the voltage difference between the inputs becomes zero due to
the circuit configuration and not only as a result of external feedback, as is the case in
VV operational amplifies. Therefore the additional small amplifier symbol points from the
noninverting to the inverting input of the graphic symbol in Fig. 5.74.
The current control of the inverted input is somewhat unusual. If current Iq flows in
Fig. 5.73b, then the current through T2 increases by Iq . This increase is transmitted by the
current mirror and current Iq remains after removing current I0 . This current is not the base
current of T5 , which may be neglected in this case, but the current generating the voltage
gain at the internal circuit resistance rCE . The functional principle is easy to understand
using the model shown in Fig. 5.74. The resulting output voltage is:
VD
Vo = Iq Z = Z
rm
4.4 V 4.4 V
0.6 V
VP 0.6 V VN Vo VP VN Vo
–5 V – – – –5 V – – –
a VV-operational amplifier b CV-operational amplifier
T1 T2 T5
T3 T4
VP 1 1
VP Vo Z= Vo
VD VD Iq rCE4
rm 100 k Ω
VN
VN
26 Ω
RE
Fig. 5.74. Schematic symbol and model of a CV operational amplifier. The values shown apply to
I0 = 1 mA
Vo Z rCE4 VA 100 V
AD = = = = = = 3846
VD rm rm VT 26 mV
Here, Z is the transimpedance after which the amplifier is named. The higher Z is, the higher
is the differential gain. In terms of circuit arrangement, we have the internal resistance at
the node with the highest impedance, the collector of T4 in this case. For the negative
feedback, the CV operational amplifier feeds a portion of the output voltage back to the
inverting input via a voltage divider, as is the case with VV operational amplifiers. But
here the internal resistance shown as rm in the model in Fig. 5.74 reduces the open loop
+5V +
+ + 4.4 V + +
T3 T4
I0 I1
1
I 0 + /2 I q
+ 0.6 V
T1 T2 (I q) T5 T6
+ 0.6 V
– –
0V 0V 0V 0V
+ –
Iq Iq +
VP + VN Vo
– 0.6 V
T1' T5' T6'
– 0.6 V
T2'
I0 – 1/ 2 Iq
I0 (0) I1
T3' T4'
– – 4.4 V – –
–5V –
+5 V +
+4.4 V
T3 T4
+ +3.8 V + +
T5 T6
I0 I1
+ 0.6 V + 0.6 V
T2 T8
T1 T7
0V – 0V 0V – 0V
+ –
VP + Iq Iq +
VN Vo
T1' T7'
T2' T8'
– 0.6 V – 0.6 V
I0 I1
– T5' T6' – –
– 3.8 V
T3' T4'
– 4.4 V
–5V –
Fig. 5.76. CV operational amplifier with cascode current mirrors for increased voltage gain
548 5 Operational Amplifiers
VV operational amplifier
Without negative feedback
rm
1 1
Vi Vo
VD Iq
rm Iq R C
1 Z
VN ≈ 0 R1 RN 1 R HF 1
Z= R = =
sC 1+ sRC sC
Vi
Output voltage Vo = Iq Z = Z = A D Vi
2rm
Vo Z R / 2rm
Gain AD = = =
Vi 2rm 1+ sRC
R 1
A D0 = for f << f c Cutoff frequency fc =
2rm 2π RC
Case distinction AD =
1 1
for f >> f c Transit frequency f T =
2s rm C 4π rm C
With negative feedback
rS
1 1
Vi Vo
VD Iq RN
rS Iq R C
1 Z
R1 1 R1
Feedback factor k= ≈
R1+ RN A 0
Vo AD 1/k 1/k
Gain A= = = ≈
Vi 1+ k A D 1+ 2rm /(k . Z) 1+ 2s rm C /k Loop gain
1 R A D0 1
A 0 ≈ = 1+ N for f << f c g0 = ~
k R1 A0 A0
Case distinction A=
1 f
= – j T for f >> f c
2s rm C f
Cutoff frequency
1 k f
Transit frequency fT = = const fc = ≈ T
4π rm C 4π rm C A0
Frequency response
A
4
10
3
10 A D10
g10 A D1 Partially compensated
10
2
g1 fc f c'
A 0 = 10 f c10 '
f c10
10 f c1 = f T
A0 = 1 f T'
1
1k 10k 100k 1M 10 M 100 M 1G f/Hz
CV operational amplifier
Without negative feedback
rm
Vi Vo
VD ≈ 0
VN ≈ Vi =
Vi
Output voltage Vo
Vo
Gain
Vi
Vi Vo
VD ≈ 0
Feedback factor
Vo AD 1 + RN /R1 1/k
Gain A= = = ≈
Vi 1+ k A D 1+ RN /Z 1+ s RN C Loop gain
gain
Loop
R
for f << f c g0 = k A D0 = =
R1 const
Case distinction for
for f >> f c
Cutoff frequency
1 A0 k
Transit frequency fT = = A0 fc = = const
2 (R1 RN)C 2 RN C 2π R nC
Frequency response
fc
fc
fc
The voltage gain resulting from (5.76) is usually insufficient, as it is further reduced
by the external resistance RE formed by the feedback voltage divider. In order to increase
the voltage gain, it is common practice to increase the internal resistance at the high-
impedance point. This is equivalent to increasing the transimpedance Z. As in the VV
operational amplifier shown in Fig. 5.25, cascode current mirrors can be used as shown in
Fig. 5.76. According to (4.27), this increases the internal resistance at the high-impedance
point by the current gain β of the transistors. The differential gain also rises by this factor:
(5.76):
Z 1 β rCE
Aoperating = = (5.77)
RE + r m 2 RE + r m
The factor of 1/2 accounts for the fact that two current sources are switched in parallel at
the high-impedance point. A drawback of the cascode current sources is the fact that the
common-mode and output control range is reduced by 0.6 V. At an operating voltage of
±5V, this gives a value of only ±3.6 V.
5.4.2
Frequency Response
Transimpedance amplifiers are only used where high bandwidth or short rise times are of
importance. Lately, wide-band VV operational amplifiers have also become available; they
are produced with the same technology and operate in AB mode as shown in Fig. 5.28. The
two amplifier types have been placed side by side in Fig. 5.77, to illustrate the differences.
The major distinction becomes visible in the model: in CV operational amplifiers there
is no impedance converter at the inverting input. Therefore, the transconductance of the
input stage is determined by the resistance at the inverting input:
Iq 1 1
gm = = =
Vi rm + R E rm + R1 ||RN
For this reason, the resistors R1 and RN must be taken into account when analyzing
the open circuit. In practice rm R1 ||RN , which means that the effect of rm may be
neglected. This means that voltage VD ≈ 0 and voltage Vi is generated across R1 and
RN . Therefore, the gain of a CV operational amplifier is clearly lower than that of a
comparable VV operational amplifier with the same resistance R at the high-impedance
point. As R is usually of the order of magnitude of 1 M, it only influences the gain at low
frequencies. The parasitic capacitances C, of a few picofarad, already cause a drop in gain
at medium frequencies. This can be accounted for in the calculation by using Z, the parallel
arrangement of R and C. With high frequencies it is the capacitance that determines the
characteristics, so that the resistance need not to be taken into account; this facilitates
the calculation. One can see that the cutoff frequencies of both amplifiers are the same.
The transit frequencies, however, are different: while that of the VV operational amplifier
is given by the inner circuit design, that of the CV operational amplifier depends on the
external circuitry.
When analyzing the CV operational amplifier with negative feedback, it is essential
to take the current at the inverting input into consideration and not to assume the feed-
back voltage divider to be without load, as it is the case with VV operational amplifiers.
Kirchhoff’s first law is applied to the inverting input to calculate the voltage gain:
5.4 Transimpedance Amplifier (CV-OPA) 551
Vo − V i Vi Vo
− + = 0
RN R1 Z
For low frequencies, this yields the same result as for the VV operational amplifier,
as shown by the comparison in Fig. 5.77. It seems strange that the current at the inverting
input does not alter the result. The reason for this is that the current Iq is small, since for a
resistance of R = 1 M even an output voltage of 5 V only requires a current of Iq = 5 mA.
The cutoff frequency of the VV operational amplifier with negative feedback is in-
versely proportional to the set gain; the gain–bandwidth product is constant equal to the
transit frequency as shown in Fig. 5.77. In CV operational amplifiers with negative feed-
back, a cutoff frequency that is independent of the adjusted gain can be achieved by keeping
RN constant and setting the gain by means of R1 . For in this case the loop gain remains
constant: decreasing R1 in order to heighten the gain increases the open-circuit gain by
lowering the current feedback to the same degree. For this reason, manufacturers usually
quote the optimum value for RN at which the loop gain just reaches the level at which
a favorable transient response is achieved. This value of RN is already incorporated into
some types.
If resistance RN of a CV operational amplifier is kept constant and the gain is set by
means of R1 , the following deviations from the VV operational amplifier can be observed,
as expressed in the formulas of Fig. 5.77:
– The bandwidth of the feedback circuit is independent of the gain selected.
– The loop gain of the feedback circuit is independent of the gain selected.
– The transit frequency of the feedback circuit is proportional to the gain selected.
A matched frequency response correction is also possible in VV operational amplifiers
(see Fig. 5.37); however, here it would be necessary to change the correction capacitance
together with the gain, but this is only possible by exchanging the OPAmp. Under this
condition the cutoff frequency remains here also constant, as shown in Fig. 5.77.
Two properties in which both circuits are equal are demonstrated by the formulas:
– The bandwidth of the feedback circuit is increased by the loop gain, compared to that
of the open amplifier, by a value equal to the loop gain.
– The transit frequency of the circuit is not altered by the feedback.
These conditions are illustrated by the frequency–response plots in Fig. 5.77.
5.4.3
Typical Applications
In order to allow the feedback of the CV operational amplifier to control the gain through
negative current feedback, it must consist of ohmic resistances. The circuit becomes un-
stable if RN or R1 is replaced by a capacitor. This is the reason why an integrator or
differentiator cannot be realized by means of a CV operational amplifier. Therefore, they
are mainly used as amplifiers with high bandwidth; for example, as video amplifiers. They
can be used as inverting or noninverting amplifiers as shown in Fig. 5.78. Resistance RN
determines the loop gain and is thus given by the amplifier to a large degree. Resistance
R1 determines the voltage gain and is of rather low impedance for higher gains. As in all
inverting amplifier, resistance R1 represents the input resistance; the noninverting mode is
therefore usually preferred.
552 5 Operational Amplifiers
Vi
Vo
Vi
Vo
Vi
Vo
RN
Vi
Vo
RN C
R1
R1'
It is important that resistance RN holds its optimum value even for gain A = 1 and that
it must not be reduced to zero (see Fig. 5.79a). Only resistance R1 may be omitted here. To
counteract the reduction in gain that occurs around the cutoff frequency, the gain can be
increased by the additional R1
C element shown in Fig. 5.79b. However, one should check
whether it is possible to achieve the same effect with a higher loop gain that results from
low resistances in the feedback voltage divider.
5.5
The Current Amplifier (CC-OPA)
The CC operational amplifier differs from the CV operational amplifier in the same way
as the VC operational amplifier differs from the VV operational amplifier; that is, in that
the impedance converter is omitted at the output.
5.5.1
The Internal Design
The basic circuit is shown in Fig. 5.80a. From a comparison with the CV amplifier in
Fig. 5.73, one can see that the only difference is the lack of the emitter follower at the
output. Assuming that the inverting input is connected to ground, the circuit can be divided
into two parts:
5.5 The Current Amplifier (CC-OPA) 553
+5V +
+ + + +4.4 V
T3 T4
+5V I0
T3 T4
+ 0.6 V I 0 + 1/2 I q
I0 T2 ( Iq )
I0 + Iq I0 + Iq T1
T2 Iq 0V – 0V Iq
+ T1 – + –
VP Iq VP + Iq
VN I os VN I os
T1'
– 0.6 V
T'2
I0 I0 I 0 – 1/2 I q
(0)
I0
–5V – –
T'3 T4'
– – 4.4 V
–5V –
a Principle b Practical design
Since the current flowing at the inverted input (the emitter of T2 ), is transferred to the
output, the CC operational amplifier is also called a current amplifier. The output current
of the circuit shown in Fig. 5.80 is equal to the input current; this means that the current
amplification factor is one. Higher amplification factors may be achieved by providing the
current mirror with a gain; values of up to kI = 8 are available.
Thus, the entire operational amplifier is no more than an expanded transistor. This is
the reason why two graphic symbols are used for CC operational amplifiers, as shown in
Figs. 5.81a,b. When they are used in circuits that are also common for VV operational
amplifiers, the graphic symbol for the operational amplifier is preferred. However, the CC
operational amplifier can also be used like a transistor, in which case the symbol for the
B
VP C
C B VP 1
Vo
VN VD Iq ro Co I os
E rm
VN
E Z
RE
transistor is more familiar. The CC operational amplifier – the current amplifier – and the
simple transistor have a lot in common:
– The collector current has the same (absolute) value as the emitter current.
– The input resistance is high at the base and low at the emitter.
– The output resistance at the collector is high.
But there are also differences, which make it easier to use a CC operational amplifier than
a transistor:
– The collector current has the opposite polarity because of the current mirror.
– The base–emitter voltage is zero: VBE = 0 due to compensation by T1 .
– The emitter and collector currents can have both polarities.
– The operating point is set internally.
For these reasons, the CC operational amplifier performs like an ideal transistor. There-
fore the manufacturer Burr Brown calls it a diamond transistor.
Due to its short internal signal path, the CC operational amplifier has particular advan-
tages for high frequencies. For this reason it is constructed in AB push–pull mode (current
on demand), to allow high output currents even with small quiescent currents. A circuit
used in practice is shown in Fig. 5.80b. A comparison with the basic circuit shows that the
current sources are replaced by a complementary circuit.
The model given in Fig. 5.81c shows the high-resistance noninverting and low-
resistance inverting inputs. The output has a high resistance. The dominating lowpass
filter is at the output. Its cutoff frequency depends on the load connected. One can see that
the short-circuit current at the output is equal to the current at the inverting input. The short-
circuit transconductance of the circuit is equal to the transconductance of transistor T2 :
Ios 1
gm = = (5.78)
VD rm
In practical applications, there is usually a resistance RE at the inverting input, which
reduces the transconductance. The operational transconductance of the circuit is then:
Ios 1
gm,op = = (5.79)
VP rm + R E
This allows the open-circuit voltage gain to be calculated:
Vo Ro,tot Ro,tot
Aop = = gm,op Ro,tot = = (5.80)
VP rm + R E RE,tot
5.5.2
Typical Applications
In most applications the behavior of the CC operational amplifier is determined by the
current feedback at the inverting input; voltage feedback is used in special cases only.
5.5 The Current Amplifier (CC-OPA) 555
IC = IE
B C
1
Vi rm IE ro RC Vo
0V
IE IE
RC E
Vi Vi RE RC Vo = V
RE i
RE
a Schematic b Model
1
0V
Vo
0V C
Vi Vi RE RC Vi R1 R1' RC Vo
C
rm
o
E
Vi Vo = Vi Vo
a Schematic b Model
IC = IE
B
1
Vi rm IE
0V
IE IE
IE
C
Vi RE Vo= Vi E 2I E
Vo
a Schematic b Model
Fig. 5.85. Use of the collector current in the CC operational amplifier as an emitter follower
IC = IE
B C
1
IE
Vi 0V RC Vi rm I E ro R C Vo
Vo
IE RE
E
RE IE
a Schematic b Model
IC = IE
V1 IC = IE
V1 0V
RC Vo 0V
R1 IE RC Vo
RE IE
V2
R2 RC R RC
V2 Vo = – V – C V Vo = (V – V 2)
R1 1 R2 2 RE 1
Common-base circuit: In the common-base circuit, the input signal is fed to the emitter
via a resistor, and the collector current generates the amplified output signal across the
collector resistance, as shown in Fig. 5.86. If we assume an ideal CC operational amplifier
with VBE = 0, the emitter current is IE = Vi /Ri ; across the collector resistance, this
current generates the voltage
RC
Vo = IC RC = − Vi
Re
The model in Fig. 5.86 is best suited for an exact analysis and yields:
ro ||RC RC
Vo = − IE (ro ||RC ) = − Vi ≈ − Vi
rm + R E RE
This is the same result as in the common-emitter circuit but with a negative sign. Compared
to the simple transistor, the common-emitter and the common-base configurations of the
CC operational amplifier produce the opposite polarity for the voltage amplification.
Since in the common-base circuit the emitter carries zero potential via a low resistance
rm , currents at this point can be added interaction-free, as at the summing point of a VV
operational amplifier. This possibility is shown in Fig. 5.87a. It is also possible to combine
the common-emitter and common-base circuits to form the subtractor shown in Fig. 5.87b.
RC RC
Vo1 Vo2
Iq Iq
T1 T2
Vi1 VD Vi2
Fig. 5.88. Differential amplifier made up of two CC
Iq RE Iq
operational amplifiers
Vi1 − Vi2 VD
Iq = =
RE RE
Since the collector current has the same value, the output voltages are:
RC RC
Vo1 = Iq RC = VD and Vo2 = −Iq RC = − VD
RE RE
The model shown in Fig. 5.89 can be used for an accurate calculation of the voltage gain.
It allows the transconductance resistances rm to be taken into account when calculating
the cross-current:
VD
Iq =
RE + 2rm
The output resistances are in parallel to the collector resistances, so that the output voltages
are:
RC ||ro RC ||ro
Vo1 = VD and Vo2 = − VD
RE + 2rm RE + 2rm
The differential amplifier has already been used in applications for the VC operational am-
plifier in Figs. 5.71 and 5.72. The external emitter resistance RE determines the transcon-
ductance of the circuit. The short-circuit transconductance is:
Ia1 1
gm = =
VD RE + 2rm
ro
rm Vo1
Vi1 1
Iq
VD Iq RC
RE
Vi2 1
rm
Vo2
ro
I2
I1
T3
V1 C V2
I2
T1 V1 Rg
L = Rg2 C T2
Rg I1
V2
Fig. 5.90. Gyrator realized with CC operational amplifiers as voltage-controlled current sources
Gyrator: Voltage-controlled current sources are particularly suitable for realizing gyra-
tors, as they enable the direct realization of the required transfer equations:
1 1
I1 = V2 I2 = V1 (5.81)
Rg Rg
These formulas can be derived directly from Fig. 5.90 if we assume that VBE = 0 and
IB = 0. To achieve the correct sign for the current, it suffices to place a simple CC
operational amplifier in the signal path from left to right; while in the opposite direction a
differential amplifier is necessary, according to Fig. 5.88. In order to achieve high-quality
gyrators, the current sources must have high output resistances. Model OPA615 (Burr
Brown) is particularly well suited for this purpose, as it contains a cascode current mirror
at the output with an output resistance in the megaohm region.
If one connects a capacitor C to one side of the gyrator, the inductance L = RG 2 C is
produced at the other side according to (12.21) on page 782. Connecting a capacitor to
both sides produces an oscillating circuit that can be damped with a parallel resistor. The
resulting circuit is identical to the filter in Fig. 5.93 if an input signal is fed to the base of
T1 and the output signal is moved from the emitter of T2 in Fig. 5.90. It is astonishing that
both filters, which are totally different in approach, produce the same circuit.
IC = IE
IE
1
IE
Vi rm I E ro Co C Vo
Vi Vi R C Vo
R VV-OPV ro R C
1
Vi Vo Vi Vo
VD gmVD R1 C1 V1 ro
1 1
Vo = IC dt = Vi dt
C RC
Of course, it is also possible to calculate the transfer function if the complex resistance of
the capacitor is considered
IC Vi
Vo = =
sC sRC
In order not to affect the operation of the circuit, it is necessary to tap the voltage at the
capacitor without a load; this normally requires an additional impedance converter.
The model shown in Fig. 5.91 can be used to study the effects of the real properties of
a CC operational amplifier used as integrator. Here too, the transconductance resistance
rm is in series with the external emitter resistance R and can be taken into account if the
external resistance is reduced accordingly.
The dominating lowpass filter ro Co limits the lower cutoff frequency of the integrator
to f low = 1/2πro (C + Co ). This limitation is common to all integrators, since with low
frequencies in ideal integrators the gain would reach infinity. The parasitic capacitance Co
presents no limitation as it is parallel to the integration capacitance C. To take this into
account, one can make the external integration capacitance C accordingly smaller. Thus,
on the basis of the model, the bandwidth is not limited toward high frequencies. Of course,
due to secondary effects, the CC integrator also has an upper cutoff frequency; however,
this value is in the very high frequency range.
By way of comparison, the features of the VV operational amplifier are much more
disadvantageous. Figure 5.92a shows the model of a VV operational amplifier that is
configured as an integrator. Above the transit frequency of the opamp fT = gm /(2π C)
the voltage becomes V1 = 0; the output resistor ro is thus at zero potential. In this case, the
model can be simplified as shown in Fig. 5.92b. The integration capacitance now acts as a
coupling capacitance and transfers the input signal to the output instead of short-circuiting
it. In this frequency range, the circuit operates solely as a voltage divider according to
Vo = Vi ro /(ro + R).
Filter: Since CC operational amplifiers can be used to achieve integrators with very good
properties for high frequencies, they are particularly suitable for active high-frequency
filters based on integrators (see Sect. 13.11 on page 831). An example of a combined
bandpass/lowpass filter of second order is shown in Fig. 5.93. It consists of two CC inte-
grators and one voltage follower. In contrast to the integrator filters with VV operational
5.5 The Current Amplifier (CC-OPA) 561
Vi
VBP VLP
Fig. 5.93. Active second-order high-frequency filter, with bandpass or lowpass output. The √
dimensions shown result in a resonance or cutoff frequency of 30 MHz with Q-factor of 1/ 2
(Butterworth). When dimensioning the circuit, the transconductance resistances of rm = 10 and
the circuit capacitances of 6 pF in parallel to the integration capacitors were taken into account.
The circuit is very accurate up to more than 300 MHz
amplifiers, the CC integrators are not inverting. Therefore, no inverter is required in the
filter loop. The transfer function can be derived from the circuit:
VLP 1
= (5.82)
Vi 1 + sCR /R1 + s 2 C 2 R 2
2
VBP sRC
= (5.83)
Vi 1 + sCR /R1 + s 2 C 2 R 2
2
1 R1
fr = Q =
2π RC R
We can see from the formulas that the resonance frequency and the Q-factor can be set
independently of one another.
IC = IE Io
IC = IE
Vi RN
Vi Vo
Vo IE
RN
IE
2 IE
Vi R1 Vi R1
Fig. 5.94. CC operational amplifier with additional voltage feedback. This configuration is called
direct feedback
562 5 Operational Amplifiers
IE IE Io
1
Vi rm IE ro Co Vo
RN
IE
V1 R1
Fig. 5.95. Model of a CC operational amplifier for calculating the voltage gain and the bandwidth
with voltage feedback
is inverted. As the feedback loop follows the shortest path possible – that is, there is no
impedance converter at the input or at the output – this is also known as direct feedback.
To calculate the voltage gain, we assume an ideal CC operational amplifier with VD =
VBE = 0. With no load at the output, the current 2IE flows through resistor R1 . The emitter
current is IE = 2Vi /2R1 . The output voltage can now be calculated:
RN RN
Vo = Vi + IE RN = Vi + Vi = 1 + Vi
2R1 2R1
The formula is very similar to that for the VV operational amplifier, apart from the factor
of two in the denominator. The formula is based on the assumption that for an unloaded
output, Io = 0. A load causes a reduction in the gain. Should this be a problem, a voltage
follower can be connected at the output.
In order to calculate for the influence of the transconductance resistance and the out-
put resistance in the CC operational amplifier, the model shown in Fig. 5.95 is used. If
Kirchhoff’s first law is applied to the emitter and the collector, then:
Vi − V 1 Vo − V1 V1
+ − = 0
rm RN R1
Vi − V1 Vo − V1 Vo
+ − = 0
rm RN ro
Consequently, the accurate value of the open-circuit voltage gain is
RN ro →∞
1+ rm =0
Vo 2 R1 RN
A = = = 1+
Vi 1 r m RN rm 2R1
1+ RN + r m + +
2ro R1 2R1
The bandwidth of the circuit shown in Fig. 5.94 is best calculated on the basis of the
model shown in Fig. 5.95. For the sake of simplicity, we presume an ideal CC operational
amplifier – with the exception of capacitance Co – which means that rS = 0 and ro = ∞.
Again, for circuit analysis Kirchhoff’s first law can be applied to the emitter and the
collector:
Vo − Vi Vi
IE + − = 0
RN R1
Vo − Vi
IE − − Vo sCo = 0
RN
5.5 The Current Amplifier (CC-OPA) 563
Rt
IC Io kI = 3 75 Ω
1
Vi k I IE Vo Vi
0V
RN RN Rt Vo
300 Ω 75 Ω
IE
Vi
R1 R1
0V
Voltage output
+
VD Vo
– +
I0 + Iq + I0 + Iq
Graphic symbol
+ Iq – Iq –
+ VP VN Vo
VD AD V D Vo
Voltage input
– I0 I0 I0
Ideal amplifier – – – –
Circuit
Vi 1 Vo =
rm + RN
+ 1 1 (1+ R ) V 1
i
Vo
VD Iq Z – 1 RN
– 1
rm
Z Voltage feedback R1
Model Vo = V
2rm D
Transimpedance amplifier
CV-OPA
+ + +
VD I0 T3 T4
Vo
Graphic symbol I0 + Iq T5
Iq
T1
+ T2 –
+ 1 VP Iq VN Vo
Current input
Iq A D VD Vo
Iq I0 I0
– – – – –
Circuit
Ideal amplifier
Vi 1 Vo =
+
+ 1 1
Vo
( 1+
RN
V
R1 i)
VD Iq Z – RN
– Iq
rm
Z Current feedback R1
Model Vo = V
2rm D
Current output
Transconductance amplifier
VC-OPA + +
+
T3 T4
VD Io
–
I0 + Iq I0 + Iq
Graphic symbol Iq
+ Iq –
+ VP VN I os
T1 T2
VD I o = gmVD
gmVD
Voltage input
– I0 I0 I0
– – –
Ideal amplifier Circuit
Vi Vo =
T1 rm T +
+ 1
3 T4
( 1+
RN
)V
R1 i
VD I os
T2 Iq Z – 1 RN
– 1
rm
VD R1
Model I os = Voltage feedback
2rm
Current amplifier
CC-OPA
+ + +
+ Io
VD Vo + T3 T4
–
–
Graphic symbols I0 + Iq
T1 Iq
+ T2 –
+ 1 VP Iq VN I os
Current input
VD Iq Io = Iq
I0 I0
– – – –
rm
Circuit
Ideal amplifier
Iq
Vi Vo =
T2 T1 +
+ 1
T3 T4
( 1+
RN
V
2R1 i)
VD Iq I os –
Z RN
– Iq
rm
Current feedback R1
Model I os = I q
Voltage output
Common name Normal operational amplifier
Systematic name VV operational amplifier
Function as controlled source Voltage-controlled voltage source VCVS
Feedback/output description Voltage feedback, voltage output VFVO
Type of feedback Voltage feedback
Applications Amplifier for low frequencies
Voltage input
Current output
Common name Transconductance amplifier
Systematic name VC operational amplifier
Function as controlled source Voltage-controlled current source, VCCS
Feedback/output description Voltage feedback, current output, VFCO
Type of feedback Voltage feedback
Applications Drivers for capacitive loads
Voltage input
Advantages Low offset voltage
Low drift
Good transient response with capacitive loads
Disadvantages Load must be known for calculation
5.6
Comparison
An overview is given of the common features and differences between the four different
operational amplifiers. The important properties are listed in Figs. 5.97 and 5.98. In the
graphic symbols, the current source symbol of types with current output is an indication
of a high-resistance output with an impressed output current. The types with current input
have an amplifier symbol between the inputs, to indicate a high-resistance noninverting
input and a low-resistance inverting input.
Every operational amplifier may be regarded as a controlled source that describes the
ideal amplifier, whereby amplifiers with low-resistance outputs are voltage sources and
those with high-resistance outputs are current sources. A high-resistance (inverting) input
forms a voltage-controlled input and a low-resistance input a current-controlled input.
The descriptions of their function as controlled sources in Fig. 5.98 provide the two-letter
designations used so far for any of the four types of operational amplifiers. From the
systematic presentation it is obvious that no other types can exist; each and every circuit
fits into the matrix of the four operational amplifiers.
The models shown in Fig. 5.97 describe the most important real characteristics of
operational amplifiers. If the element Z is realized as a parallel connection of a resistor
and a capacitor the frequency response can also be modeled. This has been used to calculate
the cutoff frequencies of the various types.
The circuit diagrams show simple realizations of the examples already discussed.
Operational amplifiers with a voltage input feature a differential amplifier at the input,
while those with a current input have a voltage follower with compensated base–emitter
voltage. Those with voltage outputs feature an emitter follower at the output that does not
exist in types with current outputs.
A particularly instructive comparison is achieved by presenting the CC operational am-
plifier, the most simple of the four amplifier types, as a (diamond) transistor and realizing
the remaining three types by adding impedance converters. This makes it clear that the CV
operational amplifier requires a voltage follower at its output, the VC operational amplifier
a voltage follower at its inverting input, and the VV operational amplifier both simultane-
ously. For this reason, it is possible to realize all four types of operational amplifiers with
CC operational amplifiers, as shown by a comparison of Figs. 5.97 and 5.99.
5.6 Comparison 569
13 (12) 13
2 (3) 2
3 (2) 3
6 1 5 6
9 Fig. 5.99. Comparison of standard
amplifiers with current output:
MAX 436 MAX 435 illustrations of the pin numbers of the
(OPA 615) dual inline case
For the purpose of comparison, the four operational amplifiers are presented as non-
inverting amplifiers in Fig. 5.97. As the feedback always leads to the inverting input it
determines the kind of feedback: If the inverting input is a voltage input (high impedance)
a voltage feedback amplifier results; if it is a current input (low impedance) a current
feedback amplifier results even though voltage feedback exists at the same time. Pure
current feedback is achieved by simply connecting the inverting input to ground via a
resistor (see Fig. 5.82–5.93). The voltage gain for voltage feedback shown in Fig 5.97 is
the same in all cases, except for the CC operational amplifier, which has a factor of two
in the denominator. The feedback output description in Fig. 5.98 also leads to the typical
systematic short names for the operational amplifiers.
The feedback loops in Fig. 5.97 show that the longest path exists in the VV operational
amplifier and the shortest path in the CC operational amplifier. This is the reason why,
in CC operational amplifiers, high frequencies cause least phase lag and thus the fewest
stability problems. For this reason, the CC operational amplifier is particularly suitable for
high frequencies. This difference is shown in Fig. 5.100. Even though both circuits require
the same amplifier and have a low output resistance, the CC operational amplifier features
a shorter feedback loop than the transimpedance amplifier where the impedance converter
is outside the feedback loop.
Vo Vo
Vi RN Vi RN
R1 R1
5.6.1
Practical Implementation
Many parasitic effects cannot be detected by circuit simulation. In particular, these include
the inductances caused by wiring, as they depend on the printed circuit board layout. Only
few simulation programs are capable of extracting these parameters from the layout and
automatically taking them into account during simulation (post-layout simulation). This
is not required for low-frequency circuits, but it becomes more and more important with
increasing frequencies above 1 MHz. Above 30 MHz, even the inductance of the cases of
integrated circuits play an important role. For this reason, SMD components are especially
advantageous for high frequencies, as their parasitic inductances are significantly lower
on account of their small dimensions. The most important issues to be observed in the
practical use of operational amplifiers are described below.
Blocking operating voltages: Operating voltages must be well blocked. Naturally, the
operating voltage leads have an inductance that increases with their length. As shown in
Fig. 5.101a, these inductances are short-circuited by capacitors in order to prevent any
voltage drop. A precondition is that the ground lead inductance of the capacitor is small in
respect to the supply inductance. One way to achieve this is with a close ground network,
or – even better – with a ground plane that has gaps for pins only. Capacitors also widely
vary in their high-frequency response. An electrolytic capacitor has a low resistance low
frequencies on account of its high capacitance. However, its resistance increases with
higher frequencies due to its parasitic inductance. To achieve low resistances even with
these frequencies, ceramic capacitors, with low inductance, are connected in parallel.
Tendency to oscillate: The circuit may oscillate, especially with capacitive loads or
when driving an amplifier below Amin . The cause of this, however, may also turn out
to be an unfavorable layout or insufficient blocking of the operating voltages. Often, the
amplitude is low and the frequency high, so that the oscillation does not become directly
apparent. An indication of oscillation is often the inaccurate operation of the circuit with
DC voltages. Therefore it is always recommended to use an oscilloscope in order to make
sure that the circuit is working properly. However, what must be borne in mind is the
fact that the input of an oscilloscope represents a capacitive load that contributes to the
tendency of the operational amplifier to oscillate. For this reason, the oscilloscope should
never be connected via a coaxial cable or a 1:1 probe, but only via a 1:10 probe whose
L
+
+ Vo
Good
–
+ t
V
– Bad
V
–
L t
+
+
capacitance is usually only a few picofarad. Grounding should be carried out via a short
lead connected to a location close to the test point.
Damping: Having established that no amplifier of a given board oscillates, the next step
is to insure that the amplifier is operated far from any oscillating condition. On the one
hand, oscillations may occur if temperature or load changes; on the other hand, there is a
general desire for a well-damped transient response. It is therefore beneficial to apply a
square-wave signal and observe the output signals on an oscilloscope. This gives a good
impression of the circuit damping. An example of a good and a bad square-wave response
are shown in Fig. 5.101b.
Cooling: Higher output currents require additional cooling of the operational amplifier.
As long as the dissipation is around 1 W, the PC board can be used as heat sink via a few
square centimeters of metal-plated.
+
V
VP RS IP +
V
Vo
Vi C
VN V
– Vo
RN
–
V R1
Input protection: The input voltages of an integrated circuit must not exceed the operat-
ing voltages, since otherwise the parasitic diodes shown in Fig. 5.102a become conductive.
Often, the maximum permissible currents are around as little as 10 mA. The moment after
turn-off, when the operating voltages drop to zero, is particularly critical, because at this
instant the maximum input voltage is only ±0.6 V. If a charged capacitor is connected to
the input, dangerous high discharge currents may flow through the diodes. The same sit-
uation occurs if a correspondingly high input signal remains. In both cases, the protective
resistance RS shown in Fig. 5.102b is very useful for limiting the current.
5.6.2
Types
There are many types of operational amplifier, that are optimized for a variety of different
applications. To provide the reader with an overview on which data can be expected, some
of the typical devices for the various applications were set out in Fig. 5.2. The list of
manufacturers shows which companies are particularly active in this field. Of course, each
of them offers many more types of operational amplifier, whose data sheets are available
via the Internet. For details of the website addresses, see Sect. 29.8 on page 1518.
The offset voltage and the quiescent input current have been given as criteria for
comparing the accuracy of DC voltages. This is followed by the gain–bandwidth product
and the slew rate, which characterize their high frequency properties. The gain–bandwidth
product has a clear meaning: it shows the bandwidth of an amplifier for the gain A = 1.
The slew rate allows the power bandwidth to be calculated; this is the frequency up to
which the full output amplitude is available. According to (5.31), this frequency is:
SR
fp =
7o
2π V
The values for the minimum or maximum operating voltage indicate which types are
suitable for low operating voltages and which types can provide high output amplitudes.
The figures stated here are the minimum and maximum voltages between the positive and
a negative operating voltage terminal for which the chip is specified. Since operational
5.6 Comparison 573
amplifiers have no ground connection, it is up to the user to decide how the voltage is
distributed between positive and negative operating voltages (see Sect. 5.2.5). Symmetrical
operating voltages of ±5 V or ±15 V are normally used. At rail-to-rail output (RRO)
amplifiers the output voltage can nearly reach the supply voltages. Thus, appreciable output
amplitudes can be obtained even at low supply voltages.
The maximum common-mode voltage and output voltage of normal operational ampli-
fiers is approximately 2 V within the operating voltages. At rail-to-rail input-output (RRIO)
amplifiers the common-mode voltage may also reach the supply voltages and even exceed
it by 200–300 mV. They are especially adapted for single supply operation.
Often, two or four amplifiers, rather than just one single operational amplifier, are
contained in one case. Figure 5.103 shows how many amplifiers are contained in one
case. We have preferred to list two-fold or four-fold operational amplifiers. Often, single
operational amplifiers are available under a similar model designation. Where the current
consumption is quoted, this always refers to one amplifier.
Universal types: These have no specific electrical properties; but the old standard types,
such as the 741 and 324 are particularly low-priced. From the value of the input current it
is possible to determine the technology used at the input of the differential amplifier: for
bipolar transistors this value is in the nanoampere range, while for field effect transistors
it is in the picoampere range.
Precision types: The most important preconditions for high accuracy with DC voltages
and low frequencies are a low offset voltage and a high differential gain. A low quiescent
input current is, of course, also desirable; the error that it causes can be reduced by low-
resistance dimensioning of the feedback resistances (see Fig. 5.61 on page 534). Due
to the low offset voltage, one must insure that the unavoidable thermal voltages do not
affect the circuit. It is important that the corresponding points of the circuit are at the
same temperature level, so that the thermal voltages compensate each other as much as
possible. In critical cases one may use special solder or make wire connections by thermal
compression.
Low-noise types: The types listed here are the operational amplifiers on the market
that have the lowest
√ noise. While amplifiers with bipolar transistors reach noise voltage
densities of 1 nV/ Hz, the best operational amplifiers with FETs at the input have values
five times as high. Nevertheless, they are advantageous for high-resistance sources due
to their noise current density, which is lower by three orders of magnitude. In any case,
feedback resistances should be set as low as possible in order to keep the noise voltages
caused by the noise current of the amplifier and the inherent noise of the resistors as low
as possible (see Fig. 5.64 on page 538).
Rail-to-rail output amplifiers (RRO): At these amplifiers the output voltage can nearly
reach the supply voltages. So appreciable output amplitudes can be obtained even at low
supply voltages. Often the common-mode range of these amplifiers includes the negative
supply voltage. Therefore these Amplifiers can be used as single supply amplifiers (see
Sect. 5.2.4 on page 498 ).
Type Manufacturer Offset- Bias Gain- Slew Operating OPs Special feature
574
LT1224 Lin. Tech. 0.5 mV 4 mA 45 MHz 400 V/ms 5/30 V 1 C Load stable
LT1365 Lin. Tech. 0.5 mV 1 mA 70 MHz 1000 V/ms 5/30 V 4
LT1812 Lin. Tech. 0.4 mV 1 mA 100 MHz 750 V/ms 5/12 V 1
LT1817 Lin. Tech. 0.2 mV 2 mA 220 MHz 1500 V/ms 5/12 V 4
LT1819 Lin. Tech. 0.2 mV 2 mA 400 MHz 2500 V/ms 5/12 V 2 low distortion
MAX4413 Maxim 0.4 mV 1.6 mA 500 MHz 220 V/ms 3/ 6 V 2
MAX4418 Maxim 0.5 mV 1.3 mA 400 MHz 200 V/ms 3/ 5 V 4 RRO
MAX4454 Maxim 0.4 mV 0.8 mA 200 MHz 240 V/ms 3/ 5 V 4 Ib = 0.6 mA
LM6172 National 0.4 mV 1.2 mA 100 MHz 3000 V/ms 10/32 V 2 Ib = 2.3 mA
LM7171 National 0.2 mV 3 mA 200 MHz 4100 V/ms 10/32 V 1
LMH6609 National 0.8 mV 2 mA 900 MHz 1400 V/ms 6/12 V 1
LMH6655 National 1 mV 5 mA 250 MHz 200 V/ms 5/12 V 2
LMH6658 National 1 mV 5 mA 270 MHz 700 V/ms 5/12 V 2 √
OPA643 Texas I. 2 mV 20 mA 800 MHz 1000 V/ms 9/11 V 1 Vnd = 2nV Hz
OPA657 Texas I. 0.1 mV 1 pA 1600 MHz 700 V/ms 9/12 V 1 FET input
OPA698 Texas I. 2 mV 3 mA 450 MHz 1100 V/ms 9/12 V 1 clamping
OPA842 Texas I. 0.3 mV 20 mA 400 MHz 400 V/ms 9/12 V 1
THS4271 Texas I. 5 mV 6 mA 1400 MHz 1000 V/ms 5/16 V 1
Some models feature a current consumption of only a few microamperes. They are partic-
ularly useful for battery operation; quite often, an on–off switch is not needed. However,
the bandwidth and the slew rate also decline with the current consumption, as the overview
shows.
High bandwidth: We can see that there is a multitude of VV operational amplifiers that
are faster than amplifiers of the 741 class by up to three orders of magnitude. This, however,
usually results in unfavorable DC data: a high offset voltage, a high input current, a low
differential gain, and high current consumption. Most wide-band operational amplifiers
are manufactured in a low voltage process for operating voltages of ±5 V, since it is easier
to make high-frequency transistors. However, the comparison shows that there are other
high-frequency amplifiers that are suitable for operation with ±15 V.
Most operational amplifiers require a relatively long recovery time to return to normal
operation after overdrive. Models with a clamping output are preferred, where overdriving
cannot be prevented. In this case, internal additions prevent transistors from entering the
saturation region; this reduces the recovery time to a few nanoseconds. Furthermore a
positive and negative output voltages limit may be set by special voltage limiting pins.
This can also protect subsequent circuits, such as A/D converters, against overdrive.
Differential output amplifiers: Newly there is a family of OPAmps that offer comple-
mentary outputs. They are especially suitable for driving symmetrical loads, for instance
AD-converters. Here the common-mode output voltage can be controlled by an additional
input so that both output voltages are always positive. Furthermore these amplifiers are the
precondition for fully symmetrical circuits.
High output voltage: There are relatively few operational amplifiers that can provide
high output voltages and allow correspondingly high operating voltages, since the standard
manufacturing processes do not support these features. For operating voltages exceeding
100 V, the relatively expensive hybrid circuits are usually used. An exception here is model
PA240.
High output currents: High output currents cause high power dissipation in operational
amplifiers. Therefore, operating voltages should not be unnecessarily high and amplifiers
should be effectively cooled. The overview shows that, as a rule, power amplifiers are
slow and have low slew rates; but there are exceptions to this rule. Here too, high-quality
products can only be produced in hybrid technology.
difference for the user is that here only ohmic feedback is possible. Correspondingly high
operating currents are required to achieve large bandwidths. At amplifiers with specially
low current consumption the supply current is given under remarks. In order to keep power
dissipation within limits, operating voltages of ±5 V are normally used; models for higher
operating voltages are an exception here. Since low-resistance loads normally have to be
driven at high frequencies, the maximum output currents are generally higher than 20 mA.
This value is specified for models that allow particularly high output currents.
VC operational amplifiers have been available for some time. The first-generation mod-
els, like the CA3080, provide data that no longer satisfy modern demands: they are too slow
and their output currents are too low. The OPA615 is CC operational amplifiers that can be
expanded to form VC operational amplifiers by means of a voltage follower (see Fig. 5.97).
In that case and in the MAX436, the transconductance of the operational amplifier can
be reduced to any given value using an external resistance. This is the prerequisite for
operation with pure current feedback, as seen in Figs. 5.82–5.91. Furthermore, all models
allow the current consumption, and thus the maximum output current, to be set by means
of an external resistance.
CC operational amplifiers are the most versatile operational amplifiers for high fre-
quencies. The fact that they are not widely used is mainly due to the habit of thinking in
terms of voltage rather than current. The advantages of CC operational amplifiers have
been explained in the example of the integrator filter; see Fig. 5.93. Figure 5.103 also
specifies the current gain, which is determined by the transfer ratio of the current mirror at
the output (see Fig. 5.80 on page 553). Here, it is also possible to set the quiescent current
using an external resistance.
Classification
The technology determines the input data and thus the accuracy of the DC voltage. This
becomes obvious when the quiescent input currents and offset voltages of operational
amplifiers are plotted, as in Fig. 5.104. Here we can see that the operational amplifiers
with FET differential amplifiers at the input have the lowest input currents but high offset
voltages. Amplifiers with an automatic zero offset have especially low offset voltages.
Vo
Broadband
10 mV
Universal
AD8056 THS4271
Mosfet Jfet 324
1mV LM7171
LMC6001 741
Fig. 5.104. Offset voltage and input bias current of various operational amplifier technolgies
5.6 Comparison 585
Some of the precision amplifiers with bipolar transistors at the input have also very low
offset voltages but significantly higher quiescent input currents. Most unfavorable are the
DC data of wide-band operational amplifiers: they feature high offset voltages as well as
high input currents.
When comparing the operational amplifiers in terms of their noise levels, two clearly
distinguishable groups can be seen, as illustrated in Fig. 5.105: operational amplifiers with
field effect transistors at the input show clearly less current noise due to their low input
current than opamps with bipolar transistor input. This is the reason why fet input opamps
are advantageous at high-resistance sources. On the other hand, one can see that operational
amplifiers with bipolar transistors have lower voltage noise than those with FETs. For this
reason, these are more advantageous for low-resistance sources (see Fig. 5.64).
The bandwidth–current diagram in Fig. 5.106 can be used to compare the dynamic
responses of operational amplifiers. In order to enlarge the bandwidth of an operational
amplifier, the transistors must be driven with higher currents; the bandwidth should there-
fore be proportional to the current. However, technology with small parasitic capacitances
can help you achieve large bandwidths even with medium currents. Correspondingly, cir-
cuits operating in current on demand AB mode (see Fig. 5.28) offer a larger bandwidth
than circuits in conventional A mode at the same quiescent current. The circuits near the
top line in Fig. 5.106 have a more favorable technology and/or circuit design with regard
to the frequency response than the models at the bottom right. From Fig. 5.106 one can
see that significant differences exist with respect to both the current at constant bandwidth
and the bandwidth at constant current. If, on the other hand, a line is drawn along which
the bandwidth is proportional to the current, then all operational amplifiers of the same
technology or circuit design will be located on a line. This shows that high-quality opera-
tional amplifiers have a bandwidth: current ratio above 100 MHz/mA, while older models
do not even reach a level of 1 MHz/mA.
Vnd Hz
0.3 nV
0.01fA 0.1fA 1fA 10 fA 100 fA 1pA 10pA 100 pA I nd Hz
fT 100 MHz
mA
OPA643
10 MHz
1GHz AD8058
AD8005 AD8045 mA
LMV651 LF356
10MHz
TLC274
MAX9916 741
1MHz 324
OP177
LT2079 LT1218
OP481
100kHz
OPA2349 OPA4336
10kHz LMC6442
1kHz
1µA 10 µA 100 µA 1mA 10 mA 100 mA I b
6.1
Transistor as Switch
In the case of linear circuits, we set the collector quiescent potential between V + and
VCE sat , thus enabling them to be driven about this operating point. The characteristic
feature of linear circuits is that the swing is kept so small that the output voltage is a linear
function of the input voltage. Consequently, the output voltage must not attain the positive
or negative limits of the swing, as this would result in distortion. With digital circuits, on
the other hand, only two operating states are employed. We are only interested in whether
a voltage is greater than a specified value VH or less than a specified value VL < VH . If
the voltage exceeds VH , it is referred to as being in the H (high) state, and if it is below
VL , it is said to be in the L (low) state.
The absolute values of the levels VH and VL depend entirely on the circuit design. For
an unambiguous interpretation, steady-state levels between VH and VL must not occur.
The circuit design implications of this will now be discussed with reference to the level
inverter in Fig. 6.1. The circuit must exhibit the following characteristics:
For Vi ≤ VL → Vo ≥ VH
and
for Vi ≥ VH → Vo ≤ VL .
This relationship must hold good, even under worst-case conditions; that is, for Vi = VL ,
Vo must not be lower than VH , and for Vi = VH , Vo must not be higher than VL . This
condition can only be satisfied by selecting suitable values for VH , VL , RC , and RB . The
following worked example should serve to indicate a possible approach.
When the transistor in Fig. 6.1 is turned off, the output voltage under no-load conditions
is equal to V + . Let us assume that the lowest output load resistance is RV = RC ; in this
case, Vo is consequently equal to V + /2. This is therefore the lowest output voltage in the
H state. To be on the safe side, we specify VH < V + /2 for a supply voltage of V + 5 V;
for example, VH = 1.5 V. In accordance with the above-mentioned requirement, the input
voltage must be in state L for V0 ≥ VH . VL is therefore defined as the highest input
voltage for which the transistor is just certain to remain in the blocking state. For a silicon
transistor, we can therefore take a value of 0.4 V if the device is at ambient temperature.
Consequently, we select VL ≤ 0.4 V. Having determined the two levels VH and VL in this
way, we must now select component values for the circuit in such a way that an output
voltage Vo ≤ VL is obtained for Vi = VH . Even under worst-case conditions, we require
a certain safety margin, namely that the output voltage shall remain below VL = 0.4 V
for Vi = VH = 1.5 V. The collector resistance RC is chosen low enough to insure that the
switching times are sufficiently short, but without making the current drain unnecessarily
high. Typically, we select RC = 5 k. We must now select a value for RB that insures that
the output voltage falls below the value VL = 0.4 V for an input voltage of Vi = 1.5 V. For
this to occur, a collector current of IC ≈ V + /RC = 1 mA has to flow. Transistors for this
kind of application normally have a current gain of B = 100. The base current required
588 6 Latching Circuits
Vo
SH
VH
SL
Vo
Vi VL
VL VH Vi
is therefore IB min = IC /B = 10µA. In order to insure that the transistor is driven into
saturation, we select IB = 100 µA; that is, it is ten times overdriven. We thus obtain
1.5 V − 0.6 V
RB = = 9 k
100 mA
Figure 6.1 shows the transfer characteristic for these parameters.
For Vi = VL = 0.4 V, the output voltage Vo = 2.5 V at full load (RV = RC ) and is
therefore 1 V above the minimum value VH = 1.5 V required. We now specify an H noise
margin S H = Vo − VH for Vi = VL . In our example, it is 1 V. Similarly, we can define
an L noise margin S L = VL − Vo for Vi = VH . In Fig. 6.1 it is identical to the voltage
difference between VL and the collector–emitter saturation voltage VCE sat ≈ 0.2 V, and
has a value of SL = 0.4 V − 0.2 V = 0.2V . The noise margins are a measure of the
reliability of the circuit performance. Their general definition is:
.
SH = Vo − VH
for a worst-case condition at the input
SL = VL − Vo
If we wish to improve the L noise margin, it is necessary to increase VL , as voltage
Vo (Vi = VH ) ≈ VCE sat cannot be reduced much further. For this purpose, we can insert
one or more diodes in front of the base, as in Fig. 6.3a. Resistor R2 serves as a drain for
the collector–base reverse current, thereby insuring that the transistor turns off reliably.
Another possibility is to simply insert a preceding voltage divider, as shown in Fig. 6.3b
or Fig. 6.3c.
Vi Vo Vi Vo Vi Vo
a b c
The output loading capability (fan-out) of the inverter in Fig. 6.1 is low. No more than
two identical inputs can be connected to one output; otherwise, the output voltage would
fall below 2.5 V in the H state.
Dynamic Characteristics
When using a transistor as a switch, the main parameter of interest is the switching time.
Within the square-wave response we differentiate between various periods, as shown in
Fig. 6.4.
We can see that the storage time tS is considerably greater than the other switching
times. It is incurred when a previously saturated transistor (VCE = VCE sat ) is turned off. If,
for the conducting transistor, VCE is greater than VCE sat , the storage time is considerably
reduced. We exploit this fact in high-speed switches and prevent VCE sat from being reached.
Digital circuits operating on this principle are known as unsaturated logic circuits. Their
design will be discussed for the relevant circuits in Sect. 7.4.5.
The time behavior of digital circuits is generally characterized by the propagation delay
time tpd :
1
tpd = (tpd L + tpd H )
2
where tpd L is the time difference between the 50% value of the input edge and the 50%
value of the falling output edge. tpd H is the corresponding time difference for the rising
output edge. This is illustrated in Fig. 6.5.
In the circuit shown in Fig. 6.1, we saw that the H level was far below the supply
voltage and was a function of the load. In order to avoid this, an emitter follower can be
connected, as shown in Fig. 6.6.
If T1 is off, the output current flows via the emitter follower T2 , thus keeping the
current in collector resistor RC low. If T1 becomes conducting, its collector potential
falls to low levels. With a resistive output load, the output voltage likewise falls. With
capacitive loading, the circuit must pick up the capacitor discharge current. As transistor
T2 is blocking in this case, diode D is inserted to allow the discharge current to flow via
conducting transistor T1 However, this increases the output voltage to about 0.8 V in the
L state.
Vi Vi
t
Vo
Vo
tD tF tS tR t
Fig. 6.4. Square-wave response of an inverter Fig. 6.5. Defining the propagation delay
tS : Storage time time
tR : Rise time tpd : propagation delay time
tD : Delay time
tF : Fall time
590 6 Latching Circuits
Vi Vo
6.2
Latching Circuits Using Saturated Transistors
Latching circuits are positive-feedback digital circuits. They differ from positive-feedback
linear circuits (oscillators) in that their output voltage does not vary continuously, but only
jumps back and forth between two fixed values. This switching process can be initiated in
various ways:
A bistable circuit changes the output state only changes when switching is initiated by
an input signal. With a flip-flop (latch), a short pulse suffices, whereas a Schmitt trigger
requires a sustained input signal.
A monostable circuit has a single stable state. Its other state is only stable for a fixed
length of time, which is determined by the circuit component values. When this time has
elapsed, the circuit automatically returns to the stable state. It is therefore also known as
a time interval switch, monostable multivibrator, or one-shot.
An astable circuit has no stable state, but constantly oscillates between states without
external triggering. It is therefore also known as a multivibrator.
The three types of circuit can be implemented using the basic configuration shown in
Fig. 6.7. The only difference is in the construction of the two coupling devices CD1 and
CD2, as listed in Fig. 6.8.
CD2 CD1
Fig. 6.8. Coupling networks required for the various types of latching circuit
6.2 Latching Circuits Using Saturated Transistors 591
S S
6.2.1
Bistable Circuits
Flip-Flops
As shown in Fig. 6.9, a bistable circuit can be implemented by connecting two inverters in
series with the coupling divices R1 , R2 and R3 , R4 providing positive feedback. We can
see that the two inverters have equal status, and so the symmetrical circuit diagram given
in Fig. 6.10 is generally preferred.
The mode of operation is as follows: a positive voltage at the set input S renders T1
conducting. This causes its collector potential to fall, thereby reducing the base current of
T2 , whose collector potential increases. This increase causes the base current of T1 to rise
via resistor R1 . The steady-state condition is therefore reached when the collector potential
of T1 has fallen to the saturation voltage. T2 then turns off and T1 is maintained on via
resistor R1 , Consequently, the voltage at the S-input can be made zero again at the end of
the switching process without inducing any further changes. The flip-flop can be returned
to its original state by applying a positive voltage pulse to the reset input R. When the
two input voltages are zero, the flip-flop retains the last state assumed. This characteristic
allows it to be used as a data memory.
If the two input voltages are simultaneously changed to the H state, both transistors
become conducting during this time. However, in this case the base currents are supplied
only by the control voltage sources and not by the adjacent transistor, as the two collector
potentials are low. Consequently, this state is not stable. If the two control voltages are
again made zero, this causes the two collector potentials to rise simultaneously. However,
as there is never complete symmetry, one collector potential rises somewhat faster than
the other. Due to the positive feedback, this difference is amplified, so that a stable state
is eventually reached in which one transistor is off and the other is on. However, as it is
impossible to predict definitively which of the two stable states the flip-flop will assume,
R S Q Q
H H (L) (L)
H L L H
L H H L
L L Q−1 Q−1
Fig. 6.11. Truth table for the RS flip-flop Q−1 and Q−1 mean that the flip-flop remains in the last
state
592 6 Latching Circuits
Vo Vo max
Vo
Vo min
Vi off Vi on Vi
Vi
Hysteresis
Schmitt Trigger
The RS flip-flop described above is caused to change state by applying a positive voltage
pulse to the base of the currently nonconducting transistor in order to turn it on. Another
possibility consists of only using one input and switching it off with a negative input
voltage. A flip-flop that operates in this way is known as a Schmitt trigger. Its simplest
circuit is shown in Fig. 6.12.
When the input voltage exceeds the upper trigger threshold Vi ON , the output voltage
jumps to the positive saturation limit Vo max . It only returns to zero when the input voltage
falls below the lower trigger threshold Vi OFF . This characteristic allows the Schmitt trigger
to be used as a square-wave converter. By way of example, Fig. 6.14 shows how a sine
wave is converted into a square wave. Due to positive feedback, the change from one state
to the other is instantaneous, even though the input voltage changes only slowly.
The transfer characteristic is illustrated in Fig. 6.13. The voltage difference between the
turn-on and turn-off levels is termed the hysteresis. This is smaller the more the difference
between Vo max and Vo min is reduced, or the greater the attenuation in the voltage divider
R1 , R2 is. Any attempt to reduce the hysteresis will adversely affect the positive feedback
in the Schmitt trigger and could result in it ceasing to be bistable. For R1 → ∞, the circuit
becomes a conventional two-stage amplifier.
Vo
Vo max
Vi on
Vo min
Vi out
Fig. 6.14. Schmitt trigger as a
Vi square-wave converter
6.2 Latching Circuits Using Saturated Transistors 593
6.2.2
Monostable Circuits
A one-shot is basically an RS flip-flop in which one of the two feedback resistors is replaced
by a capacitor, as shown in Fig. 6.15. As the capacitor blocks DC, transistor T2 is on and
transistor T1 is off under steady-state conditions.
A positive input pulse turns on transistor T1 , causing its collector potential to jump
from its steady-state value V + to zero. This jump is coupled by the capacitor RC to the
base of T2 , causing its base potential to go from 0.6 V to − V + + 0.6 V ≈ −V + , and
T2 turns off. T1 is maintained on via feedback resistor R1 , even if the input voltage has
already returned to zero as shown in Fig. 6.16.
Capacitor C is charged via resistor R connected to V + . As described in Chap. 29.3.2,
the base potential of T2 increases in accordance with the relation
Transistor T2 remains in the blocking state until VB 2 has risen to approximately +0.6 V.
We can obtain the time tON required for this sequence by substituting VB 2 ≈ 0 in (6.1),
giving
When this time has elapsed, transistor T2 begins to conduct again; that is, the circuit
flips back to its stable state. Figure 6.16 shows the relevant voltage waveforms.
The output returns to its initial state within the defined turn-on time even if the input
pulse is longer than the turn-on time. In this case, transistor T1 remains on until the input
pulse disappears, and the positive feedback has no effect. T2 does not therefore begin to
conduct instantaneously, but only in accordance with the rate of rise of VB 2 .
When the switching cycle is complete, capacitor C must be charged via RC . If the
capacitor is not fully charged until the next turn-on pulse occurs, the subsequent turn-on
time is reduced. In order to reduce the on-time by less than 1%, T1 must remain off for a
recovery time of at least 5 RC · C.
Vi
Vo
Vo
Vi
t ON
The supply voltage of the circuit should not be choosen above 5 V, as this could cause
the emitter–base breakdown voltage of T2 to be exceeded when T1 becomes conducting.
This effect would reduce the turn-on time as a function of the supply voltage.
6.2.3
Astable Circuits (Multivibrators)
If the second feedback resistor of a one-shot is also replaced by a capacitor, as in Fig. 6.17,
the two states are each stable only for a limited period of time. The circuit therefore
continuously oscillates between the two states therefore the name: multivibrator. From
(6.2), the switching times are given by
t1 = R1 C1 ln 2
and
t2 = R2 C2 ln 2
The voltage waveforms are shown in Fig. 6.18. As we can see, t1 is the off-time of T1
and t2 is the off-time of T2 . The circuit therefore always changes state when the opposite
nonconducting transistor is turned on.
The circuit designer has little margin in selecting the values of resistors R1 and R2 . On
the one hand, they must be small compared to βRC so that sufficient current flows through
them to drive the conducting transistor into saturation. On the other hand, they must be
large compared to RC so that the capacitors can be charged up to the supply voltage before
the next cycle. This condition can be expressed as
RC R1 , R2 βRC
As with the one-shot in Fig. 6.15, the supply voltage must not be selected larger than 5 V
in order not to exceed the emitter–base breakdown voltage.
Vo1
It is possible that the multivibrator in Fig. 6.17 will not start to oscillate. For example,
if one output is short-circuited, both transistors go into saturation, and this condition
continues to obtain even when the short circuit is removed.
At frequencies below 100 Hz, the capacitors are unwieldy and large; and at frequencies
above 10 kHz, the storage times of the transistors become a noticeable problem. Conse-
quently, the circuit in Fig. 6.17 is of little practical use. The precision circuits with com-
parators (Sect. 6.5.3) are preferable for low-frequency applications, whereas the emitter-
coupled multivibrators in Sect. 6.3.2 tend to be used for high-frequency applications.
6.3
Latching Circuits with Emitter-Coupled Transistors
6.3.1
Emitter-Coupled Schmitt Trigger
A noninverting amplifier can also be implemented by using a differential amplifier. By
applying positive feedback through a resistive voltage divider, we obtain the emitter-
coupled Schmitt trigger shown in Fig. 6.19. Both of its trigger thresholds are positive.
By selecting suitable component values for the circuit, we can cause current Ik to
switch from one transistor to the other when the circuit changes state, without the transis-
tors becoming saturated. This eliminates the storage time tS during switchover, and high
switching frequencies can be achieved. This principle is known as ”unsaturated logic.”
6.3.2
Emitter-Coupled Multivibrator
Due to the elimination of the storage times, considerably higher frequencies can be achieved
with emitter-coupled multivibrators than using saturated transistors. A suitable circuit is
shown in Fig. 6.21.
To explain how the circuit operates, let us assume that the amplitude of the AC voltages
present is small at all points in the circuit; say, Vpp ≈ 0.5 V. When T1 is off, its collector
potential is equal to the supply voltage, resulting in an emitter potential of V + − 1.2 V at
T2 . Its emitter current is I1 + I2 . In order to produce the required amplitude of oscillation
at R1 , the value R1 = 0.5 V/(I1 + I2 ) must therefore be selected. This gives us an emitter
Vo
Vo max
Vo
Vo min
T1 T2
Vi
Vi out Vi on Vi
t1
V + – 0.5V
V + –1.2V
Vo1 Vo2
t2
V + – 0.5V
)
V + -–1.2V
can be connected in parallel as shown by the dashed lines in Fig. 6.21. The oscillation
frequency then becomes
1 I
f = =
t1 + t2 4UD C
where UD is the forward voltage of the diodes.
Emitter-coupled multivibrators are available as monolithic integrated circuits.
IC types:
TTL SN 74 LS 624…629 f max = 20 MHz (Texas Instruments)
CMOS LTC 1799 f max = 33 MHz (Linear Technology)
CMOS LTC 6905 f max = 170 MHz (Linear Technology)
6.4
Latching Circuits Using Gates
Latching circuits can be implemented not only using transistors, but also using integrated
logic devices (gates), as described in Chap. 7.4. Readers who are unfamiliar with basic
logic functions should therefore read that chapter first.
6.4.1
Flip-Flops
Let us return to the flip-flop shown in Fig. 6.10. Transistor T1 is on when a positive voltage is
dropped across resistor R1 or resistor R2 . If we also take into account the inverter function
produced by the transistor, we can see that components R1 , R2 , T1 , and RC form a nor
gate. The same applies to the other half of the circuit. Inserting the appropriate circuit
symbols, we obtain the circuit diagram shown in Fig. 6.23, with the associated truth table
(Fig. 6.24).
6.4.2
One-Shot
The circuit shown in Fig. 6.25 provides a simple means of generating short pulses with a
duration of just a few gate propagation delay times. As long as the input variable x = 0, a
logic 0 will be produced at the output of the and gate. If x = 1, the and element produces
a logic 1 until the signal has passed through the inverter chain. When the input signal
returns to zero, the and condition is not satisfied.
R S Q Q
S
0 0 Q−1 Q−1
0 1 1 0
1 0 0 1
1 1 (0) (0)
Fig. 6.23. Flip-flop with nor gates Fig. 6.24. Truth table
598 6 Latching Circuits
ON
tON
Fig. 6.25. One-shot for short switching Fig. 6.26. Waveform diagram.
times t1 = propagation delay time of the and
Turn-on time: tON = sum of inverter gate
propagation delay times
tON
tON
The timing diagram is shown in Fig. 6.26. The duration of the output pulse is equal to
the delay in the inverter chain. This can be specified by an appropriate number of gates,
taking care to insure that there is an odd number of gates. As we can see from Fig. 6.26,
for this one-shot the trigger signal must be present at least for the duration of the output
pulse.
To obtain longer switching times, the delay chain becomes unmanageably long. In this
case, it is preferable to employ integrated one-shots (monostable multivibrators), whose
switching times are determined by an external RC network.
IC types:
TTL 74 LS 121…123, 422, 423 (Texas Instruments)
If the and gate in Fig. 6.25 is replaced by an exclusive-nor gate, we obtain a one-
shot that produces an output pulse on each edge of the input signal. Figure 6.27 shows
the relevant circuit, and Fig. 6.28 the associated waveform pattern. Under steady-state
conditions, the inputs of the exclusive-nor gate are complementary and the output signal
is zero. If the input variable x changes its state, temporarily identical input signals appear at
the exclusive-nor gate due to the delay through the inverters. During this time, the output
signal y becomes logic 1.
6.4.3
Multivibrator
A simple multivibrator comprising two inverters is shown in Fig. 6.29. In order to explain
how it works, let us assume that signal xis in the H state and that y is therefore in the L state.
This causes capacitor C to charge up via resistor R until potential V exceeds the switching
level VS of gate G1 . x then changes to the L state and y to the H state, causing potential
V to go positive by the amplitude of the output signal. The capacitor then discharges via
resistor R until the voltage falls below the switching level.
6.4 Latching Circuits Using Gates 599
The voltage waveform is shown in Fig. 6.30. Assuming that the switching level lies
halfway between the output levels, the cycle time is given by
T = 2RC ln 3 ≈ 2.2RC
Fig. 6.31. Multivibrator using a Schmitt trigger Fig. 6.32. Waveform diagram
Cycle time: (TTL) T = 1.4. . .1.8RC
(5 V-CMOS) T = 0.5. . .1.0RC
600 6 Latching Circuits
Fig. 6.33. Multivibrator with an ECL line Fig. 6.34. Internal design of the
receiver line-receiver multivibrator
Cycle time: T ≈ 3RC
6.5
Latching Circuits Using Comparators
6.5.1
Comparators
An operational amplifier without feedback, as shown in Fig. 6.35, represents the basic
circuit of a comparator. Its output voltage is given by
Vo max for V1 > V2
Vo =
Vo min for V1 < V2
The corresponding transfer characteristic is shown in Fig. 6.36. Due to the high gain,
the circuit responds to very small voltage differences V1 − V2 . It is thus suitable for the
comparison of two voltages, and operates with high accuracy.
At the zero crossing of the input voltage difference, the output voltage does not im-
mediately reach the saturation level because the transition is limited by the slew rate. For
frequency-compensated standard operational amplifiers, it can be as low as 1 V/ms. A volt-
age rise from − 12 V to +12 V therefore takes 24 ms. An additional delay is incurred due
to the recovery time needed after the amplifier has been saturated.
As the amplifier possesses no negative feedback, it also does not require any frequency
compensation. Its omission can improve the slew rate and recovery time by a factor of
about 20.
Considerably shorter response times can be attained when using special comparator
amplifiers, which are designed for use without feedback and have especially small recovery
times. However, the gain and hence the accuracy of the threshold are somewhat lower than
those of operational amplifiers. Usually, the amplifier output is directly connected to a level-
Vo
Vo max
V1
Vo V2 V1
V2
Vo min
V1
V2
V1 – V2
Fig. 6.37. Comparator with logic output Fig. 6.38. Transfer characteristic
y=1 for V1 > V2
shift circuit, which permits compatible operation with logic circuits. Its application and
characteristics are shown in Figs. 6.37 and 6.38. A number of commonly used comparators
are listed in Fig. 6.39.
Window Comparator
A window comparator can determine whether or not the value of the input voltage lies
between two reference voltage levels. Figure 6.40 shows that two comparators are used to
decide whether the input voltage is above a lower and below an upper reference voltage.
When this condition arises, both comparator amplifiers produce a logical 1 and the output
becomes y = 1. The characteristics in Fig. 6.40 illustrate the operation.
6.5.2
Schmitt Trigger
The Schmitt trigger is a comparator, for which the positive and negative transitions of the
output occur at different levels of the input voltage. Their difference is characterized by
the hysteresis Vi . As described in earlier sections, Schmitt triggers can be realized by
transistors, but in the following section some designs involving comparators are discussed.
602 6 Latching Circuits
V2
CA2 V1 V2 Vi
Vi
Vi
V1 CA1
Vi
Vo
Vo max
Vi
Vo
Vi on Vi off Vi
Vo min
V
Vo (t)
Vi (t)
Vi off
Vi on
Fig. 6.44. Voltage waveform of an inverting Schmitt trigger for a sinusoidal input voltage
R1
Vi OFF = − Vo max
R2
The output voltage jumps to Vo min as soon as Vi reaches or falls below this value. The
transition is initiated by Vi but is then determined only by the positive feedback via R2 .
The new state is stable until Vi returns to the level
R1
Vi ON = − Vo min
R2
Figure 6.47 depicts the time function of the output voltage for a sinusoidal input. Since, at
the instant of transition, VP = 0, the formulas for the trigger level have the same form as
those for the inverting amplifier.
Vo
Vo max
Vi out Vi on Vi
Vi
Vo
Vo min
V
Vo (t)
Vi (t)
Vi on
Vi off
Fig. 6.47. Voltage waveform of a noninverting Schmitt trigger for a sinusoidal input voltage
6.5.3
Multivibrators
If an inverting Schmitt trigger is configured such that the output signal is fed to the input
with a delay, we obtain a multivibrator of the type shown in Fig. 6.50.
When the potential at the N-input exceeds the trigger level, the circuit changes state,
and the output voltage jumps to the opposite output limit. The potential at the N-input
attempts to follow the transition in the output voltage, but the capacitor prevents it from
changing rapidly. The potential therefore rises or falls gradually until the other trigger
level is reached. The circuit then flips back to its initial state. The voltage wave-shapes are
V2
CA2 V1 V2 Vi
Vi
Vi
V1 CA1
Vi
Fig. 6.48. Precision Schmitt trigger Fig. 6.49. Variables as a function of the
. input voltage
Switch-on level: Vi ON = V2
for V2 > V1
Switch-off level: Vi OFF = V1
6.5 Latching Circuits Using Comparators 605
Vo (t)
Vo max
Vo off
Vo
Vo on
VN –Vo max
shown in Fig. 6.51. From Fig. 6.42, the trigger levels for Vo max = −Vo min = Vo are given
by
Vi ON = −αVo
and
Vi OFF = αVo
where α = R1 /(R1 + R2 ).
The differential equation for VN can be taken directly from the circuit diagram, as
dVN ±Vo − VN
=
dt RC
With the initial condition VN (t = 0) = Vi ON = −αVo , we obtain the solution
1 2
VN (t) = Vo 1 − (1 + α)e− RC
t
CA2
CA1
T is turned on. Capacitor C is then discharged by the resistor R2 until the lower threshold
1 +
3 V is reached, this process requiring the time
t2 = R2 C ln 2 ≈ 0.693R2 C
On reaching the lower threshold, S goes low and the flip-flop resumes its former state. Its
output goes low and transistor T is turned off. Charging of the capacitor takes place via
the series connection of R1 and R2 . The time interval needed to reach the upper trigger
level is
t1 = (R1 + R2 )C ln 2 ≈ 0.693(R1 + R2 )C
1 1 1.44
f = = ≈
t1 + t2 (R1 + 2R2 )C ln 2 (R1 + 2R2 )C
The wave-shapes of the signals y and VC are shown in Fig. 6.53. The reset input 4
allows interruption of the oscillation.
When supplying a voltage to pin 5, the trigger thresholds can be shifted. In this way, the
charging time t1 and therefore the frequency of the multivibrator can be varied. A change
in the potential V5 = 23 V + by V5 results in a relative frequency shift of
f R1 + R2 V5
≈ − 3.3 · ·
f R1 + 2R2 V +
As long as the voltage deviation is not too large, the frequency modulation is reasonably
linear.
6.5.4
One-Shots
The timer 555 is also useful for generating single pulses (one-shot), with pulse times
between few microseconds and several minutes. The required external connections are
shown in Fig. 6.54.
When the capacitor potential exceeds the upper trigger threshold, the flip-flop is re-
set; that is, the output voltage resumes the L-state. Transistor T becomes conducting and
discharges the capacitor. As the lower comparator is no longer connected to the capacitor,
this state remains unchanged until the flip-flop is set by an L-pulse at trigger input 2. The
on-time t1 is equal to the time required by the capacitor potential to rise from zero to the
upper threshold 23 V + . It is given by
t1 = R1 C ln 3 ≈ 1.1R1 C.
If a new trigger pulse occurs during this time interval, the flip-flop remains set and the
pulse is ignored. Figure 6.55 shows the signals involved.
Discharging of capacitor C at the end of t1 is not as fast as could be wished, as the
collector current of the transistor is limited. The discharge time is known as the recovery
time. If a trigger pulse occurs during this interval, the on-time is reduced and is therefore
no longer precisely defined.
CA2
CA1
Retriggerable Timer
There are cases in which the on-time is not to be counted from the first pulse of a pulse
train, as in the previous circuit, but from the last pulse of the train. Circuits that have this
characteristic are termed “retriggerable timers.” The appropriate connection of the timer
555 is shown in Fig. 6.56, where it is used only as a precision Schmitt trigger.
When the capacitor potential exceeds the upper trigger threshold, the flip-flop is reset
the output y assumes the L-state. The capacitor will not discharge, as transistor T is not
connected. The capacitor potential therefore rises to V + , this being the stable state. The
capacitor must be discharged by a sufficiently long positive trigger pulse applied to the
base of the external transistor T
. The flip-flop is set by the lower comparator and the output
y assumes the H-state. If a new trigger pulse occurs before the end of the on-time, the
capacitor is discharged again and the output remains in the H-state. It flips back only if no
new trigger pulse occurs for a time interval of at least
t1 = R1 C ln 3 ≈ 1.1R1 C
CA2
S
CA1
Vi
Vi
The circuit is therefore also called the “missing pulse detector.” The signals within the
circuit are shown in Fig. 6.57 for several consecutive trigger pulses.
Chapter 7:
Logic Families
Although, at first sight, digital equipment appears to be relatively complicated, its design is
based on the simple concept of the repeated use of a small number of basic logic circuits.
We can work out how these basic logic elements have to be linked by applying purely
formal methods to the problem. This approach is based on Boolean algebra which, when
applied specifically to digital circuit design, is known as computational algebra. In the next
few subsections, we shall summarize the basics of computational algebra.
7.1
Basic Logic Functions
Unlike a variable in conventional algebra, a logic variable can only assume two discrete
values (binary variable), generally referred to as logic or Boolean one and logic or Boolean
zero, for which the symbols 0 and 1 are used. There is no risk of confusion with the numbers
0 and 1, as it is always clear from the context whether a number or a logic value is meant.
There are three basic relations that link logic variables: conjunction, disjunction, and
negation. Using the mathematical signs of numeric algebra,
Conjunction: y = x1 ∧ x2 = x1 · x2 = x1 x2
Disjunction: y = x1 ∨ x2 = x1 + x2
Negation: y = x
A number of theorems relating these operations are listed below:
Commutative law:
x1 x2 = x2 x1 (7.1a) x1 + x2 = x2 + x1 (7.1b)
Associative law:
x1 + (x2 + x3 )
x1 (x2 x3 ) = (x1 x2 )x3 (7.2a) (7.2b)
= (x1 + x2 ) + x3
Distributive law:
x1 + x 2 x3
x1 (x2 + x3 ) = x1 x2 + x1 x3 (7.3a) (7.3b)
= (x1 +x2 )(x1 +x3 )
Absorption law:
x1 (x1 + x2 ) = x1 (7.4a) x1 + x1 x2 = x1 (7.4b)
Tautology:
xx = x (7.5a) x+x =x (7.5b)
Law of negation
xx = 0 (7.6a) x+x =1 (7.6b)
612 7 Logic Families
Double negation:
(x) = x (7.7)
De Morgans law:
x1 x2 = x 1 + x 2 (7.8a) x1 + x 2 = x 1 x 2 (7.8b)
Many of these formulas are already familiar from ordinary algebra. However, (7.3b),
(7.4a,b), (7.5a,b), and (7.10b) do not apply to algebraic numbers, and the term ”negation”
is not used at all in connection with numbers. Expressions such as 2x and x 2 do not occur
in switching algebra, due to tautology.
If one compares the equations on the left with those on the right, the important principle
of duality becomes apparent: if conjunction and disjunction and 0 and 1 are interchanged
in any identity, an identity on the other side is obtained.
Using (7.9)–(7.11), it is possible to work out the conjunction and disjunction for all
the possible values of variables x1 and x2 . Figure 7.1 shows the function or truth table for
conjunction and Fig. 7.2 the truth table for disjunction.
We can see from Fig. 7.1 that y is only 1 if x1 and x2 are 1. Consequently, conjunction
is also known as the and operation. In the case of disjunction, y is always 1 if x1 or x2 is 1.
This operation is therefore also known as the or operation. Both of these logic operations
can be extended to apply to any number of variables.
The question now is how to implement these logic operations using electrical circuits.
As logic variables can only assume two discrete values, the only candidates are circuits that
possess two clearly distinguishable operating states. The simplest means of representing
a logic variable is a switch, as shown in Fig. 7.3. An open switch can now be defined as
representing a logic “zero” and a closed switch a logic “one.” Switch S therefore represents
variable x if it is closed for x = 1. It represents variable x if it is open for x = 1.
Let us first determine which logic function is obtained when two switches, x1 and x2 ,
are connected in series, as in Fig. 7.4. The value of dependent variable y is characterized
by whether the resulting switch arrangement between the terminals is open or closed.
As we can see, current can only flow if x1 and x2 are closed; that is, both 1. The series
connection therefore represents an and operation. Similarly, an or operation is obtained
by connecting switches in parallel.
x1 x2 y x1 x2 y
0 0 0 0 0 0
0 1 0 0 1 1
1 0 0 1 0 1
1 1 1 1 1 1
Fig. 7.1. Truth table for Fig. 7.2. Truth table for
conjunction (and) y = x1 x2 disjunction (or) y = x1 + x2
7.1 Basic Logic Functions 613
Using this switching logic, we can now verify the theorems stated above. We shall
illustrate this by taking tautology as an example. In Fig. 7.5, both sides of (7.5a) have been
realized using switch arrangements. We can see that the given identity is satisfied, because
two switches connected in series and simultaneously opened and closed have the same
effect as a single switch.
As we saw in Sect. 6.1, another way of representing logic variables is by electrical
voltages. To the two distinct levels H = high and L = low we can now assign the logic
states 1 and 0. By making H = 1 and L = 0, we obtain what is known as “positive logic.”
Conversely, we can also make H = 0 and L = 1, this being termed “negative logic.”
The basic logic functions can be implemented by appropriate electronic circuits with
one or more inputs and a single output. These are generally known as ”gates.” The voltage
levels at the inputs and the types of logic operation determine the output level. As a
logic function can be implemented electronically in a variety of ways, to simplify matters
circuit symbols have been introduced which only show the logic function and give no
indication of the internal design. These symbols are given in Figs. 7.6–7.8. The complete
standard can, for instance, be found in IEC 60617-12 and IEEE Std 91-1984. To facilitate
the understanding of old-style circuit diagrams, the symbols formerly used are shown in
Figs. 7.9–7.11.
In digital circuits, we are not concerned with the voltage as a physical quantity, but
merely with the logic state that it represents. Instead of denoting the input and output
signals by V1 , V2 , and so on, they are therefore expressed directly using the logic variables
shown.
Fig. 7.6. An and circuit Fig. 7.7. An or circuit Fig. 7.8. A not circuit
Figs. 7.6–7.8. Circuit symbols to IEC 60617-12 and IEEE Std 91-1984
Fig. 7.9. An and circuit Fig. 7.10. An or circuit Fig. 7.11. A not circuit
Figs. 7.9–7.11. Old-style circuit symbols
614 7 Logic Families
7.2
Construction of Logic Functions
In digital circuit design, the task is generally presented in the form of a function table,
also known as a truth table. The objective is then to find a logic function that satisfies
this truth table. The next step is to reduce this function to its simplest form, so that it can
be implemented by a suitable combination of basic logic circuits. The logic function is
generally given in the disjunctive normal form (sum-of-products, standard product terms,
canonical products). The procedure is done in 3 steps:
1) We find all the rows in the truth table in which the output variable y is 1.
2) From each of these rows, we form the logical product (conjunction) of all the input
variables, substituting xi if the relevant variable is 1 and using x i if the variable is 0.
We thus obtain as many product terms as there are rows with y = 1.
3) We obtain the required function by forming the logical sum (disjunction) of all the
product terms found.
By way of an example, let us consider the truth table in Fig. 7.12. In rows 3, 5, and 7, we
have y =1. We must therefore form the logical products of these rows.
Row 3: K 3 = x 1 x2 x 3 ,
Row 5: K5 = x1 x 2 x 3 ,
Row 7: K7 = x1 x2 x 3
The required function is now obtained as the sum of the products (disjunction of the
conjunctions):
y = K3 + K5 + K7 ,
y = x 1 x2 x 3 + x1 x 2 x 3 + x 1 x2 x 3
This is the disjunctive normal form of the required logic function. To simplify, we now
apply (7.3a) and obtain
y = [x 1 x2 + x1 (x 2 + x2 )]x 3
Equations 7.6b) and (7.9a) yield a further simplification:
y = (x 1 x2 + x1 )x 3
From (7.3b),
y = (x1 + x2 )(x1 + x 1 )x 3
Row x1 x2 x3 y
1 0 0 0 0
2 0 0 1 0
3 0 1 0 1
4 0 1 1 0
5 1 0 0 1
6 1 0 1 0
7 1 1 0 1
8 1 1 1 0
By again applying (7.6b) and (7.9a), we finally obtain the simple result
y = (x1 + x2 )x 3
If the output variable y has more “ones” than “zeros,” a large number of product terms
is obtained. Simplification can be performed in this case by considering the complemented
(negated, barred) output variable ȳ instead of y. There are sure to be fewer “ones” than
”zeros” for this negated variable. Consequently, by expressing the logic function for the
complemented variable ȳ, we obtain from the start a smaller number of product terms; that
is, a simpler function. It only needs to be negated at the end to obtain the required function
for y. To do this, we merely interchange the operations (+) and (·), and complement all
the variables and constants individually.
7.2.1
Karnaugh Map
The Karnaugh map provides an important means of obtaining a logic function in its simplest
form. It is basically another version of the truth table. However, the values of the input
variables are not simply written one below the other, but arranged at the horizontal and
vertical sides of an area that is divided into squares like a chessboard. If the number of
input variables is even, half are written horizontally and half vertically. If the number of
input variables is odd, one side will have one more variable than the other.
The various combinations of input function values must be arranged in such a way
that only one variable changes from one square to the next. In the squares themselves
are entered the values of the output variables y associated with the input variable values
written at the sides. Figure 7.13 shows once again the truth table of the and function for
two input variables, and Fig. 7.14 shows the corresponding Karnaugh map.
As the Karnaugh map is merely a simplified version of the truth table, it can be used
for obtaining the disjunctive normal form of the associated logic function in the manner
described above. Its advantage is that it makes it easier to spot possible simplifications.
We shall illustrate this by means of the example in Fig. 7.15.
To write the disjunctive normal form, we must first, as described above, form the logical
product of all the input variables for each cell containing a 1. For the cell in the top left-hand
corner, we obtain
K1 = x 1 x 2 x 3 x 4
For the cell immediately to the right,
K 2 = x 1 x2 x 3 x 4
x1 x2 y
0 0 0
0 1 0
1 0 0
1 1 1
Fig. 7.13. Truth table for the and Fig. 7.14. Karnaugh map for the
function and function
616 7 Logic Families
x1 x2 x3 x4 y
0 0 0 0 1
0 0 0 1 1
0 0 1 0 1
0 0 1 1 1
0 1 0 0 1
0 1 0 1 0
0 1 1 0 0
0 1 1 1 0
1 0 0 0 1
1 0 0 1 0
1 0 1 0 1
1 0 1 1 1
1 1 0 0 0
1 1 0 1 0
1 1 1 0 1
1 1 1 1 1
If we finally form the sum of all the products, one possible expression is
K 1 + K 2 = x 1 x 2 x 3 x 4 + x 1 x2 x 3 x 4
This can be simplified to
K1 + K2 = x 1 x 3 x 4 (x 2 + x2 ) = x 1 x 3 x 4
This illustrates the general simplification rule for the Karnaugh map.
If a rectangle or square with 2, 4, 8, 16, ... cells contains all “ones,” the logical product
of the entire block can be obtained directly by only taking into account the input variables
that possess a constant value in all the cells of the block.
Therefore, in our example we obtain for pair B the logical product
KB = x 1 x 3 x 4
which corresponds to the function given above. It is also possible to block together those
cells located at the left and right extremities of a row, or at the top and bottom of a column.
For the block of four D in Fig. 7.15, we obtain
KD = x 1 x 2
Similarly, for the block of four C, the logical product is given by
KC = x1 x3
This still leaves the 1 in the top right-hand corner. As shown, it can be combined with the
1 at the bottom of that column to form a pair KA . Another possibility would be to combine
it with the 1 at the other end of the same row. However, the simplest solution is to note that
there is a 1 at each corner of the Karnaugh map. These can be combined to form a block
of four, and we obtain
KA
= x 2 x 4
7.3 Extended Functions 617
For the disjunctive normal form, we now obtain the already considerably simplified result:
y = KA
+ KB + KC + KD ,
y = x 2 x 4 + x 1 x 3 x 4 + x 1 x3 + x 1 x 2
7.3
Extended Functions
In the preceding discussion, we have shown that every logic function can be represented
by a suitable combination of the basic functions or, and, and not. We shall now consider
a number of derived functions that occur so frequently in circuit design that they have been
given names of their own. Their truth tables and circuit diagrams are shown in Fig. 7.16.
The nor and nand functions are the complements of the or and and functions,
respectively: nor = not or; nand = not and. Thus:
x1 nor x2 = x1 + x2 = x 1 x 2 , (7.12)
x1 nand x2 = x1 x2 = x 1 + x 2 (7.13)
In the case of the equivalence function, y = 1 if the two input variables are the same.
By writing the disjunctive normal form, we obtain from the truth table
y = x1 equiv x2 = x 1 x 2 + x1 x2
The antivalence (nonequivalence) function is a complemented equivalence function for
which y is 1 if the input variables are different. Written in the disjunctive normal form:
y = x1 antiv x2 = x 1 x2 + x1 x 2
The truth table reveals another meaning of the antivalence function: it coincides with the or
function in all values except in the case in which all the input variables are 1. It is therefore
also known as the exclusive-or (exor) function. Similarly, the equivalence function may
also be termed the exclusive-nor function (exnor).
When using integrated circuits, it is sometimes preferable to implement functions using
only nand or nor gates because they are easy to be implemented. For this purpose, the
functions are modified such that only the desired logic operations occur. A simple method
Input y = x1 + x2 y = x1 · x2 y = x1 + x2 y = x1 · x2 y = x1 ⊕ x2 y = x1 ⊕ x2
variables = x1 or x2 = x1 and x2 = x1 nor x2 = x1 nand x2 = x1 exor x2 = x1 exnor x2
x1 x2 = x1 antiv x2 = x1 äquiv x2
0 0 0 0 1 1 0 1
0 1 1 0 0 1 1 0
1 0 1 0 0 1 1 0
1 1 1 1 0 0 0 1
Gates
Function
NAND NOR
NOT
AND
OR
Fig. 7.17. Implementation of basic functions using nor and nand gates
of doing this is to replace the basic functions by nand and nor operations. For the and
function,
x1 x2 = x1 x2 = x1 nand x2 ,
x1 x2 = x 1 x 2 = x 1 + x 2 = x 1 nor x 2
This yields the possible implementations shown in Fig. 7.17 .
x1 + x2 = x 1 + x 2 = x 1 x 2 = x 1 nand x 2 ,
x1 + x2 = x1 + x2 = x1 nor x2
7.4
Circuit Implementation of the Basic Functions
In the preceding subsections, we have manipulated logic circuits without concerning our-
selves with their internal design. This approach is justified because, nowadays, digital
circuitry is based virtually exclusively on integrated circuits which, apart from their power
supply connections, possess only the inputs and outputs mentioned.
For implementation of the individual basic operations, a wide variety of circuit tech-
nologies exists, with differing characteristics in terms of power consumption, supply volt-
age, H and L level, gate propagation delay, and fanout. In order to make an appropriate
choice, it is necessary to have at least a basic understanding of the internal design of these
circuits. We have therefore summarized the principal families of circuits below.
When ICs are interconnected, a large number of gate inputs are often connected to one
output. The number of inputs of the same family of circuits that can be connected without
falling below the guaranteed noise margin is characterized by the fanout. A fanout of ten
therefore means that ten gate inputs can be connected. If the fanout is inadequate, a buffer
is used instead of a standard gate.
It is a characteristic of a gate that a specific output state is associated with each input
state. As described in Chap. 6, these states can be designated by H and L, depending on
whether the voltage is greater than VH or less than VL . The operation of a gate can be
described by a level table, as shown in Fig. 7.18. However, this does not determine which
7.4 Circuit Implementation of the Basic Functions 619
V1 V2 Vo x1 x2 y x1 x2 y
L L H 0 0 1 1 1 0
L H H 0 1 1 1 0 0
H L H 1 0 1 0 1 0
H H L 1 1 0 0 0 1
Fig. 7.18. Example of a Fig. 7.19. Truth table for Fig. 7.20. Truth table for
level table (function table) positive logic: the nand negative logic: the nor
function function
logic function the gate will implement, as we have not yet defined the assignment of levels
to logic states. Although this assignment is arbitrary, it is advisable to adopt a uniform
system within the equipment. The level/state assignment
H=
7 1, L=
70
is termed “positive logic” and produces, in our example, the truth table shown in Fig. 7.19,
which may be identified as that of the nand operation. The assignment
H=
7 0, L=
71
is known as “negative logic.” In our example, it results in the truth table shown in Fig. 7.20;
that is, the nor operation.
One and the same circuit can therefore represent either a nor or a nand circuit,
depending on the type of logic selected. The logic functions of digital circuits are normally
described in positive logic. In negative logic, the operations are reversed:
nor ⇔ nand ,
or ⇔ and ,
not ⇔ not
7.4.1
Resistor-Transistor Logic (RTL)
RTL circuits are the IC implementation of inverter circuits with saturated transistors, as
shown in Fig. 6.10, for example. If one input voltage of the RTL gate in Fig. 7.21 is in
the H state, the relevant transistor is turned on and the output goes low (L). We therefore
obtain a nor operation in positive logic. The relatively low-valued base resistors insure
that the transistors are turned fully on even for low current gains. However, the fanout
3.6 V
1.5
Vo
Fig. 7.21. RTL nor gate, type MC717
V1 3.6 V1 3.6
Power dissipation : P V = 5 mW
Gate propagation delay : tpd = 25 ns
620 7 Logic Families
Vo
V1
V2
is consequently low. The following circuits are considerably better in this respect and,
nowadays, RTL circuits are no longer used.
7.4.2
Diode-Transistor Logic (DTL)
In the DTL circuit in Fig. 7.22, the base current for the output transistor is injected via
resistor R1 if input diodes D1 and D2 are blocking; in other words, when all of the input
voltages are in the H state. In this case, transistor T1 is turned on and the output voltage
goes low. We thus obtain a nand operation in positive logic. If the same nand gates are
reconnected to the output, the output voltage in the H state is not loaded by the inputs. It
therefore assumes, in the H state, the value V+ . Due to the large gate propagation delay
caused by saturation of the transistors, DTL circuits are no longer used.
7.4.3
High-Level Logic (HLL)
Modified DTL circuits are available for use in equipment in which high noise levels cannot
be avoided. In these circuits, the double diode D3 is replaced by a Zener diode, as shown
in Fig. 7.23. This increases the switching level at the input to approximately 6 V, and a
noise margin of 5 V is obtained for a 12 V supply voltage. In order to increase the fanout,
9.1
Vo
V1
6.8
V2
2.2
HLL circuits have a push–pull stage as shown in Fig. 6.6. The switching time is artificially
increased by using low-speed transistors, and it can be increased still further by means
of an external capacitor. As a result, short spikes have no effect even if their amplitude
is greater than the noise margin. HLL circuits are also known as low-speed logic circuits
(LSL).
7.4.4
Transistor-Transistor Logic (TTL)
Basically, TTL gates operate in exactly the same way as DTL gates. The only difference
is in the design of the diode gate and amplifier. With the standard TTL gate shown in
Fig. 7.24, the diode gate is replaced by transistor T1 , which incorporates several emitters.
If all of the input levels are in the H state, the current from R1 flows via the forward-biased
base–collector diode of the input transistor to the base of T2 , turning it on. If one input is
at low potential, the relevant base–emitter diode becomes conducting and takes over the
base current of T2 . This turns T2 off and the output potential goes high.
In TTL circuits, the amplifier consists of drive transistor T2 and a push–pull output
stage (a totem-pole circuit). When T2 is conducting, T3 is also on and T4 is off. The
output is at L and transistor T3 can accept high currents originating, for example, from the
connected gate inputs. (In the L state, a current flows from the inputs!)
When T2 is off, T3 is also off. In this case, T4 is turned on and delivers an H signal
to the output. The transistor operated as an emitter follower can then supply high output
currents, and thus rapidly charge up load capacitances. Standard TTL circuits, as shown
in Fig. 7.24, are no longer used due to the gate propagation delay caused by the saturation
of the transistors.
One method to avoid saturation consists of connecting a Schottky diode in parallel
with the collector-based (Fig. 7.25). When the transistor is conducting, it provides voltage
feedback to prevent the collector–emitter voltage from falling below about 0.3 V. A TTL
gate employing ”Schottky transistors” of this type is shown in Fig. 7.26; this is actually
a simplified representation of a low-power Schottky TTL gate. A comparison with the
standard TTL gate in Fig. 7.24 shows that the values of the circuit resistors are higher by
a factor of five. The power consumption is therefore lower by a factor of five, being only
2 mW. Nevertheless, the gate propagation delay is no greater, being only 10 ns. The input
V1
V2 Vc
Vo
V1
V2
diode gate, as in DTL circuits, consists of separate diodes. The diode D required in the
output stage for level shifting (Fig. 7.24) is replaced here by the Darlington transistor T3 .
The transfer characteristic of the low-power Schottky TTL inverter (not operation) is
shown in Fig. 7.27. We can see that the switching level is around 1.1 V at the input. The
specified tolerance limits are well exceeded: at the maximum permissible L level at the
input of 0.8 V, an H level of at least 2.4 V must be present at the output. For the minimum
H level at the input of 2.0 V, the L level at the output must be no more than 0.4 V.
Open-Collector Outputs
The problem sometimes arises that a large number of gate outputs must be logically linked.
For 20 outputs, for instance, a gate with 20 inputs would be required, with 20 individual
leads leading to them. This complexity can be avoided by using gates with an open-collector
output. As shown in Fig. 7.28, their output stage consists merely of an npn transistor whose
4.8
4.0
3.2
2.4
Vo / V
1.6
0.8
Vo
emitter is connected to ground. Outputs of this kind can simply be paralleled – unlike the
push–pull output stages otherwise used – and provided with a common-collector resistor
as shown in Fig. 7.28.
The output potential therefore only goes high if all the outputs are high. Consequently,
in positive logic an and operation is produced. On the other hand, we can see that the output
voltage then goes low if one or more of the outputs assumes the L state. We therefore have
an or operation in negative logic. As the operation is realized by the external wiring, it
is referred to as a wired-and or wired-or circuit. As the gates are at low impedance in
the L state only, they are also known as active-low outputs. The wired-and operation is
represented using logic symbols as shown in Fig. 7.29.
An OR operation can also be implemented using open-collector outputs by applying
the wired-and operation to the complemented variables. De Morgan’s law states that:
y1 + y2 + . . . + yn = y 1 · y 2 · . . . · y n
The corresponding circuit is shown in Fig. 7.30.
A disadvantage of using open-collector outputs is that the output voltage rises more
slowly than with push–pull outputs, because the circuit capacitances can only charge up
via resistor RC . In this respect, open-collector TTL gates have the same disadvantages as
Fig. 7.31. Inverter with a tristate output Fig. 7.32. Circuit symbol for an inverter
with a tristate output
the RTL circuits in Fig. 7.21. There, the logical linking can likewise be interpreted as a
wired-and operation.
Tristate Outputs
There is another important application in which circuit simplification can be achieved by
paralleling gate outputs; namely, when any one of several gates connected to a signaling
line is to determine the logic state. This is then referred to as a bus system.
This end can also be attained using open-collector gates, as shown in Fig. 7.29, by
placing all of the outputs, apart from one, in the high-impedance H state. However, the
main disadvantage of the low rate of rise can be avoided in this particular application by
using gates with a tristate output instead of gates with an open-collector output. The tristate
output is a genuine push–pull output with the additional property that it can be placed in
a high-impedance state using a special control signal. This state is known is the Z state.
The basic circuit implementation is shown in Fig. 7.31. When the enable signal EN
= 1, the circuit operates as a normal inverter: for x = 0, z1 = 0 and z2 = 1; that is, T1 is
OFF and T2 is ON. For x = 1, T1 is turned on and T2 is turned off. However, if the control
variable EN = 0, the signals z1 = z2 = 0, and both output transistors are OFF. This is the
high-impedance Z state.
There are many realizations for TTL circuits that differ in speed, supply voltage and
propagation delay. The various Schottky TTL families are listed in Fig. 7.46.
7.4.5
Emitter-Coupled Logic (ECL)
We saw in Fig. 4.56 that, in a differential amplifier with an input voltage difference of
about ±100 mV, current Ik can he completely switched from one transistor to the other. The
amplifier therefore possesses two defined switching states, namely IC = Ik or IC = 0. It is
therefore also known as a current switch. If, for this switch mode, suitably low-resistance
components are selected to insure that the change in voltage across the collector resistors
remains sufficiently low, the conducting transistor can be prevented from being driven into
saturation.
Figure 7.33 shows a typical ECL gate. Transistors T2 and T3 form a differential am-
plifier. A constant potential Vref = −1.3 V is applied to the base of T3 via the voltage
divider. If all of the input voltages are in the L state, transistors T1 and T2 are turned off.
In this case, the emitter current flows via transistor T3 , producing a voltage drop across
R2 . Output voltage Vo1 is therefore in the L state, and Vo2 in the H state. When at least one
7.4 Circuit Implementation of the Basic Functions 625
Vo1
–1.3 V
Vo2
V1 V2
Fig. 7.33. ECL nor-or gate, type MC10102. The emitter resistors R5 and R6 are not incorporated
in the IC and must be connected externally if required
Power dissipation per gate: Pdg = 25 mW
Power dissipation R5 , R6 , each: PdR = 30 mW
Gate propagation delay: tpd = 2 ns
input level goes high, the output states are reversed. Positive logic gives an or operation
for Vo1 and a nor operation for Vo2 .
We shall now examine the potentials within the circuit. We will assume that the base-
emitter voltage of a conducting transistor will be 0.8 V. When transistor T3 is off, there is
hardly no voltage drop across R2 . Consequently, in this case the emitter potential of T5 is
−0.8 V. This is the output H level. If this level is applied, for example, to the base of T2 ,
the emitter potential is
VE = − 0.8 V − 0.8 V = − 1.6 V
If T1 or T2 is conducting the collector current of 4.6 mA will produce a voltage drop of
1 V across R2 . This results in an output voltage of VL = −1.8 V .
Vref must now be selected such that the input transistors are sure to be on at an input
voltage of VH = −0.8 V and off at an input voltage of VL = −1.8 V. This condition
can best be satisfied by setting Vref halfway between VH and VL ; that is, at about −1.3 V.
The complete transfer characteristic is shown in Fig. 7.34. We can see that the switching
level is −1.3 V. At the maximum permissible input L level of −1.5 V, an H level of at least
−1.0 V must be produced at the nor output. At the lowest input H level of −1.1 V, the L
level at the output must not exceed −1.65 V.
In contrast to other logic circuit families, the input voltage of ECL in the H state is
tightly constrained at the upper limit. If it exceeds −0.8 V, the relevant input transistor will
be driven into saturation. This can be seen from the bend in the transfer characteristic for
the nor output, at an input voltage of −0.4 V. As the voltage increases further, the collector
potential VC increases with the emitter potential due to the saturation of transistor T2 , and
therefore output voltage Vo2 also increases.
We can see from Fig. 7.34 that the logic levels are much closer to zero potential than to
the negative supply voltage (−5.2 V). Moreover, the magnitude of the supply voltage does
not affect the H level, as this is determined only by the base–emitter voltage of the emitter
followers. Therefore it was usual in the beginning of ECL-logic to define the positive
supply voltage as ground and use a negative emitter supply. However one can shift all
626 7 Logic Families
–0.4
–0.6
–0.8 OR
–1.0
–1.2
Vo / V
–1.4
–1.6
–1.8
–2.0
–2.0 –1.8 –1.6 –1.4 –1.2 –1.0 –0.8 –0.6 –0.4 –0.2 0
Vi / V
voltages by 5.2 V in positive direction. If ECL-circuits are operated this way one speaks of
Positive-ECL shortly PECL-circuits. Then the positive supply voltage is usually reduced
to 5.0 V or even 3.3 V in order to use the supply voltages used in the remaining circuits.
Of all the logic families, ECL circuits have the smallest gate propagation delays. Indeed,
they are even faster than Schottky TTL circuits, which are also operated unsaturated. The
difference is that the collector–emitter voltage across the conducting transistors is higher –
never less than 0.6 V. Not only does this provide a greater margin to the saturation voltage,
but it also results in a lower collector–base junction capacitance.
Another reason for the high speed of ECL circuits lies in the small signal amplitudes
of only 1 V involved in switching. The unavoidable switching capacitances can therefore
be charged rapidly. The low output resistance of the emitter followers also promotes fast
switching times. From (2.117), this is given by
ro ≈ 1/gm = VT /IC = 26 mV/7.7 mA = 3.4
The high speed of ECL circuits is obtained at the expense of high power dissipation.
For a gate in the MC10.000 series, this can be as much as 25 mW. To this must be added
the power dissipation in the emitter resistors. For an average output voltage of −1.3 V,
there is a power dissipation of 30 mW in a 510 emitter resistor; that is, more than in the
entire gate. For this reason, emitter resistors will only be connected to the outputs used.
This dissipation in the emitter resistors can be reduced to 10 mW if instead of connecting
510 resistors to the −5.2 V supply, 50 resistors are used on an additional supply
voltage of VT T = −2 V. However, the associated cost and complexity is only justifiable
for extensive ECL circuitry. Additional care must be taken to insure that the −2 V supply
voltage is generated with high efficiency in the power supply. Otherwise, the problem of
power dissipation is merely shifted from the circuit to the power supply. For this reason,
it is impractical to produce the −2 V from the −5.2 V using a series regulator.
Wired-OR Operation
By connecting ECL outputs in parallel, it is possible – as with open-collector outputs –
to implement a logic operation. This possibility is illustrated in Fig. 7.35. As the H level
is predominant when the emitter followers are connected in parallel (active high), we
obtain an or operation in positive logic. The advantage of a wired- or operation using
7.4 Circuit Implementation of the Basic Functions 627
–5.2V
ECL circuits is that the speed is not reduced. We not only save one gate, but also one
propagation delay time.
To summarize, let us enumerate once more the main considerations for using ECL
gates in high-speed logic circuits:
1) They exhibit the shortest gate propagation delays.
2) Their power consumption is independent of the switching state. No current or voltage
spikes occur during switching. Consequently, high-frequency noise injected into the
printed circuit board remains low.
3) The balanced outputs allow noise-immune differential signal transmission over com-
paratively large distances (see Sect. 7.5).
A list of the various ECL families is given in Fig. 7.46 below.
7.4.6
Complementary MOS Logic (CMOS)
CMOS logic circuits constitute a family characterized by extremely low power consump-
tion. Figure 7.36 shows an inverter circuit. It is noticeable that the circuit consists ex-
clusively of enhancement-type MOSFETS. The source electrode of the n-channel FET is
connected to ground and that of the p-channel FET to the supply voltage VDD . The two
FETs therefore operate in source connection and amplify the input voltage on an inverting
basis, with one transistor constituting the pull-up resistance for the other.
The pinch-off voltage of the two MOSFETs is about 1.5 V. For a supply voltage of
5 V, at least one of the two MOSFETs is therefore on. If we make Vi = 0, the p-channel
FET T2 is on and the n-channel FET T1 is off. The output voltage is equal to VDD .
For Vi = VDD , T2 is off and T1 is on. The output voltage becomes zero. We can see
that under steady-state conditions no current flows through the circuit. It is only during
switching that a small crossover current flows, as long as the input voltage is in the range
|Vp | < Vi < VDD − |Vp |. The crossover current IDD and the transfer characteristic are
plotted in Fig. 7.37.
Vo / V
The logic levels depend on the supply voltage selected. The permissible supply voltage
range for CMOS circuits is very large. In the case of silicon gate circuits it is between 3 V
and 6 V, while for metal gate circuits it is as much as 3–15 V. For reasons of symmetry, the
switching level is always at half the supply voltage. Consequently, for a supply voltage of
5 V the H level must be above 3.5 V, as we can see from Fig. 7.37. In order to drive a CMOS
gate input with a TTL gate output, an additional pull-up resistor is therefore required. On
the other hand, HCT (high-speed CMOS TTL) circuits are fully TTL-compatible, because
they have input transistors with an adapted pinch-off voltage.
The current consumption of the CMOS gate comprises three parts:
1. A low reverse current of only a few microamperes that flows if the input voltage is at
zero level or equal to VDD .
2. A cross current that flows temporarily through both transistors whenever the state of
the input signal changes.
3. The principal contribution occurs during charging and decharging the transistor ca-
pacitances CT .
1 2 is stored in the transistor capacitances
During the charging process the energy 2 CT VDD
CT ; simultaneously the same amount of energy is converted to heat in the charging FET.
During discharging the energy stored in the capacitance is is converted to heat in the
discharging FET. The energy W = CT VDD 2 is therefore converted to heat during a L-H-L
P V = W/t = W · f = CT · VDD
2
·f
Since the losses caused by the cross current are also proportional to the frequency they
can also be taken into account by defining a power dissipation capacity CP D as described
by the formula
PV ges
CP D = 2 ·f
VDD
This is slightly higher than the pure transistor capacitances CT .
The potential of open CMOS inputs is undefined. The open inputs must therefore be
connected to ground or VDD . This is necessary even for unused gates, because otherwise
an input potential can occur at which an undefined but high crossover current flows through
the two transistors, resulting in unexpectedly high power dissipation.
7.4 Circuit Implementation of the Basic Functions 629
= T4
Vi
Vi Vo
Fig. 7.38. Input protection circuit for Fig. 7.39. Parasitic thyristor caused by the
CMOS gates junction insulation of the MOSFET
CMOS Gates
Figure 7.40 shows a CMOS nor gate which operates on the same principle as the inverter
described above. In order to insure that the pull-up resistance is high when one of the input
voltages assumes the H state, a suitable number of p-channel FETs must be connected in
series. By replacing the parallel circuit with a series arrangement, the nor gate becomes
the nand gate shown in Fig. 7.41.
Transmission Gates
In Sect. 7.1 we saw that logic operations can also be implemented using switches. This
possibility is also utilized in MOS technology, as it frequently results in circuit simplifi-
cation. The resultant component is known as a transmission gate and is used in addition
to the conventional gates. Its circuit symbol and equivalent circuit are shown in Fig. 7.42.
The way it operates is that input and output are either connected via a low on resistance
630 7 Logic Families
V1
Vo
V2 V1
Vo
V2
Fig. 7.40. CMOS nor gate Fig. 7.41. CMOS and gate
of a conducting MOSFET or isolated. As, in effect, the two terminals are interchangeable,
the signal can be transmitted in both directions with minimal delay.
Whereas the logic level in conventional gates is always regenerated, no level regenera-
tion occurs here. The noise margin therefore deteriorates as the number of interconnected
transmission gates is increased. Consequently, they are only used in conjunction with
conventional gates.
Circuit implementation in CMOS technology is shown in Fig. 7.43. The actual switch is
formed by the two complementary MOSFETs, T1 and T2 . The drive arrangement consists
of the inverter producing complementary gate potentials. When Vcontr. = 0, VGN = 0 and
VGP = VDD . This causes the two MOSFETs to be turned off, provided that the signal
voltages V1 and V2 lie within the range zero to VDD . If, on the other hand, we make
Vcontr. = VDD , then VGN = VDD and VGP = 0. In this case there is always at least one
MOFSFET conducting, as long as V1 and V2 are within the permissible signal voltage
range.
As we shall see in Chap. 17.2.1, this configuration can also be used as an analog
switch. It differs from the transmission gate in that the gate electrodes of T1 and T2 are
not controlled with logically complementary signals, but driven with signals of opposite
polarity. This makes it possible to switch positive and negative signal voltages.
Due to their low power requirement and wide supply voltage range, CMOS circuits
are particularly suitable for battery-operated equipment. The various CMOS families are
listed in Fig. 7.46.
V1 Vcontr. V2
Vcontr. Vcontr. V1 V2
V1 V2
Fig. 7.42. Circuit symbol and Fig. 7.43. Internal design of a transmission gate
operation of a transmission gate
7.4 Circuit Implementation of the Basic Functions 631
Vo
V1 V2
Vo
V1 V2
Fig. 7.44. Standard circuit for the NMOS Fig. 7.45. NMOS nor gate with a
nor gate depletion load
7.4.7
NMOS Logic
The feature of NMOS ICs is that they consist exclusively of n-channel MOFSFETS. They
are therefore particularly easy to manufacture and, for this reason, they are mainly used in
large-scale integrated (LSI) circuits.
The NMOS nor gate shown in Fig. 7.44 is a close relative of the RTL nor gate shown
in Fig. 7.21. For technological reasons, a MOSFET is used instead of the ohmic pull-up
resistor. As with the input FETs an enhancement type is employed. To make it conduct,
a high gate potential VGG must be applied. If the output voltage in the H state is to rise
to the drain potential VDD , the auxiliary potential VGG must be selected to be higher than
VDD by at least the pinch-off voltage. In addition, a negative substrate bias VBB is often
required, in order to turn off the input FETs reliably and reduce the junction capacitances.
As we can see from Fig. 7.44, T1 , operates as a source follower for VGG . The internal
resistance ro therefore has the value l/gm . In order to achieve the high resistance values
required, it is given a considerably lower transconductance than the input FETs.
The positive auxiliary voltage VGG can be dispensed with by using a depletion-type
MOSFET for T3 . This possibility is illustrated in Fig. 7.45, in which T3 is operated as a
constant current source as in Fig. 4.126. However, the input FETs must always be of the
enhancement type, as the control voltage would otherwise have to be negative, whereas
the output voltage is always positive. Direct coupling of such gates would therefore be
impossible.
Using ion implantation, it is possible to integrate depletion and enhancement type
MOSFETs on the same chip. The negative auxiliary voltage can be eliminated by selecting
suitable pinch-off voltages, or can be generated from the positive supply voltage using a
voltage converter incorporated in the device.
NMOS logic has no more importance because all MOS chips are made in CMOS
technology today.
7.4.8
Summary
Figure 7.46 provides a list of the most commonly used logic families. In each case, the
data refer to a single logic gate. We can see that each technology is available in various
632 7 Logic Families
Fig. 7.46b. Some ECL-families. Power includes one output resistor of 50 connected to
VT T = VCC − 2V resulting in 10 mW average power dissipation
versions, which differ in terms of power dissipation and propagation delay. Clearly, the
newer families, possess a noticeably higher speed at lower power. The reason for this is
that they are dielectrically insulated and therefore have smaller switching capacitances
than the older junction-insulated types.
The power consumptions of the logic circuit families vary greatly. We can see from
Fig. 7.47 that the CMOS circuits perform well at low frequencies. However, above 1 MHz
there is little difference in power dissipation between low-power Schottky and CMOS
circuits. It is noticeable that in this frequency range the power consumption of TTL circuits
also rises. The reason for this is that a crossover current flows through the totem-pole output
stage at each switching cycle, which significantly increases the power consumption at high
frequencies. ECL circuits do not have this drawback. Consequently, apart from being more
expensive, ECL circuits offer advantages at frequencies above 300 MHz.
Digital ICs will only function properly with a well-designed power supply arrange-
ment. All of the logic circuit families generate high-frequency current pulses on the supply
lines during switching. As all of the signals are referred to ground potential, low-resistance
and low-inductance grounding of all ICs is required. This requirement is best satisfied on
a printed circuit board, by means of a ground layer. At frequencies above 50 MHz, it is
advisable to metallize one side of the board completely as a ground layer and only cut out
7.5 Connecting Lines 633
the terminals (see the next section). In order to prevent the current pulses from contam-
inating the supply voltage during switching, the latter must be fed to the ICs with very
low resistance and inductance. If the ground connection is made well, interference can be
prevented by smoothing the supply voltage with capacitors. For this purpose, 10 . . . 100 nF
ceramic capacitors are used. Electrolytic capacitors are unsuitable due to their poor high-
frequency performance. Depending on the requirements, one capacitor is assigned to
2 . . . 5 chips.
7.5
Connecting Lines
So far, we have assumed that the digital signals are transmitted undistorted from one to
another. However, at signals with low rise- and fall-times the effect of the connecting lines
is not negligible. As a rule of thumb, a simple connecting wire is no longer adequate if the
delay on that wire attains the order of magnitude of the rise time of the signal. Consequently,
the maximum length for such connections is approximately 10 cm per nanosecond rise
time.
If this length is exceeded, severe pulse deformations, reflections, and more or less
damped oscillations occur. This problem can be overcome by using lines of defined char-
acteristic impedance (coaxial cable, microstrip lines), which are terminated in their char-
acteristic (surge) impedance. This is generally between 50 and 300 .
Microstrip lines can be produced by fabricating all the connecting tracks on the un-
derside of a circuit board and fully metallizing the component side, while providing small
clearances for the insulation of the component terminals. In this way, all of the connecting
tracks on the underside become microstrip lines. If the circuit board used has a relative
permittivity εr = 5 and a thickness d = 1.2 mm, we obtain a surge impedance of 75 for
a conductor track width of w = 1 mm; see Chap. 24.2.1.
For connections from one board to another, coaxial cables can be used. However, they
have the major disadvantage that they are difficult to run via multi-pin connectors. It is much
634 7 Logic Families
+ 3.3 V + 3.3 V
110 Ω
SN65LVDS100 SN65LVDS100
110 Ω
50 Ω – 5.2 V
– 5.2 V
–2V
simpler to run the signal via a simple insulated twisted-wire pair, which can be connected
to two adjacent pins of a normal multi-pin connector. If these twisted-pair lines have
approximately 100 turns per meter, a characteristic impedance of about 110 is obtained.
The simplest method of transmitting signals on a twisted-pair line is shown in Fig. 7.48.
Due to the low-impedance termination required, the sending-end gate must be able to
deliver a correspondingly high output current. Such gates are available in IC form as “line
drivers” (buffers). It is advisable to use a Schmitt trigger gate as a receiver to reshape the
signal edges.
Signal transmission that is unsymmetrical about ground, as illustrated in Fig. 7.48, is
relatively sensitive to external disturbances, such as voltage spikes on the ground wire.
In larger systems, symmetrical signal transmission as shown in Fig. 7.49 is therefore
preferable. Here, complementary signals are transmitted on the two wires of the twisted-
pair line and a comparator is used as a receiver. In this mode of operation, the information
is contained in the polarity of the difference voltage, rather than in its absolute value. A
noise voltage therefore only causes a common-mode input, which remains ineffective due
to the common mode rejection of the comparator.
When forming the complementary signals, it must be insured that there is no time delay
between the two signals. Consequently, a special circuit with complementary outputs must
be used with TTL circuits instead of a simple inverter.
Complementary outputs of this type are inherent in ECL gates. They are therefore
particularly well suited for symmetrical data transmission. In order to exploit their high
speed capability, a simple differential amplifier with an ECL-compatible output is used as
a comparator. It is known as a line receiver. The relevant circuit arrangement is shown in
Fig. 7.50.
Chapter 8:
Combinatorial Circuits
combinatorial combinatorial
circuit circuit
Truth Table
ones
high randomly
regularity distributed
gates PLD
Fig. 8.2. Implementing combinational circuits
636 8 Combinatorial Circuits
When many complicated functions have to be realized one is soon confronted with the
notorious TTL grave. In such a case the use of programmable logic devices (PLD) is of
great advantage, since they allow the realization of even the most complicated functions by
means of a single chip in view of the fact that components containing more than 100,000
gates are available. In principle, with PLDs the logical operations are realized in the same
way as with the use of discrete gates. The only difference is that all gates needed may be
contained on a single chip and that the required connections of the gates can be configured
by programming the PLD (see Sect. 10.4 on page 711).
8.1
Number Representation
As digital circuits can only process binary – that is, two-valued – variables, the number
representation must be converted from the conventional decimal system to a binary system.
The various methods of doing this are summarized in the following sections.
8.1.1
Positive Integers in Straight Binary Code
The simplest form of representation is the straight binary code. The digit positions are
arranged in ascending powers of 2. For the straight binary representation of an N -digit
number, we therefore have:
0
N−1
ZN = zN−1 · 2N−1 + zN−2 · 2N−2 + . . . + z1 · 21 + z0 20 = zi 2 i
i=0
As in the decimal system, we simply write the digit sequence {zN−1 . . . z0 } and think in
terms of multiplying the digits by the relevant power of 2 and adding.
Example:
15253dec = 1 1 1 0 1 1 1 0 0 1 0 1 0 1 Straight binary
213 20 Weight
Octal Code
As can be seen, the straight binary representation is difficult to read. One therefore uses an
abbreviated notation by condensing binary numbers into groups of three and writing them
as decimal digits. As the resulting digits are arranged in powers of 23 = 8, this is known
as octal code.
Example: 3 5 6 2 5 Octal
15253dec = 011 101 110 010 101 Dual
212 29 26 23 20
84 83 82 81 80 Weight
8.1 Number Representation 637
Hexadecimal Code
Another commonly used abbreviated notation is obtained by combining binary digits into
groups of four. As the resulting digits are arranged in powers of 24 = 16, this is known as
hexadecimal, or simply hex, code. Each digit can assume values between 0 and 15. Since
we only have ten decimal digits, the numbers “ten” to “fifteen” are represented by the
letters A to F.
Example: 3 B 9 5 Hex
15253dec = 0011 1011 1001 0101 Dual
212 28 24 20
163 162 161 160 Weight
8.1.2
Positive Integers in BCD Code
Straight binary numbers are unsuitable for numeric input and output, as we are accustomed
to calculating in the decimal system. Binary-coded decimal (BCD) notation has therefore
been introduced, in which each individual decimal digit is represented by a binary number;
in other words, by the corresponding straight binary number. In this case we have, for
example:
1 5 2 5 3 Dec
15253dec = 0001 0101 0010 0101 0011 BCD
104 103 102 101 100 Weight
A decimal number encoded in this way could more precisely be termed a BCD number in
8421 code, or a natural BCD number. The individual decimal digits can also be represented
by other binary combinations of four or more digits. As the 8421 BCD code is the most
commonly used, it is often known simply as BCD code. We shall adopt this convention
and draw the reader’s attention to any deviations from natural BCD code.
Numbers between 0 and 15dec can be represented using a four-digit straight binary
number (a tetrad). As only ten combinations are used in BCD code, this form of represen-
tation requires more bits than straight binary code.
8.1.3
Binary Integers of Either Sign
Signed-Magnitude Representation
A negative number can be characterized quite simply by placing a sign bit s in front of the
highest-order digit. Zero means “positive” and one means “negative.” An unambiguous
interpretation is only possible if a fixed word length has been agreed upon.
638 8 Combinatorial Circuits
+ 118dec = 0 1 1 1 0 1 1 02
− 118dec = 1 1 1 1 0 1 1 02
(− 1)s 26 25 24 23 22 21 20
Two’s-Complement Representation
The disadvantage of signed-magnitude representation is that positive and negative numbers
cannot be added simply. An adder must be switched over to subtraction mode when a minus
sign occurs. With two’s-complement representation, this is unnecessary.
With 2’s-complement representation, the most significant bit is given a negative weight.
The rest of the number is represented in normal binary form. Once again, a fixed word
length must be agreed upon, so that the most significant bit is unambiguously defined. For
a positive number, the most significant bit is 0. For a negative number, the most significant
bit must be 1, because only this position has a negative weight.
Example for an 8-bit word length:
+118dec = 0 1 1 1 0 1 1 0
3 45 6
BN
− 118dec = 1 0 0 0 1 0 1 0 = −128 + 10
3 45 6
X
− 27 26 25 24 23 22 21 20 Weight
and
8.1 Number Representation 639
(2) (1)
BN = BN + 1 (8.1)
The 2’s-complement of a binary number is therefore the result of negation of all the digits
and the addition of 1.
It can easily be demonstrated that it is not necessary to deal with the sign digit separately,
but that, to change the sign, it is possible merely to form the 2’s-complement of the entire
number including the sign digit. For binary numbers in 2’s-complement representation,
the following relationship therefore holds:
(2)
−BN = BN (8.2)
This relation applies to the case in which we likewise consider only N digits in the result
and disregard the overflow digit.
Example of an eight-digit binary number in 2’s-complement representation:
118dec = 01110110
1’s-complement: 10001001
+ 1
2’s-complement: 1 0 0 0 1 0 1 0= − 118dec
Reconversion:
1’s-complement: 01110101
+ 1
0 1 1 1 0 1 1 0= + 118dec
Sign Extension
If we wish to expand a positive number to a larger word length, we simply add to the
leading zeros. In 2’s-complement, a different rule applies: we have to extend the sign bit.
Example: 8 bit 16 bit
118dec =01110110=0000000001110110
− 118dec = 1 0 0 0 1 0 1 0 = 13 1 1 1451 1 1 16 1 0 0 0 1 0 1 0
sign extension
The proof is simple. For an N-digit negative number, the sign bit has the value − 2N−1 . If
we extend the word length by one bit, we have to insert an additional leading “one”. The
added sign digit has the value −2N . The old sign digit changes its value from − 2N−1 to
+2N−1 . The two together therefore have the value:
Offset Binary
Some circuits can only process positive numbers. Therefore, they always interpret the most
significant digit as being positive. In such cases, we define the midpoint of the number
range as being represented as zero (offset binary representation).
Using an eight-digit positive binary number, the range 0–255dec can be represented;
while using an eight-digit 2’s-complement number, we have the range from −128dec to
+127dec . To change to offset binary representation, we shift the number range by adding
128 to 0 − 255. Numbers above 128 are therefore to be treated as positive, and numbers
under 128 as negative. In this case, the range midpoint “128” means zero. The addition of
128 can simply be performed by negating the sign bit in the 2’s-complement notation. The
number range is shown in Fig. 8.3.
8.1.4
Fixed-Point Binary Numbers
Like a decimal fraction, a binary fraction is defined such that the weights to the right of
the point are interpreted as negative powers of 2.
Example:
225.8125dec = 1 1 1 0 0 0 0 1 , 1 1 0 1
27 26 25 24 23 22 21 20 2−1 2−2 2−3 2−4
In general, a fixed number of digits after the point is stipulated; hence the term “fixed-point
binary digit.” Negative fixed-point numbers are given in signed-magnitude form.
When specifying a defined number of digits, it is possible, by multiplying by the
reciprocal of the lowest power of 2, to produce integers that can be processed in the
notations described. For the numeric output, the multiplication is again reversed.
8.1.5
Floating-Point Binary Numbers
By analogy with floating-point decimal numbers,
Z10 = M · 10E
a floating-point binary number is defined as
8.1 Number Representation 641
Length
Single
Double
Internal
Z2 = M · 2 E
where M is the mantissa and E is the exponent.
Example:
225.8125 decimal, fixed point
= 2.258125 E 2 decimal, floating point
= 11100001.1101 straight binary, fixed point
= 1.11000011101 E 0111 straight binary, floating point
For computation with floating-point numbers, the notation specified in IEEE Standard
P 754 is universally employed nowadays. This notation is used not only in mainframe
computers, but also in PCs and even in some signal processors, and is in many cases
supported by the corresponding arithmetic hardware. The user can choose between two
precision formats: 32-bit single precision and double precision with 64 bits. Internally,
computation is performed with 80-bit precision. These three formats are shown in Figs. 8.4
and 8.5. There are three distinct parts in each format: the sign bit S, the exponent E, and the
mantissa M. The word lengths of the exponent and mantissa are functions of the precision
selected.
In the IEEE Standard, the mantissa Mis specified by the digits m0 , m1 , m2 , . . . Usually,
the mantissa is normalized to m0 = 1:
0
k
M = 1 + m1 · 2−1 + m2 · 2−2 + . . . = 1 + mi 2−i ,
i=1
Its absolute value is therefore 1 ≤ M < 2. The digit m0 = 1 is only specified for the
internal notation; otherwise, it is hidden and must be restored for the calculation.
The exponent E is specified in IEEE format as an offset binary number, so that positive
and negative values can be defined. For the calculation, an offset amounting to half the
range must be subtracted, namely:
642 8 Combinatorial Circuits
Word 1 Word 2
Z = (− 1)S · M · 2E−offset
Taking the example of IEEE 32-bit single precision, we shall now examine this in
somewhat greater detail. The segments of a word are shown in Fig. 8.6. The most significant
bit is the sign bit S, followed by 8 bits for the exponent and 23 bits for the mantissa. The
MSB (most significant bit) of the mantissa m0 = 1 is hidden; the point is to the left of m1 .
The weight of m1 is therefore 1/2.
A 32 bit floating-point number can be split up into two words of 16 bits, 4 bytes,
or 8 nibbles each. It can therefore be expressed by eight hex characters. A number of
examples are given in Fig. 8.7. The normalized number NOR1 has an exponent of 127;
after subtracting the offset from 127, we obtain a multiplier of 20 = 1. The noted value of
the mantissa is 0.75. This, together with the hidden 1, produces the specified value +1.75.
In the second example, NOR2 , a negative number has been selected; in this case, S = 1.
The number 10 in the third example is represented in normalized form as 10 = 23 · 1.25.
We arrive at the given hex representation in the usual way, by organizing the bit string into
groups of four and using the associated hex symbols. Unfortunately, the hex representation
of IEEE numbers is very involved, because the first symbol contains the sign and part of
the exponent, and the third symbol a mixture of exponent and mantissa.
A couple of special cases are also listed in Fig. 8.7. The largest number that can be
represented in 32-bit IEEE format is
The exponents 0 and 255 are reserved for exceptions. The exponent 255 is interpreted in
conjunction with the mantissa M = 0 as ±∞, depending on the sign. If the exponent and
the mantissa are both 0, the number is defined as Z = 0. In this case, the sign is irrelevant.
8.2 Multiplexer – Demultiplexer 643
Fig. 8.7. Examples of normalized numbers and exceptions in 32-bit floating-point format
8.2
Multiplexer – Demultiplexer
Multiplexer are circuits that connect one of several data sources to a single output. Which
of the sources is selected must be determined by an address. The inverse circuit operation
which distributes data to several outputs according to their address is known as demulti-
plexer. In both circuits, addressing the selected input or output is performed by a 1-out-of-n
decoder, described below.
8.2.1
1-of-n Decoder
A 1-of-n decoder is a circuit with n outputs and ld n address inputs. The outputs yi are
numbered from 0 to (n − 1). An output therefore goes to “one” precisely when the input
binary number A is identical to the number i of the relevant output. Figure 8.9 shows the
A a1 a0 y3 y2 y1 y0
0 0 0 0 0 0 1
1 0 1 0 0 1 0
2 1 0 0 1 0 0
3 1 1 1 0 0 0
Fig. 8.8. Circuit for a 1-of-4 decoder Fig. 8.9. Truth table for a 1-of-4 decoder
y0 = a 0 a 1 , y1 = a0 a 1 , y2 = a 0 a1 , y3 = a0 a1
644 8 Combinatorial Circuits
truth table for a 1-of-4 decoder. The variables a0 and a1 represent the binary code of the
number A. The sum of the products (disjunctive normal form) of the recoding functions can
be taken directly from the truth table. Figure 8.8 shows the corresponding implementation.
When using monolithic integrated circuits, nand functions are often chosen rather
than and functions, so that the output variables are complemented. For further IC types,
see the following section on demultiplexers.
8.2.2
Demultiplexer
A demultiplexer can be used to distribute input d to various outputs. It represents an
extension of the 1-of-n decoder. The addressed output does not go to “one,” but assumes
the value of input data d. Figure 8.10 illustrates the principle by means of switches, while
Fig. 8.11 shows its implementation using gates. If we make d = 1, the multiplexer operates
as a 1-of-n decoder. Commonly used demultiplexers are listed in Fig. 8.12.
1-of-4
decoder
8.2.3
Multiplexer
The opposite of a demultiplexer is a multiplexer. Starting from the circuit in Fig. 8.10, it
can be implemented by swapping the outputs and input to give the basic circuit shown
in Fig. 8.13. This provides a particularly simple illustration of the mode of operation: a
1-of-n decoder selects from n inputs the one whose number coincides with the address and
switches it to the output. The corresponding gate implementation is shown in Fig. 8.14.
In CMOS technology, a multiplexer can be implemented using both gates and analog
switches (transmission gates). When analog switches are employed, signal transmission
is bidirectional. In this case, therefore, the multiplexer is identical to the demultiplexer,
as comparison of Figs. 8.10 and 8.13 will show. The circuit is then known as an analog
multiplexer/demultiplexer.
The or operation required in multiplexers can also be implemented using a wired- or
connection. This possibility is shown for open-collector outputs in Fig. 8.15. In positive
logic, this connection results in an and operation, therefore it is necessary to use the
complemented signals – as in Fig. 7.30.
In order to overcome the disadvantage associated with open-collector outputs, namely
the higher switching time, tristate outputs can be connected in parallel, with only one being
activated at a time. This alternative is shown in Fig. 8.16.
1-of-4
decoder
Fig. 8.15. Multiplexer with open-collector Fig. 8.16. Multiplexer with tristate
gates gates
646 8 Combinatorial Circuits
8.3
Priority Decoder
The 1-of-n code can be converted to binary code by using a priority decoder. At its outputs
a binary number appears that corresponds to the highest input number, which is logic 1.
The value of the lower-index input variables is irrelevant; hence the name priority decoder.
This property enables the circuit to convert not only the 1-of-n code but also a sum code
in which not just one variable is 1, but also the less significant bits as it is the matter at
parallel AD-converters in Chap. 18.10.1. The truth table for the priority decoder is shown
in Fig. 8.18.
IC types:
1-of-10 code: SN 74147 (TTL)
1-of-8 code, extendable: SN 74148 (TTL); MC 10165 (ECL);
MC 14532 (CMOS)
J x9 x8 x7 x6 x5 x4 x3 x2 x1 y3 y2 y1 y0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 1 0 0 0 1
2 0 0 0 0 0 0 0 1 × 0 0 1 0
3 0 0 0 0 0 0 1 × × 0 0 1 1
4 0 0 0 0 0 1 × × × 0 1 0 0
5 0 0 0 0 1 × × × × 0 1 0 1
6 0 0 0 1 × × × × × 0 1 1 0
7 0 0 1 × × × × × × 0 1 1 1
8 0 1 × × × × × × × 1 0 0 0
9 1 × × × × × × × × 1 0 0 1
8.4
Combinatorial Shift Register (Barrel Shifter)
For many arithmetic operations, a bit pattern must be shifted by one or more binary digits.
This operation is usually carried out by a shift register, as described in Sect. 9.5. A single
clock pulse results in a shift by one bit. There is a disadvantage, however, in that a sequential
controller is necessary to organize the loading of the bit pattern into the shift register and
the subsequent shifting by a given number of binary digits.
The same operation may be carried out without a clocked by employing a combinatorial
network involving multiplexers, as illustrated in Fig. 8.19. For this reason, the unclocked
shift registers involved are termed combinatorial or asynchronous shift registers or barrel
shifter that are mainly used in signal processors. If, in Fig. 8.19, the address A = 0 is
applied, then y3 = x3 , y2 = x2 , and so on, but if A = 1, then y3 = x2 , y2 = x1 , y1 = x0 ,
and y0 = x1 , due to the wiring arrangement of the multiplexers. The bit pattern X therefore
appears at the output left-shifted by one digit. As with a normal shift register, the MSB
is lost. If multiplexers with N inputs are used, a shift of 0, 1, 2, ..., (N − 1) bits can be
executed. For the example shown in Fig. 8.19, N = 4; the corresponding function table is
shown in Fig. 8.20.
If the loss of MSBs is to be avoided, the shift register may be extended by adding
identical elements, as illustrated in Fig. 8.21. For the chosen example, where N = 4, a 5-
bit number X can be shifted in this way by a maximum of 3 bits without loss of information.
The shifted number then appears at outputs y3 − y7 .
The circuit shown in Fig. 8.19 can also be operated as a ring shifter if the extension
inputs x−1 . . . x−3 are connected to the inputs x1 . . . x3 , as shown in Fig. 8.22.
IC Types:
16-bit (TTL): SN 74 AS 897 from Texas Instruments
m m m m
a1 a0 y3 y2 y1 y0
0 0 x3 x2 x1 x0
0 1 x2 x1 x0 x−1
1 0 x1 x0 x−1 x−2
1 1 x0 x−1 x−2 x−3
Fig. 8.20. Function table for the barrel shifter
648 8 Combinatorial Circuits
barrel shifter
8.5
Digital Comparators
Comparators check two numbers, A and B, against one another, the relations of interest
being A = B, A > B, and A < B. We shall first consider comparators that determine
whether two binary numbers are equal (identity comparators). The criterion for this is that
all corresponding bits of the two numbers are identical. The comparator will produce a
logic 1 at its output if the numbers are equal, and otherwise a logic 0. In the simplest case,
the two numbers consist of only one bit each; to compare them, the equiv operation (the
exclusive nor gate) may be used. Two N -bit numbers are compared bit by bit using an
equiv circuit for each binary digit, and the outputs are combined by an and gate, as shown
in Fig. 8.23.
IC Types:
2 × 8 inputs: SN 74 LS 688 (TTL) from Texas Instruments
Comparators have a wide range of application if, in addition to indicating equality,
they can also determine which of two numbers is the larger. Such circuits are known as
magnitude comparators. To enable a comparison of the magnitudes of two numbers, their
Fig. 8.24. 1-bit magnitude Fig. 8.25. Truth table for a 1-bit magnitude
comparator comparator
codes must be known, for the following, we assume that both numbers are straight binary
coded, in other words that
A = aN · 2N + aN−1 · 2N−2 + . . . + a1 · 21 + a0 · 20 .
The simplest case is again that of comparing two single-bit numbers. The formulation
of the logic functions is based on the truth table of Fig. 8.25. From these, we can directly
obtain the circuit shown in Fig. 8.24.
The following algorithm is used for comparing numbers that consist of more than one
bit: To begin with, the most significant bit (MSB) of A is compared with the MSB of B.
If they are different, these bits are sufficient to determine the result. If they are equal, the
next lower significant bit must compared, and so on. If the identity variable of digit i is
denoted by gi , as in Fig. 8.23, the magnitude comparison of an N-digit number is given
by the general relation
IC Types:
for 5-digit comparison: MC10166 (ECL)
for 8-digit comparison: SN 74 LS 682–689 (TTL).
The circuits can be cascaded serially or in parallel; the serial method is shown in
Fig. 8.26. When the three most significant bits are the same, the outputs of comparator 1
determine the result, as they are connected to the LSB inputs of comparator 2 (LSB = least
significant bit).
When comparing many-bit numbers it is better to employ parallel cascading, as shown
in Fig. 8.27, since the propagation delay time is shorter.
comparator 2 comparator 1
comparator 5
8.6
Adders
8.6.1
Half-Adder
Adders are circuits that give the sum of two binary numbers. We shall first describe adders
for straight binary numbers. The simplest case is the addition of two single-bit numbers. To
devise the logic circuit, all possible cases must first be investigated so that a logic function
table can be compiled. If two single-bit numbers A and B are to be added, the following
cases can occur:
0 + 0 = 0,
0 + 1 = 1,
1 + 0 = 1,
1 + 1 = 10
If both A and B are 1, a carry to the next higher bit is obtained. The adder must therefore
have two outputs, one for the sum and one for the carry to the next higher bit. The truth table
shown in Fig. 8.29 can be deduced by expressing the numbers A and B by the logic variables
a0 and b0 . The carry is represented by the variable cl , and the sum by the variable s0 .
By setting up the canonical products, the Boolean functions
c1 = a0 b0 and s0 = a 0 b0 + a0 b0 = a0 ⊕ b0
a0 b0 s0 c1
0 0 0 0
0 1 1 0
1 0 1 0
1 1 0 1
Fig. 8.28. Circuit for the half-adder Fig. 8.29. Truth table for the half-adder
8.6 Adders 651
are obtained. The carry thus represents an and operation, and the sum an exclusive or
operation. A circuit that implements both operations is known as a half-adder and is shown
in Fig. 8.28.
8.6.2
Full-Adder
If two straight binary numbers of more than one digit are to be added, the half-adder can
only be used for the LSB. For all other binary digits, not two but three bits must be added,
as the carry from the next lower binary digit must be included. In general, each bit requires
a logic circuit with three inputs, ai , bi , and ci , and two outputs, si and ci+1 . Such circuits
are called full-adders and can be implemented as shown in Fig. 8.30, using two half-adders.
Their truth table is given in Fig. 8.31.
For each bit, a full-adder is required, but for the LSB a half-adder is sufficient. Fig-
ure 8.32 shows a circuit suitable for adding two 4-bit numbers, A and B. Such circuits are
FA FA FA FA
available as ICs, but usually a full-adder is also used for the LSB, in order to enable the
circuit to be extended as required (SN 74 LS 83).
8.6.3
Look-Ahead Carry Logic
The computing time of the adder in Fig. 8.32 is considerably longer than that of the
individual stages, because the carry c4 can assume its correct value only after c3 has been
determined. The same applies to all the previous carries (ripple carry). To shorten the
computing time for the addition of many-bit straight binary numbers, a look-ahead carry
generator (parallel or simultaneous carry logic) can be used. In this method, all carries are
determined directly from the input variables. From the truth table in Fig. 8.31, the general
relation for the carry of stage i can be deduced:
The quantities gi and pi introduced for brevity appear as intermediate variables in the full-
adder of Fig. 8.30. Their calculation therefore requires no additional complexity. These
variables can be interpreted as follows: the quantity gi indicates whether or not the input
combination ai , bi results in a carry in stage i, and is therefore called the generate variable.
The quantity pi indicates whether the input combination causes a carry from the next
lower-order stage to be absorbed or passed on. It is therefore called the propagate variable.
From (8.3), we obtain successively the individual carries
c 1 = g0 + p 0 c0 ,
c2 = g1 + p1 c1 = g1 + p1 g0 + p1 p0 c0 ,
c3 = g2 + p2 c2 = g2 + p2 g1 + p2 p1 g0 + p2 p1 p0 c0 , (8.4)
c4 = g3 + p3 c3 = g3 + p3 g2 + p3 p2 g1 + p3 p2 p1 g0 + p3 p2 p1 p0 c0
.. ..
. .
It can be seen that the expressions become ever more complicated, but that they can
be computed from the auxiliary variables within 2 propagation delay times: one for the
products and one for the sum.
FA FA FA FA
Figure 8.33 shows the block diagram of a 4-bit adder with look-ahead carry logic. The
equations given in (8.4) are implemented in the carry generator. The complete circuit is
available on a single chip.
IC Types:
TTL: SN 74 LS 181; SN 74 S 281; SN 74 LS 381; SN 74 LS 382; SN 74 LS 681
Adder networks for more than 4 bits can be realized by cascading several 4-bit blocks.
The carry c4 would then be applied as c0 to the next block up. However, this method
is somewhat inconsistent, because the carry is parallel-processed within the blocks but
serially processed between the blocks.
To obtain short computation times, the carries from block to block must therefore also
be parallel-processed. The relationship for c, in (8.4) is thus reconsidered:
c4 = g3 + p3 g2 + p3 p2 g1 + p3 p2 p1 g0 + p3 p2 p1 p0 c0 (8.5)
3 45 6 3 45 6
G P
To abbreviate this, the block-generate variable G and the block-propagate variable P are
introduced, and
c4 = G + P c0
is obtained. The form of this equation is the same as (8.3). Within the individual 4-bit adder
blocks, only the additional auxiliary variables G and P need be computed; when these are
known, the algorithm given in (8.4), and used for the bit-to-bit carries, can also be used
for the carries from block to block. The result is the block diagram given in Fig. 8.34 for
a 16-bit adder with look-ahead carry logic. The carry logic is identical to that of the 4-bit
adder in Fig. 8.33. It can be obtained as a separate IC. When performing a 16-bit addition
with TTL circuits, the computation time is 36 ns; for Schottky-TTL circuits, it is reduced
to 19 ns.
IC carry blocks:
For four digits: SN 74182 (TTL), MC 10179 (ECL), MC 14582 (CMOS)
For eight digits: SN 74LS 882 (TTL)
Fig. 8.34. 16-bit adder with look-ahead carry logic on two levels
654 8 Combinatorial Circuits
sign sign
4-bit adder
Fig. 8.35. Subtraction of
sign two’s complement numbers
D = A−B
8.6.4
Subtraction
The subtraction of two numbers can be reduced to an addition, since
D = A − B = A + (−B) (8.6)
If the numbers are represented in 2’s-complement, for a specified word length N , we can
derive, from (8.2), the simple relation
(2) (1)
−BN = BN = BN + 1
The difference is therefore
(2)
DN = AN + BN
To calculate the difference, we therefore have to form the 2’s-complement of BN and add
it to AN . For this purpose, we must negate all the digits of BN (1’s-complement) and add
1, as stated in (8.1). The addition of AN and 1 can be performed by one and the same
adder, by utilizing the carry input. This results in the 4-bit circuit shown in Fig. 8.35.
In order to insure that the difference DN appears in the correct 2’s-complement notation,
AN and BN must likewise be entered in this format; that is, for positive numbers the highest
bits a3 and b3 must be 0.
The 181-series integrated adders described in Sect. 8.6.3 have control inputs that enable
the input numbers to be complemented. They are therefore also suitable as subtractors. As
further control inputs can be used to select logic operations for the input variables, these
devices are generally known as arithmetic logic units (ALUs).
8.6.5
Two’s-Complement Overflow
When two positive N -bit binary numbers are added, the result can be an (N +1)-bit number.
This overflow is recognizable from the fact that, from the most significant bit, a carry is
produced.
In 2’s-complement notation, the leftmost digit position is reserved for the sign. When
two negative numbers are added, a carry into the overflow position will systematically
occur, as the sign bit of the two numbers is 1. When processing 2’s-complement numbers
of either sign, the occurrence of a carry into the overflow position does not therefore
necessarily mean that an overflow has taken place.
An overflow can be detected as follows: when two positive numbers are added, the
result must also be positive. If the sum is out of range, a carry into the overflow position
8.6 Adders 655
sign
FA FA FA FA
occurs; that is, the result becomes “negative”. This indicates a positive overflow. Similarly,
a negative overflow is present if a “positive” result is obtained when two negative numbers
are added. When a positive and a negative number are added, no overflow can occur, as
the magnitude of the difference is then smaller than the numbers entered.
The occurrence of a 2’s-complement overflow can be easily detected by comparing
the carry cN−1 with the carry cN (Fig. 8.36). An overflow has taken place precisely when
these two carries are different. This case is decoded by the exclusive-or gate. This output
is available on the 4-bit arithmetic unit SN 74 LS 382.
8.6.6
Addition and Subtraction of Floating-Point Numbers
When processing floating-point numbers, the mantissa and exponent must be handled
separately. For addition, it is first necessary to adjust the exponents so that they are the same.
To do this, we take the difference of the exponents and shift the mantissa associated with
the smaller exponent by the corresponding number of bits to the right (see Sect. 8.4). Both
numbers then have the same – namely, the larger – exponent. It is passed on to the output
via the multiplexer shown in Fig. 8.37. The two mantissas can now he added or subtracted,
generally producing a nonnormalized result; that is, the leading 1 in the mantissa is not
multiplexer subtractor
subtractor
priority
shifter right
adder /
shifter left
subtractor
shifter right
in the correct position. To normalize the result, the leftmost 1 in the mantissa is located
using a priority decoder (see Sect. 8.3). The mantissa is then shifted to the left by the
corresponding number of bits and the exponent is reduced accordingly.
8.7
Multipliers
8.7.1
Multiplication of Fixed-Point Numbers
Multiplication of two straight binary numbers is best illustrated by an example. The product
13 · 11 = 143 is to be calculated:
1101 ·
1011
1101
+ 1101
+ 0000
+ 1101
10001111
This calculation is particularly easy, because only multiplications by either 1 or 0 occur.
The product is obtained by consecutive shifting of the multiplicand to the left by 1 bit at
a time, and by adding or not adding, depending on whether the corresponding multiplier
bit is 1 or 0. The individual bits are processed consecutively, and this method is therefore
known as serial multiplication.
Multiplication can be implemented by combining a shift register and an adder, although
such a circuit would require a sequential controller. If the adders are suitably staggered and
interconnected, the shifting process can be carried out by wiring. This method requires a
large number of adders, but the shift register and the sequential controller are no longer
needed. The main advantage, however, is the considerably reduced computation time since,
instead of the time-consuming clock control, only gate propagation delays are incurred.
Figure 8.38 shows a suitable circuit for a combinatorial 4 × 4 bit multiplier. For the ad-
ditions, we can usefully employ the SN 74 LS 381 chip, whereby addition can be activated
and deactivated by the mode control input, giving
A + 0 für m = 0
S=
A + B für m = 1
The multiplier is applied bit by bit to the control inputs m. The multiplicand is fed in
parallel to the four addition inputs b0 to b3 .
To begin with, we assume that the number K = 0. We then obtain the expression
S 0 = X · y0
at the output of the first element, corresponding to the first partial product in the above mul-
tiplication algorithm. The LSB of S0 represents the LSB of the product P ; it is transferred
directly to the output. The more significant bits of S0 are added in the second element to
the expression X · y1 . The resulting sum is the subtotal of the partial products in the first
and second line of the multiplication algorithm. The LSB of this sum gives the second
lowest bit of P and is therefore transferred to the output p1 . The subsequent subtotals are
8.7 Multipliers 657
controllable adder
controllable adder
controllable adder
controllable adder
Fig. 8.38. Multiplier for two 4-bit numbers. The values entered refer to the example 13 · 11 = 143.
Result: P = X · Y + K
treated accordingly. We have entered the numbers of the above example in Fig. 8.38 to
demonstrate this process.
The additional inputs k0 − k3 can be used to add a 4-bit number K to the total product
P , so that the multiplier function becomes
P = X·Y +K.
The method of expansion for larger numbers can now be understood. For each additional
bit of the multiplier Y , a further arithmetic unit is added at the bottom left of the circuit.
If the multiplicand X is to be increased, the word length is enlarged by cascading an
appropriate number of arithmetic units at each stage.
In the multiplication method described, the new partial product is always added to
the previous subtotal. This technique requires the fewest elements and results in straight-
forward and easily extensible circuitry. The computing time can be shortened if as many
summations as possible are carried out simultaneously, and if the individual subtotals are
added afterwards by a fast adder circuit. Several procedures are available, which differ
only in the adding sequence (the Wallace Tree). Another way of reducing the computing
time is to use the Booth algorithm. The multiplier bits are combined into pairs, thereby
halving the number of adders required, and the computing time is reduced accordingly.
658 8 Combinatorial Circuits
8.7.2
Multiplication of Floating-Point Numbers
To multiply floating-point numbers, the mantissas of both numbers have to be multiplied
and their exponents added, as shown in Fig. 8.39. During this process, an overflow may
occur in the mantissa. The result can be renormalized by shifting the mantissa one place
to the right and increasing the exponent by 1. In this case, denormalization as used with
the floating-point adder in Fig. 8.37 is not required; the complexity lies in the multiplier.
Formerly many hardware multiplier chips have been available. Now the multipliers
shown are part of CPUs and signal processors like those in Fig. 19.59.
adder adder + 1
overflow
shifter 1
multiplier
right
multiplier normalizer
A sequential logic system is an arrangement of digital circuits that can carry out logic
operations and, in addition, store the states of individual variables. It differs from a com-
binatorial logic system in that the output variables yj are not only dependent on the input
variable xi , but also on the previous history, which is represented by the state of flip-flops.
In what follows, we shall first discuss the design and operation of integrated flip-flops.
9.1
Integrated Flip-Flops
We described simple transistor flip-flops in Sect. 6.2.1. We shall now demonstrate the
operation of flip-flops with reference to gates. This approach defines their basic operation
irrespective of the particular technology employed.
9.1.1
Transparent Flip-Flops
By connecting two nor gates in a feedback arrangement, as shown in Fig. 9.1, we obtain
a flip-flop that has complementary outputs Q and Q and two inputs S (Set) and R (Reset).
If the complementary input state S = 1 and R = 0 is applied, we have
Q = S+Q = 1+Q = 0
and
Q = R+Q = 0+0 = 1
The two outputs therefore assume complementary states. Similarly, for R = 1 and S = 0,
the opposite output state is obtained. If we make R = S = 0, the old output state is
retained. This explains why RS flip-flops are used as memories. When R = S = 1, the
two outputs simultaneously become zero; however, the output state is no longer defined
when R and S simultaneously become zero. Consequently, the input state R = S = 1 is
generally avoided. The switching states are summarized in the truth table shown in Fig. 9.2,
with which we are already familiar from the transistor circuit in Fig. 6.10 on page 591.
S R Q Q
0 0 Q−1 Q−1
0 1 0 1
1 0 1 0
S 1 1 (0) (0)
Fig. 9.1. RS flip-flop consisting of nor Fig. 9.2. Truth table for an RS flip-flop
gates
660 9 Sequential Logic Systems
S R Q Q
0 0 (1) (1)
0 1 1 0
1 0 0 1
S 1 1 Q−1 Q−1
Fig. 9.3. RS flip-flop consisting of Fig. 9.4. Truth table for an RS flip-flop
nand gates comprising nand gates
In Sect. 7.1 on page 611 we showed that a logic equation does not change if all the
variables are negated and the arithmetic operations (+) and (·) are interchanged. Applying
this rule here, we arrive at the RS flip-flop comprising nand gates shown in Fig. 9.3, which
has the same truth table as that shown in Fig. 9.2. However, note that the input variables
are now R and S. As we shall be using the RS flip-flop comprising nand gates frequently,
we have given its truth table for input variables R and S in Fig. 9.4.
Clocked RS Flip-Flops
We frequently require an RS flip-flop that only reacts to the input state at a specific moment
in time, which is determined by an additional clock variable C. Figure 9.5 shows a statically
clocked RS flip-flop of this kind. If C =0, then R = S = 1. In this case, the flip-flop stores
the old state. For C = 1, we obtain
R = R
and S = S
Clocked D Flip-Flops
We shall now examine how the value of a logic variable D can be stored using the flip-flop
shown in Fig. 9.5. We have seen that Q = S if complementary input states are applied and
C = 1. In order to store the value of a variable D, we therefore need only make S = D
and R = D. The inverter G5 in Fig. 9.6 is used for this purpose. In the resulting data latch,
Q = D as long as clock C = 1. This may also be seen from the truth table in Fig. 9.7.
Due to this property, the clocked data latch is also known as a transparent D flip-flop. If
we make C = 0, the existing output state is stored.
We can see that nand gate G4 in Fig. 9.6 acts as an inverter for D when C = 1. Inverter
G5 can therefore be omitted, producing the practical implementation of a D latch shown
in Fig. 9.8. The circuit symbol is given in Fig. 9.9.
IC types:
74 LS 75 (TTL); 10133 (ECL); 4042 (CMOS)
9.1 Integrated Flip-Flops 661
C D Q
0 0 Q−1
0 1 Q−1
S 1 0 0
1 1 1
Fig. 9.6. Transparent D flip-flop (D latch) Fig. 9.7. Truth table for the transparent
D-flip-flop
Fig. 9.8. Practical implementation of a transparent Fig. 9.9. Circuit symbol for a
D flip-flop transparent D flip-flop
9.1.2
Flip-Flops with Intermediate Storage
For many applications, such as counters and shift registers, transparent flip-flops are un-
suitable. In these cases, flip-flops are required that temporarily store the input state and
only transfer it to the output when the inputs are inhibited. There is no clock state where
the flip-flops are transparent: this means where data are forewarded from the input to the
output. They therefore comprise two flip-flops: the master flip-flop at the input and the
slave flip-flop at the output.
Two-Edge-Triggered Flip-Flops
Figure 9.10 shows a master–slave flip-flop of this kind. It consists of two statically clocked
RS flip-flops of the type shown in Fig. 9.5. The two flip-flops are mutually inhibited by
complementary clock signals. Gate G15 is used for clock inversion. As long as clock C = 1,
S2
S1
S2
S1
the input information is read into the master. The output state remains unchanged, because
the slave is disabled.
When the clock goes to zero, the master is disabled, thereby freezing the state that
was present immediately prior to the negative-going edge of the clock signal. The slave
is simultaneously enabled, thus transferring the state of the master to the output. Data
transmission therefore occurs on the negative-going edge; however, there is no clock state
in which the input data have a direct effect on the output, as is the case with transparent
flip-flops.
The input combination R = S = 1 necessarily results in undefined behavior, because
inputs S 1 , R 1 in the master simultaneously go from 00 to 11 when clock C goes to zero. In
order to make use of this input combination, the complementary output data are additionally
applied to the input gates. The feedback circuit shown in heavy type in Fig. 9.11 is used for
this purpose. The external inputs are then designated J and K respectively. For J = K = 1
at each clock cycle complementary data are applied to the master flip-flop. So this input
combination is not longer forbidden but results in toggeling the flip-flop. This can also be
seen in the truth table in Fig. 9.13. This is the same as dividing the frequency by two, as
shown in Fig. 9.12. Consequently, J Kmaster–slave flip-flops provide a particularly simple
means of constructing counters.
However, because of the feedback, operation of the J K flip-flop is subject to an impor-
tant limitation: the truth table in Fig. 9.13 only applies if the state at the J K inputs remains
unchanged as long as clock C = 1. This is because, unlike the RS master–slave flip-flop
in Fig. 9.10, the master–slave flip-flop here can only change state once but cannot change
back, as one of the two input nand gates is always disabled by the feedback. Failure to
observe this limitation is a frequent source of errors in digital circuits.
J K Q
0 0 Q−1 .(unchanged)
0 1 0
(Q = J )
1 0 1
1 1 Q−1 (inverted)
S S
JK
standard
JK
lockout
Fig. 9.14. Timing diagram of the input and output Fig. 9.15. Circuit symbol of a
signals of JK master–slave flip flops JK master–slave flip-flop
Special types of J K master–slave flip-flops are available which are not subject to this
limitation. They are provided with data lockout: the input state is read only during the
positive-going edge. Immediately after this edge, the two input gates are disabled and no
longer react to changes in the input states [10.1]. This is shown in Fig. 9.14. Whereas with
normal J K flip-flops the J and K inputs must not change as long as clock C = 1, with
a data lockout JK flip-flop they must remain constant only during the positive-going edge
of the clock signal. The common feature of both flip-flops is that the information read
in on the positive-going edge of the clock signal does not appear at the output until the
negative-going edge. Due to this delay, the circuit symbol in Fig. 9.15 additionally has a
delay sign at the outputs.
JK flip-flops frequently have several J and K inputs leading to an internal and gate.
The internal J and K variables are then only 1 when all of the respective J and K inputs
are 1.
In addition to the JK inputs, the JK flip-flops additionally possess Set and Reset inputs,
which operate independently of the clock—that is, asynchronously. This enables master
slave flip-flops to be set or cleared. The RS inputs have priority over the JK inputs. In order
to allow clock-controlled operation, there must be R = S = 0 or R = S = 1.
Typical IC types:
TTL ECL CMOS
Standard 7476 10135 4027
Data lockout 74 LS 111
Single-Edge-Triggered Flip-Flops
Flip-flops with intermediate storage can also be implemented by connecting two transpar-
ent D flip-flops (Fig. 9.8) in series and clocking them with complementary signals. This
produces the circuit shown in Fig. 9.16. As long as clock C = 0, the master follows the
input signal and we have Q1 = D. Meanwhile, the slave stores the old state. When the
clock goes to 1, the data D present at that instant is frozen in the master and transferred
to the slave, and thus to the Q output. The information present at the D input on the
positive-going edge of the clock signal is therefore instantaneously transmitted to the Q
output. The state of the D input has no effect for the rest of the time. This can also be seen
from Fig. 9.17. Instead of waiting for the negative-going edge, as in the JK flip-flop with
data lockout, the input value appears at the output immediately. For this reason, the circuit
symbol in Fig. 9.18 also has no delay symbols. This constitutes a significant advantage, in
that the entire clock cycle is now available for forming the new D signal. If JK flip-flops
664 9 Sequential Logic Systems
S S
Fig. 9.17. Timing diagram for the input and output Fig. 9.18. Circuit symbol for
signals in the single-edge-triggered D flip-flop the single-edge-triggered D
flip-flop
are used, this process must take place while the clock is zero; that is, with symmetrical
clock pulses in half the time.
Examples of IC types:
74 LS 74 (TTL); 10131 (ECL); 4013 (CMOS)
Single-edge-triggered D flip-flops can also be operated as toggle flip-flops. For this
purpose, we make D = Q, as shown in Fig. 9.19. The output state therefore inverts at
each positive-going edge of the clock signal. This is illustrated in Fig. 9.20. If transparent
D flip-flops were used, an oscillation would be obtained while clock C = 1, instead of a
frequency division. This is caused by the transparent propagation of the signal through the
circuit, resulting in a signal inversion after every propagation delay time.
It is also possible to make the inversion dependent on a control variable by providing
feedback from either Q or Q to the D input via a multiplexer. The latter is controlled by
the toggle input T shown in Fig. 9.21. The same mode of operation is possible using the
JK flip-flop shown in Fig. 9.22, with interconnected JK inputs.
Multipurpose flip-flops can be obtained by additionally providing synchronous data
input. The multiplexer can then be given another input preceding the D input. This addi-
tional input is selected via Load input L as shown in Fig. 9.23. If L = 1, then y = D and
therefore, after the next clock signal, Q = D. When L = 0, the circuit operates in exactly
C C1 Q
1D Q
the same way as that in Fig. 9.21. The mode of operation of this multifunction flip-flop is
summarized in Fig. 9.25.
The same behavior can also be obtained using a JK flip-flop, as shown in Fig. 9.24,
when L = 1, J = D, or K = D. Therefore, after the next clock signal, Q = D. When
L = 0, we have J = K = T ; the circuit then operates like in Fig. 9.22 as toggle flip-
flop. In the case of JK flip-flops, it must be remembered that the data have to be present
before the positive-going edge of the clock signal, but only appear after the output on the
negative-going edge. With normal JK flip-flops (as in Fig. 9.11), it must also be insured
that the J and K inputs do not change as long as C = 1. During this time, the L, T , and
D inputs must therefore also remain unchanged.
Due to their versatility, the multifunction flip-flops in Figs. 9.23 and 9.24 constitute
the basic building blocks of counters.
L T Q
0 0 Q−1
0 1 Q−1
1 0 D
1 1 D
Fig. 9.25. Circuit symbol for a Fig. 9.26. Function table for a multifunction
multifunction flip-flop flip-flop
666 9 Sequential Logic Systems
9.2
Straight Binary Counters
Counters are an important group of sequential logic systems. A counter may be any circuit
which, within certain limits, has a defined relationship between the number of input pulses
and the state of the output variables. As each output variable can have only two values, for
n outputs, there are 2n possible output combinations, although often only some of these are
used. It is not important which number is assigned to which combination, but it is useful
to choose a representation that can subsequently be easily processed. The simplest circuits
are obtained for straight binary notation.
Figure 9.28 shows the relationship between the number, Z, of input pulses and the
values of output variables zi , for a four-bit straight binary counter. If this table is read from
top to bottom, two change conditions can be recognized:
1 an output variable zi always changes state when the next lower value zi−1 changes
from 1 to 0;
2 an output variable zi always changes state when all lower variables
zi−1 . . . z0 have the value 1 and a new pulse arrives.
These conditions can also be seen in the timing diagram in Fig. 9.27. Pattern 1
is the basis of an asynchronous counter (ripple counter), whereas pattern 2 yields the
synchronous counter.
Occasionally, counters are required whose output state is reduced by 1 for each count
pulse. The operational principle of such a down-counter can also be inferred from the table
in Fig. 9.28 by reading it from the bottom up. It follows that:
1a an output variable zi of a down-counter changes state whenever the next lower variable
zi−1 changes from 0 to 1;
Z z3 z2 z1 z0
23 22 21 20
0 0 0 0 0
1 0 0 0 1
2 0 0 1 0
3 0 0 1 1
4 0 1 0 0
5 0 1 0 1
6 0 1 1 0
7 0 1 1 1
8 1 0 0 0
9 1 0 0 1
10 1 0 1 0
11 1 0 1 1
12 1 1 0 0
13 1 1 0 1
14 1 1 1 0
15 1 1 1 1
16 0 0 0 0
Fig. 9.27. Output states of a straight binary Fig. 9.28. State table of a straight
up-counter, as a function of time binary counter
9.2 Straight Binary Counters 667
2a an output variable zi of a down-counter always changes state when all lower variables
zi−1 . . . z0 have the value 0 and a new clock pulse arrives.
9.2.1
Asynchronous Straight Binary Counters
A straight binary asynchronous (ripple) counter can be implemented by arranging toggle
flip-flops in a chain, as shown in Fig. 9.29, and by connecting each clock input C to the
output Q of the previous flip-flops. If the circuit is to be an up-counter, the flip-flops must
change their output states when their clock inputs C change from 1 to 0. Edge-triggered flip-
flops are therefore required; for example, JK master–slave flip-flops where J = K = 1.
The counter may be extended to any size. Using this principle, one can count up to 1,023
with only ten flip-flops.
Flip-flops triggered by the positive-going edge of the clock pulse can also be employed;
for example, single-edge-triggered D flip-flops. If they are connected in the same way as
in Fig. 9.29, down-counter operation is obtained. For up-counter operation, their clock
pulse must be inverted. This is achieved by connecting each clock input to the Q-output
of the previous flip-flop.
Every counter is also a frequency divider. The frequency at the output of flip-flop F0
is half the counter frequency. A quarter of the input frequency appears at the output of F1 ,
an eighth at the output of F2 , and so on. This property of frequency division can be seen
clearly in Fig. 9.27.
IC-types:
Length TTL ECL CMOS
4 bit 74 LS 93 10178
7 bit 4024
8 bit 74 LS 393
24 bit 4521
30 bit 74 LS 292
9.2.2
Synchronous Straight Binary Counters
It is characteristic of an asynchronous counter that the clock pulse is applied only to the
input of the first flip-flop, while the remaining flip-flops are connected to the previous
outputs. This means that the input signal of the last flip-flop does not arrive until all of the
preceding stages have changed state. Each change of the output states z0 –zn is therefore
668 9 Sequential Logic Systems
delayed by the set-up time of the previous flip-flops. For long chains and high counter
frequencies, this may result in zn changing with a delay of one or more clock cycles. After
the last clock pulse, it is therefore necessary to wait for the delay time of the entire counter
chain before the result can be evaluated. If evaluation of the counter state is required
during counting, the period of the clock pulse must not be smaller than the delay time of
the counter chain.
Synchronous counters do not have these drawbacks, as the clock pulses are applied
simultaneously to all clock inputs C. In order that the flip-flops do not all change state at
every clock pulse, controlled toggle flip-flops – as shown in Figs. 9.21 and 9.22 – are used,
which only change state when the control variable T = 1. In accordance with Fig. 9.28, a
flip-flop of a straight binary counter may only change state when all the lower-order flip-
flops are 1. To bring this about, we make T0 = 1, T1 = z0 , T2 = z0 · z1 and T3 = z0 · z1 · z2 .
The and gates required for this purpose are shown in Fig. 9.30.
Integrated synchronous counters have yet more inputs and outputs, whose function and
application will be described in further detail with reference to Fig. 9.31. The entire counter
can be initialized using the Clear input CLR (Z = 0). It can be set to any number Z = D
via the Load input. Whereas the Clear input mostly operates asynchronously like any Reset
input, both synchronous and asynchronous types are available for the load process.
Large (multiple-bit) counters can be implemented by cascading several four-bit counter
stages. The stages are connected via the ripple carry output RCO and the enable input ENT,
which can be used to inhibit the entire counter stage and the carry output. The latter must
therefore go to 1 when a count of 1111 is reached and all the lower-order stages likewise
produce a carry. For this to occur, the logic operation
RCO = EN T · z0 · z1 · z2 · z3
must be performed in each counter stage. The corresponding output gate is shown in
Fig. 9.31.
To cascade the counter stages, it is merely necessary to connect the ENT input of a stage
to the RCO output of the next lower-order stage. However, as the delays are cumulative
due to the cascaded and operations, multiple-bit counters are subject to a reduction in
the maximum possible counting frequency. In this case, it is preferable to perform the
required and operations in parallel in each counter stage. To do this, the lowest-order
stage is omitted from the serial RCO–ENT operation, and the enabling of the higher-order
stages is controlled in parallel via the ENP inputs. In this way, the parallel and operation
can be implemented without external gates, as shown in Fig. 9.32.
Typical IC types:
Length Reset TTL ECL CMOS
4 bit asynchronous 74 LS 161 A 4161
4 bit synchronous 74 LS 163 A 10136 4163
8 bit synchronous 74 LS 590
9.2.3
Up–Down Counters
A distinction is drawn between up–down counters with one clock input and a second
input which determines the mode of counting, and those with two clock inputs, one for
incrementing the count, the other for decrementing it.
670 9 Sequential Logic Systems
1 up
Fig. 9.33. Binary counter with up–down control U =
0 down
CDN
Fig. 9.34. Straight binary counter with clock-up and clock-down inputs. F0 . . . F3 are toggle
flip-flops
CUP = Clock Up CDN = Clock Down
CO = Carry Output BO = Borrow Output
As the flip-flops that are to change state receive their clock pulses at virtually the same
time, the flip-flops for the more significant bits change state simultaneously with those for
the less significant ones. The circuit therefore operates as a synchronous counter. The and
gates at the output determine the carry for up-counter operation and that for down-counter
operation. It is possible to connect another identical counter that is in itself synchronous but
delayed with respect to the first; that is, operating asynchronously. This mode of operation
is termed “semisynchronous.”
IC type:
four-bit: 74 LS 193 (TTL)
Coincidence Cancellation
The interval between two count pulses and their duration must not be smaller than the set-
up time tsu of the counter, or the second pulse would be incorrectly processed. Counters
with only one clock input can therefore count at a maximum possible frequency off f max =
1/2tsu . For the counter shown in Fig. 9.34, the situation is more complicated. Even if the
counter frequencies at the up-clock and at the down-clock input are considerably lower than
f max , the interval between an up- and a down-clock pulse may, in asynchronous systems,
be smaller than tsu . Such close or even coinciding pulses result in a spurious counter state.
This can be avoided only by preventing these pulses from reaching the counter inputs. The
state of the counter then remains unchanged, as would also be the case after one up- and
one down-clock pulse.
Such a coincidence cancellation circuit can, for example, be designed as in Fig. 9.35,
where one-shots are used [9.2]. The one-shots (monostable multivibrators) M1 and M2
convert the counter pulses CUP and CDN into the signals xU P and xDN , each having a
defined length t1 . Their trailing edges are used to trigger the two one-shots M4 and M5 ,
which in turn generate the output pulses. Gate G1 decides whether the normalized input
pulses xU P and xDN overlap. If this is the case, a positive-going edge appears at its output,
672 9 Sequential Logic Systems
M4
M1 M3
M2
which triggers one-shot M3 . Both output gates G2 and G3 are then disabled for a time
t2 , and no pulses can appear at the output. In order that pulses are safely suppressed, the
relationship must hold:
t2 > t1 + t3
Time t3 defines the duration of the output pulses. The interval between them is shortest just
before coincidence is detected; that is, t = t1 − t3 . For correct operation of the counter,
the additional conditions
t3 > tsu und t1 − t3 > tsu
must therefore be fulfilled. The shortest permissible on-times of the one-shots are thus
t3 = tsu , t1 = 2tsu and t2 = 3tsu . The maximum counter frequency at the two inputs of the
coincidence detector is then
1 1
f max = =
t2 3tsu
The coincidence cancellation circuit therefore reduces f max by a factor of 1.5. The “anti-
race clock generator” in the 40110 counter (CMOS) operates on this principle.
Subtraction Method
A considerably more elegant method consists of counting the up and down pulses in
separate counters and then subtracting the results, as shown in Fig. 9.36. Counter pulses
that coincide then produce no unwanted effects. A further advantage is that the simpler
logic circuitry of an up-counter inherently permits a higher clock frequency.
The carry bit of the subtractor cannot be used to indicate the mathematical sign, as a
positive difference would be misinterpreted as being negative if one of the two counters has
SUB
Fig. 9.36. Straight binary up–down counter that is insensitive to coincident clock pulses
9.3 BCD Counters 673
an overflow and starts at zero. However, the result is obtained with the correct sign if the
difference is interpreted as a four-digit 2’s-complement number. Bit d3 therefore exhibits
the correct sign, provided that the difference does not exceed the permissible range of − 8
to +7.
9.3
BCD Counters
9.3.1
Asynchronous BCD Counters
The table in Fig. 9.28 shows that a three-bit counter can count up to seven, and a four-bit
counter up to 15. In a counter for straight BCD numbers, a four-bit straight binary counter
used as a decade counter is required for each decimal digit. This decade counter differs
from the normal straight binary counter in that it is reset to zero after every tenth count
pulse, and it produces a carry. This carry bit controls the decade counter for the next higher
decimal digit.
With BCD counters, a decimal display of the count is achieved much more easily than
for the straight binary counter, as each decade can be separately decoded and displayed as
a decimal digit.
In straight BCD code, each decimal digit is represented by a four-bit straight binary
number, the bit weightings of which are 23 , 22 , 21 and 20 . It is therefore also known as
the 8421 code. The state table of a decade counter employing the 8421 code is shown in
Fig. 9.38. By definition, it must be identical with that in Fig. 9.28 up to the number 9, but
the number 10 = 10dec. is again represented by 0000. The associated timing diagram for
the output variables is shown in Fig. 9.37.
Obviously, additional logic circuitry is required to reset the counter at every tenth input
pulse. However, gates may be saved by using JK flip-flops with several J and K inputs, as
shown in Fig. 9.39. In contrast to the normal straight binary counter shown in Fig. 9.29, the
circuit operates as follows: flip-flop F1 may not change state at the tenth counting pulse,
Z z3 z2 z1 z0
23 22 21 20
0 0 0 0 0
1 0 0 0 1
2 0 0 1 0
3 0 0 1 1
4 0 1 0 0
5 0 1 0 1
6 0 1 1 0
7 0 1 1 1
8 1 0 0 0
9 1 0 0 1
10 0 0 0 0
Fig. 9.37. Timing diagram of the Fig. 9.38. State table for 8421 code
output states of an 8421-code counter
674 9 Sequential Logic Systems
even though z0 changes from 1 to 0. From Fig. 9.29, we deduce a simple criterion for this
case: z1 must be kept at 0 if z3 is 1 prior to the clock signal. To achieve this, the J input of
F1 is connected to z3 . The condition that z2 must remain 0 at the tenth pulse is therefore
automatically satisfied.
The second difference with respect to a straight binary counter is that z3 changes from
1 to 0 at the tenth pulse. However, if the clock input of F3 were connected to z2 as in a
normal straight binary counter, z3 would be unable to change after the eighth counting
pulse, since flip-flop F1 is disabled by the feedback signal. The clock input of F1 , must
therefore be connected to the output of the flip-flop that is not disabled by the feedback
signal, in this case z0 .
On the other hand, the J inputs must be controlled so that they prevent flip-flop F3 from
changing state with every clock of z0 . Figure 9.38 indicates that z3 must not go to 1 unless
both z1 and z2 are 1 prior to the clock signal. This may be achieved by connecting the two
J inputs of F3 to z1 and z2 respectively. Then, at the eighth counting pulse, z3 = 1. Since
z1 and z2 become zero simultaneously, z3 resumes the state z3 = 0 as soon as possible;
that is, at the tenth counting pulse, when z0 has its next transition from 1 to 0. Figure 9.38
indicates that this is precisely the right instant.
IC types:
4 bit 74 LS 90 (TTL) 10138 (ECL)
2 × 4 bit 74 LS 390 (TTL)
9.3.2
Synchronous BCD Counters
The synchronous decade counter in Fig. 9.40 has largely similar circuitry to the syn-
chronous straight binary counter in Fig. 9.31. As with the asynchronous decade counter,
two additional features are again required to insure that, at the transition from 9dec = 10012
to 0dec = 00002 , flip-flop F3 changes state and not flip-flop F1 . The disabling of F1
is achieved in Fig. 9.40 via the feedback path of Q3 and the change of state of F3 by
additionally decoding the 9 at the toggle control input.
9.4
Presettable Counters
Presettable counters are circuits that produce an output signal when the number of input
pulses equals a predetermined number M. The output signal can be used to trigger any
desired process and is employed to stop the counter or reset it to its initial state. If the
counter is allowed to continue counting after reset, it operates as a modulo-M counter, the
counting cycle of which is determined by the preselected number M.
The most obvious method of implementing a presettable counter consists of comparing
the count Z with the preselected number M, as in Fig. 9.41. For this purpose, we can use an
identity comparator, as described in Sect. 8.5 on page 648. If Z = M after M clock pulses,
y becomes 1 and the counter is cleared (Z = 0). The equality signal y is present for the
duration of the clearing process. With an asynchronous CLR input, this time only amounts
to a few gate propagation delays. For this reason a synchronous clear input is preferable;
then the equality signal is present for precisely one clock period. The counter in Fig. 9.41
therefore returns to zero after M + 1 clock pulses. It thus represents a modulo-(M + 1)
counter.
The comparator in Fig. 9.41 can be dispensed with by using the LOAD inputs generally
provided in synchronous counters (Fig. 9.31). The circuits in Figs. 9.42 and 9.43 make use
of this possibility. The counter in Fig. 9.42 is loaded with the number P = Z max − M.
After M clock pulses, the maximum count Z max is therefore reached, which is internally
decoded and produces a carry RCO = 1. If this output is connected to the LOAD input
as shown in Fig. 9.42, the preset number P is reloaded with clock pulse M + 1. Once
Fig. 9.42. Modulo-(M + 1) counter with a Fig. 9.43. Modulo-(M + 1)-counter with a
parallel input of P = Z max − M for parallel input of M for Z = 0, using a
Z = 15 down-counter
again, we have a modulo-(M + 1) counter. For straight binary counters, the number P is
particularly easy to determine: it is equal to the 1’s-complement of M (see Sect. 8.1.3).
The counter in Fig. 9.43 is loaded with the preset number M itself. It then counts down
to zero. At zero, a carry RCO is generated (see Fig. 9.33), which can be used to reload the
counter.
9.5
Shift Registers
Shift registers are chains of flip-flops that allow data applied to the input to be advanced by
one flip-flop with each clock pulse. After passing through the chain, the data are available
at the output with a delay, but are otherwise unchanged.
9.5.1
Basic Circuit
The shift-register principle is illustrated in Fig. 9.44. On the first clock pulse, the infor-
mation, D, present at the input is read into flip-flop F1 . On the second clock pulse, it is
passed on to flip-flop F2 ; simultaneously, new information is read into flip-flop F1 . As an
example, Fig. 9.45 illustrates the mode of operation for a four-bit shift register. We can
see that the shift register is filled serially with input data after four clock pulses. These are
then available in parallel at the four flip-flop outputs Q1 . . . Q4 , or they can be extracted
serially once more at output Q4 on subsequent clock pulses. All flip-flops with interme-
diate storage can be used. Transparent flip-flops are unsuitable, because the information
applied to the input would immediately pass through all flip-flops to the last flip-flop when
the clock goes to 1 for the first time.
CLK Q1 Q2 Q3 Q4
1 D1 − − −
2 D2 D1 − −
3 D3 D2 D1 −
4 D4 D3 D2 D1
5 D5 D4 D3 D2
6 D6 D5 D4 D3
7 D7 D6 D5 D4
9.5.2
Shift Registers with Parallel Inputs
If, as in Fig. 9.46, a multiplexer is connected in front of each D input, it is possible to
switch over to parallel data input via the Load input. On the next clock pulse, data d1 , to
d4 , are loaded in parallel and appear at Q1 . . . Q4 . This allows not only serial-to-parallel
conversion but also parallel-to-serial conversion.
A shift register with parallel load inputs can also be operated as a bidirectional shift
register. For this purpose, the parallel load inputs are connected to the output of the next
flip-flop on the right. If LOAD = 1, the data are shifted from right to left.
Typical IC types:
Length TTL ECL CMOS
4 bit 74 LS 194 A 10141 40194
8 bit 74 LS 164, 299 4014
16 bit 74 LS 673 4006
9.6
Processing of Asynchronous Signals
Sequential logic circuits can be realized either in asynchronous or synchronous – that is,
clocked – mode. Asynchronous operation normally requires simpler circuits but creates a
number of problems: it must be insured that the spurious transitions (hazards) which may
gSm
S
U
appear temporarily because of the difference in propagation delay times are not decoded
as valid states. In synchronous system, the conditions are far more simple. Any transition
within the system can only take place at the triggering edge of a clock pulse. The clock
pulse therefore indicates when the system is in the steady-state condition. It is advisable to
construct the system so that all changes consistently occur on one edge of the clock pulse.
If, for instance, all the circuits are triggered by the rising edge, the system is certain to be
in the steady-state condition at the next rising edge of the clock if the clock frequency is
not too high.
As a rule, external data fed to the system are not synchronized with its clock. In order
that they may be processed synchronously, they must be synchronized by special circuits,
some examples of which are described below.
9.6.1
Debouncing of Mechanical Contacts
If a mechanical switch is opened or closed, vibrations usually generate a pulse train. A
counter then detects an undefined number of pulses instead of the single pulse intended. One
way of avoiding this is to use mercury-wetted contacts, although this is rather expensive.
A simple method of electronic debouncing by means of an RS flip-flop is illustrated in
Fig. 9.47. When the switch U is in its lower position (break contact) R = 0 and S = 1;
that is, x = 0. When the switch is operated, a pulse train initially occurs at the S input,
because the break contact is opened. Since R = S = 1, this being the storing condition, the
output remains unchanged. After the complete opening of the break contact, a pulse train
is generated by the opposite make contact. With the very first pulse, R = 1 and S = 0, the
flip-flop is set to x = 1. This state is stored during the bouncing that follows. The flip-flop
changes to its off state only when the lower break contact is touched again. The timing
diagram in Fig. 9.48 illustrates this behavior.
9.6.2
Edge-Triggered RS Flip-Flops
A flip-flop with RS inputs is set as long as S = 1 and reset as long as R = 1. Both inputs
must not be 1 simultaneously. To achieve this, we can generate short R or S pulses. A
simpler possibility is shown in Fig. 9.49. Here, the input signals are fed to the inputs of
positive-edge-triggered D flip-flops. This insures that only the instant of the positive-going
9.6 Processing of Asynchronous Signals 679
edge is important, and the rest of the clock pulse is immaterial. When a positive-going Set
edge occurs, Q1 = Q2 . This results in the exclusive-or operation
y = Q1 ⊕ Q 2 = 1
9.6.3
Pulse Synchronization
The simplest method of synchronizing pulses employs D-type flip-flops. As shown in
Fig.9.50, the external unsynchronized signal x is applied to the D-input, and the system
clock Φ to the C-input. In this manner, the state of the input variable x is monitored and
transferred to the output on the positive-going edge of the clock pulse. As the input signal
can also change during the positive-going edge of the clock pulse, metastable states may
occur in flip-flop F1 . Additional flip-flop F2 has therefore been provided to prevent errors
occurring in output signal y. Flip-flops that operate according this principle are called
“metastable resistant” as the type 74 ACT 11478.
Figure 9.51 shows a typical timing diagram. Any pulse too short to be registered by
the leading edge of a clock pulse is ignored. This case is also shown in Fig. 9.51. If such
short pulses are not to be lost, they must be read into an intermediate store before being
transferred to the D flip-flop. The D flip-flop F1 in Fig. 9.52 serves this purpose. It is set
asynchronously via the S-input when x becomes 1. With the next positive-going clock
edge, y = 1. If, at this moment, x has already returned to zero, flip-flop F1 is reset by the
same edge. A short pulse x is thus prolonged until the next clock edge occurs and cannot
therefore be lost. This property may also be seen in the example in Fig. 9.53.
9.6.4
Synchronous One-Shot
It is possible, using the circuit in Fig. 9.54, to generate a pulse that is synchronized with
the clock. The pulse length equals one clock period and is independent of the length of the
trigger signal x.
If x changes from 0 to 1, Q1 = 1 at the positive-going edge of the next clock pulse;
that is, y = 1. On the subsequent leading edge, Q2 becomes 0 and y becomes 0 again. This
state remains unchanged until x has been zero for at least one clock period. Short trigger
pulses that are not registered by the leading edge of a clock pulse are lost, as with the
synchronizing circuit in Fig. 9.50. If they too are to be considered, an additional flip-flop,
as shown in Fig. 9.52, must store the pulses until they are transferred to the main flip-flop.
The timing diagram in Fig. 9.55 shows an example of operation.
A synchronous one-shot for on-times longer than one clock period can be realized quite
simply by using a synchronous counter, as is shown in Fig. 9.56. If the trigger variable x
is at 1, the counter is loaded in the parallel-in mode on the next clock pulse. The following
clock pulses are used to count to the maximum output state Z max . At this number, the
carry output RCO = 1. The counter is then inhibited via count-enable input ENP; the
output variable y is 0. The ordinary enable input ENT cannot be employed for this purpose,
as it not only affects the flip-flops but also the RCO directly, and this would result in an
unwanted oscillation.
A new cycle is started by parallel read-in. Immediately after loading, RCO becomes
zero and y is then 1. The feedback from RCO to the and gate at the x-input prevents a new
loading process unless the counter has reached the state Z max . By this time, x should have
returned to 0; if not, the counter is loaded again – that is, it is operating as a modulo-(M +1)
counter, as shown in Fig. 9.42.
The timing diagram is shown in Fig. 9.57 for an on-time of seven clock pulses. If a
four-bit straight binary counter is employed, it must, for this particular on-time, be loaded
with P = 8. The first clock pulse is needed for the loading process and the remaining six
pulses for counting up to 15.
9.6.5
Synchronous Edge Detector
A synchronous edge detector gives an output signal that is synchronized with the clock
pulse whenever the input variable x has changed. For the implementation of such an
arrangement, we consider the one-shot circuit shown in Fig. 9.54. It produces an output
pulse whenever x changes from 0 to 1. In order that a pulse is also obtained at the transition
from 1 to 0, the and gate must be replaced by an exclusive-or gate, producing the circuit
shown in Fig. 9.58. Its characteristics are illustrated by the timing diagram in Fig. 9.59.
9.6.6
Synchronous Clock Switch
The problem often arises of how to switch the clock on and off without interrupting the
clock pulse generator. In principle, an and gate could be used for this purpose, but this
would result in the first and the last pulse being of undefined length if the switching signal
is not clock-synchronized. This effect can be avoided by employing a D-type flip-flop for
the synchronization, as is shown in Fig. 9.60. If EN = 1, at the next leading pulse edge
Q = 1 and therefore Φ
= 1. The first pulse of the switched clock Φ
always has the full
length because of the edge-triggering property.
The leading pulse edge cannot be used to switch the control flip-flop off since Q would
go to zero a moment after the transition, which would result in a very short output pulse.
The flip-flop is therefore cleared asynchronously via the reset input when EN and Φ are
0, this being achieved by the nor gate at the R-input. As is obvious from Fig. 9.61, only
full-length clock pulses can reach the output of the and gate.
9.7
Systematic Design of Sequential Circuits
9.7.1
State Diagram
To enable the systematic design of sequential circuits, it is necessary to obtain a clear
description of the problem. The starting point is the block diagram in Fig. 9.62.
In contrast to a combinatorial logic system, the output variables yi depend not only
on the input variables xi but also on the previous history of the system. All of the system
logic variables affecting the transition to the next state, apart from the input variables, are
called state variables zn . To insure that they can become effective on the arrival of the next
clock signal, they are stored in the state variable memory for one clock pulse.
The number of input variables xi is called the input vector:
X = {x1 , x2 . . . xl }
The number of output variables yj is called the output vector:
Y = {y1 , y2 . . . ym }
The number of state variables zn is called the state vector:
Z = {z1 , z2 . . . zn }
We shall denote the various states through which the sequential logic system passes by Sz .
To simplify the notation, the state vector is preferably read as a straight binary number and
the corresponding decimal number is simply written as a subscript.
The new state S(tk+1 ) is determined both by the old state S(tk ) and by the input variables
(qualifiers) xi . The sequence in which the states occur can therefore be influenced using the
qualifiers X. The appropriate assignment is made by a combinatorial logic system: if the old
9.7 Systematic Design of Sequential Circuits 683
combinatorial
logic system
S0 S1
memory
for state
variables
S3 S2
Fig. 9.62. Basic arrangement of a sequential logic system Fig. 9.63. Example of a state diagram
called Mealy machine. If Y depends of Z only its a State 0: Initial state
Moore machine. State 1: Branching state
State 2: Wait state
Input vector: X State 3: Temporary state
Output vector: Y
State vector: Z
Clock: Φ
state vector Z(tk ) is applied to its inputs, the new state vector Z(tk+1 ) appears at its output.
The corresponding system state must obtain until the next clock pulse. Consequently, the
state vector Z(tk+1 ) must not be transmitted to the outputs of the flip-flops until the next
clock pulse. For this reason, edge-triggered flip-flops must be used.
There are a few important special types of sequential logic circuit. For example, a
special case arises when the state variables can be used directly as outputs. A second
simplification occurs when the sequence of states is always the same, in which case no
input variables are required. We have made use of these simplifications for the counters.
A general description of the state sequence is provided by a state diagram, as shown in
Fig. 9.63. Each state SZ of the system is illustrated by a circle. The transition from one state
to another is shown by an arrow. The symbol on the arrow indicates under which condition
a transition is to occur. For the example in Fig. 9.63, state S(tk ) = S1 , is followed by state
S(tk+1 ) = S2 if x1 = 1. For x1 = 0, however, S(tk+1 ) = S0 . An unmarked arrow denotes
an unconditional transition.
For a synchronous sequential circuit, there is an additional condition that a transition
will only occur at the next clock pulse edge, and not at the precise moment at which the
transition condition is fulfilled. As this restriction applies to all transitions in the system, it
is usually not entered in the state diagram, but indicated in the description. We deal below
only with synchronous sequential circuits, as their design is less problematic.
If the system is in state SZ and no transition condition is fulfilled that might lead out of
this state, the system remains in state SZ . This obvious fact can sometimes be emphasized
by entering an arrow that starts and ends at SZ (the wait state). Such a case is illustrated at
state S2 in Fig. 9.63.
When the power supply is switched on, a sequential circuit must be set to a defined
initial state. This is the “pon” condition (“power on”). Its signal is produced by a special
logic circuit; it is 1 for a short time after switch-on of the supply and is otherwise 0. This
signal is generally used to clear the state variable memory by applying it to the RESET
inputs of the flip-flops.
684 9 Sequential Logic Systems
S0
S0 S1 S1
no
yes
S3 S2 S2
no
yes
S3
9.7.2
Example for a Programmable Counter
We shall demonstrate the design process for a counter, the counting cycle of which is either
0, 1, 2, 3 or 0, 1, 2, depending on whether the control variable x is 1 or 0. The appropriate
state diagram is given in Fig. 9.65. As the system can assume four stable states, we require
Z(tk ) Z(tk+1 )
5 63 4 5 63 4
x z1 z0 z1
z0
y
0 0 0 0 1 0
0 0 1 1 0 0
0 1 0 0 0 1
0 1 1 0 0 0
1 0 0 0 1 0
1 0 1 1 0 0
1 1 0 1 1 0
1 1 1 0 0 1
ROM address ROM contents
Fig. 9.65. State diagram of a counter Fig. 9.66. Truth table for the state diagram
with a programmable counting cycle shown in 9.65
3 for x = 0
Counting cycle =
4 for x = 1
9.7 Systematic Design of Sequential Circuits 685
combinatorial
logic system Fig. 9.67. Sequential circuit for
implementing the programmable
counter
two flip-flops for storage of the state vector Z, which consists of two variables, z0 and z1 .
Since these variables immediately indicate the state of the counter, they are simultaneously
used as output variables. In addition, a carry y should be produced when the counter state
is 3 for x = 1, or 2 for x = 0.
We thus obtain the circuit shown in Fig. 9.67, with the truth table given in Fig. 9.66.
The left-hand side of the table shows all the possible bit combinations of the input and
state variables. The state diagram in Fig. 9.65 shows which is the next system state for
each combination. This is shown on the right-hand side of the table. The respective values
of the carry bit y are also added.
If a ROM (read-only memory) is used to realize the combinatorial system, the truth
table of Fig. 9.66 can be directly employed to program the memory, the state and input
variables being used as address variables. The new value Z
of the state vector Z and the
output variable y are stored at the appropriate addresses. Hence, to implement our example,
we require a ROM for eight words of three bits.
Fig. 9.68. Programmable counter using a combinatorial system consisting of gates. This circuit
would also result as a configuration of a PLD
686 9 Sequential Logic Systems
The truth table shown in Fig. 9.66 supplies the Boolean functions:
z1
= z0 z1 + xz0 z1 ,
z0
= z0 z1 + xz0 ,
y = xz0 z1 + xz0 z1
Figure 9.68 shows the realization of this combinatorial system by means of gates. However
one should not use discrete gate because of the multitude of chips required. A PLD (pro-
grammable logic device) would be preferable because with it a single chip realization is
possible that contains the registers for the state variables also. The application of a ROM or
PLD also has another decisive advantage, namely flexibility: the ROM or PLD need only
be reprogrammed to provide a circuit with different properties, without further changes.
The use of gates for a sequential circuit is thus recommended only in certain simple
cases; for instance, in the standard counters described in the previous sections.
9.8
Dependency notation
The new standard for digital circuit symbols (IEC 60617-12 and IEEE Std 91-1984) does
not merely involve replacing the earlier round symbols by square ones. It also consti-
tutes a significant advance by introducing the so-called dependency notation which allows
complex circuits to be represented in a readily comprehensible manner.
The underlying concept is the use of precisely defined labeling rules extending beyond
the gate symbol itself to indicate how specific variables affect other variables. Controlling
terminals are differentiated from controlled terminals. A controlled terminal can in turn
also act as a controlling terminal for other terminals.
Various types of dependency have been standardized. These are denoted by specific
letters as shown in Fig. 9.69. The relevant letter is written inside the circuit symbol next to
the controlling terminal. The letter is followed by an identification number which is also
entered at all the terminals affected by the relevant operation.
Symbol Meaning
G and
V or
N Exclusive-or (controllable negation)
Z Unaltered transmission
C Clock, Time
S Set
R Reset
EN Enable
M Mode
L Load
T Toggle
A Addres
CT Content (e.g. of a counter)
By way of example, Fig. 9.70 shows the extension of a driver gate to form and and
gate using the dependency notation. Similarly, Figs. 9.71 and 9.72 show the extension to
or or exor gates.
A terminal can be controlled simultaneously by several other terminals. In this case
the various identification numbers are separated by commas, as in Fig. 9.73. The relevant
operations must be carried out successively from left to right.
As an example, Fig. 9.74 shows how a control terminal acts on several other terminals.
A bar over the identification number indicates that the variable in question must be linked
with the negated control variable.
As shown in Fig. 9.75, several terminals can be combined to form one control variable.
In this case the identification number is a straight binary number resulting from the weight-
ing written inside the brace. The number range in question is entered after the function
0
symbol. The notation 3 means 0 to 3. In the example, input x0 is only effective if control
inputs a0 and a1 represent the straight binary number 0.
From the examples given so far, it is clear that controlled inputs are designated only
by identification numbers. However, there are cases in which a mnemonic designation of
a terminal is desirable for other reasons, e.g. D for data. In such cases the identification
numbers are placed before the designation letter.
Figure 9.76 gives an example of the use of various modes (M) and the effect and
control action of a content (CT ). The example shown is of an up-down counter with
parallel loading inputs. Depending on mode, the clock CLK has various effects.
The notation 2,4+ at the clock input means that the count is incremented (+) when
mode 2 is present (LOAD = 0, U P = 1) and EN ABLE = 1. Similarly, in mode 0 the
count is decremented. The condition for this is 0,4-. The various modes of a terminal are
simply written alongside one another, separated by obliques.
In the third mode, the clock initiates parallel data transfer at the D inputs (loading).
The notation 1,5 D means that the parallel loading process in mode 1 is taking place in
y = a0 a1 x0
+ a0 a1 x1
+ a0 a1 x2
+ a0 a1 x3
Fig. 9.73. Multiple control Fig. 9.74. Control of Fig. 9.75. Control block
of an input several inputs, using two with several control
2-to-1 multiplexers as an variables, using a 4-to-1
example multiplexer as an example
688 9 Sequential Logic Systems
synchrony with the clock. Similarly the notation 1D would mean a clock-independent (i.e.
asynchronous) data transfer.
The carry output CO is controlled by the counter content. It is “1" if the content is 15
when counting up (2,4 CT = 15) or if the content “0" when counting down (0,4 CT = 0).
Chapter 10:
Semiconductor Memories
Semiconductor memories fall into two main categories, as shown in Fig. 10.1: table mem-
ories and function memories. With table memories, an address A is defined in the range
0 ≤ A ≤ n = 2N − 1
The word width of the address is between N = 5 and N = 22, depending on the size
of memory. Data can be stored at each of the 2N addresses. The data word width is
m = 1−16 bits. Figure 10.2 shows an example for N = 3 address bits and m = 2 data bits.
The memory capacity K = m · 2N is specified in bits, and also in bytes (K/8) for data
word widths of 8 or 16 bits. When using several memory chips, both the address space and
the word width can be increased by any amount. This will allow any tables such as truth
tables, computer programs, or results of measurements (numbers) to be stored.
Function memories store logic functions instead of tables. Each variable of a truth table
can be expressed as a logic function. Written in standard product terms, the logic function
of variable d0 in Fig. 10.2 becomes
d0 = a 2 a 1 a 0 d00 + a 2 a 1 a0 d10 + · · · + a2 a1 a0 d70
If d0 contains no regularity, and the zeros and ones are therefore statistically distributed,
we obtain n/2 – in this case, four – nonvanishing product terms. This situation occurs, for
instance, when programs are being stored. In this case, the implementation of the logic
function is more complex than its storage in a table.
Semiconductor memories
A D
a2 a1 a0 d1 d0
0 0 0 0 d01 d00 D0
1 0 0 1 d11 d10 D1
2 0 1 0 d21 d20 D2
3 0 1 1 d31 d30 D3
4 1 0 0 d41 d40 D4
5 1 0 1 d51 d50 D5
6 1 1 0 d61 d60 D6 Fig. 10.2. Table of a memory
7 1 1 1 d71 d70 D7 Address word width N = 3 bit
Data word width m = 2 bit
However, if a truth table is used as the starting point, extensive simplification is often
possible for the logic function due to the underlying regularity. One such case is when
there are only very few ”ones.” For example, if in the function d0 only d70 = 1, we only
require a single conjunction d0 = a2 a1 a0 . Another case is when the logic functions can
be simplified using Boolean algebra. For example, if d0 = a1 in Fig. 10.2, an extremely
simple function is obtained, even though it contains four “ones.” In such cases, the use of
function memories generally produces much better solutions than storage in a table.
Table memories are subdivided into two distinct categories, namely RAMs and ROMS.
RAM is a general designation for read–write memories. The contents of the memory can
be both read and written during normal operation. The abbreviation actually stands for
Random Access Memory. “Random access” means that any data word within the memory
can be accessed at any time, in contrast to shift register memories, in which data can only
be read from the memory in the same sequence in which they were written into it. As shift
register memories are no longer very important, the term RAM has become a generic term
for memories with read–write capabilities. This is somewhat misleading, in that ROMs
also allow random access to any data word.
ROM is an abbreviation for Read-Only Memory. This designation identifies memory
ICs that retain their contents when there is no supply voltage, even without backup batteries.
In normal operation, data are only read from such memories but not written into them.
Special equipment is normally required for writing the data, the storage procedure being
referred to as programming. The sub-categories listed in Fig. 10.1 differ in the type of
programming employed, which is described in greater detail below.
10.1
Random Access Memories (RAMs)
10.1.1
Static RAMs
A RAM is a memory device in which data can be stored at a specified address and subse-
quently read out from that address (random access). For technological reasons, the indi-
vidual memory cells are not arranged linearly but in a square matrix. To select a particular
memory cell, address A is decoded, as shown in Fig. 10.3, by a row and column decoder.
10.1 Random Access Memories (RAMs) 691
row decoder
column decoder
Fig. 10.3. Internal structure of a RAM: an example showing 16-bit memory capacity
Din = Data input Dout = Data output
CS = Chip Select R/W = Read/Write
we = write enable
In addition to its address inputs, a RAM has an extra data input Din , a data output Dout ,
a read/write pin R/W , and a chip select (CS) or chip enable (CE) pin. The latter is used
for multiplexing more than one memory operated via a common data line (a bus system).
When CS = 0, the data output Dout assumes high impedance and thus has no effect on
the data line. To allow this, the data output is always implemented as a tristate gate.
During the write process (R/W = 0), the output gate is likewise switched to high
impedance by an additional logic operation. This allows Din to be connected to Dout ,
enabling data to be transmitted in both directions via the same line (a bidirectional bus
system).
Another logic gate prevents a switchover to the write state (we = 1), if the chip is
not selected CS = 0. This prevents data from being written accidentally until the relevant
memory has been selected.
Figure 10.3 shows the logic operations mentioned above. Lines din , dout , and we
(write enable) are connected to each memory cell internally, as illustrated schematically
in Fig. 10.4. Data should only be written into the memory cell when address condition
xi = yi = 1 is satisfied and also we = 1. This logic operation is performed by gate
G1 . The contents of the memory cell must only reach the output if the address condition
is satisfied. This operation is performed by gate G2 , which has an open-collector output.
When the cell is not addressed, the output transistor is off. The outputs of all the cells are
692 10 Semiconductor Memories
–1
CS
internally wire-anded together and connected to the memory output Dout via the tristate
gate shown in Fig. 10.3.
Unless the supply voltage is switched off, the memory contents are retained until they
are modified by a write command. Such memories are referred to as “static,” to distinguish
them from “dynamic” memories, in which the contents have to be refreshed at regular
intervals to prevent them from being lost.
The circuit symbol for a RAM is shown in Fig. 10.5. As we can see, there are N address
inputs. These are decoded by the address decoder in such a way that the precise memory
cell is selected (out of 2N ) that corresponds to the address applied. The read–write signal
R/W is only activated when chip select CS = 1. The tristate output is therefore activated
for R/W = 1 and CS = 1; for R/W = 0 it is high impedance. For this reason, the
data input and output can be internally interconnected in the memory IC. This produces a
bidirectional data port whose direction of operation is determined by the R/W signal.
Frequently, not just a single bit but an m-digit word is stored at an address as shown in
Fig. 10.2. The storage of entire words may be seen as an extension of the block diagram in
Fig. 10.3 in the third dimension. The additional bits are then stacked in additional memory
layers above the base layer shown; their control lines x, y, and we are connected in parallel,
and their data lines form the input or output word.
Timing Considerations
For satisfactory operation of the memory, a number of timing conditions must be observed.
Figure 10.6 shows the sequence of a write operation. To prevent the data from being written
into the wrong cell, the write command must not be applied until a certain time has elapsed
10.1 Random Access Memories (RAMs) 693
Address valid
Data valid
Address valid
Data valid
after definition of the address. This time is called the address setup time tAS . The duration
of the write pulse must not be less than the minimum value tWP , (the write pulse width).
The data are read in at the end of the write pulse. They must be valid – that is, stable –
for a minimum period of time prior to this. This time is called tDW (Data Valid to End of
Write). In a number of memories, the data and addresses must also be present for a further
time tH after the end of the write pulse (the Hold Time). As can be seen from Fig. 10.6,
the time required to execute a write operation is expressed as
tW = tAS + tWP + tH
This is referred to as the Write Cycle Time.
The read operation is shown in Fig. 10.7. After the address is applied, it is necessary
to wait for time tAA until the data at the output are valid. This time is referred to as the
Address Access Time, or simply the Access Time. A list of some static RAMs is given in
Fig. 10.8.
10.1.2
Dynamic RAMs
As we wish to maximize the number of cells in a memory, every effort must be made to
implement them as simply as possible. They normally consist of just a few transistors;
in the case of static CMOS RAMs, a six-transistor cell is normally used. In the simplest
case, even the flip-flop is omitted and replaced by a single MOSFET and a capacitor
694 10 Semiconductor Memories
row decoder
row
memory matrix
a0 ... a9
to store 1 bit as a charge. This makes a single-transistor cell possible. However, as the
charge is only retained for a short time, the capacitor must be recharged at regular intervals
(approximately every 2–70 ms). This operation is known as refresh, and the memories are
called dynamic RAMs, DRAMs.
This disadvantage is offset by several advantages. Dynamic memories can provide
about four times more storage capacity on the same printed circuit board area, with the
same current drain and at the same cost.
To save on pins, with dynamic memories the address is entered in two stages and
buffered in the IC. The block diagram of a 1 Mbit RAM is shown in Fig. 10.9. In the first
step, address bits a0 –a9 are stored in the row-address latch with the RAS signal. In the
second step, address bits a10 –a19 are loaded into the column-address latch with the CAS
signal. This makes it possible to accommodate a 1 Mbit memory in an 18-pin package.
Figure 10.10 gives some examples for DRAM chips, in Fig. 10.11 some modules are
listed.
696 10 Semiconductor Memories
a0 ... a9
row
address latch
multiplexer
refresh counter
arbiter
control block
timing and
refresh
time base
precharge time
AS
address access time
RAS
CAS
counter is incremented by one every 8 ms, then after 1024 · 8 ms ≈ 8 ms all of the row
addresses will have been applied once, as required. With cycle stealing the processor is
stopped every 8 ms for one cycle and a refresh step is performed. For this purpose, the
refresh time base shown in the block diagram in Fig. 10.12 divides the frequency of
the clock signal CLK, so that the timing and control block receives a refresh command
every 8 ms.
When a refresh cycle is initiated, the status of the refresh counter is transferred via the
multiplexer to the memory and the RAS signal is temporarily set to 1. The counter is then
incremented by 1. During the refresh cycle, the memory user is inhibited by a wait signal.
This means that the ongoing process is stopped every 8 ms for 100 ns; that is, slowed down
likewise, by 2%.
3) Transparent, or hidden, refresh. With this method, a refresh step is also performed
every 8 ms, but the refresh controller is synchronized to the CPU in such a way that, instead
of inhibiting the memory user, the refresh is performed at the precise instant at which the
CPU cannot access the memory. This means that no time is lost. If any overlapping of an
external access with the refresh cycle cannot be totally eliminated, an additional priority
decoder (arbiter) can be employed, as shown in Fig. 10.12. It acknowledges an external
request with a wait signal until the current refresh cycle is complete and then executes the
request.
10.2
RAM Expansions
10.2.1
Two-Port Memories
Two-port memories are special RAMs that allow two independent processes to access
common data. This enables data to be exchanged between the two processes. To be able
to do this, the two-port memory must have two separate sets of address, data, and control
lines, as shown in Fig. 10.14. Implementation of this principle is subject to limitations,
since it is basically impossible to write into the same memory cell simultaneously from
both ports.
The “Read-While-Write” memories overcome this problem by only reading from one
of the two ports and only writing at the other. Figure 10.15 shows that memories of this
type have two separate address decoders, which allow simultaneous writing to one address
while reading from another.
698 10 Semiconductor Memories
two-port memory
arbiter
If reading and writing are to take place at both ports of a two-port memory, an access
conflict can generally only be avoided by preventing simultaneous memory access. To do
this, the address, data, and control lines can be made available via multiplexers to the port
accessed, as shown in Fig. 10.16. In many cases, the two processes accessing the memory
are synchronized to prevent simultaneous access. If this is not possible, a priority decoder
(arbiter) can be used which, in the event of access overlap, temporarily stops one of the two
processes by a wait signal. Some integrated two-port memories are listed in Fig. 10.17.
However, their capacities are limited. In order to implement large two-port memories, it is
advisable to use normal RAMs in conjunction with a dual-port RAM controller.
10.2.2
RAMs as Shift Registers
RAMs can be operated as shift registers if the addresses are applied cyclically. The counter
shown in Fig. 10.18 is used for this purpose. For each address, the stored data are first read
out and new data are then read in. The timing diagram is shown in Fig. 10.19. The positive-
going edge of the signal increments the counter. If the CLK signal is simultaneously used
as the R/W signal, the memory contents are then read out and stored in the output flip-
flop on the negative-going edge. While CLK = 0, the new data Din are written into the
memory cell that has just been read out. In this case, the minimum clock cycle is shorter
10.2 RAM Expansions 699
1 If access time is specified the RAM is asynchronous, if a frequency is specified the RAM is
synchronous.
Fig. 10.17. Examples of two-port memories
than the sum of the read and write cycle times, since the address remains constant. It is
equal to what is known as the “Read-Modify-Write Cycle Time.”
The difference between this type of shift register and the normal type (see Sect.9.5)
is that only the address, which acts as a pointer to the fixed data, is shifted, not the data
themselves. The advantage of this method is that normal RAMs can be employed, and these
are obtainable with memory capacities far greater than those of normal shift registers. Even
dynamic RAMs can be used without a refresh controller if all raw addresses are scanned
within the refresh time.
At high shift frequencies, slow low-cost RAMs can be used if several data bits are
processed in parallel, and if a serial-to-parallel converter is provided at the input and a
parallel-to-serial converter at the output, in order to obtain the required shift frequency.
Counter
10.2.3
First-In-First-Out Memories (FIFO)
A FIFO is a special type of shift register. The common feature is that the data appear at
the output in the same order as they were read in: the first word read in is also the first one
read out. With a FIFO, as opposed to a shift register, this process can take place completely
asynchronously; that is, the read-out clock is independent of the read-in clock. FIFOs are
therefore used for linking asynchronous systems.
Operation is very similar to that of a delay line: the data do not move at a fixed rate
from input to output, but only remain in the register long enough for all the previous data
to be read out. This is shown schematically in Fig. 10.20. With first-generation FIFOs, the
data were actually shifted through a register chain, as illustrated in Fig. 10.20. On entry, the
data were passed on to the lowest free memory location and shifted onward from there to
the output by the read clock. One disadvantage of this principle was the long fall-through
time. This is particularly noticeable when the FIFO is empty, as the input data then have to
pass through all the registers before being available at the output. This means that even the
smallest FIFOs exhibit fall-through times of several microseconds. Other disadvantages
include the complex shift logic and the large number of shift operations, thereby precluding
a current-saving implementation in CMOS technology.
To overcome these drawbacks, in the second-generation FIFOs it is no longer the data
that are shifted, but merely two pointers that specify the input and output addresses in a
RAM. This is illustrated in Fig. 10.21. The input counter points to the first free address
Ain , and the output counter to the last occupied address Aout . Both pointers therefore rotate
during ongoing data input and output.
The distance between the two pointers indicates how full the FIFO is. When Ain −
Aout = A max , the FIFO is full. No more data must then be entered, as this would mean
overwriting data that have not yet been read out. When Ain = Aout , the FIFO is empty. No
output
pointer
occupied
input
pointer
Fig. 10.21. FIFO as a ring memory
10.2 RAM Expansions 701
Read-while-write
memory
input output
counter subtractor counter
comparator
data must now be read out, as this would mean receiving old data for a second time. An
overflow or empty condition can only be avoided if the average data rates for input and
output are identical. To achieve this, it is necessary to monitor the occupancy of the FIFO,
and to attempt to control the source or sink in such a way that the FIFO is on average
half full. The FIFO can then accommodate short-term fluctuations, assuming that it has a
sufficient storage capacity.
The design of a FIFO is shown in Fig. 10.22. It is similar to the RAM shift register
in Fig. 10.18. Read-while-write memories with separate address inputs (see Fig. 10.15)
are particularly suitable here, as reading and writing can occur asynchronously. The more
recent FIFOs, examples of which are listed in Fig. 10.23, operate on this principle.
Fig. 10.24. FIFO implementation with standard DRAMs using the sequential flow controller
IDT72T6360
10.2.4
Error Detection and Correction
When data are stored in RAMs, two different types of error can occur: permanent errors
and transient errors. The permanent errors (hard errors) are caused by faults in the ICs
themselves or in the associated controller circuits. The transient errors (soft errors) only
occur randomly and are therefore not reproducible. They are mainly caused by α-radiation
of the package, which may not only discharge memory capacitors in dynamic RAMs, but
also cause flip-flops in static RAMs to change state. Transient errors can also result from
noise pulses generated inside or outside the circuit.
The occurrence of memory errors can have far-reaching consequences. Thus a single
error in a computer memory might not only produce an incorrect result, but even cause
the program to crash completely. Methods have therefore been developed to indicate the
occurrence of errors. In order to do this, one or more check bits must be processed in
addition to the actual data bits: the more check bits used, the greater the number of errors
that can be detected or even corrected.
10.2 RAM Expansions 703
Parity Bit
The simplest method of error detection consists of using a parity bit p that is added to the
data word D. Even or odd parity can be defined. For an even parity check, the parity bit is
set to zero if the number of ones in the data word is even. It is set to one if this number is
odd. This means that the total number of ones transmitted in a data word including parity
bits is always even or, for odd parity, always odd.
The even-parity bit can also be interpreted as the sum (modulo-2) of the data bits. This
checksum can be calculated as the exclusive-or of the data bits.
The implementation of a parity generator is shown in Fig. 10.25. The exclusive-or
gates can be in any sequence. It is chosen such that the sum of the delay times involved
remains as small as possible.
For error detection purposes, the parity bit is stored together with the data bits. When the
data are read out, the parity can then be regenerated as shown in Fig. 10.26 and compared
with the stored parity bit by an exclusive-or operation. If they differ, an error has occurred
and the error output becomes e = 1. This allows each single-bit error to be detected.
However, no correction is possible, since the bit containing the error cannot be located.
If several bits contain errors, an odd number of errors can be detected, whereas an even
number cannot.
Hamming Code
The principle of the Hamming code consists of using several check bits in order to refine
error detection, to the point at which it is possible not just to detect single-bit errors but
also to pinpoint their location. Once the error bit in a binary code is located, it can be
corrected by complementing it.
parity 1
gene- d8
rator Fig. 10.26. Data memory
with parity checking (using
eight-bit data words as an
example). The parity bit d8
e must be stored as an
additional data bit
704 10 Semiconductor Memories
Fig. 10.27. Minimum number of check bits required to detect and correct a single-bit error
The question of how many check bits are required for this purpose is easily answered:
with k check bits, 2k different bit locations can be identified. With m data bits, the resultant
total word length is m + k. An additional check bit combination is required to indicate
whether the data word received is correct. This yields the condition
2k ≥ m + k + 1 .
The most important practical solutions are listed in Fig. 10.27. It can be seen that the
relative proportion of check bits is smaller the greater the word length.
We shall now examine the procedure for determining the check bits, using a 16-bit
word as an example. Figure 10.27 shows that to safeguard 16 bits we need five check bits;
that is, a total word length of 21 bits. In accordance with Hamming, the individual check
bits are evaluated as parity bits for different parts of the data word. In our example, we
therefore require five parity generators. Their inputs are allocated to data bits in such a way
that each data bit is connected to at least two of the five generators. If a data bit is now read
incorrectly, there is a difference only between those parity bits affected by that particular
data bit. Using this method, we therefore obtain a 5-bit error word E, the syndrome word,
instead of a parity error bit. This word can assume 32 different values, which allows us to
pinpoint the error bit. It can be seen that the identification of a single-bit error is unique
only if a different parity bit combination is selected for each data bit location. If a difference
in just one parity bit is detected, only the parity bit itself can be in error, since the parity
bit combination chosen means that, for an incorrect data bit, at least two parity bits have
to differ. If all of the data bits and parity bits are read without error, the calculated parity
bits match those stored and the syndrome word becomes E = 0.
An example of the assignment of the five parity bits to the individual data bits is given
in Fig. 10.28. This shows that data bit d0 affects parity bits p0 and p1 , data bit d1 , affects
parity bits p0 and p2 , and so on. As required, each data bit affects a different combination
of parity bits. To simplify the circuitry, the combinations have been distributed in such a
way that each parity generator has eight inputs.
During reading (R/W = 1) the syndrome generator in Fig. 10.29 compares the stored
parity word P
with the parity word P
syndrome
Fig. 10.29. Data memory with error correction (using 16-bit data words as an example)
it out again. If the error persists, it is a permanent error. In this case, the syndrome word
is read out, as this allows the memory IC involved to be located, and the IC number –
together with the frequency of failure – are listed in a table. This table can then be scanned
at regular intervals so that defective chips can be replaced. This enables the reliability of
a memory with EDC (Error Detection and Correction) to be continually increased.
10.3
Read-Only Memories (ROMs)
The term ROM refers to table memories that are normally only read. They are therefore
suitable for storing tables and programs. Their advantage is that the memory content is
retained when the supply voltage is disconnected. Their disadvantage is that putting data
into the table is more difficult than with RAMs. The categories shown in Fig. 10.1 (MROM,
PROM, EPROM, and EEPROM) differ with respect to the input procedure.
10.3.1
Mask-Programmed ROMs (MROMs)
With these devices, the memory content is entered during the final manufacturing stage,
using a specific metallization mask. This process is only cost-effective for large production
quantities (from approximately 1,000,000 upward) and generally requires several months
for implementation. MROMs are for instance used in hand-held calculators.
10.3.2
Programmable ROMs (PROMs)
A PROM is a read-only memory whose content is programmed-in by the user. The pro-
grammable components are formerly fuses, which are implemented in the ICs by means of
exceptionally thin metallization links. Diodes are also used, which can be shorted by over-
loading them in the reverse-bias direction. The latest programmable elements for PROMs
are special MOSFETs with an additional “floating gate.” This is charged up during pro-
gramming, causing a change of the pinch-off voltage of the MOSFET. As the floating gate
has all-round SiO2 insulation, it can be guaranteed to retain its charge for 10 years.
We shall now describe the internal design of a PROM, using the example of the fuse-
type PROM in Fig. 10.31. For technological reasons, the individual memory cells are not
arranged linearly, but as a square matrix exactly as with RAMs. A particular memory cell
is addressed by applying a logical 1 to the appropriate column and row connection. For this
purpose, the address vector A = (a0 . . . an−1 ) must be decoded accordingly. The column
and row decoders used operate as 1-of-x decoders.
The memory cell selected is activated by the and gate at the intersection of the selected
row or column line. The oring of all the memory cell outputs produces output signal D.
In order to obviate the need for a gate with 2n outputs, wired-or logic is used. In the case
of open-collector outputs, this can be implemented by wired-anding the negated signals.
This method has already been described in Fig. 7.30.
In its basic state, each memory cell addressed generates output signal D = 1. To
program a zero, the fusible link at the output of the desired cell is blown. For it one selects
10.3 Read-Only Memories (ROMs) 707
read line
row decoder
column decoder
Fig. 10.31. Internal structure of a PROM: an example showing 16-bit memory capacity
the address of the corresponding cell, turning on the output transistor of the nand gate.
A powerful current pulse, which is just large enough to blow the fusible link at the nand
gate output, is then injected into the read line. A timing sequence precisely defined by the
manufacturer must be observed. Special programming devices are therefore used, which
can be tailored to the particular type of memory.
In the case of PROMs, not just 1 bit but an entire 4-bit or 8-bit word is usually stored
at an address. These devices therefore possess a corresponding number of data outputs.
Specifying a memory capacity as, say, 1 k × 8 bits means that the memory contains 1,024
8-bit words. The contents are specified in the form of a programming table. By way of an
example, Fig. 10.32 shows its organization for a 32 × 8-bit PROM. The circuit symbol for
a PROM is like that for the RAM in Fig. 10.5. Here, the R/W input becomes the read line
as programming input and the data inputs are omitted.
10.3.3
UV-Erasable PROMs (EPROMs)
An EPROM (erasable PROM) is a PROM that can not only once be user-programmed,
but that can also be erased using ultraviolet light. A MOSFET that incorporates an addi-
tional floating gate is used exclusively as the memory element. This is charged up during
programming, thereby changing the pinch-off voltage of the transistor. With EPROMs,
however, this charge can be erased again in about 20 minutes by irradiation with UV light.
For this purpose, the package is provided with a quartz glass window. Due to the additional
complexity involved in producing this package, EPROMs are more expensive than non-
window PROMs, even though they are realized using the same technology. Consequently,
EPROMs are useful during the developmental stage of new equipment, but the equivalent
PROMs are preferred for mass production.
708 10 Semiconductor Memories
Inputs Outputs
x4 x3 x2 x1 x0 d7 d6 d5 d4 d3 d2 d1 d0
0 0 0 0 0
0 0 0 0 1
0 0 0 1 0
0 0 0 1 1
≈ ≈ ≈ ≈ ≈ ≈ ≈ ≈ ≈ ≈ ≈ ≈ ≈ ≈
1 1 1 1 0
1 1 1 1 1
Fig. 10.32. Example of a programming table for a PROM containing 32 8-bit words
EPROMs are programmed on a word-by-word basis; that is, for the usual 8-bit organi-
zation one byte at a time. Even in the case of older EPROMs (e.g., the 2716; 2 k ×8 bit), the
programming procedure is still simple. One applies a programming voltage of Vpp = 25 V,
together with the required address and the bit pattern to be programmed. Then a program-
ming command that lasts for 50 ms is applied to store the data. Programming can then
be terminated or the process can be repeated for another address, using the associated bit
pattern. In the case of a 2-kbyte EPROM, the programming of the entire device takes about
2 min. However, it would take almost 2 hours to program a 128 kbyte memory. As this
is clearly not tolerable, it was found to be necessary to modify the technology and pro-
gramming algorithms for larger EPROMs. All fast programming algorithms are based on
the fact that most of the cells of an EPROM can be programmed in considerably less than
50 ms. However, as “slower” bytes occur from time to time, it is impossible to generally
reduce the programming time. Instead, a variable programming pulse length is employed.
The “fast” or “intelligent” programming algorithm that is generally used nowadays is
shown in Fig. 10.33. The programming voltage, Vpp = 12.5 V, is applied and the supply
voltage is raised to VCC = 6 V. The higher supply voltage speeds up the programming
process, as the transistors assume lower impedance, and this also constitutes the worst-case
condition for verification purposes. The address then becomes A = 0 and the associated
data are applied. Now follows the procedure for programming this byte. For this purpose,
an auxiliary counter is set to n = 0. A 1 ms programming command is then issued. After
the auxiliary counter has been incremented, the memory content is read out to check
whether programming has already been successful. If not, up to another 24 programming
commands are issued. If the byte has still not been programmed, the chip is deemed to be
defective. Normally, only a few programming pulses are required. However, it is still not
certain that the floating gate has sufficient charge to last for 10 years. In order to make
sure, the charge is tripled. This is done by overprogramming for 3n · 1 ms.
The first byte is thus programmed and the process can be repeated for the next address,
using new data. At the end of programming, we switch back to the read mode and once
more check that the entire memory content is in order. The fast programming algorithm
reduces the programming time for a 1 Mbit EPROM from around 2 h to less than 10 min.
By reducing the programming pulse duration to 100 ms, times of less than 1 min can be
achieved with some EPROMs.
10.3 Read-Only Memories (ROMs) 709
Start
Address = 0, Data
Counter n = 0
Program 1ms no
n = 25
n=:n+1 yes
A=:A+1
no
Programmed?
yes
Overprogram 3n . 1ms
no
Last address?
yes
Set VPP = VCC = 5 V
All data no
OK?
yes
Completed Defective
10.3.4
Electrically Erasable PROMs (EEPROMs)
The term EEPROM (electrically erasable PROM) refers to a PROM which, unlike the
EPROM, can also be erased electrically. With the more recent types, the voltage converter
for generating the programming voltage and the timer for determining the programming
pulse duration are incorporated in the memory chip. In order to program a byte, just
the address and the data need to be applied. If programming is then initiated by a write
command, the EEPROM stores the address and data internally, and immediately releases
the address and data lines. The subsequent process takes place autonomously on the chip.
The old byte is first erased and the new byte is then programmed. This process is internally
monitored to insure that the programmed charge is adequate. The process lasts for 1–
710 10 Semiconductor Memories
Fig. 10.34. Examples of EEPROMs. Supply voltage is 3.3 V for all types. A single byte can be
written or erased
10 ms; that is, it has the same order of magnitude as with EPROMs. Some EEPROMs
are capable of storing not just one byte but an entire “page” containing 16–64 bytes in
one programming process. To do this, the page is read into an internal RAM before the
programming command is issued. This gives effective programming times of 30 ms per
byte.
However, despite these simple, fast erase and write procedures, one should not be
tempted to use an EEPROM as a RAM. The number of possible write cycles is in fact
limited: no byte must be written more than 104 . . . 106 times (depending on the type). With
a programming time of 1 ms, the end of the operational life of a byte or a page may be
reached in as little as 10 s if continuous programming is performed.
Some examples of EEPROMs are listed in Fig. 10.34. With many EEPROMs, as with
most memories, the power dissipation is reduced when they are not selected by CS = 0.
The smallest power dissipation obviously occurs when the supply voltage is disconnected
completely. This does not result in the data being lost – as in the case of all ROMs – but
the access time after application of the supply voltage is increased due to the transient
response of the read amplifiers. For this reason, it is inadvisable to switch on the supply
voltage only when the memory is accessed.
Flash EEPROMs are intermediates between EPROMs and EEPROMs. Like EEP-
ROMs, they can be erased electrically, but not byte by byte, as in the case of EPROMS –
here, the entire block of 1 . . . 32 kbyte is erased at once. This is the reason for their name.
They are erased much more simply than EPROMs: a single erase pulse lasting for a few
seconds is required. It is not necessary to take the package out of the circuit and put it
into an eraser unit for 20 min. Flash EEPROM technology is simpler than that for stan-
dard EEPROMs. Correspondingly, large integration densities can therefore be achieved,
consistent with low prices. Some examples are given in Fig. 10.35
A comparison of the write and read performance of the various ROM types with that of
RAMs is shown in Fig. 10.36. We can see that the strength of RAMs is their fast write and
read processes, which can be repeated any number of times. With all the ROM variants,
writing is subject to more or less severe limitations, although all ROMs have the advantage
of retaining their contents even in the absence of a supply voltage. This characteristic
can be achieved for RAMs by adding a buffer battery. As we can see from Fig. 10.8, the
current drain of many CMOS RAMs is generally lower than the self-discharge of a battery.
10.4 Programmable Logic Devices (PLDs) 711
Fig. 10.35. Examples of Flash EEPROMs. Only a whole block can be written or erased. The block
size amounts 1 . . . 32 kbyte depending on type. Flash EEPROMs are mainly used in memory cards
for USB-sticks and memory cards
RAM ROM
MROM PROM EPROM EEPROM
Write
No. any once once … 100 times 104 . . . 106 times
Time 10 . . . 200 ns months minutes minutes milliseconds
Read
No. any any any any any
time 10 . . . 200 ns approx. 100 ns 10 . . . 300 ns 30 . . . 300 ns 30 . . . 300 ns
Fig. 10.36. Comparison of RAMs and ROMs in terms of their write and read performance
Hence, data retention of 10 years can be insured also at RAMs by using appropriate backup
batteries.
10.4
Programmable Logic Devices (PLDs)
PLDs are used for storing logic functions. The categories in Fig. 10.1 show that there
are three variants: PLEs, PALs, and PLAs. The differences between them are in respect
of programming flexibility. PALs (programmable array logic) are the easiest to program.
They are therefore particularly popular and are available in a wide range of designs. PLAs
(programmable logic arrays) are basically more flexible, but it is more complicated to
712 10 Semiconductor Memories
Fig. 10.37. Simplified representation of the and and or operations. The crosses indicate which
input is connected. An unconnected input has no effect, since it is 1 for the and operation and 0 for
the or operation
program them. They have therefore ceased to be of major importance. Gate Arrays consist
of a see of gates or primitive logical functions the connection of which is configured by
programming. GAs are programmed by the manufacturer, therefore they are used with large
quantitites only. Very popular are the Field Programmable Gate Arrays FPGAs because
they are user programmable like PLDs.
When logic functions are implemented using the standard product terms, it is first
necessary to generate all and (product) terms of the input variables and then form the
sum of the products. In order to be able to show these operations clearly, we use the
simplified representation in Fig. 10.37. The internal design of PLAs and PALs is shown in
Fig. 10.38. The input variables and the intersecting inputs of and gates form a matrix that
enables all the required logical products to be formed. In a corresponding second matrix,
the outputs of the and gates are ored together in order to form the required logical sums.
This requires only one or gate per output variable. In the case of a PLA both matrices are
user-programmable. In the case of a PAL the or matrix is fixed by the manufacturer, and
only the and matrix can be programmed.
A PROM can also be understood as a function memory if the address decoder, which has
a truth table like that shown in Fig. 10.39, is interpreted as an and matrix. For every address
applied, only a single and operation is 1, namely that corresponding to the address applied.
There are therefore n = 2N product terms, whereas the PLAs and PALs have substantially
fewer. The or-matrix represents the truth-table here.
PROMs designed for implementing logic functions are also known as PLEs (pro-
grammable logic elements). The differences become apparent by considering the example
in Fig. 10.38. All of the connections that are not required for these functions have been
programmed “open”. Figure 10.38 shows that all the required logical products are formed
in the and matrices of the PLA and PAL. In the case of the PLA, it is even possible to
use a product, which is required several times, twice in the or matrix. This freedom is not
available in the case of the (simple) PAL, as whose or matrix is not programmable.
With a PROM, in each case the particular product line corresponds to the input com-
bination that is 1. Consequently, in the or matrix it is necessary to program connections
for all combinations that are 1 in the truth table. We can see from Fig. 10.39 that a PROM
is the image of the truth table, whereas the PLA and PAL represent the logic functions.
A PROM can be used to store any kind of truth table, whereas only a limited number of
products and sums are available in a PLA or PAL. For this reason, it is not possible to
realize any truth table in a PLA or PAL, but only those that can be converted into simple
logic functions. This requires utmost simplification of the functions using Boolean algebra
and, if necessary, transformation from and into or operations using De Morgan’s Law,
10.4 Programmable Logic Devices (PLDs) 713
OR matrix,
programmable
OR matrix,
fixed
OR matrix,
programmable
decoding x =
Fig. 10.38. Comparison of the structures of PLA, PAL, and PROM/PLE devices.
An example for y0 = x2 + x̄0 x1 and y1 = x̄0 x1 + x̄0 x̄2 + x0 x̄1 x2
in order to utilize the PALs as efficiently as possible. Nowadays, this is no longer done
manually, but using computer-aided design programs, which can be run on any PC. Their
application is described in greater detail in Sect. 10.4.2.
714 10 Semiconductor Memories
Z x2 x1 x0 y1 y0
0 0 0 0 1 0
1 0 0 1 0 0
2 0 1 0 1 1
3 0 1 1 0 0
4 1 0 0 0 1
5 1 0 1 1 1
6 1 1 0 1 1 y0 = x2 + x 0 x1
7 1 1 1 0 1 y1 = x 0 x1 + x 0 x 2 + x0 x 1 x2
Fig. 10.39. Example of a truth table and its logic functions for the example in Fig. 10.38
10.4.1
Programmable Logic Array (PAL)
PALs are the most important representatives of programmable logic devices (PLDs). They
are available in a wide range of variants, all of which are based on the principle shown
in Fig. 10.38. The differences are in the implementation of the output circuit. The most
commonly used variants are listed in Fig. 10.40. Each different type is designated by the
relevant letter shown.
The high (H) output represents the basic type shown in Fig. 10.38. In the case of the
low (L) type, the output is negated.
The sharing (S) output has features in common with the PLAs. Here, the or matrix
is also partly programmable: two adjacent or gates can share the and operations that are
available to them. This makes it possible to use a product term twice in PAL.
With many PALs, an output can also be used as an input or programmed as a bidirec-
tional port (B). This is the purpose of the tristate gate at the output, whose ENABLE is
itself a logic function.
An important application of PALs is in sequential logic systems. In order to obviate
the need for additional chips, the required registers (R) are incorporated into the PALs.
They have a common clock terminal to enable the construction of synchronous systems.
In addition, the output signals are generally fed back internally to the and matrix, thereby
eliminating external feedback circuitry (see Fig. 9.62) and saving on pins.
Using the optimum PAL for each application would require a large number of different
types – as shown in Fig. 10.40. In order to reduce the variety of types, PALs with a
programmable output structure are common today. One such variable “macrocell” (V) is
also shown in Fig. 10.40. It is built around a multiplexer that can be used to select any one
of four different operating modes. These are defined by programming the function bits f0
and f1 . The different operating modes are listed in Fig. 10.41. Bit f0 determines whether
or not the output is negated. Bit f1 switches between combinatorial and registered mode.
It also determines, via a second multiplexer, whether feedback is taken from the output or
from the register. We can see that most output structures of PALs can be implemented in
this way using a single type.
10.4 Programmable Logic Devices (PLDs) 715
H High
L Low
S Sharing
B Bidirec-
tional
R Register
V Variable
10.4.2
Computer-Aided PLD Design
In order to “personalize” a PAL, it is first necessary to specify which connections are to be
programmed, and then to carry out the programming in a second step. Now that software
packages are available which will run on any PC, PAL design is no longer a manual process.
The various design phases are shown in Fig. 10.42. There are usually various input formats,
of which only the most commonly used will be presented here. The logic function or truth
716 10 Semiconductor Memories
Truth Simulator
table Testing
Input Logic function JEDEC
file Minimizing file
State
Fitting Programming
diagram
device
Schematic Syntax errors
diagram Library Logic function
Utilization
Fig. 10.42. Computer-aided PLD design. You can use ispLEVER with a tutorial in Chap. 29.2 on
page 1459
table is entered using a text editor. When designing sequential logic systems, it is also
possible to start from a state diagram and specify the transition conditions.
A particularly effective method is to enter the schematic diagram. For this, a library
can be used, in which the most common TTL functions are already defined as macros.
In addition to gates and flip-flops, the library also provides multiplexers, demultiplexers,
adders, comparators, counters and shift registers. This is not only useful for converting
an old design that incorporates TTL devices into a PLD design, but it also simplifies the
design of new circuits in which the TTL devices are used merely as conceptual models.
The input method is supported by a schematic editor.
After input by whatever method, all the data are converted to logic functions and a
syntax check is performed. The logic functions are then minimized in accordance with the
rules of Boolean algebra. However, this does not yet guarantee that they fit the relevant
PLD in their optimum form. For fitting, and operations, for instance, are converted to
or operations using De Morgan’s Law. The programming data (the fuse map) are finally
stored in a standard format, the JEDEC file.
For the programming process a programmer is connected to the PC and the JEDEC file
is downloaded. Newer programmers are controlled entirely by the PC and have no controls
of their own. Manually entering the programming data in form of a fuse map is outdated
and no longer practical due to the large data quantity. Before choosing a new PLD type it is
reasonable to check whether it is supported by the programmer. It is not sufficient to have
a universal programmer with software control for the voltages and currents of each pin
if the corresponding programming software is not available. These problems do not exist
with newer PLDs which are in-circuit or in-system programmable. They are provided with
an interface that can be connected directly to the PC. Here, the programming hardware
is located on the PLD. This not only renders the programmer or a corresponding socket
adapter unnecessary, but also allows to program and update the PLD on the circuit board.
Even though most PLDs can be erased, it is useful to check prior to programming
whether or not the designed circuit has the desired properties. For this purpose, a functional
simulator is used to determine whether the output functions respond to the input signals and
the clock pulses as intended. The resulting time diagrams are similar to those on the screen
of a logic analyser; this, however, does not take the signal delay times into consideration.
10.4 Programmable Logic Devices (PLDs) 717
To test the dynamic response requires a timing simulator, which is much more complicated
since it must take into account the data and architecture of the particular PLD. At this point
you may see whether the fitter needs to cycle twice through the PLD in order to generate
a complicated function.
Almost every manufacturer offers a software package for circuit design with their
own PLDs. Often, program versions with limited functionality can be obtained free of
charge. As a programming language they use HDL or VHDL. For graphic input they offer
schematic often in conjunction with a TTL-library.
Programs that are independent of the manufacturer, as listed in Fig. 10.43, offer the
great advantage that the manufacturer can be changed without having to re-enter the design.
Also beneficial is the fact that they use the same user interface for products of different
manufacturers. This, however, requires a device fitter of the respective manufacturer which
maps the design in accordance with the architecture of the used module.
10.4.3
Survey of Types Available
Today when designing digital circuits with more than a few 7400 gates it is advantageous
to use PLDs for the implementation. The simple PALs listed in Fig. 10.44 already contain
300 to 500 gates. Even if only half of them can be used, a single chip design is already
possible with it in many cases. A whole conglomeration of TTLs can be replaced by a
single chip. Keeping the number of chips low offers several advantages:
– PCBs can be kept small, saving space and money,
– one PLD is usually less expensive than the total of the components otherwise required
– the reliability increases, since the internal PLD connections are more reliable than those
on the PCB
– design changes can often be performed by simply reprogramming the PLD.
All PLDs listed in this survey feature configurable output cells of type V = variable as
shown in Fig. 10.40 on page 715. The most important parameter describing the architecture
718 10 Semiconductor Memories
is the number of macro cells. This is a measure of the number of logical functions that can
be formed. Each input of a PLD and each macro cell feed one signal to the and matrix.
This results in the relation:
.
Number of inputs Number of
=
in the and matrix inputs + macro cells
The number of matrix outputs specifies the possible product terms of the PLD. If these are
distributed uniformly to the macro cells the following relation applies:
.
Maximum number of Number of matrix outputs
=
product terms per function Number of macro cells
In the 16 V 8 chip with 64 matrix outputs this is 64/8 = 8. This is sufficient for most
functions, but not necessarily for every case. In order to allow implementation of very
complicated functions it is sometimes possible to share the product terms of the neighbour
(product term sharing) or to use expander product terms which can be assigned to any
function. In this respect they then behave similar to PLAs.
The architecture column in Fig. 10.44 also specifies the number of macro cells in
addition to the number of terminals that can be used only as inputs. The output can also
be configured as an input, as can be seen for the case of the variable output cell. The
architecture of a PAL can be seen from its type designation:
Number of inputs to the matrix V Number of macro cells
22 V 10 is the most commonly used type. There are two options:
– “in system programming” i.e. programming logic integrated in the chip
– “zero power”, i.e. the static power consumption is zero.
Large PLDs are realized by a matrix of small PLDs on one chip that can be connected
by a programmable interconnect matrix as shown in Fig. 10.45. The product-term allocator
of the device fitter determines which product term is realized by what PLD and how the
PLDs are interconnected. If a pinout is specified it determines the choice. These composite
PLDs are called complex PLDs shortly CPLDs. Some types are collected in Fig. 10.46.
Most PLDs show a surprisingly high power dissipation even in the static state, even
though they are CMOS circuits. This is caused by pull-up resistors in wired and functions.
But there are also PLDs with a power consumption that is proportional to the frequency.
These consume practically no current in the static state. In Fig. 10.45 these are identified
10.4 Programmable Logic Devices (PLDs) 719
I/O I/ O
PLD 1 PLD 2
PLD 3 PLD 4
Fig. 10.46. Examples for CPLD-familes. All types are in system programmable
as “zero power” types. The current consumption these types is especially low, even during
operation.
If a few bits have to be stored in a sequential logic this can be achieved by using the
registers of the macro cells. But for larger data quantities this method is not practical; then
an external RAM must be connected. With some of the newer CPLDs this is not necessary,
since RAMs are available on chip; the storage capacity is shown in Fig. 10.46.
10.4.4
User Programable Gate Arrays
One group of logic circuits that can also be programmed by the user is that if the Field Pro-
grammable Gate Arrays (FPGAs). They contain many Configurable Logic Blocks, CLBs
that can be connected by column- and row-interconnect busses as shown in Fig. 10.47. The
720 10 Semiconductor Memories
interconnect
CLB CLB CLB
I/O Block
I/O
column
row interconnect
I / O Block
I/ O
CLBs consist of configurable gates and flip-flops like macrocells in PLDs. Their structure
is much simpler than that of macrocells but their number is much greater as can be seen
in Fig. 10.48. FPGAs can therefore be used for complex designs. They are particularly
advantageous when the architecture of a PLD is not suitable for the desired functionality.
For the circuit designer there is little difference in designing a PLD or FPGA. Some
manufacturers offer the same design software for PLDs and FPGAs. But the device fitter is
much more complex for FPGAs. In a first step it must place the functions in the CLBs and
in a second step it must route the connections. Which connections are time-critical can be
specified, since there are usually several data paths of different delay times available and
short connections are faster than those forming several corners. Here, a timing simulation
which takes account of the wiring (Post Layout Simulation) is essential.
In most FPGAs the configuration of the routing is stored in a RAM on the chip. Dur-
ing power-up this is automatically loaded via a serial interface from a separate EPROM.
Fundamentally different is the programming of FPGAs in antifuse technology. Here, the
connections are made by programming. Thus it is the opposite of the Fusible Link technol-
ogy used in PROMs in which the connection is interrupted by the programming process.
As with a fusible link, a programmable link – antifuse – cannot be erased.
FPGAs are often used for digital signal processing. For these applications some man-
ufacturers offer pre-programmed multipliers, adders, and RAM blocks on the chips that
can be fitted in the design.
10.4 Programmable Logic Devices (PLDs) 721
General Applications
Chapter 11:
Operational Amplifier Applications
Most analog signal processing is done today with circuits using operational amplifiers
because opamps are available with good data for little money. Most signals that must be
processed in electronic circuits arise in analog form and are needed after processing in
analog form also. Therefore analog signal processing is first choice.
The accuracy of analog processing is limited to 0.1 to 1%. If higher accuracy is needed,
digital processing is advantageous. The same is true if the required processing is complex
or nonlinear or needs memory. To use the advantages of digital signal processing, the
signals must be digitized before digital processing and converted back to analog signals
afterwards. Therefore, Analog-Digital Converters (ADCs) and Digital-Analog Converters
(DACs) as described in Chap. 18 are needed. At the input of ADCs and the output of DACs,
the analog signal must be amplified and filtered. Therefore, analog signal processing is
necessary also in this case.
In the following sections, the most important families of operational circuits are clas-
sified and described. They are circuits for the four fundamental arithmetic operations,
for differential and integral operations, and for the synthesis of transcendental or any
other chosen functions. In order to illustrate the operating principles of these circuits as
clearly as possible, we initially assume ideal characteristics of the operational amplifiers
involved. When using real operational amplifiers, restrictions and additional conditions
must be observed in the choice of the circuit parameters, and these are treated thoroughly
in Chap. 5. We want to discuss in more detail only those effects that play a special role in
the performance of the particular circuit.
11.1
Summing Amplifier
When connected as an inverting amplifier, an operational amplifier can be used for the
addition of several voltages. As Fig. 11.1 indicates, the input voltages are connected via
series resistors to the N-input of the operational amplifier. Since this node represents virtual
ground, Kirchhoff’s current law (KCL) directly yields the relation for the output voltage:
V1 V2 Vn Vo
+ + ··· + + = 0
R1 R2 Rn RN
If a DC voltage is added to the signal voltage in the manner described, the inverting
summing amplifier can also be used as an amplifier with a wide-range zero adjustment.
Vn
V2 Fig. 11.1. Inverting summing amplifier
V1 Output voltage:
RN RN RN
Vo −Vo = V1 + V2 + · · · + Vn
R1 R2 Rn
726 11 Operational Amplifier Applications
11.2
Subtracting Circuits
11.2.1
Reduction to an Addition
A subtraction operation can be reduced to the problem of an addition by inverting the signal
to be subtracted. This requires the circuit shown in Fig. 11.2. The operational amplifier
OA1 inverts the input voltage V2 ; the output voltage is then
Vo = αP V2 − αN V1 (11.1)
If both resistance ratios are equal
αP = αN = α
then the output voltage is the amplified difference
Vo = α(V2 − V1 ) = AD (V2 − V1 ).
If the resistance ratios are not equal because of tolerances, we get an outpout voltage even
if both input voltages are equal
V1 = V2 = VCM (11.2)
resulting in
Vo = αP VCM − αN VCM = VCM (αP − αN ).
With
1 1
αP = α + α and αN = α − α
2 2
we obtain
Vo = VCM α = VCM ACM .
From this we can calculate the common mode rejection
AD α 1
G = = = . (11.3)
ACM α tolerance
The common-mode rejection ratio thus equals the reciprocal of the relative matching of
the individual gains. If resistors with a tolerance of 1% = 0.01 are used, a common-mode
rejection of G = 1/0.01 = 100 can be expected.
V1 RN/αN
V2 RN/αP
V2
Vo
OA1 OA2
11.2.2
Subtraction Using a Single Operational Amplifier
To calculate the output voltage of the subtracting amplifier in Fig. 11.3, we may use the
principle of superposition. We therefore write
Vo = k1 V1 + k2 V2
For V2 = 0, the circuit is an inverting amplifier, where Vo = −αN V1 . It follows k1 = −αN .
For V1 = 0, the circuit represents a noninverting amplifier that has a voltage divider
connected at its input. The potential
RP αP
VP = V2 = V2
RP + RP /αP 1 + αP
is thus amplified by the factor (1 + αN ), and this results in the output voltage
αP
Vo = (1 + αN )V2
1 + αP
If both resistor ratios are the same – that is, if αN = αP = α – it follows:
Vo = αV2
and that k2 = α. We now obtain the output voltage for the general case, in the form
Vo = α(V2 − V1 )
Should the ratios of the resistors at the P- and N-inputs not be precisely equal to α, the
circuit will not evaluate the precise difference between the input voltages. In this case,
1 + αN
Vo = αP V 2 − α N V1
1 + αP
To calculate the common-mode rejection ratio we again use the formulation of (11.3), and
obtain
AD 1 (1 + αN )αP + (1 + αP )αN
G = = ·
ACM 2 (1 + αN )αP − (1 + αP )αN
With αN = α − 21 α and αP = α + 21 α, the expression may be rewritten and expanded
into a series. Neglecting higher-order terms, we obtain
α 1
G ≈ (1 + α) ≈ AD (11.4)
α tolerance
V1
RP Vo
V2
RP/αP RP
For constant α, the common-mode rejection ratio is inversely proportional to the tolerance
of the resistor ratios. If the resistor ratios are identical, G = ∞, although this applies
to ideal operational amplifiers only. In order to obtain a particularly high common-mode
rejection ratio under real conditions, RP may be varied slightly. In this way, α can be
adjusted and the finite common-mode rejection ratio of the operational amplifier can be
compensated for.
Equation (11.4) also shows that the common-mode rejection ratio for a given resistor
matching tolerance α/α is approximately proportional to the chosen differential gain
AD = α. This is a great improvement over the previous circuit.
This may be best illustrated by an example. Two voltages of about 10 V are to be
subtracted one from the other. Their difference is no more than 100 mV. This value is to
be amplified and is to appear at the output of the subtraction amplifier as a voltage of 5 V,
with an accuracy of 1%. The differential gain must therefore be AD = 50. The absolute
error at the output must be smaller than 5 V · 1% = 50 mV. If we assume the favorable case
of the common-mode gain representing the only source of error, we then find it necessary
to limit the common-mode gain to
50 mV
ACM ≤ = 5 · 10−3
10 V
that is,
50
G≥ = 104 =
7 80 dB
5 · 10−3
For the subtracting amplifier in Fig. 11.3, this requirement can be met by a relative resistor
matching tolerance of α/α = 0,5%, as follows from (11.4). For the subtraction circuit
of Fig. 11.2, however, (11.3) yields a maximum tolerable mismatch of 0.01%.
Figure 11.4 shows an expansion of the subtracting amplifier for any number of addi-
tional summing and/or subtracting inputs. The determining factor for the proper functioning
of the circuit is that the specified coefficient condition is satisfied.
If this is not achieved with the coefficients αi and αi
specified, the voltage 0 can be
added or subtracted using the missing coefficient required to satisfy the equation.
In order to deduce the relationships given below the caption to Fig. 11.4, we apply
Kirchhoff’s current law to the N-input:
0m
Vi − V N Vo − V N
+ = 0
RN RN
i=1
αi
Vm
V2
Fig. 11.4. Multiple-input subtracting amplifier
V1 8
n 8
m
V1' Output voltage: Vo = αi
Vi
− αi Vi
Vo i=1 i=1
V2' 8
n 8
m
Condition for coefficients: αi
= αi
Vn' i=1 i=1
11.3 Bipolar-Coefficient Circuit 729
Hence
0
m 0
m
αi Vi − V N αi + 1 + V o = 0
i=1 i=1
11.3
Bipolar-Coefficient Circuit
The circuit shown in Fig. 11.5 allows the multiplication of the input voltage by a constant
factor, the value of which can be set between the limits +n and −n by the potentiometer
R2 . If the slider of the potentiometer is positioned as far to the right as possible, then
q = 0 and the circuit operates as an inverting amplifier with gain A = −n. In this case,
the resistor R1 /(n − 1) is ineffective, since there is no voltage across it.
Vi Vo
Fig. 11.5. Bipolar-coefficient circuit
Output voltage: Vo = n(2q − 1)Vi
730 11 Operational Amplifier Applications
0
1 1 q
2
For q = 1, the full input voltage Vi is at the P-input. The voltage across R1 /n is
therefore zero, and the circuit operates as a noninverting amplifier that has the gain
R1
A = 1+ = +n
R1 /(n − 1)
A = n(2q − 1)
It is thus linearly dependent on q and can be easily adjusted – for instance, by means of a
calibrated multi-turn potentiometer. The factor n determines the range of the coefficient.
The smallest value is n = 1; in this case, the resistor R1 /(n − 1) may be omitted.
11.4
Integrators
One of the most important applications of the operational amplifier in analog computing
circuits is as an integrator. Its output voltage can be expressed in the general form
t
Vo (t) = K Vi (t˜)d t˜ + Vo (t = 0)
0
11.4.1
Inverting Integrator
The inverting integrator shown in Fig. 11.7 differs from the inverting amplifier in that the
feedback resistor RN is replaced by the capacitor C. The output voltage is then expressed
by
t
Q 1
Vo = = IC (t˜)d t˜ + Q0
C C
0
11.4 Integrators 731
where Q0 is the charge on the capacitor at the beginning of the integration (t = 0). As
IC = −Vi /R, it follows:
t
1
Vo = − Vi (t˜)d t˜ + Vo 0
RC
0
The constant Vo 0 represents the initial condition: Vo 0 = Vo (t = 0) = Q0 /C. It has to be
set to a defined value by the additional measures described in the next section.
Let us now look at two special cases. If the input voltage Vi is constant, the output
voltage is
Vi
Vo = − t + Vo 0
RC
which increases linearly with time. This circuit is thus very well suited to the generation
of triangular and sawtooth voltages.
7i cos ωt, the output voltage becomes
If Vi is a cosinusoidal alternating voltage vi = V
t 7
1 7i cos ωt˜d t˜ + Vo 0 = − Vi sin ωt + Vo 0
Vo (t) = − V
RC ωRC
0
The amplitude of the alternating output voltage is therefore inversely proportional to the
angular frequency ω. When the amplitude–frequency response is plotted using log–log
coordinates, the result is a straight line with a slope of −6 dB per octave. This characteristic
is a simple criterion for determining whether a circuit behaves like an integrator.
The behavior in the frequency domain can also be determined directly with the help
of complex calculus:
Vo Z 1
A = = − C = − (11.6)
Vi R s RC
Hence, it follows for the ratio of the amplitudes:
7o
V 1
= |A| =
7i
V ωRC
as shown before.
With regard to the frequency compensation, it must be noted that, unlike all of the
circuits that have been discussed previously, the feedback network causes a phase shift.
This means that the feedback factor becomes complex:
V
s RC
k = N
= (11.7)
Vo Vi =0 1 + s RC
For high frequencies, k approaches k = 1 and the phase shift becomes zero. Therefore, in
this frequency range the same conditions obtain as for a unity-gain inverting amplifier (see
732 11 Operational Amplifier Applications
Chap. 5). The frequency compensation that is necessary in the latter case must therefore
also be used for the integrator circuit. Internally compensated amplifiers are normally
designed for this application, and are therefore suitable for use as integrators.
The frequency range that is usable for integration can be seen in Fig. 11.8, to give a
typical example. The integration time constant chosen is τ = RC = 100 µs. It is apparent
that by making this choice, a maximum loop gain of |g| = |k AD | ≈ 600 is attained,
which corresponds to an output accuracy of about 1/|g| ≈ 0.2%. In contrast to that of the
inverting amplifier, the output accuracy falls not only at high, but also at low, frequencies.
For the real operational amplifier, the input bias current IB and the offset voltage VO
may be very troublesome, as their effects accumulate with time. If the input voltage Vi is
reduced to zero, the capacitor carries the error current
Vo
+ IB
R
An error current of IB = 1 mA causes the output voltage to rise at a rate of 1 V/s if C = 1 mF.
Equation (11.8) indicates that, for a given time constant, the contribution of the input bias
current is smaller, the larger is the value of C chosen. The contribution of the offset voltage
Vi
Vo
Fig. 11.9. Integrator with input bias current
compensation. Capacitor C1 , shorts noise voltages at the
P-input
11.4 Integrators 733
remains constant. Because there is a limit to the size of C, it is important to at least make
certain that the effect of IB does not exceed that of V0 . This is the case if
V0 V0 C
IB < =
R τ
If a time constant of τ = 1 s has to be achieved with a capacitance of C = 1 mF, an
operational amplifier that has an offset voltage of VO = 1 m V should possess an input
bias current smaller than
1 mF · 1 mV
IB = = 1 nA
1s
Operational amplifiers with bipolar input transistors rarely have such low input bias cur-
rents. Their undesirable effects can be reduced by not connecting the P-input to ground
directly, but via a resistor that also has the value R. The voltage IB R is then dropped
across both resistors and the error current through capacitor C is therefore zero. The only
remaining error current is now the difference between the input bias currents – that is, the
offset current – which, however, is generally small in comparison.
In the case of FET operational amplifiers, the input bias current is usually negligible.
They are therefore preferred if the integration time constants are large, even though their
offset voltages are often much larger than for operational amplifiers with bipolar input
transistors.
Leakage currents through the capacitor may be a further source of error. As electrolytic
capacitors have leakage currents that are of the order of microamperes, they cannot be
used as integration capacitors. It is therefore necessary to use foil capacitors, which makes
capacitances of over 10 mF very bulky.
11.4.2
Initial Condition
An integrator can often be used only if its output voltage Vo (t = 0) can be set independently
of the input voltage. Using the circuit shown in Fig. 11.10, it is possible to stop integration
and set the initial condition.
If the switch S1 is closed and S2 is open, the circuit operates like that in Fig. 11.7; the
voltage V1 is integrated. If switch S1 is now opened, the charging current becomes zero in
the case of an ideal integrator, and the output voltage remains at the value that it had at the
time of switching. This may be of use if we want to interrupt computation; for example, in
order to read the output voltage at leisure. To set the initial condition, S1 is left open and
S2 is closed. The integrator becomes an inverting amplifier, with an output voltage of
RN
Vo = − V2
R2
V2
S2
Fig. 11.10. Integrator that has three modes of
S1 operation: integrate, hold, and set initial
condition
V1 RN
Vo Initial condition : Vo (t = 0) = − V2
R2
734 11 Operational Amplifier Applications
V2
OA2
VMC2 Vo
OA1
VMC1
V1
Vn
Fig. 11.12. Summing integrator
V2
Output voltage:
V1
t
1 V1 V2 Vn
Vo Vo = − + + ··· + d t˜ + Vo 0
C R1 R2 Rn
0
However, the output assumes this voltage only after a certain delay, which is determined
by the time constant RN C.
Figure 11.11 shows one possibility for replacing the switches by electronic compo-
nents. The two FETs T1 and T2 replace the switches gm1 and gm2 shown in Fig. 11.10.
They are conducting (on) if the corresponding mode control signal voltage is greater than
zero. For sufficiently negative control voltages they are in the off state. The precise op-
eration of the FET switches, and of the diodes D1 − D6 , is described in more detail in
Chap. 17.2.1.
The voltage follower OA2 reduces the delay time constant for setting the initial con-
dition, from the value RN C to the small value of RDS on C.
11.4.3
Summing Integrator
Just as the inverting amplifier can be extended to become a summing amplifier, so an inte-
grator can be developed into a summing integrator as shown in Fig. 11.12. The relationship
given for the output voltage can be derived directly by applying KCL to the summing point.
11.4.4
Noninverting Integrator
For integration without polarity reversal, an inverting amplifier can be added to the inte-
grator. Another solution is shown in Fig. 11.13. The circuit basically consists of a lowpass
filter as the integrating element. A NIC, having an internal resistance of −R, is connected in
11.5 Differentiators 735
parallel with the filter and simultaneously acts as an impedance converter (see Chap. 12.5).
To calculate the output voltage, we apply KCL to the P-input and obtain
Vo − V P Vi − VP dVP
+ −C = 0
R R dt
Hence, with VP = VN = 21 Vo we arrive at the result
t
2
Vo = Vi (t˜)d t˜
RC
0
It is to be noted that the input voltage source must have a low impedance; otherwise,
the stability condition for the NIC is not fulfilled. The operational amplifier evaluates
differences between large quantities, and therefore this integrator does not have the same
precision as the basic circuit shown in Fig. 11.7.
11.5
Differentiators
11.5.1
Basic Circuit
If the resistor and capacitor of the integrator in Fig. 11.7 are interchanged, we obtain the
differentiator shown in Fig. 11.14. The application of KCL to the summing point yields
the relationship
dVi Vo
C + = 0,
dt R
dVi
Vo = −RC (11.9)
dt
7i sin ωt, we obtain the output voltage
Thus, for sinusoidal alternating voltages Vi = V
7i cos ωt
Vo = −ω RC V
11.5.2
Practical Implementation
The practical implementation of the differentiator circuit shown in Fig. 11.14 presents
certain problems, since the circuit is prone to oscillations. These are caused by the feedback
network, which – at higher frequencies – gives rise to a phase lag of 90 ◦ , as the feedback
factor is
VD 1
k = = (11.12)
Vo 1 + s RC
This lag is added to the phase lag of the operational amplifier, which in the most favor-
able case is already 90 ◦ . The remaining phase margin being zero, the circuit is therefore
unstable. This instability can be overcome if the phase shift of the feedback network at
high frequencies is reduced by connecting a resistor R1 in series with the differentiating
capacitor, as shown in Fig. 11.15. This measure need not necessarily reduce the usable
frequency range, since the reduction in loop gain limits the satisfactory operation of the
differentiator at higher frequencies.
For the cutoff frequency f1 of the lowpass element R1 C, it is advisable to choose
the value for which the loop gain becomes unity. To find this value, we consider a fully
compensated amplifier, the amplitude–frequency response of which is shown as a dashed
Vi
VD Vo
(optimum compensation)
(full compensation)
Fig. 11.16. Example for the frequency response of the loop gain f1 = fT /2πτ where
τ = RC and fT is the unity gain bandwidth
line in the example of Fig. 11.16. The phase margin at frequency fl is then approximately
45 ◦ . Since, in the vicinity of frequency f1 , the amplifier has a feedback factor of less
than unity, an increase in the phase margin can be obtained by reducing the frequency
compensation, and hence a transient behavior of near-critical damping can be approached.
To optimize the compensation capacitor Ck , a triangular voltage is applied to the input
of the differentiator and Ck is reduced so that the rectangular output voltage is optimally
damped.
11.5.3
Differentiator with High Input Impedance
The input impedance of the differentiator described exhibits capacitive behavior, which in
some cases can lead to difficulties; for example, an operational amplifier circuit used as an
input voltage source can easily become unstable. The differentiator shown in Fig. 11.17 is
better in this respect. Its input impedance does not fall below the value of R, even at high
frequencies.
The operation of the circuit is best illustrated as follows. Alternating voltages of low
frequency are differentiated by the RC network at the input. In this frequency range, the
operational amplifier corresponds to a noninverting amplifier, and has a gain of A = 1.
Alternating voltages of high frequency pass the input RC element unchanged and are
differentiated by the feedback amplifier. If both time constants are equal, the effects of
differentiation at low frequencies and at high frequencies overlap and produce a smooth
changeover.
With regard to stabilization against likely oscillations, the same principles apply as for
the previous circuit. The damping resistor R1 is shown by the dashed line in Fig. 11.17.
11.6
Solving Differential Equations
There are many problems that can be described most easily in the form of differential
equations. One obtains the solution by using the analog computing circuits described
above to model the differential equation, and by measuring the resulting output voltage.
In order to avoid stability problems, the differential equation is transformed in such a way
that only integrators are required, rather than differentiators.
We will illustrate this method using the example of a linear second-order differential
equation:
y
+ k1 y
+ k0 y = f (x) (11.13)
In the first step, the independent variable x is replaced by the time variable t:
t
x =
τ
Since
dy dt
y
= · = τ ẏ and y
= τ 2 ÿ
dt dx
the differential equation (11.13) becomes
τ 2 ÿ + k1 τ ẏ + k0 y = f (t/τ ) (11.14)
In the second step, the equation is solved for the undifferentiated quantities:
k0 y − f (t/τ ) = −τ 2 ÿ − k1 τ ẏ
Thirdly, the equation is multiplied throughout by the factor −1/τ and integrated:
1
− [k0 y − f (t/τ )]dt = τ ẏ + k1 y (11.15)
τ
In this way, an expression is formed on the left-hand side of (11.15), which can be computed
by a simple summing integrator. Its output voltage is termed the state variable, zn , where
n is the order of the differential equation; here, n = 2. Therefore,
1
z2 = − [k0 y − f (t/τ )]dt (11.16)
τ
In this equation, the output variable y is initially taken as being known.
By inserting (11.16) in (11.15), we arrive at
z2 = τ ẏ + k1 y (11.17)
This differential equation is now treated in the same way as (11.14), and therefore we
obtain
z2 − k1 y = τ ẏ,
1 (11.18)
− [z2 − k1 y]dt = −y
τ
11.7 Function Networks 739
Fig. 11.18. Signal flow graph for solving the differential equation
t
τ 2 ÿ + k1 τ ẏ + k0 y = f
τ
11.7
Function Networks
The problem often arises that a function, V2 = f (V1 ) is to be calculated, where f is a
nonlinear function of V1 , so that for example
V1
V2 = VA log
VB
or
V1
V2 = VA sin
VB
740 11 Operational Amplifier Applications
There are three possibilities for realizing such relationships. One can either make use
of a physical effect that complies with the function required, or one can approximate the
function by a piecewise linear approximation or by a power series expansion. Below, we
give some examples of these methods.
11.7.1
Logarithm
A logarithmic amplifier must produce an output voltage that is proportional to the logarithm
of the input voltage. It is therefore possible to make use of the diode characteristic
⎛ ⎞
VAC
IA = IS ⎝e nVT − 1⎠ (11.21)
Vi Vi
Vo Vo
The output voltage of the transistor logarithmic amplifier shown in Fig. 11.21 is therefore:
Vi
Vo = −VBE = −VT ln
IS R 1
As well as eliminating the correction factor n, the circuit shown in Fig. 11.21 has two
further advantages. First, no distortion due to the collector–base leakage current occurs,
as VCB = 0. In addition, the magnitude of the current gain does not affect the result,
as the base current flows away to ground. When suitable transistors are employed, the
collector current can be varied from the picoampere to the milliampere region; that is, over
nine decades. However, operational amplifiers with very low input currents are needed to
exploit this range to the full.
Since the transistor T increases the loop gain by its own voltage gain, the circuit is
prone to oscillations. The voltage gain of the transistor stage can be reduced quite simply
by connecting a resistor RE between the emitter and the amplifier output, as shown in
Fig. 11.22. This limits the voltage gain of the transistor by means of current feedback to
the value R1 /RE . The resistance RE must not, of course, give rise to voltage saturation of
the operational amplifier at the largest possible output current. The capacitor C can further
improve stability of the circuit by differentiate action in the feedback. It must be noted,
however, that the upper cutoff frequency decreases proportionally to the current because
of the nonlinear transistor characteristic.
Enhanced performance is achieved if the transistor is operated from a high-impedance
current source. The loop gain is then gm · R1 , where gm is the transconductance of the
drive circuit. As it is independent of the collector current, frequency compensation can
be optimized over the entire current range. Operational amplifiers which have a current
output – for example, CA 3080 is available as integrated “transconductance amplifier.”
The disadvantage of these types, however, is that they have a relatively large input bias
current.
V1
Vref
Vi OA2
Vo V1
OA1
Diode D in Fig. 11.22 prevents the operational amplifier from being overdriven in
the event of negative input voltages. This insures that transistor T is not damaged by an
excessively high emitter–base reverse voltage and it shortens the recovery time.
One disadvantage of the logarithmic amplifier is its strong temperature dependence.
The reason for this is that VT and ICS vary markedly with temperature. For a temperature
rise from 20 ◦ C to 50 ◦ C, VT increases by 10%, while the reverse current IS multiplies
tenfold. The influence of the reverse current can be eliminated by computing the difference
between two logarithms. We employ this principle in Fig. 11.23, where the differential
amplifier stage T1 , T2 is used to find the logarithm. In order to examine the operation
of the circuit, we determine the current sharing in the differential amplifier stage. From
Kirchhoff’s voltage law (KVL), it follows:
V1 + VBE 2 − VBE 1 = 0
IC 1 = IS e VT
VBE 2
IC 2 = I S e VT
and therefore
1V
IC 1
= e VT (11.27)
IC 2
From Fig. 11.23, we get the additional equations
Vi Vref R2
IC 2 = IC 1 = V1 = Vo ,
R1 R1 R3 + R 2
if R2 is not chosen to be too large. By substitution, we obtain the output voltage
R3 + R 2 Vi
Vo = −VT ln (11.28)
R2 Vref
The value of R4 does not appear in the result. Its resistance is chosen so that the voltage
across it is smaller than the maximum possible output voltage swing of amplifier OA2.
11.7 Function Networks 743
Logarithmic amplifiers that provide an output voltage of 1 V per decade are frequently
required. To determine the sizes of R2 and R3 for this special case, we rewrite (11.28) in
the form
R3 + R 2 1 Vi Vi
Vo = −VT · · lg = − 1 V lg
R2 lg e Vref Vref
With VT = 26 mV, the resulting condition is
R3 + R2 1 V · lg e
= ≈ 16.7
R2 VT
If we select R2 = 1 k, then R3 = 15.7 k.
With regard to the frequency compensation of both amplifiers, the same argument
holds as for the previous circuit. C1 and C2 are the additional compensation capacitors.
The temperature effect of VT can be offset by letting resistor R2 have a positive, or R3 a
negative, temperature coefficient of 0.3%K −1 . A realization is found in the MAX 4206
from Maxim and the LOG 112 from Burr-Brown. Another solution is to maintain the
differential amplifier at constant temperature using a transistor array with two additional
transistors. One of these is then used as a temperature sensor and the other as a heater. A
suitable transistor array is, for example, the MAT 04 from Analog Devices.
11.7.2
Exponential Function
Figure 11.24 shows an exponential function amplifier whose design is analogous to that
of the logarithmic amplifier in Fig. 11.21. When a negative voltage is applied to the input,
the current flowing through the transistor is given by (11.25),
VBE Vi
−
IC = IS e VT = IS e VT ,
and the output voltage is therefore
Vi
−
Vo = IC R1 = IS R1 e VT .
As with the logarithmic amplifier shown in Fig. 11.23, the temperature stability can
be improved by using a differential amplifier. The appropriate circuit is represented in
Fig. 11.25. Again, from (11.27),
1 V
IC 1
= e VT
IC 2
From Fig. 11.25 we deduce the equations:
Vo Vref R2
IC 1 = IC 2 = V1 = Vi
R1 R1 R3 + R 2
V1
Vo
Vref OA1
OA2
Vi
11.7.3
Computation of Power Functions Using Logarithms
The computation of power expressions of the form
y = xa
11.7 Function Networks 745
Vi Vi
VT Vref
Vi Vref VT Vo
can be performed for x > 0 by means of logarithmic and exponential function amplifiers
because
x a = (eln x )a = ea ln x
The basic arrangement for such a circuit is shown in Fig. 11.26. The equations mentioned
apply to the logarithmic-function amplifier of Fig. 11.23 and the exponential-function
amplifier of Fig. 11.25, where R2 = ∞ and R3 = 0. We therefore obtain the output
voltage
Vi
aVT ln
Vref a
Vi
Vo = Vref e VT = Vref
Vref
The logarithm and the exponential function can be obtained using a single integrated circuit
if so-called multifunction converters are used, such as the AD 538 from Analog Devices.
Involution (raising to the power) by means of logarithms is in principle defined for
positive input voltages only. However, from a mathematical point of view, bipolar input
signals are also permitted for whole-number exponents a. This case can be realized by
using the multipliers described in Sect. 11.8.
11.7.4
Sine and Cosine Functions
The output of a sine-function network should approximate the expression
7 π Vi
Vo = Vo sin · (11.30)
2 V 7i
7i ≤ Vi ≤ +V
within the range −V 7i . For small input voltages
7o · π · Vi
Vo = V
2 V 7i
7o so that near the origin, Vo = Vi . This is the case for
It is advisable to choose a value for V
7o = 2 · V
V 7i (11.31)
π
For small input voltages, the sine-function network must accordingly have unity gain,
whereas at higher voltages the gain must decrease. Figure 11.27 represents a circuit that
fulfills these conditions, based on the principle of piecewise approximation.
For small input voltages, all the diodes are reverse biased, and Vo = Vi , as required.
If Vo rises above Vi , diode D2 becomes forward biased. Vo then increases more slowly
746 11 Operational Amplifier Applications
V1 V2 V3
.
.
. .
Vi .
Vo
.
.
– V1 – V2 – V3
than Vi , because of the voltage divider formed by Rv and R4 . If Vo becomes larger than
V2 , the output of the network is additionally loaded with R5 , so that the rise in voltage is
slowed down even more. Diode D3 finally produces the horizontal tangent at the top of
the sine curve. Diodes D
1 − D3
have the corresponding effects for the negative part of
the since function. Considering that diodes do not become conducting suddenly, but have
an exponential characteristic, low harmonic distortions of Vo can be obtained with only a
small number of diodes.
In order to determine the parameters of the network, we begin by choosing the break-
points of the approximation curve. It can be shown that the first n odd harmonies disappear
if 2n breakpoints are assigned to the following values of the input voltage:
2k 7
Vik = ± Vi , 0<k≤n (11.32)
2n + 1
According to (11.30) and (11.31), the corresponding output voltages are
27 πk
Vok = ± V i sin , 0<k≤n (11.33)
π 2n + 1
Therefore, the slope of the line segment above the kth breakpoint is given as
Vo(k+1) − Vok 2n + 1 π(k + 1) πk
mk = = sin − sin (11.34)
Vi(k+1) − Vik π 2n + 1 2n + 1
For the highest breakpoint, when k = n, the slope becomes zero, as was stipulated earlier
in the qualitative description. The slope m0 must be chosen to be equal to unity.
For reasons of symmetry, no even harmonics appear. With the r.m.s. values of the odd
harmonics present in the waveform, we obtain a theoretical distortion factor of 1.8% if
2n = 6 breakpoints are chosen, this being reduced to 0.8% for 2n = 12. However, as real
diode characteristics do not have sharp breakpoints, the actual distortion is considerably
lower. This is illustrated by the following example.
11.7 Function Networks 747
Vi 2 V1
V1 Vi V2 . V2
Vi . Vi Vo
Vi
the error becomes zero for x = 0, ±0.96, and ±π/2. Between these values, the absolute
error is less than 0.57% of the amplitude. The harmonic distortion is 0.6%. It can be reduced
to 0,25% by a slight variation of the coefficients, and is therefore somewhat smaller than
for the piecewise approximation method using 2 × 3 breakpoints. The lack of breakpoints
is particularly advantageous when the signal is to be differentiated.
For a practical circuit, we define
π Vi Vo
x = · and y =
2 V7i 7o
V
7i = V
Furthermore, we select V 7o and thus obtain, from (11.35),
Vi3 π Vi
Vo = 1.543 Vi − 0.543 7
≈ Vi sin ·
72
V 2 V 7i
i
The block diagram for this operation is represented in Fig. 11.29; the input voltage ampli-
7i is equal to the computing unit E for the multipliers. We shall discuss the analog
tude V
multipliers required in the next section.
Differential Amplifiers
Another method of approximating a sine wave is based on the fact that the function tanh x
has a similar shape for small x. This function can be easily generated using a differential
amplifier stage, as shown in Fig. 11.30. It was shown in Sect. 11.7.1 that, for a differential
amplifier, using (11.27),
i V
IC 1
= e VT and IC 1 + IC 2 ≈ Ik
IC 2
Therefore,
Vi
e VT − 1 Vi
IC 1 − I C 2 = Ik = Ik tanh (11.36)
Vi 2VT
+1e VT
The operational amplifier forms the difference between the two collector currents, such
that
Vo = R2 (IC 1 − IC 2 )
It follows:
Vi
Vo = Ik R2 tanh (11.37)
2VT
11.7 Function Networks 749
Vo
Vi
The quality of the sine approximation is dependent on the peak value V 7i chosen. For
7 7
Vi = 2.8 VT ≈ 73 mV, the error becomes minimal and Vo is 0.86 Ik R2 . However, the error
is still 3%. It can be reduced to 0.02% by providing the differential amplifier with two
additional appropriately biased transistors. This is the operating principle of the AD 639
from Analog Devices, which can be used to produce all the other trigonometric functions
as well as the sine function.
Cosine Function
The cosine function can be generated for values 0 ≤ x ≤ π by means of the sine function
networks that have previously been described. The input voltage Vi , which should be
between zero and Vi max , is first converted to an auxiliary voltage:
V1 = Vi max − 2Vi (11.38)
As can be seen in Fig. 11.31, this equation is already a linear approximation of the cosine
function. For the necessary rounding-off of the curve near the maximum and the minimum,
we apply V1 to the input of a sine function network. As is obvious from Fig. 11.32, the
addition of a summing amplifier is all that is needed to convert a sine function network
into a cosine function network.
V1
Vi max
Vi max Vi
V2 / Vi max
Vi
Vi max Fig. 11.33. Shape of the auxiliary
V1 / Vi max voltages for generating the sine
and cosine functions for
−π ≤ x ≤ π
Voltage V1 approximates the cosine function. For Vi > 0, it is identical to the voltage V1
in Fig. 11.31. For Vi < 0, it is symmetrical about the y-axis. We can therefore use (11.38)
by replacing Vi by |Vi |, and obtain
V1 = Vi max − 2|Vi | (11.39)
The relationships for the sine function are somewhat more complicated, since we must
differentiate between three cases:
⎧
⎪
⎪ −2(Vi + Vi max ) for − Vi max ≤ Vi ≤ − 21 Vi max (11.42a)
⎪
⎨
V2 = 2Vi for − 21 Vi max ≤ Vi ≤ 21 Vi max (11.42b)
⎪
⎪
⎪
⎩
−2(Vi − Vi max ) for 2 Vi max ≤ Vi ≤ Vi max
1
(11.42c)
Such functions are best implemented using the general precision function network,
which is described below.
11.7.5
Arbitrary Function Networks
In Fig. 11.27, a diode network was used for the piecewise linear approximation of functions.
Calculation of the circuit parameters is only possible up to an approximation, because the
forward voltage of the diodes and the loading of the voltage divider chains must be taken
11.7 Function Networks 751
–U
–Vk1 V1
V2 = k1V1 +U
V3
OA1 OA2
–U
V4 = k 0Vi
Vi
OA4 Vo
OA3
OA5 OA6
–Vk2 V6 = k2V5
V5
+U
Fig. 11.34. Arbitrary function network. U = unit reference voltage, for example U = 10 V
into account. Furthermore, the sign of the slope of each linear segment is already defined
by the structure of the network. Therefore, such a circuit can only be optimized for one
particular function, and its parameters cannot easily be changed.
Figure 11.34, on the other hand, represents a circuit that allows the breakpoint and slope
of each individual segment to be set precisely, using a separate potentiometer. The part of
the circuit formed by operational amplifiers OA1 and OA2 permits a segment for positive
input voltages to be formed, while operational amplifiers OA5 and OA6 are effective for
negative input voltages. Amplifier OA4 determines the slope about the origin. The circuit
can be extended for any number of segments by adding further sections that are identical
to those mentioned.
Amplifiers OA2, OA4, and OA6 are connected as bipolar-coefficient circuits, as in
Fig. 11.5 for n = 1. Their gain can be adjusted to values between −1 ≤ k ≤ +1 by the
associated potentiometers; and their output voltages are added by the summing amplifier
OA3. An additional DC voltage can be added by means of potentiometer P3 .
Near zero input voltage, only amplifier OA4 contributes to the output voltage:
V4 = k0 Vi
Both voltages V1 and V5 are zero in this case, because diodes D1 and D4 are reverse biased,
and amplifiers OA1 and OA5 have a feedback path via the conducting diodes D2 and D3 .
When the input voltage becomes greater than Vk 1 , diode D1 is forward biased, and we
obtain
V1 = −(Vi − Vk 1 ) for Vi ≥ Vk 1 ≥ 0
Amplifier OA1 therefore operates as a half-wave rectifier, with a positive bias voltage Vk 1 .
Operational amplifier OA5 behaves correspondingly for negative input voltages:
V5 = −(Vi − Vk 2 ) for Vi ≤ Vk 2 ≤ 0
752 11 Operational Amplifier Applications
V
Vi max
V V
Vo
Vi
Vi max
V
Hence, we obtain the general relationship for the slope of the output voltage Vo as
⎧
⎪
⎪ −k0 + k1 + · · · + km for Vi > Vk m > 0
⎪
⎪
⎨ −k0 + k1 for Vi > Vk 1 > 0
Vo
m = = 10 · −k0 for Vk 2 < Vi < Vk 1 (11.41)
Vi ⎪
⎪
⎪
⎪ −k 0 + k 2 for V i < Vk2 < 0
⎩
−k0 + k2 + · · · + kn for Vi < Vkn < 0
As an example, we shall demonstrate the implementation of the voltage wave-shape
V2 /Vi max in Fig. 11.33. A positive breakpoint at Vk 1 = 21 Vi max and a negative break-
point at Vk 2 = − 21 Vi max are required. According to (11.40b), the slope of the segment
through the origin must have the value m = +2; therefore, k0 = −0.2. Above the positive
breakpoint, the slope must be −2. For this region, we take, from (11.41),
m = 10(−k0 + k1 )
and therefore obtain k1 = −0.4, and correspondingly k2 = −0.4. The shapes of the output
voltage functions that result from this process are shown in Fig. 11.35.
Even if no calibrated potentiometers are available, the network output can be given the
desired shape in a simple way, using the following procedure. Initially, all the breakpoint
voltages and slopes are set to their maximum values and the input voltage is made zero.
This insures that |Vi | < |Vki |. Only the zeroing potentiometer P3 affects the output; it is
used to adjust the output voltage Vo (Vi = 0) to the desired value. In the next step, Vi is
made equal to Vk1 and P4 is set so that Vo (Vi = Vk1 ) assumes the level required. The factor
k0 is now defined. P1 is then adjusted to a point at which the output voltage just begins to
change; this occurs when the setting of P2 corresponds to Vk1 . Now, Vi is set to the value
of the next higher breakpoint (or to the end of the range, if there is no higher breakpoint),
and P2 is adjusted so that Vo , attains the desired value. In this way, k1 is defined. The
remaining breakpoints and slopes are dealt with in the same manner.
In cases in which no calibrated potentiometers are needed for the adjustment of the
segment slopes, the circuit may be simplified. One can replace the bipolar-coefficient cir-
cuits by simple potentiometers that are connected to a multiple-input subtracting amplifier,
as shown in Fig. 11.36. The subtracting amplifier consists of operational amplifiers OA2
and OA3, and is based on the principle of Fig. 11.2.
11.8 Analog Multipliers 753
–U +U
–Vk1 V1 V3
–U
OA1
OA2
Vi
OA5
–Vk2
Vo
V5 OA3
+U
11.8
Analog Multipliers
So far, we have described circuits for addition, subtraction, differentiation, and integration.
Multiplication could be carried out only if a constant factor was involved. Below, we deal
with the most important principles for the multiplication and division of two variable
voltages.
11.8.1
Multipliers with Logarithmic Amplifiers
Multiplication and division can be reduced to an addition and subtraction of logarithms:
xy
= exp[ln x + ln y − ln z]
z
The function can be implemented by using three logarithmic amplifiers, one exponential
function amplifier, and one adder/subtractor circuit. The latter can be eliminated by using
the inputs of the differential amplifier for the exponential function amplifier in Fig. 11.25
to perform the subtraction, and by considering the fact that the terminal for the reference
voltage can be used as an additional signal input.
The logarithmic amplifiers shown in Fig. 11.37 produce the expressions
Vy Vz
V1 = −VT ln and V2 = −VT ln
IS R 1 IS R1
The exponential function generator therefore provides an output voltage of the form
V2 −V1
Vx Vy
Vo = Vx e VT =
Vz
754 11 Operational Amplifier Applications
Vx
OA3 OA4 Vo
Vy Vz
OA1 OA2
We can see that, in this case, not only the reverse saturation currents IS but also voltage
VT are eliminated, thereby obviating the need for temperature compensation. However,
it is essential that the four transistors have the same characteristics and are at the same
temperature. They must therefore be integrated on one chip.
An inherent disadvantage of this method is that all input voltages must be positive, and
may not even be zero. Such multipliers are called one-quadrant multipliers.
Multipliers such as those shown in Fig. 11.37 can be implemented using multifunction
converters such as the AD 538 (Analog Devices).
11.8.2
Transconductance Multipliers
As shown in Chap. 2.11 on page 42, the transconductance of a transistor is defined as
dIC IC
gm = =
dVBE VT
and is therefore proportional to the collector current.
The variation of the collector current is then proportional to the product of the variation
in the input voltage and the quiescent collector current. This property is made use of for
multiplication in the differential amplifier shown in Fig. 11.38.
The operational amplifier evaluates the difference between the collector currents:
Vo = Rz (IC 2 − IC 1 ) (11.42)
Applying a negative voltage Vy and setting Vx to zero, the currents through both transistors
are equal, and the output voltage remains zero. If Vx , is made positive, the collector current
through T1 rises and that of T2 falls; the output voltage is negative. Correspondingly, Vo
becomes positive when Vx is negative. The resulting difference in collector currents is
greater, the larger the emitter current, i.e. the higher the value |Vy |. It can therefore be
assumed that Vo is at least approximately proportional to Vx · Vy . For a more precise
11.8 Analog Multipliers 755
Vo
calculation, we determine the current sharing within the differential amplifier stage. As
was shown in Sect. 11.7.4, (11.36) states that
Vx
IC 1 − IC 2 = IE tanh (11.43)
2VT
A power series expansion up to the fourth order gives
Vx Vx3
IC 1 − IC 2 = IE − (11.44)
2VT 24VT3
Hence,
Vx
IC 1 − IC 2 ≈ IE · for |Vx | VT (11.45)
2VT
If |Vy | VBE , then
Vy
IE ≈ −
Ry
Substitution in (11.45) gives, in conjunction with (11.42), the result
Rz Vx Vy
Vo ≈ · (11.46)
Ry 2VT
If the error in (11.46) is not to exceed 1%, the voltage Vx must be |Vx | < 0.35 VT ≈ 9 mV.
Because of the small value of Vx , transistors T1 and T2 must be closely matched to prevent
the offset voltage drift from affecting the result.
For the correct operation of the circuit, it is necessary that Uy is always negative,
while the voltage Ux , may have either polarity. Such a multiplier is called a two-quadrant
multiplier.
There are several properties of the transconductance multiplier shown in Fig. 11.38 that
can be improved. In deducing the output equation (11.46), we had to use the approximation
that |Vy | VBE ≈ 0.6 V. This condition can be dropped if resistor Ry is replaced by a
controlled current source for which IE is proportional to Vy .
A further disadvantage of the circuit shown in Fig. 11.38 is that |Vx | must be limited
to small values in order to minimize the error. This can be avoided by not applying Vx ,
directly but, rather, its logarithm.
An expansion to a four-quadrant multiplier – that is, a multiplier for input voltages of
either polarity – is possible if a second differential amplifier stage is connected in parallel,
756 11 Operational Amplifier Applications
Vz
Vo
VD1 VD2
V1
Vx Vy
the emitter current of which is controlled by Vy , in opposition to that of the first transistor
pair.
All these aspects are considered in the four-quadrant transconductance multiplier
shown in Fig. 11.39. The differential amplifier stage T1 , T2 is the same as that of Fig. 11.38.
It is supplemented symmetrically by the differential amplifier T1
, T2
. Transistors T5 , T6
form a differential amplifier with current feedback. The collectors represent the outputs of
two current sources that are controlled by Vy simultaneously but in opposition, as required:
Vy Vy
I5 = I8 + , I6 = I8 − (11.47)
Ry Ry
For the difference between the collector currents of the two differential amplifier stages
T1 , T2 and T1
, T2
we obtain, by analogy with the previous circuit,
V1 Vy V1
I1 − I2 = I5 tanh = I8 + tanh (11.48)
2VT Ry 2VT
V1 Vy V1
I1
− I2
= I6 tanh = I8 − tanh (11.49)
2VT Ry 2VT
As before, the operational amplifier evaluates the difference between the collector currents
according to
I = (I2 + I1
) − (I2
+ I1 ) = (I1
− I2
) − (I1 − I2 ) (11.50)
By subtracting (11.48) from (11.49), it follows:
2Vy V1
I = − tanh (11.51)
Ry 2VT
where Vy may now have either polarity. By expanding this expression into a series, we can
see that the same approximation to multiplication is involved as for the previous circuit.
11.8 Analog Multipliers 757
We shall now examine the relationship between V1 and Vx . Two transistors are con-
nected as diodes (transdiodes), D1 and D2 , and these are used to form the logarithm of the
input signals:
I4 I3
V1 = VD 2 − VD 1 = VT ln − VT ln
ICS ICS
Hence,
Vx
I7 −
I4 Rx
V1 = VT ln = VT ln (11.52)
I3 Vx
I7 +
Rx
Substitution in (11.51) gives the current difference:
2Vx Vy
I = (11.53)
Rx Ry I7
From this, the operational amplifier configured as a current subtractor forms the output
voltage
2Rz Vx Vy
Vo = I Rz = · V x Vy = (11.54)
Rx Ry I7 U
where U = Rx Ry I7 /2Rz is the computing unit. This is usually chosen to be 10 V. Good
temperature compensation is attained, since VT cancels out. Either (11.53) or (11.54) is
obtained without recourse to power expansion, and therefore a considerably larger range
of input voltages Vx , is permissible. The limits of the input range are reached when one of
the transistors in the controlled current source is turned off. Therefore,
|Vx | < Rx I7 and |Vy | < Ry I8
If the currents I7 are controlled by a further input voltage V7 , simultaneous division and
multiplication is possible. However, the usable range for I7 is limited, because I7 influences
all the quiescent potentials within the multiplier and also the permissible range for Vx .
A simpler way of achieving division is to open the connection between Vo and Vz ,
and to link the voltages Vy and Vo , instead. Because of the resulting feedback, the output
voltage assumes a value such that I = Vz /Rz . Therefore, from (11.53),
2Vx Vy Vz
I = =
Rx Ry I7 Rz
Thus the new output voltage is
Rx Ry I7 Vz Vz
Vo = Vy = · = U (11.55)
2Rz Vx Vx
However, stability is only guaranteed if Vx , is negative; otherwise, the negative feedback
becomes positive. The signal Vz , on the other hand, can have either polarity, and therefore
the circuit is a two-quadrant divider. The limitation on the sign of the denominator is not
peculiar to this arrangement, but is common to all divider circuits.
Transconductance multipliers operating on the principle shown in Fig. 11.40 are avail-
able as monolithic integrated circuits. The achievable accuracy is 0.1% referred to com-
puting unit U ; that is, for a computing unit of U = 10 V. As we shall see in Sect. 11.8.5,
the simple types require four trimmers in order to achieve this degree of accuracy. High
758 11 Operational Amplifier Applications
accuracy types are already laster trimmed by the manufacturer, so that external adjustment
is generally unnecessary.
The 3 dB bandwidth is of the order of 1 MHz and beyond, at which frequency the
computing error is already 30%. As a deviation of this magnitude is unacceptable in the
majority of applications, a better reference point is the frequency at which the output
voltage is reduced by 1%.
Vx
Vo = U
Vz
Assuming that Vz > 0 and |Vx | < Vz , two auxiliary voltages that are always positive,
1 1
V1 = Vz − Vx V2 = Vz + Vx (11.56)
2 2
can be generated. The logarithms of these two voltages are computed according to the
block diagram of Fig. 11.41, each by means of the simple logarithmic amplifier shown in
11.8 Analog Multipliers 759
V1 V1 V3
Vx Vz Vx – VT
IS R1
V3–V4
2VT Vo
V2
Vz Vx – VT
V2 IS R1 V2
Vz
Fig. 11.21. Using a differential amplifier stage, as in Fig. 11.38, the hyperbolic tangent of
the difference between the output voltages V3 and V4 is calculated so that
VT ln(V2 /V1 )
Vo = Rz IE tanh
2VT
Therefore, using (11.56),
Rz IE Vx
Vo = ·
2 Vz
With this method, an accuracy of 0.1% of the computing unit U can be obtained over a
dynamic range of 1 : 1, 000.
11.8.3
Multipliers Using Electrically Controlled Resistors
A voltage can be multiplied by a constant using a simple voltage divider. Analog multipli-
cation is possible if, by employing closed-loop control, if one insures that this constant is
proportional to a second input voltage.
The principle of such a circuit is illustrated in Fig. 11.42. The arrangement contains two
identical coefficient elements Kx and Kz , the output voltages of which are proportional to
their input voltages. Their constant of proportionality k can be controlled by voltage V1 .
Due to the feedback via Kz , the output voltage V1 of the operational amplifier assumes a
level such that kVz = Vy , and this results in k = Vy /Vz . If the voltage Vx , is applied to
the second coefficient element Kx , its output voltage becomes
Vx Vy
Vo = kVx =
Vz
Vy
Vz
Vz OA1
V1
Fig. 11.42. Multiplier using controlled coefficients
Vx Vx Vy
Vx Vo Vo = for Vz > 0
Vz
760 11 Operational Amplifier Applications
Vy
Vz
Vz
OA1
Vo
OA2
Vx Vo
Vx
Voltage Vz , must be larger than zero, so that the negative feedback does not become positive.
Whether voltage Vy is allowed to go both positive and negative depends on the design of
the coefficient elements. If the latter permits bipolar coefficients, Vy may also be bipolar.
The voltage Vx , can be bipolar in any case. It has the additional advantage that it is not
transferred through balancing amplifier OA1. Consequently, very high bandwidths can be
achieved for Vx .
FETs can be employed as electrically controlled resistors, as in the circuit shown in
Fig. 11.43. Amplifier OA1 operates as a controller for adjusting the coefficients. Its output
voltage causes the resistance RDS to vary, so that
αVz Vy
+ = 0
RDS R4
Hence,
Vz
RDS = −αR4
Vy
The output voltage of operational amplifier OA2 is
R3 R 3 V x Vy
Vo = −α Vx = ·
RDS R4 Vz
In order that FETs may be operated as resistors, the voltage across them must be kept
below approximately 0.5 V. Voltage dividers R1 , R2 provide the necessary attenuation.
The resistors R5 make the voltage–current characteristic of the FET more linear, as is
described in Sect. 3.1.3. In order to prevent any reactive effect of the control voltage V1 on
the input signals Vz and Vx , the circuit incorporates the two additional source followers,
T3 and T4 respectively. The magnitude of their gate–source voltage is irrelevant, as it is
controlled by operational amplifier OA1. It is merely essential for them to be well matched.
Double FETs should therefore be used.
To insure negative feedback in the control circuit, Vz must be positive. As only positive
coefficients can be realized using the simple coefficient elements shown in Fig. 11.43, Vy
must always be negative to allow balancing. However, Vx , can have either polarity.
11.8 Analog Multipliers 761
In order to achieve a high degree of accuracy, FETs T1 and T2 should be well matched
over a wide range of resistances. This requirement can only be met using monolithic dual
FETs.
11.8.4
Adjustment of Multipliers
A multiplier should conform to the expression
Vx Vy
Vo =
U
where U is the computing unit; for example, −10 V. In practice, there is a small offset
voltage superposed on any terminal voltage. Therefore, in general,
1
Vo + Vo 0 = (Vx + Vx 0 )(Vy + Vy 0 )
U
Thus,
Vx Vy V y Vx 0 + V x V y 0 + V x 0 V y 0
Vo = + − Vo 0 (11.57)
U U
The product Vx , Vy must be zero whenever Vx or Vy is zero. This is only possible if the
parameters Vx O , Vy O , and Vo O become zero independently. Therefore, three trimmers
are essential for compensating the offset voltages. A suitable trimming procedure is as
follows. First, Vx is made zero. Then, according to (11.57),
Vy Vx 0 + Vx 0 Vy 0
Vo = − Vo 0
U
When voltage Vy is now varied, the output voltage also changes because of the term Vy Vx O .
The nulling circuit for Vx is adjusted in such a way that, despite variation of Vy , a constant
output voltage is obtained; Vx O is then zero.
In a second step, Vy is nulled and Vx is varied. In the same way as above, the offset of
Vy can now be compensated. Thirdly, Vx , and Vy are made zero and the third trimmer is
adjusted such that the output offset Vo O becomes zero.
A fourth trim potentiometer may often be necessary for adjusting the constant of
proportionality, U , to the desired value.
11.8.5
Expansion to Four-Quadrant Multipliers
There are cases in which one- and two-quadrant multipliers have to be operated with
input voltages of a polarity for which they are not designed. The most obvious remedy
would then be to invert the polarity of the input and output of the multiplier whenever the
impermissible polarity combination occurs. However, this method involves a large number
of components and is not particularly fast. It is more convenient to add constant voltages
Vxk and Vyk to the input voltages Vx and Vy , so that the resulting input voltages remain
positive. Then, for the output voltage, we get
(Vx + Vxk )(Vy + Vyk )
V3 =
U
762 11 Operational Amplifier Applications
V1
Vx 0.5
0.5 V1 V2 V3
U Vx Vy
0.5 U Vo
Vy 0.5 U U
V2
Hence,
V x Vy Vxk Vyk Vxk Vyk
= V3 − Vy − Vx −
U U U U
It follows that a constant voltage – and also two voltages, each of which is proportional to
an input voltage – must be subtracted from the output voltage of the multiplier.
The block diagram of the resulting arrangement is shown in Fig. 11.44. The constant
voltage and coefficients (Vxk = 0.5 U ) are selected such that the range of control is fully
exploited. If the input voltage Vx is within −U ≤ Vx ≤ +U , the range for the voltage
V1 = 0.5 Vx + 0.5 U is 0 ≤ V1 ≤ U and V2 = 0.5 Vy + 0.5 U is 0 ≤ V2 ≤ U . Therefore,
the output voltage obtained is:
1
2 (Vx + U ) · 21 (Vy + U ) Vx Vy
Vo = 4 · − V x − Vy − U =
U U
11.8.6
Multiplier as a Divider or Square Rooter
Figure 11.45 illustrates a method by which a multiplier without a division input can be used
as a divider. Because of negative feedback, the output voltage of the operational amplifier
finds a level such that
V o Vz
= Vx
U
Thus, the circuit evaluates the quotient Vo = U Vx /Vz , but only as long as Vz > 0. For
negative denominators, the feedback is positive.
A multiplier can be employed as a square-rooter by operating it as a squarer and
inserting it in the feedback loop of an operational amplifier, as shown in Fig. 11.46. The
output voltage settles at a level such that
Vo2
= Vi , hence Vo = U Vi
U
Vo Vz
Vz U
Vo Fig. 11.45. Multiplier used as a
divider
Vx Vx
Vo = U for Vz > 0
Vz
11.9 Transformation of Coordinates 763
Vo2
U
Vo Fig. 11.46. Multiplier used as
square-rooter
Vi
Vo = U Vi for Vi > 0
Correct operation is insured only for positive input and output voltages. Difficulties may
arise if the output becomes momentarily negative; for example, at switch-on. In such a
case, the squarer causes a phase inversion in the feedback loop so that positive feedback
occurs, and the output voltage becomes more negative until it reaches the negative level of
output saturation. The circuit is then said to be in “latch-up” and is inoperable. Therefore,
a diode must be used to insure that the output voltage cannot become negative.
11.9
Transformation of Coordinates
Cartesian as well as polar coordinates play an important role in many technical applications.
Therefore, in this section we discuss some circuits that allow transformations from one
coordinate system to the other.
11.9.1
Transformation from Polar to Cartesian Coordinates
In order to implement the transformation equations
x = r cos ϕ,
(11.58)
y = r sin ϕ
V V1
V U V1 Vr
U
U Vy
Vr
V2 Vr Vx
V V2
U U
U
11.9.2
Transformation from Cartesian to Polar Coordinates
Inversion of the transformation equation (11.58) yields
'
r = x 2 + y 2 or Vr = Vx2 + Vy2 , (11.60)
and
y U Vy
ϕ = arctan or Vϕ = arctan (11.61)
x π Vx
respectively. The magnitude, Vr , of the vector can be computed according to the block
diagram shown in Fig. 11.48, using two squarers and one square-rooter. A more simple
circuit, which also has a larger input voltage range, can be deduced by applying a few more
mathematical operations. From (11.60),
Vr2 − Vy2 = Vx2 ,
(Vr − Vy )(Vr + Vy ) = Vx2
Hence,
Vx2
Vr = + Vy
Vr + V y
The implicit equation for Vr can be implemented by means of a multiplier with a division
input, as shown in Fig. 11.49. The summing amplifier S1 evaluates the expression
V1 = Vr + Vy
and, therefore,
Vx2
V2 =
Vr + V y
Vx2
Vx U
V1
UV1 Fig. 11.48. Computation of
Vr the vector magnitude
Vy Vy2 '
U Vr = Vx2 + Vy2
11.9 Transformation of Coordinates 765
S2
VVxx22 V2
Vx V1 Vr VVxx22++ VVyy2
V1
S1 Fig. 11.49. Simplified
Vy
circuit for computing the
vector magnitude
In order to obtain Vr , this voltage V2 is added to the input voltage Vy using the summing
amplifier S2 .
The fact that voltage Vy must always be positive may be easily explained by reference
to the special case Vx = 0. We thus have V2 = 0 and Vr = Vy . This is the correct solution
only for positive values of Vy . In addition, as practical dividers cannot handle a sign change
in the denominator, it is necessary to form the absolute value for bipolar values of Vy using,
for example, the circuit shown in Fig. 20.20. This does not limit the vector calculation, as
the intermediate variable Vy2 is positive in each case.
The simplest implementation of a vector meter is one in which multiplication and
division are performed via logarithms, because both operations can be performed using a
single circuit, as shown in Fig. 11.37. However, in this case it is also necessary to form the
absolute value of Vx .
This is not required when using transconductance multipliers, as these generally allow
four-quadrant operation. However, in this case separate circuits are required for multipli-
cation and division. It is advisable, as shown in Fig. 11.50, to perform division before
multiplication, as the dynamic range would otherwise be reduced due to the presence of
the variable Vx2 .
Vx
Vx V3 Vx Vx2
V1 V2 =
V1 V3 V1
Fig. 11.50. Transconductance multipliers to compute the vector magnitude according to the
method in Fig. 11.49
Chapter 12:
Controlled Sources and Impedance Converters
In linear network synthesis, not only passive components are used, but also idealized active
elements such as controlled current and voltage sources. In addition, idealized converter
circuitry, such as the negative impedance converter (NIC), the gyrator, and the circulator,
is often employed. In the following sections, we describe the most common ways of
implementing these circuits.
12.1
Voltage-Controlled Voltage Sources
A voltage-controlled voltage source is characterized by having an output voltage V2 that is
proportional to the input voltage V1 . It is therefore nothing more than a voltage amplifier.
Ideally, the output voltage should be independent of the output current and the input current
should be zero. Hence, the transfer characteristics are:
I1 = 0 · V1 + 0 · I2 = 0,
V2 = Av V1 + 0 · I2 = Av V1
In practice, the ideal source can only be approximated. Considering that the reaction of
the output on the input is usually negligibly small, the equivalent circuit of a real source
is as shown in Fig. 12.1, and its transfer characteristics are:
1
I1 = r V1 + 0 · I 2
i (12.1)
V 2 = A v V 1 − r o I2
The internal voltage source shown is assumed to be ideal. The input resistance is ri and
the output resistance is ro .
Voltage-controlled voltage sources of low output resistance and of defined, but ad-
justable, gain have already been described in Chap. 5 in the form of inverting and nonin-
verting amplifiers. They are shown for the sake of completeness in Figs. 12.2 and 12.3.
It is easy to obtain output resistances of far less than 1 and therefore to approach the
ideal behavior fairly closely. It should be noted, however, that the output impedance is
somewhat inductive; in other words, it rises with increasing frequency (see Chap. 5).
The input resistance of an electrometer amplifier is very high. At low frequencies, one
easily attains values in the gigaohm range, and hence very nearly ideal conditions. The high
(incremental) input resistance must not lead us to overlook the additional errors that may
arise due to the constant input bias current IB , particularly when the output resistance of
the signal source is high. In critical cases, an amplifier with FET input should be employed.
V1 V1 V2
i
Fig. 12.1. Low-frequency equivalent circuit of a
voltage-controlled voltage source
768 12 Controlled Sources and Impedance Converters
V1
V2
V1
V2
For low-resistance signal sources, the inverting amplifier circuit in Fig. 12.2 can be used,
as its low input resistance R1 then causes no error. The advantage is that no inaccuracies
can arise due to common-mode signals.
12.2
Current-Controlled Voltage Sources
The equivalent circuit of a current-controlled voltage source, shown in Fig. 12.4, is identical
to that of the voltage-controlled voltage source in Fig. 12.1. The only difference between the
two sources is that the input current is now the controlling signal. It should be influenced
by the circuit as little as possible, a condition fulfilled in the ideal case when ri = 0.
Disregarding the effect of the output on the input, the transfer characteristics are:
V1 = ri I1 + 0 · I2 V1 = 0
V2 = RI1 − ro I2 ⇒ V2 = RI1 (12.2)
(real) (ideal, ri = ro = 0)
When implementing this circuit as shown in Fig. 12.5, we use the fact that the summing
point of an inverting amplifier represents virtual ground. Because of this, the input resis-
tance is low, as required. The output voltage becomes V2 = −RI 1 if the input bias current
of the amplifier is negligible compared with I1 . If very small currents I1 are to be used
as control signals, an amplifier with FET input must be employed. Additional errors may
occur due to the offset voltage—and these will increase if the output resistance Rg of the
signal source is low, since the offset voltage is amplified by a factor of (1 + R/Rg ).
For the output impedance of the circuit, the same conditions hold as for the previous
circuit, where the loop gain g is dependent on the output resistance Rg of the signal source
ro Rg
Zo = with g = k AD = A
g R + Rg D
A current-controlled voltage source with a floating input is discussed in Sect. 20.2.1.
12.3 Voltage-Controlled Current Sources 769
V1 V2 V1
i
V2
Fig. 12.4. Low-frequency equivalent circuit Fig. 12.5. Current-controlled voltage source
of a current-controlled voltage source
Ideal transfer characteristic: V2 = −RI1
R
Input impedance: Zi =
AD
ro
Output impedance: Zo =
g
12.3
Voltage-Controlled Current Sources
For ri → ∞ and ro → ∞, one obtains the ideal current source. The parameter gm is the
forward transconductance or transfer conductance.
12.3.1
Current Sources for Floating Loads
In inverting and electrometer amplifiers, the current through the feedback resistor is I2 =
V1 /R1 and is therefore independent of the voltage across the feedback resistor. Both circuits
can thus be used as current sources if the load RL is inserted in place of the feedback resistor,
as shown in Figs. 12.7 and 12.8.
The same conditions obtain for the input impedance as for the corresponding voltage-
controlled voltage sources in Figs. 12.2 and 12.3.
V1 i gmV1 V2
o
Fig. 12.6. Low-frequency equivalent circuit of a
voltage-controlled current source
770 12 Controlled Sources and Impedance Converters
= I1
V1 Vo
VN
V2 V2
V1
Vo
For a finite open-loop gain AD of the operational amplifier, the output resistance
assumes finite values only, as the potential difference VD = VP − VN does not remain
precisely zero. To determine the output resistance, we take the following relationships
from Fig. 12.7,
V1 − VN Vo
I1 = I2 = VN = − V2 = VN − Vo
R1 AD
and obtain
V1 V2 V1 V2
I2 = − ≈ −
R1 R1 (1 + AD ) R1 AD R1
and therefore, for the output resistance,
∂V2
ro = − = AD R1 (12.5)
∂I2
which is thus proportional to the differential gain of the operational amplifier.
Since the open-loop gain AD of a frequency-compensated operational amplifier has a
fairly low cutoff frequency (e.g., fcA ≈ 10 Hz for the 741 type), one must take into account
that AD is complex even at low frequencies. Equation (12.5) must then be rewritten in its
complex form:
AD fT
Z o = AD R1 = R1 ≈ −j R1 (12.6)
ω f
1+j ωcA
This output impedance may be represented by a parallel connection of a resistor ro and a
capacitor Co , as the following rearrangement of (12.6) shows:
9
1 9 1
Zo = = ro 9
9 j ωC , where (12.7)
1 jω
+ o
AD R1 AD R1 ωcA
1 1
ro = AD R1 and Co = = .
AD R1 ωcA 2π R1 fT
12.3 Voltage-Controlled Current Sources 771
With an operational amplifier that has AD = 105 and fT = 1 MHz, one obtains, for
R1 = 1 k, the result ro = 100 M and Co = 159 pF. For a frequency of 10 kHz, the
magnitude of the output impedance is reduced to 100 k.
The same considerations apply to the output impedance of the circuit shown in Fig. 12.8.
As far as their electrical data are concerned, the two current sources in Figs. 12.7 and 12.8
are well suited for many applications. However, they have a serious technical disadvantage:
the load RL must be floating – that is, it must not be connected to a fixed potential –
otherwise the amplifier output or the N-input is short-circuited. The following circuits
overcome this restriction.
12.3.2
Current Sources for Grounded Loads
The principle of the current source in Fig. 12.9 is based on the fact that the output current
is measured as the voltage drop across R1 . The output voltage of the operational amplifier
finds a value such that this voltage is equal to a given input voltage. In order to determine
the output current, we apply KCL to the N-input, the P-input, and the output. Thus,
Vo − VN VN V1 − VP V2 − VP
− = 0 + = 0
R2 R3 R1 + R 2 R3
Vo − V2 VP − V2
+ − I2 = 0
R1 R3
Since VN = Vp , we obtain the output current:
V1 R22 − R32
I2 = + V2
R1 R1 R3 (R2 + R3 )
We can see that the output current for R2 = R3 is independent of the output voltage. The
output resistance then becomes ro = ∞ and the output current is I2 = V1 /R1 . In practice,
resistance R1 is made low so that the voltage across it remains of the order of a volt.
Resistance R2 is selected large in comparison with R1 so that the operational amplifier and
the voltage source V1 are not unnecessarily loaded. The output resistance of the current
source can, at low frequencies, be adjusted to infinity even for a real operational amplifier,
by slightly varying R3 . The internal resistance Rg of the controlling voltage source is
V1
Vo
V2
V1
V2
OA1 OA2
12.3.3
Precision Current Sources Using Transistors
Simple single-ended current sources employ a bipolar transistor or a field effect transistor
to supply loads that have one terminal connected to a constant potential. Such circuits have
been described in the introductory Chap. 4.1.1 on page 277.
Their disadvantage is that the output current is affected by VBE or VGS and therefore
cannot be precisely defined. An operational amplifier can be used to eliminate this effect.
Figure 12.11 shows the relevant circuits for a bipolar transistor and for a FET. The output
voltage of the operational amplifier finds a value such that the voltage across the resistance
12.3 Voltage-Controlled Current Sources 773
V2 V2
V1 V1
R1 equals V1 . Obviously, this holds for positive voltages only, as otherwise the transistors
are off. Since the current through R1 is V1 /R1 , the load current is:
V1 B V1 1
for the bipolar transistor: I2 = ≈ 1−
R1 1 + B R1 B
V1
and for the FET: I2 =
R1
The difference between these currents is due to the fact that, in the bipolar transistor, part
of the emitter current escapes through the base. As the current transfer ratio B is dependent
on VCE , the current IB also changes with the output voltage V2 . As shown in Sect. 4.1.1 on
page 277, this effect limits the output resistance to the value βrCE , even if the operational
amplifier is assumed to be ideal.
The effect of the finite current transfer ratio can be reduced if the bipolar transistor is
replaced by a Darlington circuit. It can be virtually eliminated by using a FET, because
the gate current is extremely small. The output resistance of the circuit in Fig. 12.11b is
limited only by the finite gain of the operational amplifier. It can be determined by the
following relationships, obtained directly from the circuit for V1 = constant:
dVDS ≈ −dV2
dVGS = dVG − dVS = −AD R1 dI2 − R1 dI2 ≈ −AD R1 dI2
Using the basic equation (3.9),
1
dI2 = gm dVGS + dVDS
rDS
we obtain for the output resistance
dV2
ro = − = rDS (1 + AD gm R1 ) ≈ µAD R1 (12.8)
dI2
It is thus greater by a factor µ = gm · rDS ≈ 50 than that of the corresponding current
source in Fig. 12.8, which uses an operational amplifier without a field effect transistor.
774 12 Controlled Sources and Impedance Converters
V2
Using the same values as in the example given for the circuit in Fig. 12.8, we obtain the
very high output resistance of approximately 5 G. Because of the frequency dependence
of the open-loop gain AD , this value holds only for frequencies below the cutoff frequency
fcA of the operational amplifier. For higher frequencies, we must take into account the fact
that the differential gain is complex and, instead of (12.8), obtain the output impedance
9
AD 1 9 1
Z o = AD µR1 = µR1 = = r0 99 j ωC (12.9)
ω 1 jω
1+j + o
ωcA µAD R1 µAd R1 ωcA
A comparison with (12.6) and (12.7) shows that this impedance is equivalent to connecting
a resistance ro = µAD R1 in parallel with a capacitance Co = 1/µAD R1 ωcA . For the
practical example mentioned, we obtain Co = 3 pF. The drain-gate capacitance of the
FET, which is of the order of a few picofarads, will appear in parallel.
The circuit shown in Fig. 12.11b can be modified by connecting the input voltage
directly to R1 and by grounding the P-input terminal. Figure 12.12 shows the resulting
circuit. To insure that the FET is not turned off, V1 must always be negative. In contrast to
the circuit in Fig. 12.11b, the control voltage source is loaded by the current I1 = I2 .
When a current source is required, the output current of which flows in the opposite
direction to that in the circuit of Fig. 12.11b, the n-channel FET is simply replaced by a
V1 V1
V2 V2
Fig. 12.13. A current source using a Fig. 12.14. A current source with a
p-channel FET quasi-p-channel FET
V1 V1
Output current: I2 = − for V1 < 0 Output current: I2 = − for V1 < 0
R1 R1
Output resistance: ro = µAD R1 Output resistance: ro = AD R1
12.3 Voltage-Controlled Current Sources 775
p-channel FET, as shown in the circuit given in Fig. 12.13. If no p-channel FET is available,
the arrangement in Fig. 12.14 can also be used. In contrast to the previous circuits, the
source terminal here serves as the output. However, this does not affect the output current,
since it is controlled, as before, by the voltage across R1 . Here the inputs of the opamp must
be exchanged because the FET acts as an inverting amplifier. As the source electrode is the
current output the amplifier has to transform the low output resistance ro = 1/gm of the
source follower. Another disadvantage of the arrangement is that the output of the opamp
has to follow the load voltage V2 and therefore it needs a large output voltage sewing. In
the previous circuits the opamp must only compensate the voltage drop of the gate-source
voltage.
V+ V−
I2 = ID1 − ID2 = + = 0 for V + = −V −
4R1 4R1
For positive input voltages V1 , current ID2 increases by V1 /4R1 , whereas ID1 decreases
by the same amount. Therefore, one obtains a negative output current
V1
I2 = −
2R1
For negative input voltages, ID2 decreases while ID1 becomes larger, this resulting in a
positive output current.
The circuit has a rather poor zero stability. This is because the output current is itself the
difference between the two relatively large currents, ID1 and ID2 , which are additionally
affected by changes in the supply voltages.
OA1
V1 V2
OA2
V3 V3
OA2
OA1
V1
V2
OA3
V4 V4
Fig. 12.16. A bipolar FET current source for large output currents
R2
Output current: I2 = − V1
R1 R3
The circuit shown in Fig. 12.16 is considerably better in this respect. It differs from the
former circuit in that it uses a different kind of control. The two output stages are controlled
by the currents I3 and I4 , which flow in the supply terminals of the amplifier OA1. For the
drain currents,
V3 R2 V4 R2
ID1 = = I3 , ID2 = = I4 (12.10)
R1 R1 R1 R1
The output stages therefore operate as current mirrors. Hence, the output current is:
R2
I2 = ID1 − ID2 = (I3 − I4 ) (12.11)
R1
Amplifier OA1 operates as a voltage follower. Therefore, the voltage across resistor
R3 is equivalent to the input voltage V1 . Thus, its current is:
I5 = V1 /R3 (12.12)
For further processing of this signal, use is made of the fact that the operational amplifier
can be regarded as a current node for which, by applying KCL, the sum of the currents
must equal zero. As the input currents are negligible and as there is usually no ground
connection, the following relationship holds with very good accuracy:
I5 = I3 − I4 (12.13)
Substitution in (12.12) and (12.11) yields the output current:
R2 V1
I2 = V1 = for R2 = R3
R1 R 3 R1
For zero input, I5 = 0 and I3 = I4 = IQ , where IQ is the quiescent current flowing
in the supply leads of the amplifier OA1. It is small in comparison with the maximum
possible output current, I5 , of the amplifier. For a positive input voltage, I3 ≈ I5 I4 .
12.3 Voltage-Controlled Current Sources 777
The output current I2 is then supplied virtually only from the upper output stage, whereas
the lower stage is disabled. The inverse is true for a negative input voltage difference. The
circuit is therefore of the class AB push–pull type. Since the quiescent current in the output
stage is
R2
ID 1 Q = ID 2 Q = IQ (12.14)
R1
and is small relative to the maximum output current. The output current at zero input
signal is now determined by the difference of small quantities. This results in a very good
zero-current stability. A further advantage is the high efficiency of the circuit, this being
of special interest if the circuit is to be designed for high output currents. For this reason,
the device selected for OA1 is an operational amplifier with a low quiescent current drain
IQ .
In the case of the circuit shown in Fig. 12.16, it is particularly advisable to use power
MOSFETS. As they are of the enhancement type, their gate potentials are within the supply
voltage range, thereby obviating the need for positive and negative auxiliary voltages for
operational amplifiers OA2 and OA3 respectively. However, amplifiers must be used whose
common-mode input and output voltage range extends to the positive or negative supply
voltage. Therefore one should use rail-to-rail amplifiers for OA2 and OA3 as in the example
in Fig. 12.16.
If resistor R3 in Fig. 12.16 is not grounded but is connected to the output of a second
voltage follower, the input voltage difference will determine the output current.
12.3.4
Floating Current Sources
In the previous sections, we have discussed two types of current sources. Neither load
terminal in the circuits of Figs. 12.7 and 12.8 may be connected to a fixed potential. Such
a load is called “off-ground” or “floating,” and is illustrated in Fig. 12.17a. For this kind of
operation, the load may in practice consist only of passive elements, since for active loads
there is normally a connection to ground via the supply. Grounded loads can be supplied
by a current source based on the arrangement shown in Fig. 12.17b. Its practical design is
shown in Figs. 12.9–12.16.
I I I
RL RL RL
A A
If one or other load terminal is to be connected to any desired potential without the
current being affected, a floating current source is required. It can be constructed using two
grounded current sources which supply equal but opposite currents, as shown in Fig. 12.18.
It can be realized especially easy with the MAX435.
12.4
Current-Controlled Current Sources
The equivalent circuit of the current-controlled current source in Fig. 12.19 is identical to
that of the voltage-controlled current source shown in Fig. 12.6. The only difference is that
the input current is now the controlling signal and should be affected as little as possible
by the circuit. This is the case for the ideal condition in which ri = 0. When the effect of
the output on the input is neglected, the transfer characteristics are:
V1 = ri I1 + 0 · V2 ⇒ V1 = 0
(12.15)
I2 = AI I1 − r1o · V2 ⇒ I2 = AI I1
In Figs. 12.7 and 12.12, we show two voltage-controlled current sources of finite input
resistance. They can be operated as current-controlled current sources that have virtually
ideal characteristics if the resistor R1 is made zero; then, I2 = I1 .
Current-controlled current sources allowing polarity reversal of the output currents are
of particular interest. They are called current mirrors (also see Fig. 4.14 on page 281), and
one example is shown in Fig. 12.20. It is based on the voltage-controlled current source
shown in Fig. 12.11b, and the current-to-voltage conversion is effected by the additional
resistor R1 . However, this results in nonideal conditions for the input resistance.
The maximum freedom in specifying the circuit parameters is obtained if a circuit from
Sect. 12.2 is used for the current-to-voltage conversion and one of the voltage-controlled
current sources described in Sect. 12.3 is connected in series.
V1 i AI I 1 V2
o
12.5
NIC (Negative Impedance Converter)
There are cases in which negative resistances or voltage sources that have negative internal
resistances are required. By definition, the resistance R = +V /I , if the arrows for the
current and the voltage point in the same direction. If the voltage V across and the current
I through a two-terminal network have opposite signs, the quotient V /I then becomes
negative. Such a network is said to have a negative resistance. Negative resistances can,
in principle, be implemented only by the active circuits known as NICs. There are two
types: the UNIC, which reverses the polarity of the voltage without affecting the direction
of the current; and the INIC, which reverses the current without changing the polarity
of the voltage. The implementation of the INIC is particularly simple. Its ideal transfer
characteristics are:
V1 = V2 + 0 · I2 I 1 = 0 · V2 − I 2 (12.16)
These equations can be implemented as shown in Fig. 12.21, by a voltage-controlled
voltage source and a current-controlled current source. However, both functions can also
be performed by a single operational amplifier, as shown in Fig. 12.22.
For the ideal operational amplifier, VP = VN , and therefore V1 = V2 , as required. The
output potential of the amplifier has the value
Vo = V2 + I2 R
Hence, the current at port 1 is, as required,
V2 − Vo
I1 = = −I2
R
For this deduction, we have tacitly assumed stability of the circuit. However, since it
simultaneously employs positive and negative feedback, the validity of this assumption
must be examined separately in each case. To do so, we determine what proportion of the
output voltage affects the P-input and the N-input respectively. Figure 12.23 shows the
INIC in a general application, where R1 and R2 are the internal resistances of the circuits
connected to it. The feedback of the voltage
R1
VP = Vo
R1 + R
is positive, and that of
R2
VN = Vo
R2 + R
Vo
V1 V2 V1 V2
V2 = V1
Fig. 12.21. Model for an INIC, using Fig. 12.22. INIC using a single amplifier
controlled sources
780 12 Controlled Sources and Impedance Converters
Vo Vo
V1
V2
V1 V2
V0
Output voltage: V2 = V0 + I2 R1
dV2
Output resistance: ro = − = −R1
dI2
Negative resistances can be connected in series and in parallel just like conventional
resistors, and the same laws apply. For example, a voltage source with negative output
resistance can be used to compensate for the resistance of a long line so that, at the end of
the line, the voltage V0 is obtained with zero source resistance.
12.6
Gyrator
The gyrator is a converter circuit by which any impedance can be converted into its dual-
transformed counterpart; for example, a capacitance can be changed into an inductance.
The graphic symbol for a gyrator is shown in Fig. 12.26. The ideal transfer characteristics
are:
1
I1 = 0 · V1 + V
Rg 2
1
(12.18)
I2 = V + 0 · V2
Rg 1
where Rg is the gyration resistance. Hence, the current at one port is proportional to
the voltage at the other port. For this reason, the gyrator can be constructed from two
voltage-controlled current sources that have high input and output resistances, as shown
schematically in Fig. 12.27. A practical realization is shown in Fig. 5.90 on page 559.
Another method of implementing a gyrator is based on the combination of two INICs
and is illustrated in Fig. 12.28. To determine the transfer characteristics, we apply KCL to
the P- and N-inputs of OA1 and OA2 and obtain:
V1 V1 V2 V1
V2 V2
Fig. 12.26. Symbol for the gyrator Fig. 12.27. Implementation of a gyrator
using two voltage-controlled current
sources
782 12 Controlled Sources and Impedance Converters
OA1 OA2
V1 – V2
V1 V2
Fig. 12.28. Gyrator using two INICs
V3 − V1 V1
For node P1 : − + I1 = 0
Rg Rg
V 3 − V1 V2 − V1
For node N1 : + = 0
Rg Rg
V4 − V2 V1 − V2
For node P2 : + − I2 = 0
Rg Rg
V4 − V2 V2
For node N2 : − = 0
Rg Rg
By eliminating V3 and V4 , the transfer characteristics become
V2 V1
I1 = and I2 = ,
Rg Rg
which are the desired relationships as given in (12.18).
Some applications of the gyrator are described below. In the first example, a resistor
R2 is connected to the right-hand port. Since the arrows for I2 and V2 point in the same
direction, I2 = V2 /R2 , following Ohm’s law. Insertion of this relationship into the transfer
characteristics gives
V2 Rg V2
V1 = I2 Rg = and I1 =
R2 Rg
Port 1 therefore behaves like a resistance that has the value
V1 Rg2
R1 = = (12.19)
I1 R2
and thus it is proportional to the reciprocal of the load resistance connected to port 2.
The conversion of resistances is also valid for impedances and, according to (12.19),
gives
Rg2
Z1 = (12.20)
Z2
This relationship indicates an interesting application of the gyrator: if a capacitor with
value C2 is connected to one side, the impedance measured on the other side is
!
Z 1 = Rg2 · j ωC2 = j ωL1
which is the inductance
L1 = Rg2 · C2 (12.21)
The importance of the gyrator is due to the fact that it can be used to emulate large low-loss
inductances. The appropriate circuit is depicted in Fig. 12.29. The two free terminals of
12.6 Gyrator 783
V1 V1
Fig. 12.29. Emulation of an
inductance
V1 V1
V1 V2 V1 V2
counterpart of the middle two-port then appears between the two outer ports. To deduce the
transfer characteristics, the product of the chain matrices is calculated. The four-terminal
network to be transformed has the chain matrix:
A11 A12
(A) =
A21 A22
From (12.18), we obtain the following relationship for the gyrator:
V1 0 Rg V2
= (12.22)
I1 1/Rg 0 I2
3 45 6
(Ag )
12.7
Circulator
A circulator is a circuit with three or more ports, the graphic symbol for which is shown in
Fig. 12.33. It has the characteristic that a signal applied to one of the terminals is transferred
in the direction of the arrow. If a terminal is open, the signal passes unchanged, whereas at a
short-circuited terminal the polarity of the signal voltage is inverted. If a resistor R = Rg is
connected between one terminal and ground, the signal voltage appears across this resistor
and, in this case, will not be passed on to the next terminal.
A circuit that has these properties is shown in Fig. 12.35. It can be seen that it consists
of three identical stages, one of which is shown again in Fig. 12.34. Several cases can be
distinguished:
12.7 Circulator 785
Rg
Rg
Vi Vo
Rg
V1 V2 V3 V1
Fig. 12.33. Symbol for a circulator Fig. 12.34. One stage of a circulator
From these characteristics, the operation of the circuit in Fig. 12.35 can be easily under-
stood. Let us assume that voltage V1 is applied to port 1, that a resistor Rg is connected
at port 2, and that port 3 is left open. We already know that the output voltage of OA2
becomes zero. OA3 has unity gain because of the open terminal 3 and its output voltage
is therefore also zero. OA1 consequently operates as a noninverting amplifier with a gain
of 2, so that its output voltage is 2V1 . Half of this voltage (V1 ) appears at terminal 2, since
this is terminated in Rg . Other special cases can be analyzed in an identical manner.
If a more general case is to be considered, the transfer characteristics of the circulator
are used to determine the properties of the circuit. For this purpose, we apply KCL to the
P- and N-inputs:
P-inputs N-inputs
V6 − V1 V6 − V1 V4 − V1
+ I1 = 0 + = 0
Rg Rg Rg
V 4 − V2 V4 − V2 V5 − V2
+ I2 = 0 + = 0
Rg Rg Rg
V 5 − V3 V5 − V3 V6 − V3
+ I3 = 0 + = 0
Rg Rg Rg
786 12 Controlled Sources and Impedance Converters
I3 I1 I2
Rg Rg Rg
I1 I2 I3
V1 – V2 V2 – V3 V3 – V1
V1 V2 V3
1 1 1
I1 = Rg
(V2 − V3 ) I2 = Rg
(V3 − V1 ) I3 = Rg
(V1 − V2 ) (12.24)
It is obvious from (12.24) that a circulator can also be implemented by three voltage-
controlled current sources with differential input, as shown in Fig. 12.36. A practical
realization with CC-operational amplifiers is shown in Fig. 12.37.
Figure 12.38 shows a circulator employed as an active hybrid set for telephone circuits.
It consists of a circulator that has three ports, all of which are terminated in the transfer
resistance Rg , The signal from the microphone is relayed to the exchange and does not
reach the speaker. The signal from the exchange is transferred to the speaker, but not to
the microphone. The cross-talk attenuation is largely determined by the degree to which
the terminating resistances are matched.
13.1
Basic Theory of Lowpass Filters
Simple lowpass and highpass filters are discussed in Sects. 29.3.1 and 29.3.2, the circuit of
the simplest lowpass filter being shown again in Fig. 13.1. The ratio of the output voltage
to the input voltage can be expressed using (29.3.1) as
Vo 1
A(j ω) = =
Vi 1 + j ωRC
and is called the frequency response of the circuit. Replacing j ω by j ω + σ = s gives the
transfer function:
L{Vo (t)} 1
A(s) = =
L{Vi (t)} 1 + s RC
This is the ratio of the Laplace-transformed output and input voltages for signals of any
time dependence. On the other hand, the transition from the transfer function A(s) to the
frequency response A(j ω) for sinusoidal input signals is made by setting σ to zero.
In order to present the problem in a more general form, it is useful to normalize the
complex frequency variable s by defining
s
sn =
ωc
Hence, for σ = 0
jω f
sn = = j = j ωn
ωc fc
The circuit shown in Fig. 13.1 has the cutoff frequency fc = 1/2π RC. Therefore,
sn = s RC and
1
A(sn ) = (13.1)
1 + sn
For the absolute value of the transfer function – that is, for the amplitude ratio with
sinusoidal input signals – we obtain with ωn = ω/ωc = f/fc
1
|A(j ωn )|2 =
1 + ωn2
If ωn 1 – that is, f fc – then |A| = 1/ωn ; this corresponds to a reduction in gain of
20 dB per frequency decade.
Vi Vo
a Fourth order
b Tenth order
. . . .
Fig. 13.2. Comparison of the amplitude–frequency responses for different filter types.
Curve 1: Lowpass filter with critical damping. Curve 2: Bessel lowpass filter.
Curve 3: Butterworth lowpass filter. Curve 4: Chebyshev lowpass filter with 3 dB ripple
If a sharper cutoff is required, N lowpass filters can be connected in series. The transfer
function is then expressed in the form:
1
A(sn ) = (13.2)
(1 + α1 sn )(1 + α2 sn ) . . . (1 + αn sn )
where the coefficients α1 , α2 , α3 , …are real and positive. For ωn 1, |A| ∼ 1/ωnN is
proportional to 1/ωN ; the gain therefore falls off at N ·20 dB per decade. It can be seen that
the transfer function possesses N real negative poles. This is characteristic of N th -order
passive RC lowpass filters. If decoupled lowpass filters of identical cutoff frequencies are
cascaded, then
'
N√
α1 = α2 = α3 = . . . = α = 2−1
this being the condition for which critical damping occurs. Each individual lowpass filter
then has a cutoff frequency that is a factor 1/α higher than that of the filter as a whole.
The transfer function of a lowpass filter has the general form:
A0
A(sn ) = (13.3)
1 + c1 sn + c2 sn2 + . . . + cn snN
where c1 , c2 , . . . , cN are positive and real. The order of the filter is equal to the highest
power of sn . It is advantageous for filter design if the denominator polynomial is written
13.1 Basic Theory of Lowpass Filters 789
Vo
Vi
in factored form. If complex poles are also permitted, a separation into linear factors as
shown in (13.2) is no longer possible, and a product of quadratic expressions is obtained:
A0
A(sn ) = (13.4)
(1 + a1 sn + b1 sn )(1 + a2 sn
2 + b2 sn2 ) . . .
where ai and bi are positive and real. For odd orders N , the coefficient b1 is zero.
There are several different theoretical aspects with respect to which the frequency
response can be optimized. Any such aspect leads to a different set of coefficients ai and
bi . As will be seen, conjugate complex poles arise. They cannot be realized by passive RC
elements, as a comparison with (13.2) shows. One way of implementing conjugate complex
poles is to use LRC networks. For high frequencies, the design of the necessary inductances
usually presents no difficulties, but in the low-frequency range large inductances are often
required. These are unwieldy and have poor electrical properties. However, the use of
inductances at low frequencies can be avoided by the addition of active elements (e.g.,
operational amplifiers) to the RC network. Such circuits are called active filters.
Let us first compare the most important optimized frequency responses, the technical
realizations of which are discussed in the following sections.
Butterworth lowpass filters have an amplitude–frequency response that is flat for as
long as possible and falls off sharply just before the cutoff frequency. Their step response
shows a considerable overshoot, which increases for higher-order filters.
Chebyshev lowpass filters have an even steeper roll-off above the cutoff frequency. In
the passband, however, the gain varies with a ripple of constant amplitude. For a given
order, the decrease above the cutoff frequency is steeper the larger the permitted ripple.
The overshoot in the step response is even greater than for the Butterworth filters.
Bessel lowpass filters have the optimum square-wave response. The prerequisite for this
is that the group delay is constant over the largest possible frequency range; in other words,
that the phase shift in this frequency range is proportional to the frequency. The amplitude–
790 13 Active Filters
Order
2 4 6 8 10
Critical damping
Normalized rise time tr /Tc 0.344 0.342 0.341 0.341 0.340
Normalized delay time td /Tc 0.172 0.254 0.316 0.367 0.412
Overshoot % 0 0 0 0 0
Bessel
Normalized rise time tr /Tc 0.344 0.352 0.350 0.347 0.345
Normalized delay time td /Tc 0.195 0.329 0.428 0.505 0.574
Overshoot % 0.43 0.84 0.64 0.34 0.06
Butterworth
Formalized rise time tr /Tc 0.342 0.387 0.427 0.460 0.485
Normalized delay time td /Tc 0.228 0.449 0.663 0.874 1.084
Overshoot % 4.3 10.8 14.3 16.3 17.8
Chebyshev 0.5 dB ripple
Normalized rise time tr /Tc 0.338 0.421 0.487 0.540 0.584
Normalized delay time td /Tc 0.251 0.556 0.875 1.196 1.518
Overshoot % 10.7 18.1 21.2 22.9 24.1
Chebyshev 1 dB ripple
Normalized rise time tr /Tc 0.334 0.421 0.486 0.537 0.582
Normalized delay time td /Tc 0.260 0.572 0.893 1.215 1.540
Overshoot % 14.6 21.6 24.9 26.6 27.8
Chebyshev 2 dB ripple
Normalized rise time tr /Tc 0.326 0.414 0.491 0.529 0.570
Normalized delay time td /Tc 0.267 0.584 0.912 1.231 1.555
Overshoot % 21.2 28.9 32.0 33.5 34.7
Chebyshev 3 dB ripple
Normalized rise time tr /Tc 0.318 0.407 0.470 0.519 0.692
Normalized delay time td /Tc 0.271 0.590 0.912 1.235 1.557
Overshoot % 27.2 35.7 38.7 40.6 41.6
Fig. 13.4. Comparison of lowpass filters. The rise time and delay time are normalized to the
reciprocal cutoff frequency Tc = 1/fc
frequency response of Bessel filters does not fall off as sharply as that of Butterworth or
Chebyshev filters.
Figure 13.2 shows the amplitude–frequency responses of the four described filter types
for the fourth and tenth orders. It can be seen that the Chebyshev lowpass filter has the most
abrupt transition from the passband to the stopband. This is advantageous, but it has the
side effect of a ripple in the amplitude–frequency response in the passband. As this ripple
is gradually reduced, the behavior of the Chebyshev filter approaches that of a Butterworth
filter. Both kinds of filter show a considerable overshoot in the step response, as can be seen
in Fig. 13.3. Bessel filters, on the other hand, have only a negligible overshoot. Despite
their unfavorable amplitude–frequency response, they will always be used where a good
step response is important. A passive RC lowpass filter exhibits no overshoot; however, the
relatively small improvement over the Bessel filter involves a considerable deterioration in
13.1 Basic Theory of Lowpass Filters 791
the amplitude–frequency response. In addition, the corners in the step response are much
rounder than for the Bessel filter. The table in Fig. 13.4 compares the rise times, delay
times, and overshoots. The rise time is the time in which the output signal rises from 10%
to 90% of its final-state value. The delay time is that in which the output signal increases
from 0 to 50% of the final-state value.
It can be seen that the rise time does not depend to any great extent on the order; nor
does it depend on the type of the filter. Its value is approximately tr = 1/3fg , as shown in
Sect. 29.3.1. On the other hand, as the order increases, so the delay time and the overshoot
increase. The Bessel filters are an exception in that the overshoot decreases for orders
higher than four.
It will be seen later that the same circuit allows the implementation of these filter types
for a particular order; the values of resistances and capacitances determine the type of
filter. In order that the circuit parameters can be defined, the frequency responses of the
individual filter types must be known for each order. We shall therefore discuss these in
more detail in the following sections.
13.1.1
Butterworth Lowpass Filters
From (13.3), the absolute value of the gain of an nth -order lowpass filter has the general
form:
A20
|A|2 = (13.5)
1 + k2 ωn2 + k4 ωn4 + · · · + k2n ωn2N
Odd powers of ω do not occur, since the square of |A|2 must be an even function.
Below the cutoff frequency of the Butterworth lowpass filter, the function |A|2 must
be maximally flat. Since for this range ωn < 1, this condition is best fulfilled if |A|2 is
dependent only on the highest power of ωn . The reason for this is that, for ωn < 1, the
lower powers of ωn contribute most to the denominator and therefore to the decrease in
gain. Hence, for Butterworth-filters
A20
|A|2 =
1 + k2n ωn2N
The coefficient k2N is defined by the “normalizing condition”; namely, that the gain at
ωn = 1 is reduced by 3 dB. Thus
A20 A20
=
2 1 + k2N
k2N = 1
Therefore, the square of the gain of N th -order Butterworth lowpass filters is given by
A20
|A|2 = (13.6)
1 + ωn2N
To implement a Butterworth lowpass filter, a circuit must be designed in which the square
of the gain has the form given above. However, the circuit analysis initially gives the
complex gain A, rather than the square of the gain, |A|2 . It is therefore necessary to know
the value of the complex gain involved in (13.6). This is found by calculating the absolute
792 13 Active Filters
N
1 1 + s√n
2 1 + 2sn + sn2
3 1 + 2sn + 2sn2 + sn3 = (1 + sn )(1 + sn + sn2 )
4 1 + 2,613sn + 3,414sn2 + 2,613sn3 + sn4 = (1 + 1,848sn + sn2 )(1 + 0,765sn + sn2 )
value of (13.3) and by comparing the coefficients with those of (13.6). In this way, the
desired coefficients c1 , . . . , cn can be defined. The denominators of (13.3) are then the
Butterworth polynomials, the first four orders of which are shown in Fig. 13.5.
It is possible to determine the poles of the transfer function analytically. By combining
the conjugate complex poles, we immediately obtain the coefficients, ai and bi , of the
quadratic expressions in (13.4):
even order N:
(2i − 1)π N
ai = 2 cos for i = 1... ,
2N 2
bi = 1
odd order N :
(i − 1)π N +1
a1 = 1, ai = 2 cos for i = 2... ,
N 2
b1 = 0 bi = 1
The coefficients of the Butterworth polynomials up to tenth order are shown in Fig. 13.14.
It can be seen that the first-order Butterworth lowpass filter is a passive lowpass filter that
has the transfer function of (13.1). The higher Butterworth polynomials possess conjugate
complex zeros. A comparison with (13.2) shows that such denominator polynomials cannot
be implemented by passive RC networks, because in the case of the latter, all of the zeros
are real. In such cases, the only choice is to use LRC circuits, with all their disadvantages,
or active RC filters. The frequency response of the gain is shown in Fig. 13.6.
. . . .
13.1.2
Chebyshev Lowpass Filters
At low frequencies, the gain of a Chebyshev lowpass filter has the value A0 , but it varies
below the cutoff frequency, having a predetermined ripple. Polynomials that have a constant
ripple within a defined range (an equal ripple) are the Chebyshev polynomials,
cos(N arccos x) for 0 ≤ x ≤ 1
TN (x) =
cosh(N Arcosh x) for x > 1,
the first four of which are shown in Fig. 13.7. For 0 ≤ x ≤ 1, |T (x)| oscillates between 0
and 1; for x > 1 rises steadily. In order to obtain the equation for a lowpass filter from the
Chebyshev polynomials, one defines
kA20
|A|2 = (13.7)
1 + ε 2 TN2 (x)
The constant k is chosen such that, for x = 0, the square of the gain |A|2 becomes A20 ; that
is, k = 1 for odd N and k = 1 + ε 2 for even N . The factor a is a measure of the ripple,
and is given by
A max
= 1 + ε2
A min
and
√ .
A max = A0 1 + ε 2
for even orders
A min = A0
and
.
A max = A0 √
: for odd orders
A min = A0 1 + ε2
The appropriate values are listed for different ripples in Fig. 13.8. In principle, the complex
gain can be calculated from |A|2 and hence the coefficients of the factored form can be
determined. However, it is possible to derive the poles of the transfer function directly
from those of the Butterworth filters. By combining the conjugate complex poles, the
coefficients ai , and bi in (13.4) are determined:
even order N :
1 ⎫
bi
= ⎪
⎪
(2i − 1)π ⎪ ⎪
⎪
cosh2 γ − cos2 ⎬ N
2N for i = 1 . . .
⎪
⎪ 2
(2i − 1)π ⎪ ⎪
⎪
⎭
ai = 2bi · sinh γ · cos
2N
N
1 T1 (x) = x
2 T2 (x) = 2x 2 − 1
3 T3 (x) = 4x 3 − 3x
4 T4 (x) = 8x 4 − 8x 2 + 1
Parameter Ripple
0.5 dB 1 dB 2 dB 3 dB
A max /A min 1.059 1.122 1.259 1.413
k 1.122 1.259 1.585 1.995
ε 0.349 0.509 0.765 0.998
odd order N:
b1
= 0
a1
= 1/ sinh γ
1 ⎫
bi
= ⎪
⎪
(i − 1)π ⎪
⎪
cosh2 γ − cos2 ⎪
⎬
N N +1
for i = 2 . . .
⎪
⎪ 2
(i − 1)π ⎪
⎪
⎪
⎭
ai
= 2bi
· sinh γ · cos
N
1 1
where γ = Arsinh .
N ε
If the coefficients ai
and bi
found in this way are used to replace ai and bi in (13.4),
Chebyshev filters are obtained. However, sn is then not normalized with respect to the 3 dB
cutoff frequency ωc but, rather, to the frequency ωx which the gain assumes the value A min
for the last time.
For an easy comparison of the different filter types, it is useful to normalize sn to the
3 dB cutoff frequency. The variable sn is replaced by αsn and √the normalizing constant α
is determined such that the gain, for sn = j , has the value 1/ 2 = 7 −3 dB. The quadratic
expressions in the denominator of the complex gain are then
(1 + ai
αsn + bi
α 2 sn2 )
Hence, by comparing the coefficients with those of (13.4),
ai = αai
and bi = α 2 bi
The coefficients ai and bi are shown in the table of Fig. 13.14 up to the tenth order, and
for ripple values of 0.5, 1, 2, and 3 dB. The frequency response of the gain is shown in
Fig. 13.9 for ripple values of 0.5 and 3 dB. Figure 13.10 makes possible a direct comparison
of fourth-order Chebyshev filters that have different amounts of ripple. It can be seen that
the differences in the frequency response in the stopband are very small, and that they
become even smaller for higher orders. It is also obvious that even the Chebyshev filter
response that has a small ripple of 0.5 dB emerges from the passband much more steeply
than that of the Butterworth filter.
The transition from the passband to the stopband can be made even steeper. To accom-
plish this, zeros are introduced into the amplitude–frequency response above the cutoff
frequency. One way of optimizing the design is to give the amplitude–frequency response
a constant ripple in the stopband as well. Such filters are called Cauer and Elliptic filters.
13.1 Basic Theory of Lowpass Filters 795
a 0.5 db ripple
b 3 dB ripple
. . . .
. . . .
Their transfer function differs from the ordinary lowpass filter equation in that the numera-
tor is a polynomial, instead of the constant A0 . For this reason, “steepened” lowpass filters
cannot be designed using the simple circuits of Sect. 13.4. However, in Sect. 13.11 we
discuss a universal filter with which any numerator polynomial can be implemented.
796 13 Active Filters
13.1.3
Bessel Lowpass Filters
As previously shown, Butterworth and Chebyshev lowpass filters have a considerable
overshoot in their step response in Figs. 13.3/13.4. An ideal square-wave response is
achieved by filters that have a frequency-independent group delay; that is, that have a
phase shift that is proportional to the frequency. This behavior is best approximated by
Bessel filters, which are sometimes also called Thomson filters. The approximation consists
of selecting the coefficients such that the group delay below the cutoff frequency ωn = 1
is dependent as little as possible on ωn . This procedure is equivalent to a Butterworth
approximation of the group delay; that is, the optimization of a maximally flat group
delay.
From (13.4), with sn = j ωn , the gain of a second-order lowpass filter is given by
A0 A0
A = =
1 + a1 sn + b1 sn
2 1 + j a1 ωn − b1 ωn2
Therefore the phase shift is
a 1 ωn
ϕ = − arctan (13.8)
1 − b1 ωn2
The group delay is defined
dϕ
tgr = −
dω
To simplify further calculations, we introduce the normalized group delay
tgr
Tgr = tgr ωc = 2π tgr · fc = 2π (13.9a)
Tc
where Tc is the reciprocal of the cutoff frequency. We thus obtain
dϕ dϕ
Tgr = −ωc · = − (13.9b)
dω dωn
and, using (13.8)
a1 (1 + b1 ωn2 )
Tgr = (13.9c)
1 + (a12 − 2b1 )ωn2 + b12 ωn4
In order to find the Butterworth approximation of the group delay, we use the fact that:
1 + b1 ωn2
Tgr = a1 · for ωn 1
1 + (a12 − 2b1 )ωn2
This expression becomes independent of ωn if the coefficients of ωn2 in the numerator and
denominator are identical. The condition for this is that
b1 = a12 − 2b1
or (13.10)
1 2
b1 = a
3 1
13.1 Basic Theory of Lowpass Filters 797
N
1 1 + sn
1
2 1 + sn + 3 sn2
2 1
3 1 + sn + 5 sn2 + 15 sn3
3 2 1
4 1 + sn + 7 sn2 + 21 sn3 + 105 sn4
ci
= c
i(2N − i + 1) i−1
The denominators of (13.3) obtained in such a way are the Bessel polynomials, and are
shown in Fig. 13.11 up to the fourth order. However, it should be noted that, in this
representation, the frequency sn is not normalized with respect to the 3 dB cutoff frequency,
but to the reciprocal of the group delay for ωn = 0. This is of little use for the design
of lowpass filters. We have therefore recalculated the coefficients ci for the 3 dB cutoff
frequency, as in the previous section, and in addition we have broken the denominator
down into quadratic expressions. The coefficients ai and bi of (13.4) thus obtained are
listed in Fig. 13.14 for up to the tenth order. The frequency response of the gain is plotted
in Fig. 13.12.
In order to demonstrate the amount of phase distortion of other filters in comparison
with the Bessel filters, in Fig. 13.13 we have illustrated the frequency response of the phase
shift and of the group delay for fourth-order filters. These curves can best be calculated
from the factored transfer function in (13.4) by adding together the phase shifts of each
individual second-order filter stage and by adding the individual group delays. For a filter
of a given order, the following relations are derived from (13.8) and (13.9c):
0 ai ω
ϕ = − arctan
1 − bi ω 2
i
and
0 ai (1 + bi ωn2 )
Tgr =
i
1 + (ai2 − 2bi )ωn2 + bi2 ωn4
0
Tgr0 = ai
i
798 13 Active Filters
. . . .
. . . .
Fig. 13.13. Comparison of the frequency response of the group delay and the phase shift of
fourth-order filters
Curve 1: Lowpass filter with critical damping. Curve 2: Bessel lowpass filter.
Curve 3: Butterworth lowpass filter. Curve 4: Chebyshev lowpass filter with 0.5 dB ripple.
Curve 5: Chebyshev lowpass filter with 3 dB ripple
13.1 Basic Theory of Lowpass Filters 799
N i ai bi fci /fc Qi
N i ai bi fci /fc Qi
Bessel filters
1 1 1.0000 0.0000 1.000 –
2 1 1.3617 0.6180 1.000 0.58
3 1 0.7560 0.0000 1.323 –
2 0.9996 0.4772 1.414 0.69
4 1 1.3397 0.4889 0.978 0.52
2 0.7743 0.3890 1.797 0.81
5 1 0.6656 0.0000 1.502 –
2 1.1402 0.4128 1.184 0.56
3 0.6216 0.3245 2.138 0.92
6 1 1.2217 0.3887 1.063 0.51
2 0.9686 0.3505 1.431 0.61
3 0.5131 0.2756 2.447 1.02
7 1 0.5937 0.0000 1.684 –
2 1.0944 0.3395 1.207 0.53
3 0.8304 0.3011 1.695 0.66
4 0.4332 0.2381 2.731 1.13
8 1 1.1112 0.3162 1.164 0.51
2 0.9754 0.2979 1.381 0.56
3 0.7202 0.2621 1.963 0.71
4 0.3728 0.2087 2.992 1.23
9 1 0.5386 0.0000 1.857 –
2 1.0244 0.2834 1.277 0.52
3 0.8710 0.2636 1.574 0.59
4 0.6320 0.2311 2.226 0.76
5 0.3257 0.1854 3.237 1.32
10 1 1.0215 0.2650 1.264 0.50
2 0.9393 0.2549 1.412 0.54
3 0.7815 0.2351 1.780 0.62
4 0.5604 0.2059 2.479 0.81
5 0.2883 0.1665 3.466 1.42
N i ai bi fci /fc Qi
Butterworth filters
1 1 1.0000 0.0000 1.000 –
2 1 1.4142 1.0000 1.000 0.71
3 1 1.0000 0.0000 1.000 –
2 1.0000 1.0000 1.272 1.00
4 1 1.8478 1.0000 0.719 0.54
2 0.7654 1.0000 1.390 1.31
5 1 1.0000 0.0000 1.000 –
2 1.6180 1.0000 0.859 0.62
3 0.6180 1.0000 1.448 1.62
6 1 1.9319 1.0000 0.676 0.52
2 1.4142 1.0000 1.000 0.71
3 0.5176 1.0000 1.479 1.93
7 1 1.0000 0.0000 1.000 –
2 1.8019 1.0000 0.745 0.55
3 1.2470 1.0000 1.117 0.80
4 0.4450 1.0000 1.499 2.25
8 1 1.9616 1.0000 0.661 0.51
2 1.6629 1.0000 0.829 0.60
3 1.1111 1.0000 1.206 0.90
4 0.3902 1.0000 1.512 2.56
9 1 1.0000 0.0000 1.000 –
2 1.8794 1.0000 0.703 0.53
3 1.5321 1.0000 0.917 0.65
4 1.0000 1.0000 1.272 1.00
5 0.3473 1.0000 1.521 2.88
10 1 1.9754 1.0000 0.655 0.51
2 1.7820 1.0000 0.756 0.56
3 1.4142 1.0000 1.000 0.71
4 0.9080 1.0000 1.322 1.10
5 0.3129 1.0000 1.527 3.20
N i ai bi fci /fc Qi
N i ai bi fci /fc Qi
N i ai bi fci /fc Qi
N i ai bi fci /fc Qi
13.1.4
Summary of the Theory
We have seen that the transfer functions of all lowpass filters have the form
A0
A(sn ) = ; (13.11)
(1 + ai sn + bi sn2 )
i
The order N of the filter is determined by the highest power of the frequency variable s,
respectively, sn in (13.11) when the denominator is expanded. It defines the slope of the
asymptote of the amplitude–frequency response as having the value −N ·20 dB per decade.
The rest of the amplitude–frequency response curve for a particular order is determined by
the type of filter. Of special interest are the Butterworth, Chebyshev, and Bessel filters, all
of which have different values for the coefficients ai and bi in (13.11). The values of the
coefficients are summarized in Fig. 13.14 for up to the tenth order. In addition, the 3 dB
806 13 Active Filters
cutoff frequency of each individual filter stage in respect to the whole filter is given by the
ratio
√ - '
fci 2
= 2bi − ai2 + ai4 bi + 8bi2
fc 2bi
Although this value is not needed for the design, it is useful for checking the correct
operation of the individual filter stages. Also listed are the pole-pair quality factors Qi
of the individual filter stages. By analogy
√ with the Q-factors of the bandpass filters in
Sect. 13.6.1, they are defined as Qi = bi /ai . The larger the pole-pair Q-factor, the larger
is the transient oscillation of the step response. Filters with real poles have Q-factors of
Q ≤ 0.5.
The frequency response of the gain, phase shift, and group delay can be calculated
using the coefficients ai and bi of the factored transfer function:
A20
|A|2 = ; * + (13.12)
1 + (ai2 − 2bi )ωn2 + bi2 ωn4
i
0 ai ωn
ϕ = − arctan (13.13)
1 − bi ωn2
i
0 ai (1 + bi ωn2 )
Tgr = (13.14)
i
1 + (ai2 − 2bi )ωn2 + bi2 ωn4
13.2
Lowpass/Highpass Transformation
In the logarithmic representation, the amplitude–frequency response of a lowpass filter is
transformed into that of the analogous highpass filter response by plotting its mirror image
about the cutoff frequency; that is, by replacing ωn by 1/ωn and sn by 1/sn . The cutoff
frequency remains the same, and A0 changes to A∞ . Equation (13.11) then becomes
A∞ A∞ sn2
A(sn ) = = ;% & (13.15)
; ai bi bi + ai sn + sn2
1+ + 2 i
i sn sn
In the time domain, the performance cannot be transformed, as the step response shows
a basically different behavior. This can be seen in Fig. 13.15, where an oscillation about
the final-state value occurs even in highpass filters that have critical damping. The analogy
with the corresponding lowpass filters still applies inasmuch as the transient oscillation
decays more slowly the higher the pole-pair Q-factors are.
13.3 Realization of First-Order Lowpass and Highpass Filters 807
Vo
Vi
.
13.3
Realization of First-Order Lowpass and Highpass Filters
According to (13.11), the transfer function of a first-order lowpass filter has the general
form
A0
A(sn ) = (13.16)
1 + a 1 sn
It can be implemented by the simple RC network shown in Fig. 13.1. From Sect. 13.1,
it follows for this circuit:
1 1
A(sn ) = =
1 + s RC 1 + ωc RCsn
The low-frequency gain is defined by the value A0 = 1, but the parameter a1 can be
chosen freely. Its value is found by comparing the coefficients
a1
RC =
2πfc
As can be seen from the table in Fig. 13.14, all filter types of first order are identical and
have the coefficient a1 = 1. When higher-order filters are implemented by cascading filter
stages of lower orders, first-order filter stages may be required for which a1 = 1. The
reason for this is that individual filter stages have, as a rule, a cutoff frequency that is
different from that of the filter as a whole, namely fc 1 = fc /a1 .
The simple RC network shown in Fig. 13.1 has the disadvantage that its properties
change when it is loaded. Therefore an impedance converter (buffer) is usually connected
808 13 Active Filters
Vi
Vo
Fig. 13.16. A first-order lowpass filter
with an impedance converter
(R2 + R3 )/R3
A(sn ) =
1 + ωc R1 C1 sn
in series. If it is given voltage gain A0 , the low-frequency gain can then be chosen freely.
An appropriate circuit is presented in Fig. 13.16.
In order to arrive at the corresponding highpass filter, the variable sn in (13.16) must
be replaced by 1/sn . In the circuit itself, the conversion is achieved by simply exchanging
R1 and C1 .
Somewhat simpler circuits for first-order lowpass and highpass filters are obtained if
the filtering network is included in the feedback loop of the operational amplifier. The
corresponding lowpass filter is shown in Fig. 13.17. For the actual design, the cutoff
frequency, the low-frequency gain A0 , which in this case is negative, and the capacitance
C1 need to be defined. Comparing the coefficients to those of (13.16), it follows:
a1 R2
R2 = and R1 = −
2πfc C1 A0
1
R1 = and R2 = −R1 A∞
2πfc a1 C1
The transfer functions given for the previous circuits apply only to the range of fre-
quency for which the open-loop gain of the operational amplifier is large with respect to
the absolute value of A. This condition is difficult to fulfill for high frequencies, as the
magnitude of the open-loop gain falls at a rate of 6 dB per octave because of the necessary
frequency compensation. For a standard operational amplifier of the 741-type at 10 kHz,
|AD | is only about 100.
Vi Vi
Vo Vo
Fig. 13.17. First-order lowpass filter with Fig. 13.18. First-order highpass filter with
an inverting amplifier an inverting amplifier
R2 /R1 −sn R2 /R1
A(sn ) = − A(sn ) =
1 + ωc R2 C1 sn 1
+ sn
R1 C1 ωc
13.4 Realization of Second-Order Lowpass and Highpass Filters 809
13.4
Realization of Second-Order Lowpass and Highpass Filters
According to (13.11), the transfer function of a second-order lowpass filter has the general
form
A0
A(sn ) = (13.17)
1 + a1 sn + b1 sn2
As can be seen from the table in Fig. 13.14, the optimized transfer functions of the second
and higher orders have conjugate complex poles because Q > 0.5. Such transfer functions
cannot be implemented by passive RC networks, as discussed in Sect. 13.1. One possi-
ble way of realizing these circuits is to use inductances, and this is demonstrated in the
following example.
13.4.1
LRC Filters
The conventional implementation of second-order filters involves using LRC networks,
as shown in Fig. 13.19. Comparison of the coefficients with those of (13.17) yields
a1 b1
R = and L =
2πfc C 4π 2 fc2 C
From Fig. 13.14, the coefficients of a second-order Butterworth lowpass filter are a1 =
1.414 and b1 = 1.000. For a given cutoff frequency of fc = 10 Hz and a capacitance of
C = 10 mF, the remaining design parameters are R = 2.25 k and L = 25.3 H. Obviously,
such a filter is extremely difficult to implement because of the size of the inductance.
However, this can be avoided by emulating the inductance with an active RC circuit. The
gyrator shown in Fig. 12.32 is useful for this purpose, although its implementation involves
a considerable number of components.
The desired transfer functions can be put into practice much more simply without
inductance emulation by connecting suitable RC networks around operational amplifiers.
13.4.2
Filters with Multiple Negative Feedback
An active second-order RC lowpass filter is shown in Fig. 13.20. By comparing the coef-
ficients with those of (13.17), we obtain the relations
A0 = −R2 /R1
R 2 R3
a 1 = ωc C 1 R 2 + R 3 +
R1
b1 = ωc2 C1 C2 R2 R3
Vi Vo 1
A(sn ) =
1 + ωc RCsn + ωc2 LCsn2
810 13 Active Filters
Vi
Vo
Fig. 13.20. Active second-order lowpass filter with multiple negative feedback
R2 /R1
A(sn ) = −
R2 R3
1 + ωc C1 R2 + R3 + sn + ωc2 C1 C2 R2 R3 sn2
R1
For the actual specification, the values of the resistors R1 and R3 , for example, can be
predetermined; the parameters R2 , C1 , and C2 can then be calculated from the above
equations. Such a determination is possible for all positive values of a1 and b1 , so that any
desired type of filter can be realized. The gain at zero frequency, A0 , is negative. At low
frequencies, therefore, the filter inverts the signal.
In order to achieve the desired frequency response, the circuit elements must not have
too large a tolerance. This requirement is easy to fulfill for resistors, as they can be obtained
off the shelf with 1% tolerance, in the E 96 standard series. The situation is different with
capacitors, which, as a rule, are only available off the shelf in the E 6 series. It is therefore
advantageous in filter design to predetermine the capacitors and calculate the values for
the resistors. We therefore solve the design equations for the resistances and obtain:
'
a1 C2 − a12 C22 − 4C1 C2 b1 (1 − A0 )
R2 =
4πfc C1 C2
R2
R1 =
−A0
b1
R3 = 2 2
4π fc C1 C2 R2
In order that the value for R2 will be real, the condition
4b1 (1 − A0 )
C2 ≥ C1
a12
must be fulfilled. The most favorable design is obtained if the ratio C2 /C1 is chosen
not much larger than is prescribed by this condition. The filter parameters are relatively
insensitive to the tolerances of the components, and therefore the circuit is particularly
suited to the realization of filters having high Q-factors.
13.4.3
Filter with Single Positive Feedback
Active filters can also be designed using amplifiers with positive feedback. However, the
gain must be fixed at a precise value by an internal negative feedback (a “controlled
source”). The voltage divider R3 , (α − 1)R3 shown in Fig. 13.21 provides this negative
13.4 Realization of Second-Order Lowpass and Highpass Filters 811
V1
Vi
Vo
Fig. 13.21. Active second order lowpass filter with single positive feedback
α
A(sn ) = 1 2
1 + ωc C1 (R1 + R2 ) + (1 − α)R1 C2 sn + ωc2 R1 R2 C1 C2 sn2
feedback and determines the internal gain as having the value Vo /V1 = α. The positive
feedback is provided by capacitor C2 .
Dimensioning can be substantially simplified by certain specializations right from
the start. One possible specialization is to select the internal gain α = 1. This makes
(α − 1)R3 = 0, and both R3 resistances can be omitted. Such operational amplifiers, with
a unit feedback factor, are available as integrated voltage followers. A simple impedance
converter or buffer – for instance, an emitter or source follower – is often sufficient. In this
way, filters in the MHz range can also be realized. For the special case in which α = 1,
the transfer function is given by
1
A(sn ) =
1 + ωc C1 (R1 + R2 )sn + ωc2 R1 R2 C1 C2 sn2
Defining C1 and C2 and comparing the coefficients with those of (13.17) gives
A0 = 1
'
a 1 C2 ∓ a12 C22 − 4b1 C1 C2
R1/2 =
4πfc C1 C2
In order to arrive at values that are real, the condition
must be fulfilled.As for the filter with multiple negative feedback, the most favorable design
is obtained here also if the ratio C2 /C1 is chosen to be not much larger than is prescribed
by this condition. The MAX270 family from Maxim contains two filters of second order
with Butterworth characteristics. The cutoff frequency can be externally adjusted by a 7 bit
word in 128 steps in the range 1 . . . 25 kHZ.
Another interesting specialization is achieved when using identical resistors and iden-
tical capacitors, i.e. R1 = R2 = R and C1 = C2 = C. In order to realize the different
filter types it is then necessary to vary the internal gain α. The transfer function is then
α
A(sn ) =
1 + ωc RC(3 − α)sn + (ωc RC)2 sn2
812 13 Active Filters
Vi
Vo
Fig. 13.23. An active second-order highpass filter with single positive feedback
Vo αsn2
A(sn ) = =
Vi 1 R2 (C1 + C2 ) + R1 C2 (1 − α)
2
+ sn + sn2
R1 R2 C1 C2 ωc R1 R2 C1 C2 ωc
13.5 Realization of Higher-Order Lowpass and Highpass Filters 813
13.5
Realization of Higher-Order Lowpass and Highpass Filters
In cases in which the filter characteristic is insufficiently steep, filters of higher orders
must be employed. For this purpose, first- and second-order filters are cascaded, thereby
multiplying the frequency responses of the individual filters. It would, however, be wrong
to cascade two second-order Butterworth filters, for example, to obtain a fourth-order
Butterworth filter. The resulting filter would have a different cutoff frequency and also a
different filter characteristic. The coefficients of the individual filters must therefore be
chosen such that the product of the frequency responses gives the desired optimized filter
type.
To simplify the design of the individual filters, we have factored the polynomials of
the different filter types. The coefficients a1 and b1 of the individual filter stages are given
in Fig. 13.14. Each filter section can be implemented by one of the second-order filters
described previously. It is merely necessary to replace the coefficients a1 and b1 by ai
and bi . To calculate the circuit parameters from the given formulas, the desired cutoff
frequency of the resulting total filter must be inserted. As a rule, the individual filter stages
possess cutoff frequencies that are different from that of the filter as a whole, as can be
seen in Fig. 13.14. Odd-order filters contain a factor in which bi = 0. The corresponding
filter stage can be implemented by one of the first-order filters described if a1 is replaced
by ai . In this case also, the cutoff frequency fc of the total filter must be inserted. Because
of the defined value of ai , this filter stage automatically has the cutoff frequency fci given
in Fig. 13.14.
In principle, the sequence in which the individual filter stages are cascaded is not
significant, as the resulting frequency response is the product of the individual stages which
remains the same. In practice, however, there are several design considerations governing
the best sequence of the filter stages. One such aspect is the permissible voltage swing, for
which it is useful to arrange the filter stages according to their cutoff frequencies, with the
one having the lowest cutoff frequency at the input; otherwise, it may become saturated
while the output of the second stage is still below the maximum permissible voltage
swing. The reason for this is that the filter stages that have the higher cutoff frequencies
also invariably have the higher pole-pair Q, and therefore show a rise in gain in the vicinity
of their cutoff frequency. This can be seen in Fig. 13.24, where the amplitude–frequency
responses of a tenth-order 0.5 dB Chebyshev lowpass filter and of its five individual stages
are shown. It is obvious that the permissible voltage swing is highest if the filter stages
that have low cutoff frequencies are at the input of the filter cascade.
Another aspect that may have to be considered for a suitable arrangement of the filter
stages is noise. In this case, just the reverse sequence is the most favorable, as then the
filters with the lower cutoff frequencies at the end of the filter chain reduce the noise
introduced by the input stages.
The design process is demonstrated for a third-order Bessel lowpass filter. It is to be
constructed from the first-order lowpass filter shown in Fig. 13.16 and the second-order
lowpass filter shown in Fig. 13.21, for which the special case of α = 1 (described in
Sect. 13.4.3) is chosen. The low-frequency gain is defined as unity. To achieve this, the
impedance converter in the first-order filter stage must have a gain of α = 1. The resulting
circuit is represented in Fig. 13.25.
814 13 Active Filters
. . . .
Fig. 13.24. Amplitude–frequency responses of a tenth-order Chebyshev filter with 0.5 dB ripple
and of its five individual filter stages
The desired cutoff frequency is fc =100 Hz. For the calculation of the first filter stage,
we predetermine C1 = 100 nF and obtain, according to Sect. 13.3, with the coefficients
from Fig. 13.14:
a1 0.7560
R11 = = = 12.03 k
2πfc C11 2π · 100 Hz · 100 nF
For the second filter stage, we set C22 = 100 nF and obtain, in accordance with Sect. 13.4.3,
the condition for C21 :
a22 (0.9996)2
C21 ≤ C22 = 100 nF · = 52.3 nF
4b2 4 · 0.4772
We choose the nearest standard value C21 = 47 nF and arrive at
'
a2 C22 ∓ a22 C22 2 − 4b C C
2 21 22
R21/22 =
4πfc C21 C22
R21 = 11,51 k, R22 = 22,33 k
For third-order filters, it is possible to omit the first operational amplifier. The simple
lowpass filter of Fig. 13.1 is then connected in front of the second-order filter. Due to the
mutual loading of the filters, a different method of calculation must be employed, which is
considerably more difficult than in the decoupled case. Figure 13.26 shows such a circuit,
which has the same characteristics as that shown in Fig. 13.25.
.
. .
Vi
OA1 Vo
OA2
Fig. 13.25. Third-order Bessel lowpass filter with a cutoff frequency of fc = 100 Hz
13.6 Lowpass/Bandpass Transformation 815
. . .
Fig. 13.26. Simplified
Vi third-order Bessel lowpass
Vo filter with a cutoff frequency
of fc = 100 Hz
13.6
Lowpass/Bandpass Transformation
In Sect. 13.2 it was shown how, by transforming the frequency variable, a given lowpass
frequency response could be converted to the corresponding highpass frequency response.
The frequency response of a bandpass filter can be created using a very similar transfor-
mation; that is, by replacing the frequency variable sn in the lowpass transfer function by
the expression
1 1
sn → sn + (13.18)
ωn sn
By means of this transformation, the amplitude response of the lowpass filter in the range
0 ≤ ωn ≤ 1 is converted into the pass range of a bandpass filter between the center
frequency ωn = 1 and the upper cutoff frequency ωn, max . On a logarithmic frequency
scale, it also appears as a mirror image below the center frequency. The lower cutoff
frequency is then ωn, min = 1/ωn, max . This process is illustrated in Fig. 13.27.
The normalized bandwidth ωn = ωn, max −ωn, min can be chosen freely. The described
transformation results in the bandpass filter having the same gain at ωn, min and ωn, max as
the corresponding lowpass filter at ωn = 1. If the lowpass filter, as in the table given in
Fig. 13.14, is normalized with respect to the 3 dB cutoff frequency, ωn represents the
normalized 3 dB bandwidth of the bandpass filter. Since ωn = ωn, max − ωn, min and
ωn, max · ωn, min = 1, we obtain, for the normalized 3 dB cutoff frequencies,
1 1
ωn, max/ min = (ωn )2 + 4 ± ωn .
2 2
13.6.1
Second-Order Bandpass Filters
The simplest bandpass filter is obtained by applying the transformation equation (13.18)
to a first-order lowpass filter, where
A0
A(sn ) =
1 + sn
Therefore, the transfer function of the second-order bandpass filter results
A0 A0 ωn sn
A(sn ) = = (13.19)
1 1 1 + ωn sn + sn2
1+ sn +
ωn sn
The interesting parameters of bandpass filters are the gain Ar at the resonant frequency
fr , and the quality factor Q. It follows directly from the given transformation characteristic
that Ar = A0 . This can be easily verified by making ωn = 1; that is, sn = j in (13.19).
Since Ar is real, the phase shift at resonant frequency is zero.
As for a resonant circuit, the Q-factor is defined as the ratio of the resonant frequency
fr to the bandwidth B. Therefore,
fr fr 1 1
Q = = = = (13.20)
B f max − f min ωn, max − ωn, min ωn
Inserting this in (13.19) gives
(Ar /Q)sn
A(sn ) = (13.21)
1
1 + sn + sn2
Q
This equation is the transfer function of a second-order bandpass filter; it allows direct
identification of all parameters of interest.
With sn = j ωn , we obtain from (13.21) the frequency response of the amplitude and
the phase:
These two functions are shown in Fig. 13.28 for the Q-factors 1 and 10.
13.6.2
Fourth-Order Bandpass Filters
The amplitude–frequency response of second-order bandpass filters becomes more peaked
the larger is the Q-factor selected. However, there are many applications where the curve
must be as flat as possible in the region of the resonant frequency, but must also have a
steep transition to the stopband. This optimization problem may be solved by applying
the lowpass/bandpass transformation to higher-order lowpass filters. It is then possible to
choose freely not only the bandwidth ωn , but also the most suitable type of filter.
13.6 Lowpass/Bandpass Transformation 817
. . .
Fig. 13.28. Frequency response of the amplitude and the phase of second-order bandpass filters
with Q = 1 and Q = 10 or ωn = 1 and ωn = 0.1
.1 . .
Fig. 13.29. Frequency response of the amplitude and the phase for bandpass filters with a
bandwidth of ωn = 1
Curve 1: A fourth-order Butterworth bandpass filter. Curve 2: A fourth-order Chebyshev bandpass
filter with a 0.5 dB ripple. Curve 3: A second-order bandpass Q = 1 for comparison
818 13 Active Filters
second-order bandpass filters, which are not tuned to precisely the same center frequency.
This method is called “staggered tuning.” For the design of the individual bandpass filter
stages, we split the numerator of (13.25) into two factors that contain sn , and obtain
(Ar /Qi )(αsn ) (Ar /Qi )(sn /α)
A(sn ) = · 2 (13.28)
αsn
1+ + (αsn )2 1 + 1 sn + sn
Qi Qi α α
Comparing the coefficients with those of (13.25) and (13.21), we obtain the parameters of
the two individual bandpass filters:
13.7 Realization of Second-Order Bandpass Filters 819
fr Q Ar
√
1st filter stage fm /α Qi Qi ωn Am /b1 (13.29)
√
2nd filter stage fm · α Qi Qi ωn Am /b1
where fm is the center frequency of the resulting bandpass filter and Am is the gain at this
frequency. The factors α and Qi are given by (13.26) and (13.27).
The determination of the parameters of the individual filter stages will be demonstrated
by means of an example. A Butterworth bandpass filter is required, with a center frequency
of 1 kHz and a bandwidth of 100 Hz. The gain at the center frequency is required to be
Am = 1. To begin with, we take the coefficients of a second-order Butterworth lowpass filter
from the table in Fig. 13.14: a1 = 1.4142 and bi = 1.000. As ωn = 0.1, (13.26) gives
α = 1.0360. Equation (13.27) yields Qi = 14.15 and, from (13.29), Ar = 1.415, fr1 =
965 Hz, and fr2 = 1.036 kHz.
13.7
Realization of Second-Order Bandpass Filters
The cascade connection of a highpass and a lowpass filter of first order, as shown in
Fig. 13.30, gives a bandpass filter with the transfer function:
1 1 αs RC
A(s) = · =
s RC
1+
1 1 + α 2
1+ 1+ s RC + (s RC)2
α αs RC α
With the resonant frequency ωr = 1/RC, we obtain the normalized form. Comparison of
the coefficients with those of (13.21) gives the Q-factor:
α
Q =
1 + α2
For α = 1, Q max = 21 , which is the highest Q-factor that can be achieved by cascading
first-order filters. In this case the cutoff frequencies of the lowpass and the highpass are
the same. For higher Q-factors, the denominator of (13.21) must have complex zeros, but
R /a
Vi
Ra Vo
Lowpass wC = a Highpass wC = 1
RC aRC
Fig. 13.30. Bandpass filter consisting of a first-order lowpass filter and a highpass filter
αsn
A(sn ) =
1 + α2
1+ sn + sn2
α
820 13 Active Filters
C
R sn
L
Vi Vo A(sn ) =
C
1+R sn + sn2
L
Fig. 13.31. LRC bandpass filter
such a transfer function can only be implemented by LRC circuits or by special active
RC circuits, which are discussed below.
13.7.1
LRC Bandpass Filter
A common method of designing selective filters that have high quality factors is the use
of resonant circuits. Figure 13.31 shows such a circuit, the transfer function of which is
s RC
A(s) =
1 + s RC + s 2 LC
√
With the resonant frequency ωr = 1/ LC, we obtain the normalized expression as given
in Fig. 13.31. Comparison of the coefficients with those of (13.21) gives
1 L
Q = and Ar = 1
R C
For high frequencies, the inductances required can be easily implemented and suffer little
loss. In the low-frequency range, the inductances become unwieldy and have poor electrical
performance. If, for example, a filter having the resonant frequency fr = 10 Hz is to be
implemented using the circuit shown in Fig. 13.31, the inductance L = 25.3 H is required if
a capacitance of 10 mF is selected. As has already been shown in Sect. 13.4.1 for the lowpass
and highpass filters, such inductances can be emulated; for example, by using gyrators.
In most cases, the desired transfer function of (13.21) can be put into practice much more
easily by inserting suitable RC networks into the feedback loop of an operational amplifier.
13.7.2
Bandpass Filter with Multiple Negative Feedback
The principle of multiple negative feedback can also be applied to bandpass filters. The
appropriate circuit is shown in Fig. 13.32. As can be seen by comparison with (13.21),
the coefficient of sn2 must be unity. Therefore, the resonant frequency is given by
1 R1 + R 3
fr = (13.30)
2π C R1 R2 R3
Inserting this relation into the transfer function and comparing the remaining coefficients
13.7 Realization of Second-Order Bandpass Filters 821
Vi
Vo
Q
R2 = = 3.18 M
πfr C
R2
R1 = = 159 k
− 2Ar
−Ar R1
R3 = = 79.5
2Q2 + Ar
The open-loop gain of the operational amplifier must, at the resonant frequency, still be
large compared to 2Q2 = 20 000.
One advantage of the circuit is that it has no tendency to oscillate at the resonant fre-
quency, even if the circuit elements do not quite match their theoretical values. Obviously,
this is true only if the operational amplifier is correctly frequency compensated; otherwise,
high-frequency oscillations can occur.
13.7.3
Bandpass Filter with Single Positive Feedback
The application of single positive feedback results in the bandpass circuit shown in
Fig. 13.33. The negative feedback via the resistors R1 and (k − 1)R1 fixes the internal
gain at the value k. Comparison of the coefficients with those of (13.21) yields the equa-
tions given for determining the circuit parameters.
A disadvantage is that Q and Ar cannot be chosen independently of one another. The
advantage, however, is that the Q-factor may be altered by varying k without at the same
time changing the resonant frequency.
1
Resonant frequency: fr = 2πRC
k
Gain at fr : Ar = 3−k
Vi 1
Vo Q-factor: Q = 3−k
Condition: 1≤k<3
For k = 3, the gain is infinite, and an undamped oscillation occurs. The adjustment of
the internal gain therefore becomes more critical the closer it approaches the value 3.
13.8
Lowpass/Bandstop Filter Transformation
Selective rejection of a particular frequency requires a filter whose gain is zero at the
resonant frequency and rises to a constant value at higher and lower frequencies. Such
filters are called rejection filters, bandstop filters, or notch filters. To characterize the
selectivity, the rejection quality factor is defined as Q = fr /B, where B is the 3 dB
bandwidth. The larger the Q-factor of the filter, the more steeply the gain fails off in the
vicinity of the resonant frequencyfr .
As in the case of the bandpass filter, the amplitude–frequency response of the band-
rejection filter can be derived from the frequency response of a lowpass filter by using a
suitable frequency transformation. To accomplish this, the variable sn is replaced by the
expression
ωn
sn → (13.33)
1
sn +
sn
where ωn = 1/Q is the normalized 3 dB bandwidth. By means of this transformation,
the amplitude of the lowpass filter in the range 0 ≤ ωn ≤ 1 is converted to the pass band
of the band-rejection filter between 0 ≤ ωn ≤ ωn,c1 . In addition, it appears as a mirror
image above the resonant frequency, when plotted on a logarithmic scale. At the resonant
frequency ωn = 1, the transfer function is zero. As with the bandpass filter, the order of the
filter is doubled by the transformation. It is of particular interest to apply the transformation
to a first-order lowpass filter. This results in a notch filter of second order, which has the
transfer function:
A0 (1 + sn2 ) A0 (1 + sn2 )
A(sn ) = = (13.34)
1 + ωn sn + sn2 1
1 + sn + sn2
Q
From this, we obtain the relations for the amplitude–frequency response and the phase–
frequency response:
A0 |(1 − ωn2 )| ωn
|A| = , ϕ = arctan
1 Q(ωn2 − 1)
1 + ωn2 − 2 + ωn4
Q2
The corresponding curves are shown in Fig. 13.34 for the rejection quality factors 1 and 10.
The denominator of (13.34) is identical to that of (13.21) for bandpass filters. It
follows from (13.21) that a maximum Q-factor of only Q = 21 can be attained with
passive RC circuits; for higher Q-factors, LRC networks or special active RC circuits
must be employed.
824 13 Active Filters
. . .
Fig. 13.34. Frequency response of the amplitude and the phase for second-order bandstop filters
with Q = 1 and Q = 10
13.9
Realization of Second-Order Bandstop Filters
13.9.1
LRC Bandstop Filter
A well-known method of implementing rejection filters involves using series resonant
circuits, as shown in Fig. 13.35. At the resonant frequency, the arrangement represents a
short-circuit and the output voltage is zero. The transfer function of the circuit is
1 + s 2 LC
A(s) =
1 + s RC + s 2 LC
√
Hence, the resonant frequency is ωr = 1/ LC, and we obtain the normalized form as
given in Fig. 13.35. The rejection quality factor is found by comparing the coefficients
with those of (13.35):
1 L
Q =
R C
1 + sn2
Vi Vo A(sn ) =
C
1+R sn + sn2
L
Vi
Vo
1
Fig. 13.36. Active parallel-T bandstop filter Resonant frequency: fr =
2πRC
k(1 + sn2 ) Gain: A0 = k
A(sn ) = 1
1 + 2(2 − k)sn + sn2 Rejection Q-factor: Q=
2(2 − k)
Condition: 1≤k<2
This holds for lossless inductors only, because only then the output voltage falls to zero. In
addition, the same limitations for the use of inductances apply here as for bandpass filters.
13.9.2
Active Parallel-T Bandstop Filter
As shown in Sect. 29.3.6, the parallel-T filter represents a passive RC rejection filter. From
(29.3.24), its rejection Q-factor is Q = 0.25, which can be increased by incorporating
the parallel-T filter in the feedback loop of an amplifier. One possible implementation is
shown in Fig. 13.36.
For high and low frequencies, the parallel-T filter transfers the input signal unchanged.
The output voltage of the impedance converter is then kVi . At the resonant frequency,
the output voltage is zero. In this case, the parallel-T filter behaves as if resistor R/2
were connected to ground. Therefore, the resonant frequency fr = 1/2π RC, remains
unchanged.
From the transfer function, the filter data given below Fig. 13.36 can be directly de-
duced. If the voltage follower has unit gain, Q = 0.5. For an increase in gain, Q rises
toward infinity as k approaches 2.
A precondition for the correct operation of the circuit is the precise adjustment of the
resonant frequency and the gain of the parallel-T filter. This is difficult to achieve for higher
Q-factors, since varying one resistance always affects both parameters simultaneously. The
active Wien–Robinson bandstop filter is more favorable in this respect.
13.9.3
Active Wien–Robinson Bandstop Filter
As shown in Sect. 29.3.5, the Wien–Robinson bridge also behaves like a notch filter.
However, its Q-factor is not much higher than that of the parallel-T filter, but it can also
be increased to any desired value by incorporating the filter in the feedback loop of an
826 13 Active Filters
Vi
OA1 V1 OA2 Vo
1
Fig. 13.37. An active Wien–Robinson bandstop filter Resonant frequency: fr =
2πR2 C
β
(1 + sn2 ) Gain: A0 = −
β
1+α 1+α
A(sn ) = −
3 1+α
1+ sn + sn2 Rejection Q-factor: Q=
1+α 3
operational amplifier. The corresponding circuit is shown in Fig. 13.37, and its transfer
function is obtained from the relation for the Wien–Robinson bridge,
1 + sn2
Vo = V
1 + 3sn + sn 1
which directly yields the filter data given below Fig. 13.37. For the actual design of the
circuit, the parameters fr , A0 , Q, and C are defined and the remaining parameters are then
1
R2 = , α = 3Q − 1 and β = − 3A0 Q
2πfr C
In order to tune the filter to the resonant frequency, the capacitors Care changed in steps
and the two resistors R2 can be varied continuously using potentiometers. If the resonant
frequency is not fully suppressed due to a slight mismatch of the bridge components, the
final adjustment can be made by slightly varying resistor 2R3 .
13.10
Allpass Filters
13.10.1
Basic Principles
The filters discussed so far are circuits for which the gain and phase shift are frequency
dependent. In this section, we examine circuits for which the gain remains constant and
only the phase shift is dependent on the frequency. These are called allpass filters, and they
are used for phase correction and signal delay.
Initially, we show how the frequency response of an allpass filter can be derived from
the frequency response of a lowpass filter. To do this, the constant factor A0 in the numerator
of (13.11) is replaced by the conjugate complex denominator and, in this way, constant
13.10 Allpass Filters 827
where
0 ai ωn
ϕ = − 2α = − 2 arctan (13.36)
1 − bi ωn2
i
The use of allpass filters for signal delay is of particular interest. Constant gain, a
condition that is always fulfilled by allpass filters, is one prerequisite for undistorted signal
transfer. The second prerequisite is that the group delay of the circuit is constant for all
frequencies considered. The filters that best fulfill this condition are Bessel lowpass filters,
for which the group delay is Butterworth-approximated. Therefore, in order to obtain a
“Butterworth allpass filter,” the Bessel coefficients must be inserted into (13.35).
It is advisable, however, to renormalize the frequency responses thus obtained, as the
3 dB cutoff frequency of the lowpass filters is meaningless in this case. For this reason,√ we
recalculate the coefficients a1 and b1 so that the group delay at ωn = 1 is reduced to 1/ 2
of its low-frequency value. The coefficients obtained in this way are shown in Fig. 13.38
for filters of up to the tenth order.
The group delay is the time interval by which the signal is delayed in the allpass filter.
According to the definition given in (13.9b), it can be determined from (13.36),
dϕ 0 ai (1 + bi ωn2 )
Tgr = tgr · ωc = 2π tgr · fc = − = 2 (13.37)
dωn 1 + (ai2 − 2bi )ωn2 + bi2 ωn4
i
which √ is also given for each order in Fig. 13.38. In addition, the pole-pair quality factor
Qi = bi /ai is given. As it is unaffected by the renormalization, it has the same values
as for the Bessel filters.
To enable the correct operation of the individual filter stages to be checked, we have
also shown the ratio fi /fc in Fig. 13.38. Here, fi is the frequency at which the phase of
the particular filter stage approaches the value −180◦ for a second-order, or −90◦ for a
first-order, filter stage. This frequency is considerably easier to measure than the cutoff
frequency of the group delay.
The frequency response of the group delay is shown in Fig. 13.39 for allpass filters of
first to tenth order.
The following example shows the steps in the design of an allpass filter. A signal
that has a frequency spectrum of 0–1 kHz is to be delayed by tgr 0 = 2 ms. In order that
the phase distortion is not too large, the cutoff frequency of the allpass filter should be
fc ≥ 1 kHz. From (13.37), it is therefore necessary that
Tgr 0 ≥ 2 ms · 2π · 1 kHz = 12.566.
828 13 Active Filters
N i ai bi fi /fc Qi Tgr 0
1 1 0.6436 0.0000 1.554 – 1.2872
2 1 1.6278 0.8832 1.064 0.58 3.2556
3 1 1.1415 0.0000 0.876 – 5.3014
2 1.5092 1.0877 0.959 0.69
4 1 2.3370 1.4878 0.820 0.52 7.3752
2 1.3506 1.1837 0.919 0.81
5 1 1.2974 0.0000 0.771 – 9.4625
2 2.2224 1.5685 0.798 0.56
3 1.2116 1.2330 0.901 0.92
6 1 2.6117 1.7763 0.750 0.51 11.5579
2 2.0706 1.6015 0.790 0.61
3 1.0967 1.2596 0.891 1.02
7 1 1.3735 0.0000 0.728 – 13.6578
2 2.5320 1.8169 0.742 0.53
3 1.9211 1.6116 0.788 0.66
4 1.0023 1.2743 0.886 1.13
8 1 2.7541 1.9420 0.718 0.51 15.7607
2 2.4174 1.8300 0.739 0.56
3 1.7850 1.6101 0.788 0.71
4 0.9239 1.2822 0.883 1.23
9 1 1.4186 0.0000 0.705 – 17.8656
2 2.6979 1.9659 0.713 0.52
3 2.2940 1.8282 0.740 0.59
4 1.6644 1.6027 0.790 0.76
5 0.8579 1.2862 0.882 1.32
10 1 2.8406 2.0490 0.699 0.50 19.9717
2 2.6120 1.9714 0.712 0.54
3 2.1733 1.8184 0.742 0.62
4 1.5583 1.5923 0.792 0.81
5 0.8018 1.2877 0.881 1.42
Fig. 13.38. Allpass filter coefficients for a maximally flat group delay
Tgr
10
20
9
8
16
7
12 6
5
8 4
3
4 2
N= 1
0
0.01 0.03 0.1 0.3 1 3 10 30 !n
Fig. 13.39. Frequency response of the group delay for orders 1–10
13.10 Allpass Filters 829
Vi Vo
From Fig. 13.38, it can be seen that a filter of at least seventh order is needed, for which
Tgr0 = 13.6578. To make the group delay exactly 2 ms, the cutoff frequency must be
selected, in accordance with (13.37), as
Tgr 0 13.6578
fc = = = 1.087 kHz
2π · tgr 0 2π · 2 ms
13.10.2
Realization of First-Order Allpass Filters
The circuit in Fig. 13.40 exhibits a gain of +1 at low frequencies and a gain of −1 at high
frequencies; in other words, the phase shift changes from 0 to −180◦ . The circuit is an
allpass filter if the magnitude of the gain is also unity for the middle frequency range. To
examine this, we consider the transfer function shown in Fig. 13.40. The absolute value
of the gain is indeed constant and unity. Comparing the coefficients with those of (13.35)
gives
a1
RC =
2πfc
Using (13.37), the low-frequency value of the group delay is therefore
tgr 0 = 2RC
The first-order allpass filter shown in Fig. 13.40 is very well suited for use as a phase shifter
over a wide range of phase delays. By varying the value of resistor R, the phase delay can
be adjusted to values between 0 and −180◦ without affecting the amplitude. The phase
shift is
ϕ = − 2 arctan(ωRC)
13.10.3
Realization of Second-Order Allpass Filters
The second-order allpass filter transfer function can be implemented, for example, by
subtracting the output voltage of a bandpass filter from its input voltage. The transfer
function of this circuit is then
Ar
1 − Ar
sn + sn
2
sn 1+
Q Q
A(sn ) = 1 − =
1 1
1 + sn
+ sn
1 + sn
+ sn
2 2
Q Q
830 13 Active Filters
It can be seen that, for Ar = 2, the transfer function of an allpass filter is obtained. It is
normalized not to the cutoff frequency of the allpass, but to the resonant frequency of the
bandpass filter. For a correct normalization, we take
ωc = βωr
and obtain:
s βs
sn
= = = βsn
ωr ωc
Hence, the transfer function is:
β
1−sn + β 2 sn2
Q
A(sn ) =
β
1 + sn + β 2 sn2
Q
Comparing the coefficients with those of (13.35) yields
β
a1 = and b1 = β 2
Q
The data of the required bandpass filter are therefore
Ar = 2
:
fr = fc b1
:
Q = b1 a1 = Q1
As an example, let us consider implementation using the bandpass filter shown in Fig. 13.32.
As the Q-factors are relatively small, resistor R3 may be omitted and, instead, the gain
adjusted by resistor R/α in Fig. 13.41. The component values are obtained by comparing
the coefficients of the transfer function with those of (13.35):
a1 b1 a12 1
R1 = , R2 = and α = =
4πfc C πfc Ca1 b1 Q21
From the transfer function, a further application of the circuit in Fig. 13.41 can be deduced.
If 2R1 − αR2 = 0 a bandstop filter circuit is obtained.
Vi
OA1 Vo
OA2
13.11
Adjustable Universal Filters
As shown previously, the transfer function of a second-order filter element has the general
form
d0 + d1 sn + d2 sn2
A(sn ) = (13.38)
c0 + c1 sn + c2 sn2
The filter families described so far can be deduced from (13.38) by assigning special values
to the coefficients of the numerator:
lowpass filter: d1 = d2 = 0;
highpass filter: d0 = d1 = 0;
bandpass filter: d0 = d2 = 0;
bandstop filter: d1 = 0, d0 = d2 ;
allpass filter: d0 = c0 , d1 = −c1 , d2 = c2
The numerator coefficients may have either sign, whereas the coefficients of the denomi-
nator must always be positive for reasons of stability. The pole-pair Q-factor is defined by
the denominator coefficients
√
c 0 c2
Qi = (13.39)
c1
Vi
Vi
High- Low-
stop VBS pass VHP VBP pass VLP
Fig. 13.43. Universal filter of second order with independently adjustable parameters. State
Variable Filter, Biquad. Integration time constant τ = RC
R12 R1 R3 2
s
V LP R2 R4 V HP R2 R4 n
= =
Vi R1 R1 2 2 2 Vi R3 R3
1+ τ ωc sn + τ ωc sn + sn + sn2
R4 R3 R1 τ 2 ωc2 R4 τ ωc
(Lowpass filter) (Highpass filter)
R12 R1 R1 2 2 2
− τ ωr sn − 1+ τ ωr sn
V BP R2 R4 V BS R2 R3
= =
Vi R1 R1 2 2 2 Vi R1 R1 2 2 2
1+ τ ωr sn + τ ωr sn 1+ τ ωr sn + τ ωr sn
R4 R3 R4 R3
(Bandpass filter) (Bandstop filter)
frequency is determined solely by the product τ = RC. As these variables do not occur in
the equations for A and Q, the frequency can be varied without affecting A and Q. These
two parameters can be set independently using resistors R2 and R4 .
Also when operated as a bandpass or bandstop filter the resonance frequency, the gain
and the quality can be varied without mutually affecting each other, since the resonant
frequency is determined solely by the product τ = RC. Since these parameters do not
appear in the equations for A and Q the frequency can be varied without changing A and/or
Q. These two parameters can be adjusted by the resistances R2 and R4 independently of
each other.
As can be seen in Fig. 5.92 on page 560, integrators with VV operational amplifiers
are not very suitable for high frequencies. CC integrators are much better. Figure 5.93 on
page 561 shows an example.
Universal filters are available as integrated circuits requiring only a few external re-
sistors to determine the filter type and the cutoff frequency. Some examples are listed in
Fig. 13.44. By comparison with the very popular SC filters described in Sect. 13.12, the
continuous filters offer the advantage that they need no clock and therefore have no clock
noise.
divider can then be realized by low-resistance potentiometers. This method may also be
used for the resistors R1 and R2 .
If a filter parameter is to be voltage controlled, the voltage divider can be replaced by
an analog multiplier, where the control voltage is connected to the second input, as shown
in Fig. 13.45. The effective resistance is then
U
Requiv = R0 ·
Vcont
where Vcont is the controlling voltage. If two such circuits are inserted instead of the two
frequency-determining resistors R, the resonant frequency of the bandpass filter becomes
1 Vcont
fr = ·
2π R0 C U
and is proportional to the control voltage.
For numeric control of filter parameters – for example, via a computer – digital-to-
analog converters DACS can be used instead of analog multipliers. These provide an output
voltage that is proportional to the product of the applied number and the reference voltage:
Z
Vo = Vref
Z max + 1
The types preferred for use in filters are those whose reference voltage can assume any
positive or negative value. Consequently, the multiplying DACs with CMOS switches
described in Sect. 18.4 are particularly suitable for this purpose, but as they possess con-
siderable resistor tolerances, they cannot simply be inserted as series resistors in Fig. 13.43.
However, the effect of the absolute resistance value can be eliminated by providing a fol-
lowing operational amplifier with a feedback path via a resistor incorporated in the DAC.
Vx Vx Vcont
Vcont U
Q DAC 1 DAC 2
Vi A
The resulting circuit for digital frequency adjustment is shown in Fig. 13.46. The two
integrators are preceded by a DAC. The resulting integration time constant is
τ = RC(Z max + 1)/Z (13.41)
If the number Z equals the maximum value Z max – that is, all the bits are one – virtually
the same resonant frequency is obtained as in the circuit shown in Fig. 13.43.
In comparison with Fig. 13.43, the arrangement of the feedback loops is modified
somewhat, because the DACs in conjunction with the associated operational amplifiers
and following integrators form a noninverting integrator. However, the resulting transfer
functions are quite similar. It is particularly simple to select component values if we select
τ ωc = 1; that is, fc = 1/2πτ :
Lowpass filter: Highpass filter: Bandpass:
given R1 given R1 given R1
R3 = R1 /bi R3 = R1 bi R3 = R1
R4 = R1 /ai R4 = R3 /ai R4 = R1 Q
R2 = −R1 /A0 R2 = −R3 /A∞ R2 = −R1 Q/Ar
Substituting the integration time constant in (13.41), we can see that the cutoff and resonant
frequencies become proportional to the number Z:
1 1 Z
fc = = ·
2π τ 2πRC Z max + 1
836 13 Active Filters
The outputs of the D/A converters must have a large dynamic range in order to be able
to adjust the frequency over a wide band. To prevent any DC errors in the circuit, it is
necessary to use operational amplifiers with a low offset voltage. Suitable opamps can
be found in Fig. 5.103 on page 574. Suitable DACs include the AD 7528 (8-bit) or the
AD 7537 (12-bit) from Analog Devices, as these incorporate two D/A converters with a
common computer interface.
A considerably simpler way of implementing a frequency adjustable filter is to use
switched capacitor filters of the type described in Sect. 13.12. But they possess some
restrictions resulting from switching interference.
13.12
Switched Capacitor Filters
13.12.1
Principle
The active filters described above require an active component in the form of an operational
amplifier, as well as passive elements in the form of capacitors and resistors. Normally,
filters with a variable cutoff frequency are only realized by varying the capacitors or
resistors (see Fig. 13.46). However, it is also possible to simulate a resistor by means of a
switched capacitor. The principle involved is shown in Fig. 13.47.
If the switch connects the capacitor to the input voltage, capacitor C receives the charge
Q = CS · V . In the other switch position, the capacitor delivers the same charge again. In
each switching period it therefore transfers the charge Q = CS · V from the input to the
output of the circuit. This produces an average current flow of I = CS ·V /TS = CS ·V ·fS .
Comparing this relation with Ohm’s Law, the basic equivalence between the switched
capacitor and an ohmic resistor can be expressed in the form
I = V /Requiv = V · CS · fS with Requiv = 1/(CS · fS )
Note the proportional relationship between the switching frequency and the equivalent
conductance. It is this property that is utilized in switched capacitor (SC) filters.
13.12.2
SC Integrator
The switched capacitor can replace the ohmic resistor in the conventional integrator shown
in Fig. 13.48. The result is the SC integrator of Fig. 13.49. In this circuit the integration
time constant
C η
τ = C · Requiv = = (13.42)
CS · f S 2πfS
V V
Vi Vi
Vo Vo
can be set by the switching frequency fS . The capacitance ratio C/CS = η/2π is perma-
nently preset by the manufacturer; the parameter η can be obtained from the data sheet,
and is generally between 50 and 200.
However, the use of switched capacitors has yet more advantages: the implementation
of a noninverting integrator in conventional technology requires an inverting integrator
with a preceding or following voltage inverter. With the SC integrator, the polarity of
the input voltage can be changed simply by connecting the capacitor, which has been
charged up to the input voltage to be sampled, with reversed terminals to the operational
amplifier input during the following charge transfer phase. Reversal of the terminals can
be effected as shown in Fig. 13.50, using an additional changeover switch S2 that switches
simultaneously with S1 .
The charging and discharging of the capacitor CS does not occur instantaneously, but
exponentially due to the unavoidable resistances in the switches. Instantaneous charging
and discharging would also be completely undesirable, neither the input voltage source
nor the operational amplifier could supply the required currents. On the other hand, these
parasitic resistances also determine the maximum switching frequency, as a complete
charge–discharge cycle is impossible at too high a switching frequency.
13.12.3
First-Order SC Filter
The two basic SC integrators circuits can be extended by a feedback resistor in parallel to
C to produce a first-order lowpass filter similar to that shown in Fig. 13.17. However, a
different basic structure is generally selected for the monolithic version. This consists of an
Vi Vo Vi Vo
C η
τ = C f = 2πf
S S S
Vi
V LP −R1 /R2
Vi
= τ ωc R1
1+ · sn
R3
VHP VLP V HP −sn R3 /R1
Vi
= R3
+ sn
τ ωc R1
Fig. 13.51. First-order highpass and lowpass filter
In the case of first-order filters, for which a, 1 in accordance with Fig. 13.14, we therefore
have R3 = R1 . The gains of the lowpass and highpass filters are therefore equal, and thus
the circuit represents complementary highpass and lowpass filters.
13.12.4
Second-Order SC Filters
Second-order SC filters are mainly designed as “biquad” structures, as shown in Fig. 13.46.
As noninverting integrators are once again employed, we obtain the same structure and
identical transfer functions (monolithic IC universal filters always contain this biquad
structure). Unlike in the case of the RC filter, the integration time constant τ as formulated
in (13.42) is determined by the switching frequency fS selected.
To determine the transfer function, we derive the following relations from the circuit
in Fig. 13.52:
R3 R3 R3
VHP = − Vi − VBP − VLP
R2 R4 R1
1 1
VBP = VHP VTP = VBP
τs τs
From these, we can calculate the specified transfer function for the individual filters.
If the switching frequency is again made equal to the ηth multiple of the cutoff frequency
(or resonant frequency), we obtain τ ωc = 1, and the following design equations:
13.12 Switched Capacitor Filters 839
Integration constant
C η
τ = C F = 2πf
S S S
Vi
Fig. 13.52. SC biquad to realize second-order highpass, lowpass, and bandpass filters
V LP −R1 /R2 V HP −sn2 R3 /R2
= =
Vi R1 τ ωc R1 τ 2 ωc2 2 Vi R3
+
R3
sn + sn2
1+ sn + sn
R4 R3 R1 τ 2 ωc2 R4 τ ωc
(Lowpass) (Highpass)
V BP −sn τ ωr R1 /R2
= (Bandpass)
Vi R1 τ ωr R1 τ 2 ωr2 2
1+ sn + sn
R4 R3
When the component values of one filter type have been defined, the other two do not,
of course, necessarily have the same filter parameters. For the cutoff frequencies (or the
resonant frequency), the following relation applies:
:
fc LP b1 = fr BP = fc HP b1
As b1 = 1 for second-order filters, the three frequencies coincide. In this case, the
gains are given by
A0 = Ar /Q = A∞
13.12.5
Implementation of SC Filters with ICs
SC filters are of course implemented not with discrete components but using integrated
circuits that contain capacitors and operational amplifiers in addition to the switches. This
not only is simpler for the user but also offers significant advantages, as will be shown
below.
SC filter ICs employ the two-switch arrangement illustrated in Fig. 13.50, because this
arrangement compensates for the effect of stray capacitances. The changeover switches
are realized in the form of a transmission gate and are driven by an internal clock generator
that provides nonoverlapping timing signals. This insures that no charge is lost during
switching.
As we can see, the capacitance ratio C/Cs together with the switching frequency fs
determine the integration time constant. The basic advantage of an IC implementation is
that capacitance ratios with 0.1% tolerance can be produced. The use of monolithic SC
filters therefore provides well-reproducible accuracies. In addition, the time constant is
temperature-invariant, as the two capacitors exhibit identical temperature dependence if
both are integrated on the same chip. Reproducible time constants, which are otherwise
difficult and costly to achieve in IC technology, can be easily provided using SC devices.
To achieve this, the only ratio of the two capacitances must be appropriately selected.
13.12.6
General Considerations for Using SC Filters
Despite the clearly superior characteristics of modern SC circuit design, the use of these
components is subject to certain limitations, as these are actually sampling systems. Every
time Shannon’s sampling theorem is violated, one has to contend with unwanted mixing
products in the base frequency band (aliasing). Consequently, the input system must not
contain any frequency components above half the switching frequency fS . In order to insure
this, analog filtering is generally required at the input, which must introduce sufficient
attenuation (some 70 . . . 90 dB) at 21 fS . As the typical sampling frequency of SC filter ICs
is approximately 50 . . . 100 times the cutoff frequency, a second-order analog filter that
acts as an anti-aliasing filter is normally adequate for this purpose.
The output signal of an SC filter always has a staircase waveform, as the output voltage
only changes at the switching instant. It therefore contains spectral components associ-
ated with the switching frequency. Consequently, an analog smoothing filter must also be
provided at the output at high requirements.
13.12.7
A Survey of Available Types
The SC filters that are available nowadays mainly contain complete functional blocks
comprising SC integrators, summing circuits, and the associated (controllable) oscillators
for clock generation. Their arrangement on the chip is either permanently fixed by masks
(filters with fixed characteristics) or the components can be combined by the user as
required (universal filters with variable characteristics). As the universal filters require
external circuitry, the IC package involved must have more pins. The number of pins thus
limits the complexity of the filter and, therefore, universal filter ICs are for low orders only.
13.12 Switched Capacitor Filters 841
However, dual types in a single package are generally available, which can then be simply
cascaded, thus allowing fourth-order filters to be realized on a single chip.
Switching by means of the clock frequency produces background noise in the filters,
which limits the signal-to-noise ratio to about 70 . . . 90 dB (Fig. 13.53). This constitutes a
disadvantage compared with “continuous” RC filters.
Most manufacturers provide filter design programs free of charge which can be down-
loaded from their homepage. This is the most convenient method for dimensioning freely
reconfigurable filters.
Chapter 14:
Signal Generators
In this chapter we shall describe circuits that generate sinusoidal signals. In the case of LC
oscillators, the frequency is determined by a tuned circuit, in the case of crystal-controlled
oscillators a piezoelectric crystal is used, and with the Wien and differential-equation oscil-
lators, RC networks are the frequency-determining components. The function generators
primarily produce a triangular signal, which can be converted into sinusoidal form using
a suitable function network.
14.1
LC Oscillators
The simplest method of generating a sine wave is to use an amplifier to eliminate the
damping of an LC resonant circuit. In the following section, we deal with some of the
basic aspects of this method.
14.1.1
Condition for Oscillation
The principle of an oscillator circuit is shown in Fig. 14.1. The amplifier multiplies the
input voltage by the gain A, and thereby causes a parasitic phase shift a between V2 and
V1 . The load resistance RL and a frequency-dependent feedback network – for example, a
resonant circuit – are connected to the amplifier output. The voltage feedback is therefore
V 3 = k V2 and the phase shift between V 3 and V 2 is denoted by β.
To establish whether the circuit can produce oscillations, the feedback loop is opened.
An additional resistor Ri is introduced at the output of the feedback network, representing
the input resistance of the amplifier. An alternating voltage V 1 is applied to the amplifier
and V 3 is measured. The circuit is capable of producing oscillations if the output voltage
is the same as the input voltage. Hence, the necessary condition for oscillation is:
V1 = V3 = k A V1
The loop gain must therefore be
g = kA = 1 (14.1)
from which two conditions can be deduced; that is,
|g| = |k| · |A| = 1 (14.2)
feedback
amplifier feedback
VV11 network V3
V2 L i Fig. 14.1. Basic arrangement of
˛ ˇ
an oscillator
844 14 Signal Generators
V2 V1
and
α + β = 0, 2π, . . . (14.3)
Equation (14.2) is the amplitude condition, which states that a circuit can oscillate only if
the amplifier eliminates the attenuation due to the feedback network. The phase condition
of (14.3) states that an oscillation can arise only if the output voltage is in phase with the
input voltage. Details of the oscillation – for example, the frequency and waveform – can
only be obtained if we have additional information on the feedback network. To this end,
let us consider the LC oscillator in Fig. 14.2 as an example.
The noninverting amplifier multiplies the voltage V1 (t) by the gain A. As the output
resistance of the amplifier is low, the resonant circuit is damped by the parallel resistor R.
To calculate the feedback voltage, we apply KCL to node 1 and obtain
V2 − V1 1
− C V̇1 − V1 dt = 0
R L
As V2 = AV1 , it follows:
1−A 1
V̈1 + V̇1 + V1 = 0 (14.4)
RC LC
This is the differential equation of a damped oscillation. To abbreviate, we define
1−A 1
γ = and ω02 =
2RC LC
and therefore
V̈1 + 2γ V̇1 + ω02 V1 = 0
the solution of which is given by
'
V1 (t) = V0 · e−γ t sin( ω02 − γ 2 t) (14.5)
One must differentiate between three cases:
1) γ > 0; that is, A < 1.
The amplitude of the AC output voltage decreases exponentially; the oscillation is
damped.
2) γ = 0; that is, A = 1.
The result is a sinusoidal oscillation with the frequency ω0 = √ 1 and with constant
LC
amplitude.
3) γ < 0; that is, A > 1.
The amplitude of the AC output voltage rises exponentially.
14.1 LC Oscillators 845
With (14.2) we have the necessary condition for an oscillation. This can now be de-
scribed in more detail: For A = 1, a sinusoidal output voltage of constant amplitude and
the frequency
1
ω = ω0 = √
LC
is obtained. With reduced feedback the amplitude falls exponentially, while with increased
feedback the amplitude rises exponentially. To insure that the oscillation builds up after the
supply has been switched on, the gain A must initially be larger than unity. The amplitude
then rises exponentially until the amplifier begins to saturate. Because of the saturation,
A decreases until |g| = |k| · A| reaches the value 1; the output, however, is then no
longer sinusoidal. If a sinusoidal output is required, an additional gain control circuit must
insure that A = 1 before the amplifier saturates. For the high-frequency range, resonant
circuits with high Q-factors are usually simple to implement. The voltage of the resonant
circuit is still sinusoidal, even if the amplifier saturates. For this frequency range, additional
amplitude control is usually not required, and the voltage across the resonant circuit is then
taken as the output voltage.
14.1.2
Meissner Oscillator
The feature of a Meissner circuit is that the feedback is provided by a transformer. A capaci-
tor C, together with the transformer primary winding, forms the frequency-determining res-
onant circuit. Three Meissner oscillators, each in common-emitter connection, are shown
in Figs. 14.3–14.5. At the resonant frequency,
1
ω0 = √
LC
the amplified input voltage appears at the collector with maximum amplitude and with a
phase shift of 180◦ . Part of this alternating voltage is fed back via the secondary winding.
To fulfill the phase condition, the transformer must effect a further phase inversion of
180◦ . This is achieved by AC-grounding the secondary winding at the end that has the
same voltage polarity as the collector end of the primary winding. The dots on the two
Vz
Fig. 14.3. Biasing by a Fig. 14.4. Biasing by current Fig. 14.5. Current
constant base current feedback feedback for a negative
supply voltage
846 14 Signal Generators
windings indicate which winding ends have the same polarity. The turns ratio is selected
such that the magnitude of the loop gain k A at the resonant frequency is always larger
than unity. Oscillation then begins when the supply is switched on, its amplitude rising
exponentially until the transistor saturates. Saturation reduces the mean value of the gain
until |k A| = 1 and the amplitude of the oscillation remains constant. Two saturation effects
can be distinguished, that at the input and that at the output. The output saturation arises
when the collector–base junction is forward biased. This is the case for the circuits shown
in Figs. 14.3 and 14.5 when the collector potential goes negative. The maximum amplitude
of the oscillation is therefore V̂C = V + . The maxima of the collector potential are then
V̂CE max = 2V + . This affects the choice of the transistor. For the circuit in Fig. 14.4, the
maximum amplitude is smaller than V + , the reduction being due to the Zener voltage.
With heavy feedback, input saturation can also occur. Large input amplitudes arise,
which are rectified at the emitter–base junction. The capacitor C1 is therefore charged, and
the transistor conducts only during the positive peaks of the AC input voltage.
In the circuit of Fig. 14.3, a few oscillations may be sufficient to charge the capacitor
C1 to such a high negative voltage that the oscillation stops altogether. It restarts only after
the base potential has risen to +0.6 V with the relatively large time constant R1 C1 . In this
case, a sawtooth voltage appears across C1 . This arrangement has therefore often been
used as a sawtooth generator, such a circuit being known as a blocking oscillator.
To prevent the circuit from operating in the blocking oscillator mode, the input satura-
tion must be reduced by choosing fewer turns at the base side. In addition, the resistance of
the base biasing circuit should be kept as low as possible [14.1]. This is difficult to achieve
for the circuit shown in Fig. 14.3, since the base current would then be excessively high.
Biasing by series feedback, as shown in Figs. 14.4 and 14.5, is therefore preferable.
14.1.3
Hartley Oscillator
The Hartley oscillator resembles a Meissner oscillator. The only difference is that the
transformer is replaced by a tapped winding (an auto-transformer). The inductance of this
winding, together with the parallel-connected capacitor, determines the resonant frequency.
A Hartley oscillator in common-emitter connection is shown in Fig. 14.6.An alternating
voltage is applied to the base via capacitor C2 ; it is in phase opposition to the collector
voltage, so that positive feedback occurs. The amplitude of the feedback voltage can be
adjusted to the required value by appropriate positioning of the tap. As with the Meissner
oscillator in Fig. 14.5, the collector quiescent current is determined by the series feedback
resistor R1 .
For the Hartley oscillator in Fig. 14.7, the transistor is operated in common-base con-
nection. Therefore, a voltage must be taken from the inductor L by means of capacitor C1 ,
in phase with the collector voltage.
14.1.4
Colpitts Oscillator
A characteristic of the Colpitts circuit is the capacitive voltage divider, which determines
the fraction of the output voltage that is fed back. The series connection of the capacitors
acts as the oscillator capacitance; that is,
Ca Cb
C =
Ca + C b
The common-emitter circuit of Fig. 14.8 corresponds to the circuit in Fig. 14.6, but requires
an additional collector resistor R3 for applying the positive supply voltage.
The common-base connection is again much simpler, as can be seen in Fig. 14.9. It
corresponds to the Hartley oscillator of Fig. 14.7.
14.1.5
Emitter-Coupled LC Oscillator
A simple way of realizing an oscillator is to use a differential amplifier, as shown in
Fig. 14.10. As the base potential of T1 is in phase with the collector potential of T2 ,
positive feedback can be attained by directly connecting the two terminals. The loop gain
Fig. 14.11. Push–pull oscillator with Fig. 14.12. Push–pull oscillator with
inductive feedback capacitive feedback
14.1.6
Push–Pull Oscillators
Push–pull circuits are used in power amplifiers in order to achieve higher output powers
and better efficiency. For the same reason, these circuits can also be employed for the
design of oscillators. One such design, consisting basically of two Meissner oscillators in
which the transistors T1 and T2 are alternately conducting, is shown in Fig. 14.11.
As the base potential of one transistor is in phase with the collector potential of the
other, the secondary winding normally required for phase inversion can be omitted. This
version is shown in Fig. 14.12. The positive feedback is provided by the capacitive voltage
dividers C1 and C2 . The parallel resistive voltage dividers provide the bias.
In addition to providing greater output power, both circuits also generate fewer har-
monics than the single-ended oscillators.
The most suitable amplifier for an oscillator with a parallel resonant circuit is a voltage
controlled current source. For this purpose we use a CC-Opamp in Fig. 14.13. It does
not load the resonant circuit because it offers both a high input and output resistance.
The losses in the series resistance of the inductor R are compensated by an appropriate
transconductance of the amplifier with R1 . In order to reduce distortions by hard clipping
of the amplifier output R2 is added. The internal structure of a CC-Opamp in Fig. 5.80
shows a push-pull output stage.
14.2
Crystal Oscillators
The frequency of the LC oscillators described is not sufficiently constant for many appli-
cations, as it depends on the temperature coefficients of the capacitance and inductance
of the resonant circuit. Considerably more stable frequencies can be achieved by using
quartz crystals. Such a crystal can be excited by electric fields to vibrate mechanically
and, when provided with electrodes, behaves electrically like a resonant circuit that has a
high Q-factor. The temperature coefficient of the resonant frequency is very small. The
frequency stability that can be attained by a crystal oscillator is of the order of
f
= 10−6 . . . 10−10
f
14.2.1
Electrical Characteristics of a Quartz Crystal
The electrical behavior of a quartz crystal can be described by the equivalent circuit in
Fig. 14.14. The two parameters C and L are well defined by the mechanical properties
of the crystal. The resistance R is small and characterizes the damping. C0 represents the
value of the capacitance formed by the electrodes and leads. Typical values for a 4 MHz
crystal are
giving a Q-factor of
-
1 L
Q = = 26000
R C
To calculate the resonant frequency, we initially determine the impedance of the quartz
crystal. From Fig. 14.14, and neglecting R,
1 + s 2 LC
Zq = (14.6)
s(C0 + C) + s 3 LCC0
Fig. 14.14. Equivalent circuit of a quartz Fig. 14.15. Tuning the series resonant
crystal frequency
850 14 Signal Generators
It can be seen that there is a frequency for which Z q = 0 and another for which Z q = ∞.
The quartz crystal therefore has a series and a parallel resonance. To calculate the series
resonant frequency fs , the numerator of (14.6) is set to zero, and thus
1
fS = √ (14.7)
2π LC
The parallel resonant frequency is calculated by setting the denominator to zero:
1 C
fP = √ 1+ (14.8)
2π LC C0
As can be seen, the series resonant frequency is dependent only on the well-defined product
LC, whereas the parallel resonant frequency is influenced by the electrode capacitance
C0 , which is far more susceptible to variations.
The frequency of a quartz oscillator must often be adjustable within a small range. This
can be achieved by simply connecting a capacitor CS in series with the quartz crystal, as
shown in Fig. 14.15; CS must be large compared to C.
To calculate the shift in the resonant frequency, we determine the impedance of the
series connection. Using Eq. (14.6), we obtain
C + C0 + CS + s 2 LC(C0 + CS )
Z
q = (14.9)
sCS (C0 + C) + s 3 LCC0 CS
Setting the numerator to zero gives the new series resonant frequency:
1 C C
fS = √ 1+ = fS 1 + (14.10)
2π LC C0 + C S C0 + C S
By expanding this into a power series, we arrive at the approximation
C
fS
= fS 1 +
2(C0 + CS )
The relative shift in frequency is therefore
f C
=
f 2(C0 + CS )
The parallel resonant frequency is not changed by CS , as the poles of (14.9) are independent
of CS . A comparison of (14.10) and (14.8) shows that, for CS → ∞, the series resonant
frequency cannot be raised to a value higher than that of the parallel resonant frequency.
14.2.2
Fundamental Frequency Oscillators
In the Pierce oscillator shown in Fig. 14.16, the crystal in conjunction with capacitors CS
and C1 forms a series resonant circuit with a series capacitance of
1 1 1
= +
CS ges CS C1
The resonant circuit is excited via the collector. Assuming that the current in the oscillatory
circuit is large compared to the excitation current, antiphase signals will be produced at
C1 and CS , resulting in positive feedback.
14.2 Crystal Oscillators 851
Vo
Vo
Fig. 14.16. Pierce oscillator with an amplifier Fig. 14.17. Pierce oscillator with a CMOS
in common-emitter connection inverter as an amplifier
Vo1 Vo1
Nowadays, the amplifiers are usually CMOS inverters. The resulting circuit is shown in
Fig. 14.17. Not only does it require fewer components, but it also imposes little damping
on the crystal, as its input resistance is high. The resistor fixes the operating point at
Vi = Vo ≈ 21 Vb . Its resistance may be very high, as virtually no input current flows.
The crystal-controlled oscillator shown in Fig. 14.18 operates in the same way as the
emitter-coupled multi-vibrator shown in Fig. 6.21. The amount of positive feedback can
be adjusted via the transconductance of the transistors, using the emitter resistors. It is
selected such that the circuit begins to oscillate reliably, but is not excessively overdriven.
The output voltage difference, and hence the current flowing through the crystal, are then
virtually sinusoidal. An automatic gain control device of this kind is incorporated, for
example, in type MC 12061.
A precision crystal oscillator that allows grounded crystals to be employed is shown in
Fig. 14.19. In order not to impair the Q-factor of the crystal, the circuit must be driven at the
Vo
lowest possible impedance (series resonance). Emitter follower T1 is used for this purpose.
The current I flowing through the crystal is translated into a voltage VC2 = I R2
in transistor T2 , which is configured as a current–voltage converter. Positive feedback is
provided via the emitter follower T4 and the base of T1 . The reduced transconductance of
T1 and hence the loop gain of the circuit is at its maximum at the series resonant frequency
of the crystal. Attenuator R5 , R6 is adjusted such that the AC voltage across the crystal
is only a few tens of millivolts. The power dissipation in the crystal is then so small that
the frequency stability is unimpaired. It is preferable to use an electrically controllable
attenuator – for example, a transconductance multiplier – which is set to the correct value
using an amplitude control circuit. This also insures reliable start-up of the oscillator,
and the output voltage has a good sinusoidal waveform. Complete crystal-oscillators are
offered in a wide variety in the frequency range from 32 kHz to 50 MHz. Therefore the
chips in Fig. 14.20 are seldom used today.
14.2.3
Harmonic Oscillators
Crystals for frequencies above 30 MHz are difficult to manufacture. To obtain high fre-
quencies of this kind with crystal accuracy, one can either stabilize an LC oscillator via a
PLL (see Sect. 22.4.5) using a low-frequency crystal, or excite a crystal at a harmonic of
its characteristic frequency.
If we examine the frequency response of the crystal impedance shown in Fig. 14.21,
we can see that it also possesses resonance points at odd-numbered harmonics. However,
the circuits considered so far are unsuitable for operating a crystal at a harmonic. For such
a circuit, we require an amplifier that provides the maximum gain close to the desired
frequency. This can be achieved by using an additional LC resonant circuit.
If the positive feedback of the Hartley oscillator shown in Fig. 14.7 is obtained via a
crystal, we obtain the circuit shown in Fig. 14.22. The LC resonant circuit is tuned to the
required harmonic. The gain is then a maximum for this frequency, and the crystal will
Q Q
Fig. 14.22. Hartley oscillator with a crystal Fig. 14.23. Colpitts oscillator with a crystal
Vo
tend to be excited at the corresponding harmonic. The suitably modified Colpitts oscillator
of Fig. 14.9 is shown in Fig. 14.23.
A harmonic oscillator can also be implemented using the emitter-coupled oscillator in
Fig. 14.10. For this purpose, a crystal is inserted in the positive feedback loop, as shown in
Fig. 14.24. This arrangement provides positive feedback with the required crystal harmonic
at the resonant frequency of the LC resonant circuit. The simplest way of implementing
the high-frequency amplifier required is to use an ECL gate. In this case, a line receiver
provides a particularly effective solution, as the reference potential VBB is accessible. If
the resonant circuit is connected as shown in Fig. 14.24, the amplifier will be optimally
biased. Capacitor C1 serves merely to short-circuit VBB at high frequencies. The resultant
output voltage is very nearly sinusoidal. If a square-wave ECL signal is required, it is
merely necessary to connect another line receiver at the output.
14.3
Wien–Robinson Oscillator
In the low-frequency range, LC oscillators are less satisfactory because the inductances and
capacitances required become unpleasantly large. Consequently, oscillators are preferred
in which RC networks are used to determine the frequency.
854 14 Signal Generators
Vi
V1 V2
Fig. 14.25. Frequency response of the phase shift Fig. 14.26. Detuned Wien–Robinson
Curve 1: Wien–Robinson bridge for ε = 0.01 bridge
Curve 2: Resonant circuit for Q = 10
Curve 3: Passive bandpass filter Q = 13
VD
Vo
the phase shift is not limited to ± 90◦ , but to ± 180◦ . Harmonics are thereby strongly
damped. One disadvantage of the Wien–Robinson bridge is that the attenuation at the
resonant frequency becomes higher the smaller is the value of ε. In general, the attenuation
at resonant frequency is
V̂D ε
= k≈
V̂i 9
and, in our example, k ≈ 900 1
. To fulfill the amplitude condition of an oscillator, the
amplifier must compensate for the attenuation. Such an oscillator circuit is represented in
Fig. 14.27.
If the amplifier has open-loop gain AD , the detuning factor ε must have the value
ε = 9k = 9/AD to fulfill the amplitude condition kAD = 1. If ε is slightly larger, the
oscillation amplitude increases until the amplifier begins to saturate. If ε is smaller, there
will be no oscillation. However, it is impossible to adjust the resistors R1 and R1 /(2 + ε)
with the required precision. Therefore, one of the two resistances must be controlled
automatically, depending on the output amplitude. This is the purpose of the field effect
transistor T in Fig. 14.27. As shown in Sect. 3.1.3, the channel resistance RDS behaves like
a controllable ohmic resistor if VDS remains sufficiently small. To insure that VDS does
not become too large, only part of VN is applied to the FET and the rest appears at R2 .
The sum of R2 and RDS must have the value R1 /(2 + εT ). The smallest possible value of
RDS is RDS on , and hence R2 must be smaller than
1
R2 < R1 − RDS on
2
When the supply voltage is switched on, VG is initially zero and therefore RDS = RDS on .
If the above design condition is fulfilled, the sum of R2 and RDS is then smaller than 21 R1 .
At the resonant frequency of the Wien–Robinson bridge, there is therefore a relatively large
signal VD . As a consequence, oscillation begins and the amplitude increases. The output
voltage is rectified by the voltage doubler D1 , D2 . The gate potential thereby becomes
negative and RDS increases. The amplitude of the output voltage rises until
R1 R1
RDS + R2 = =
2+ε 9
2+
AD
856 14 Signal Generators
Vo
Vo
Rectifier
A
A
Vref
The distortion factor of the output voltage is mainly dependent on the linearity of the FET
output characteristic, which can be greatly improved if part of the drain–source voltage is
added to the gate potential, as shown in Fig. 3.13. This is the purpose of the two resistors
R3 and R4 . The capacitor C3 insures that no direct current will flow into the N-input of the
operational amplifier, as this would cause an output offset. In practice, R3 ≈ R4 but, by
fine adjustment of R3 ,the distortion factor can be reduced to a minimum and values below
0.1% can be achieved.
If R is made adjustable, the frequency can be continuously controlled. Good matching
of the two resistors is important, since the greater their mismatch, the more efficient the
amplitude control must be. The maximum value of R should be low enough to insure
that there is no noticeable voltage arising from the input bias current of the operational
amplifier. On the other hand, R must not be too low, or the output will be overloaded.
To adjust the frequency within a range of 1 : 10, fixed resistors with the value R/10 are
connected in series with the potentiometers R. If, in addition, switches are used to select
different values for the capacitances C, a range of output frequencies from 10 Hz to 1
MHz can be covered by this circuit. To insure that the amplitude control does not cause
distortion even at the lowest frequency, the charge and discharge time constants, R5 C1
and R6 C2 respectively, must be larger than the longest oscillation period by a factor of at
least 10.
The data of the field effect transistor T determine the output amplitude. The stability of
the output voltage amplitude is not particularly good, since some deviation in amplitude
is required to effect a noticeable change in the resistance of the FET. It can be improved
by amplifying the gate voltage, and such a circuit is shown in Fig. 14.28.
The rectified AC output voltage is applied to OA2, which is configured as a modified
PI-controller (Fig. 22.7). This adjusts the gate potential of the FET in such a way that
the mean value of |V o | equals Vref . The controller time constant must be large compared
to the oscillation period, or the gain will change even within a single cycle, resulting in
considerable distortion. Therefore, a pure PI-controller cannot be used; it is better to have a
modified type in which a capacitor is connected in parallel with R6 . The alternating voltage
at R6 is thereby short-circuited even at the lowest oscillator frequency. The proportional
controller action comes into effect only below this frequency.
14.4 Differential-Equation Oscillators 857
14.4
Differential-Equation Oscillators
Low-frequency oscillations can be generated by programming operational amplifiers to
solve the differential equation of a sine wave. From Sect. 14.1.1,
V̈o + 2γ V̇o + ω02 Vo = 0 (14.11)
the solution of which is
'
Vo (t) = V̂o e−γ t sin( ω02 − γ 2 t) (14.12)
Since operational amplifiers are better suited to integration than differentiation, the differ-
ential equation is rearranged by integrating it twice:
Vo + 2γ Vo dt + ω02 Vo dt 2 = 0
This equation can be implemented using two integrators and an inverting amplifier, and
several arrangements are available. One of these, which is particularly suitable for an
oscillator, is shown in Fig. 14.29. For this circuit, the damping is given by γ = −α/20RC
and the resonant frequency f0 = 1/2πRC. Hence, from (14.12), the output voltage is:
⎛ ⎞
α 2 t
t α
Vo (t) = V̂o e 20RC sin ⎝ 1 − ⎠ (14.13)
400 RC
It is evident that the damping of the oscillation is adjusted by α. At the right-hand stop
of potentiometer P, α = 1. At the left-hand stop, α = − 1; in the middle, α = 0. The
damping can thus be varied between positive and negative values. For α = 1, the oscillation
amplitude is increased, within 20 cycles, by the factor e; for α = − 1, it is decreased by
the factor 1/e. For α = 0, the oscillation is undamped, although this is the case only for
ideal conditions. In practice, for α = 0, a slightly damped oscillation will occur, and to
achieve constant amplitude, α must be given a small positive value. The adjustment is so
critical that the amplitude can never be kept constant over a longer period, and one must
therefore introduce automatic amplitude control. As for the Wien–Robinson oscillator
shown in Fig. 14.28, the amplitude at the output can be measured by a rectifier and α
controlled, depending on the difference between this amplitude and a reference voltage.
As shown previously, the controller time constant must be large compared to the period of
V1 V2 Vo
A A A
Vo
Fig. 14.29. Second-order differential equation for sine-wave generation. The circuit can also be
seen as a biquad-filter with Q = ∞
Resonant frequency: f0 = 1/2πRC
858 14 Signal Generators
V1 V2 Vo
A A A
V3 Vo
U
V12
U
Vo2
U
A V3
Vref
Fig. 14.30. Differential-equation oscillator and supplementary circuit for precise amplitude
control. U is the compting unit of the multipliers for instance U = 10 V
Frequency: f0 = 1/2πRC, Amplitude: V̂o2 = U Vref
oscillation to insure that amplitude control causes no distortion. This requirement becomes
increasingly difficult to fulfill for frequencies below 10 Hz.
The difficulties arise from the fact that a whole cycle of the oscillation must be allowed
to occur before the correct amplitude is known. Such problems can be eliminated if the
amplitude is measured at every instant of the oscillation period. This is possible for the
circuit in Fig. 14.29 where, in the case of an undamped oscillation,
1
Vo = V̂o sin ω0 t and V1 = − Vo dt = V̂o cos ω0 t
τ
The amplitude can now be determined at any instant by calculation of the vector magnitude
Vo2 + V12 = V̂o2 (sin2 ω0 t + cos2 ω0 t) = V̂o2 (14.14)
It is obvious that the expression Vo2 + V12 is dependent only on the amplitude of the output
signal and not on its phase. A pure DC voltage is therefore obtained, which requires no
filtering and which can be directly compared with the reference voltage.
A differential-equation oscillator whose amplitude is controlled in this way is shown in
Fig. 14.30. Analog multipliers M1 and M2 form the squares of V1 and Vo respectively. To
these two portions, the reference voltage at the summation point of automatic gain control
(AGC) amplifier OA4 is now added, giving an output voltage V3 such that
V12 V2 Vref
+ o − = 0
U R2 U R2 R2
Applying (14.14), this is the case for an amplitude of V̂o2 = U Vref . The time constant
of the control amplifier is determined by RC network R3 C1 . The design of the circuit is
described in Chap. 22.
Voltage Vo V3 /U is present at the output of multiplier M3 . This voltage is applied to
resistor 10R instead of potentiometer P in Fig. 14.29, making α = V3 /U . If the amplitude
14.5 Function Generators 859
increases, V̂o2 > U Vref and both V3 and α become negative; that is, the oscillation is
damped. If the amplitude decreases, V3 becomes positive and the oscillation is exited.
Apart from providing a good method of controlling the amplitude, sine-wave generation
by solving a second-order differential equation offers a further advantage in that it allows
virtually ideal frequency modulation. If this is to be accomplished for LC oscillators, the
value of L or C must be varied. However, this will alter the energy of the oscillator and
hence the oscillation amplitude, and parametric effects arise. For the differential-equation
method, however, the resonant frequency can be changed by varying the two resistors R
without affecting the oscillator energy.
As each of the two resistors is connected to virtual ground, multipliers connected in
front of them can be used to modulate the frequency. The multipliers then produce the
output voltages:
VSt VSt
Vo
= Vo and V1
= V1
U U
This is equivalent to an increase in the resistance R by the factor U/VSt , so that the resonant
frequency is given by
1 VSt
f0 = ·
2π RC U
14.5
Function Generators
As we have seen, the amplitude control involved in the generation of low-frequency sine
waves is rather cumbersome. It is much easier to use a Schmitt trigger and an integrator
to generate a triangular alternating voltage. A sine wave can then be produced if the sine
function network of Sect. 11.7.4 is employed. Since, with this method, a triangular wave,
a square wave, and a sinusoidal wave are obtained simultaneously, circuits based on this
principle are called function generators. The block diagram of such a circuit is shown in
Fig. 14.31.
The principle consists in applying a constant voltage to an integrator. This voltage is
either positive or negative depending on the direction in which the integrator output voltage
is to be changed. If the integrator output voltage reaches the switch-on or switch-off level
of the following Schmitt trigger, the sign at the integrator input is inverted. This produces
a triangular voltage at the output, which oscillates between the trigger levels.
860 14 Signal Generators
Analog
switch Vsqu
Vtri
Sinus
Network Vsine
14.5.1
Basic Arrangement
Two approaches are possible, which differ in the way in which integration is implemented.
In the circuit shown in Fig. 14.32, +Vi or −Vi is applied to an integrator, depending on
the position of the analog switch. With the circuit in Fig. 14.33, current +Ii or −Ii is
impressed on capacitor C via an analog switch. This also results in a rise or fall in the
linear voltage. In order not to falsify the triangular voltage across the capacitor due to
loading, an impedance converter is generally required. The advantage of the method in
Fig. 14.33, however, is that it is easier to implement the impedance converter and current
switch for higher frequencies.
+Vi
–Vi
Vtri Vsqu
+ Ii
Impedance
Converter
– Ii
Vtri Vsqu
Vtri Vsqu
A Cmp
14.5.2
Practical Implementation
The simplest method of implementation is to start from the principle illustrated in
Fig. 14.32, and to use the output voltage of the Schmitt trigger itself as the input volt-
age for the integrator. The resulting circuit is shown in Fig. 14.34. The Schmitt trigger
supplies a constant output voltage, which is integrated by the integrator. If the integrator
output voltage reaches the trigger level of the Schmitt trigger, the voltage Vsqu to be in-
tegrated instantaneously changes sign, causing the integrator output to reverse direction
until the other trigger level is reached. In order to insure that the absolute values of the
positive and negative slopes are the same, the comparator must have a symmetrical output
voltage ±Vsqu max . In accordance with Sect. 6.5.2, the amplitude of the triangular voltage
is given by
R1
V̂tri = Vsqu max
R2
The oscillation period is equal to four times the time required by the integrator to change
from zero to V̂tri . It is therefore
R1
T = 4 RC
R2
An example of a practical implementation of the circuit principle of Fig. 14.33 is shown
in Fig. 14.35. The controlled current switch consists of transistors T1 − T3 . As long as
control signal x = L, the capacitor is discharged via T1 with current I . If the triangular
voltage falls below − 1 V, the precision Schmitt trigger from Fig. 6.48 changes state, and
x = H . This turns T3 off and current source T2 is turned on. The latter supplies twice as
much current as T1 , namely 2I . This means that capacitor C is charged with current I and
T1 needs not be turned off.
If the triangular voltage exceeds the upper trigger level of +1 V, the Schmitt trigger
reverts to the state x = L, and capacitor C is discharged again.
The dual comparator NE 521 from On semiconductor is particularly suitable for im-
plementing the precision Schmitt trigger, as it already contains the two gates required. This
comparator additionally features particularly short switching times of only some 8 ns, en-
abling frequencies of up to several MHz to be generated. The impedance converter shown
in Fig. 14.33 is only required if the triangular voltage is to be loaded. The subsequent
comparators place virtually no load on the triangular voltage.
862 14 Signal Generators
Cmp2 Vsqu
Cmp1
Vtri
Fig. 14.35. High-speed function generator with a current switch and a precision comparator
I 0.6
Frequency: f = = , Amplitude: V̂tri = 1 V
4V̂tri · C RC
14.5.3
Function Generators with a Controllable Frequency
Using the principle illustrated in Fig. 14.32, the frequency can be controlled quite easily
by varying the voltages +Vi and −Vi . An example of a function generator of this type
is shown in Fig. 14.36. The voltages +Vi or −Vi are available at low impedance at the
outputs of OA1 and OA2. These voltages are applied to the integrator input via transistor
T1 or transistor T2 , depending on the switching state of the Schmitt trigger. If the output
voltages of the comparator are greater than ±Vi , the two transistors operate as saturated
emitter followers and therefore only have a voltage drop of a few millivolts, as described
in Sect. 17.2.3.
+Vi
A
Vi
+
–Vi
A Cmp1
–Vi Vtri Vsqu
A
The Schmitt trigger once again determines the amplitude of the triangular signal, which
is given by
R1
V̂tri = Vsqu max
R2
The rate of change of the triangular voltage is:
Vtri Vi
= ±
t RC
The period is equal to four times the time required by the integrator to go from zero to
V̂tri . The frequency is therefore
Vi R2 1 Vi
f = = · ·
4RC V̂tri 4R1 RC Vsqu max
in other words, it is proportional to the input voltage Vi and so the circuit is suitable for
use as a voltage–frequency converter. If we select
Vi = Vi 0 + Vi
we obtain a linear frequency modulation.
If accuracy and stability of the amplitude and frequency are essential, care must be
taken to insure that they are not dependent on Vsqu max . This is easily achieved by using
a precision Schmitt trigger, as shown in Fig. 14.35. However, we then need an additional
amplifier to generate the bipolar signals that are required for driving T1 and T2 . In this
case, it is simpler to replace the bipolar transistors by CMOS analog switches.
Vsqu
Vtri
Vi
Vi
If we now want to change the symmetry without changing the frequency, we have to
increase the absolute value of one potential and reduce that of the other, so that
1 1
T = t1 + t2 = 2RC V̂tri + (14.15)
V1 |V2 |
remains constant. This condition can also be easily satisfied by using the drive circuit
shown in Fig. 14.38. Its output potentials are given by
1 1 1 1
+ = [R3 + (1 − α)R4 + R3 + αR4 ] = [2R3 + R4 ]
V1 |V2 | Vi R3 Vi R 3
As required, this expression holds irrespective of the duty cycle selected. By substitution
in (14.15), we obtain the frequency:
R3 Vi
f = ·
2RC[2R3 + R4 ] V̂tri
The duty cycle t1 /T or t2 /T can be set between
R3 R3 + R4
and
2R3 + R4 2R3 + R4
using potentiometer R4 . With R4 = 3R3 , we obtain values between 20% and 80%.
Function generators that not only produce triangular and square-wave voltages but also
contain a sine-wave function network are available in IC form. Some types are collected
in Fig. 14.39. The use of these circuits constitutes the simplest method of implementing
function generators. However, the signal quality and the usable frequency range are limited.
14.5.4
Simultaneously Producing Sine and Cosine Signals
The easy amplitude stabilization of function generators can also be utilized for the simul-
taneous generation of sine and cosine signals. Taking the triangular signal of a function
generator, its changes in sign can be determined by means of a comparator. The resulting
14.5 Function Generators 865
Cmp2
Cmp1 A A
Vsqu sin Vtri sin Vsqu cos Vtri cos
Fig. 14.40. Function generator for producing triangular and square-wave signals that are
phase-shifted by 90◦
R2 1 R1
Frequency f = , Amplitude: V̂tri = Vsqu max
4R1 RC R2
square wave is shifted by 90◦ with respect to the square wave at the input of the integrator.
Using a second integrator, the resulting square-wave signal can be converted to a triangular
signal, which is also phase-shifted by 90◦ with respect to the original triangular signal.
A simple version of this principle is illustrated in Fig. 14.40. Operational amplifier
OA1 and comparator Cmp1 constitute a function generator of the type shown in Fig. 14.34.
Comparator Cmp2 produces the phase-shifted square-wave signal and integrator OA2 the
associated triangular signal.
However, the circuit would not operate without a feedback path via R3 : integrator OA2
would run away due to the unavoidable symmetry and offset errors. This can be avoided
by inserting an additional resistor R3 . Voltage Vtri sin can be shifted via this resistor toward
positive or negative values, thereby also allowing the duty factor of Vsqu cos to be varied.
The feedback via R3 virtually cancels out the DC voltage superimposed at output Vtri cos .
It is not immediately apparent why the triangular voltage present at output Vtri cos ,
and fed back via R3 , does not impair the operation of function generator Cmp1, OA1.
The reason may be seen in Fig. 14.41, which shows that the triangular voltage Vtri cos is
zero at the peak values of Vtri sin and therefore does not alter the switching instant of the
comparator Cmp1. This can only happen due to a superimposed DC voltage.
Fig. 14.41. Time diagram of triangular and square-wave voltages shifted by 90◦
Chapter 15:
Power Amplifiers
Power amplifiers are designed to provide large output powers, with the voltage gain playing
only a minor role. Normally, the voltage gain of a power output stage is near unity and
the power gain is thus mainly due to the current gain of the circuit. The output voltage
and current must be able to assume positive and negative values. Power amplifiers with
unidirectional output current are known as power supplies. They are discussed in Chap. 16.
15.1
Emitter Follower as a Power Amplifier
The operation of the emitter follower has already been described in Sect. 2.4.2. Here, we
define some of the parameters that are of particular interest for its application as a power
amplifier. First, we calculate the load resistance for which the circuit in Fig. 15.1 delivers
maximum power without distortion. If the output is negative, RL carries some of the current
flowing through RE . The limit for control of the output voltage is reached when the current
through the transistor becomes zero. This is the case for the output voltage
RL
Vo min = − · Vb
RE + R L
If the output voltage is to be controlled sinusoidally around 0 V, its amplitude must not
exceed the value
RL
V̂o max = · Vb
RE + R L
Vi
L Vo
1 V̂o2max Vb2 RL
PL = = .
2 RL 2(RE + RL )2
dPL
With dRL = 0 it follows that, for RL = RE , the maximum output power,
Vb2
PL max =
8RE
is attained. This result is surprising in that one would normally expect maximum power
output when the load resistance equals the output resistance ro of the voltage source.
However, this is only the case for a constant open-circuit voltage. Here, the open-circuit
voltage is not constant, as it must be reduced when RL is small.
In the next step, we determine the power consumption within the circuit for any output
voltage amplitude and any load resistance. For a sinusoidal voltage, the power
1 V̂o2
PL =
2 RL
is supplied to the load resistance RL . The power dissipation of the transistor is
T
1 Vo (t) Vo (t) + Vb
PT = (Vb − Vo (t)) + dt.
T RL RE
0
Vb2 1 V̂o2
PE = + .
RE 2 RE
The circuit therefore draws from the power supplies a total power of
Vb2
Ptot = PL + PT + PE = 2
RE
This is a surprising result, since it shows that the total power of the circuit is independent
of the drive voltage and of the load, and that it remains constant as long as the circuit is not
overdriven. The efficiency η max is defined as the ratio of the maximum obtainable output
power to the power consumption at full voltage swing. The results for PL max and Ptot give
η max = 161
= 6.25%. The following characteristics are typical for this circuit:
15.2
Complementary Emitter Followers
The output power of the emitter follower shown in Fig. 15.1 is limited in that the resistor
RE restricts the maximum output current. A considerably higher output power and a better
efficiency can be attained if RE is replaced by a second emitter follower, as shown in
Fig. 15.2.
15.2.1
Complementary Class-B Emitter Follower
For positive input voltages, T1 operates as an emitter follower and T2 is reverse biased; and
vice versa for negative drive. The transistors thus carry the current alternately, each for half
a period. Such a mode of operation is known as push–pull class-B operation. For Vi = 0,
both transistors are turned off and therefore no quiescent current flows in the circuit. The
current taken from the positive or negative power supply is thus the same as the output
current. The circuit therefore has a considerably better efficiency than the normal emitter
follower. A further advantage is that, at any load, the output can be driven between ±Vb as
the transistors do not limit the output current. The difference between the input and output
voltages is determined by the base–emitter voltage of the current-carrying transistor. As
it changes only very slightly with the load, Vo ≈ Vi , irrespective of the load current.
The output power is inversely proportional to the resistance RL and has no extremum,
and therefore no matching is required between the load resistance and any internal circuit
resistance. The maximum power is determined instead by the permissible peak currents
and the maximum power dissipation of the transistors, which, for full sinusoidal drive, is
given by
V̂o2
PL = .
2RL
Vi
L
Vo
L
. Pges
.
PL
.
.
.
PT1
.
Fig. 15.3. Power and loss curves for the
. . . . . . . . . . Vo complementary emitter follower as a function
of the output amplitude
T /2
1 Vo (t)
PT 1 = (Vb − Vo (t)) dt.
T RL
0
PL PL π V̂o V̂o
η = = = · ≈ 0.785 .
Ptot 2P T 1 + PL 4 Vb Vb
It is therefore proportional to the output amplitude and, for a full output swing (V̂o = Vb ),
it attains a value of η max = 78.5%.
The power dissipation of the transistors reaches its maximum not at the full output
voltage swing, but at
2
V̂o = Vb ≈ 0.64 Vb .
π
This follows from the extremum condition:
dP T 1
= 0.
dV̂o
In this case, the power dissipation for each transistor is
1 Vb2 Vb2
P T max = ≈ 0.1 .
π 2 RL RL
The curves for the output power, the power dissipation, and the total power, as functions
of the relative output voltage swing V̂o /Vb , are given in Fig. 15.3.
15.2 Complementary Emitter Followers 871
15.2.2
Complementary Class-AB Emitter Followers
Figure 15.4 represents the transfer characteristic Vo = Vo (Vi ) for push–pull class-B oper-
ation, as described for the previous circuit. Near zero voltage, the current in the forward-
biased transistor becomes very small and the transistor impedance increases. The output
voltage at the load therefore remains nearly zero, as indicated by a region in the character-
istic near the origin. This gives rise to distortion of the output voltage, known as crossover
distortion. If a small quiescent current flows through the transistors, their impedance near
the origin is reduced and the transfer characteristic shown in Fig. 15.5 is obtained. The
transfer characteristic of each individual emitter follower is shown by dashed lines. It can
be seen that the crossover distortion is considerably reduced. If the quiescent current is
made as large as the maximum output current, the resulting mode of operation would be
called push–pull class-A, by analogy with Sect. 17.1. However, the crossover distortion
is already greatly reduced if only a fraction of the maximum output current is permitted
to flow as a quiescent current. This mode is called push–pull class-AB operation, and its
crossover distortion is so small that it can be easily reduced to tolerable values by means
of feedback.
Vo Vo
–0.6 V
0 .6 V Vi Vi
Fig. 15.4. Crossover for push–pull class-B Fig. 15.5. Crossover for push–pull
operation class-AB operation
872 15 Power Amplifiers
V1 V3
Vi Vo Vi Vo
V2
Additional distortion may arise if positive and negative voltages are amplified at dif-
ferent gains. This is the case if the complementary emitter followers are driven from a
high-impedance source and the transistors have different current transfer ratios. If strong
feedback is undesirable, the transistors must be selected such that they have identical
current transfer ratios.
The basic circuit for the realization of class-AB operation in shown in Fig. 15.6. To
obtain a small quiescent current, a DC voltage of about 1.4 V is applied between the base
terminals of T1 and T2 . If the two voltages V1 and V2 are the same, the quiescent potential
of the output is approximately equal to that of the input. The bias voltage can also be
supplied by a single voltage source, V3 = V1 + V2 , as represented in Fig. 15.7. In this
case, the potential difference between the output and the input is about 0.7 V.
The main problem with class-AB operation is keeping the required quiescent current
constant over a wide range of temperatures. As the transistors get warmer, the quiescent
current increases, which may itself further increase the temperature and finally lead to
destruction of the transistors. This effect is known as positive thermal feedback. The
increase in the quiescent current can be avoided if the voltages V , and V2 are each reduced
by 2 mV for every degree of temperature rise. For this purpose, diodes or thermistors can
be mounted on the heat sinks of the power transistors.
However, temperature compensation is never quite perfect, as the temperature dif-
ference between the junction and the case is usually considerable. Therefore, additional
stabilization is required in the form of resistors R1 and R2 , which provide current feedback.
This becomes more effective as the resistances chosen increase. As the resistors are con-
nected in series with the load, they reduce the available output power and must therefore
be selected to be small compared to the load resistance. This dilemma can be avoided by
using Darlington circuits, as will be shown in Sect. 15.3.
15.2.3
Generation of the Bias Voltage
One method of providing a bias voltage is shown in Fig. 15.8. The voltage of V1 = V2 ≈
0.7 V across each of the diodes D1 and D2 just allows a small quiescent current to flow
through the transistors T1 and T2 . To attain a higher input resistance, the diodes can be
replaced by emitter followers, this resulting in the circuit shown in Fig. 15.9.
Figure 15.10 represents a driver arrangement that allows adjustment of the bias voltage
and its temperature coefficient over a wide range. Feedback is applied to transistor T3 by
15.2 Complementary Emitter Followers 873
V1
Vi Vo Vi Vo
V2
Fig. 15.8. Bias voltage generation by diodes Fig. 15.9. Bias voltage generation by
transistors
means of voltage divider R5 , R6 . For a negligible base current, its collector–emitter voltage
has the value
R5
VCE3 = VBE3 1 +
R6
To obtain the desired temperature coefficient, R5 is, in practice, a resistor network that
contains an NTC resistor mounted on the heat sink of the output transistors. In this way,
the quiescent current can to a large extent be made temperature-independent, even though
the temperature of the case is lower than that of the output transistor junctions.
In circuits that use diodes for bias voltage generation, no current can flow from the
input into the base of the output transistors. The base current for the output transistors must
therefore be supplied from the constant-current sources. The constant current I1 must be
larger than the maximum base current of T1 and T2 , so that diodes D1 and D2 (or transistors
T3 and T4 ) will not he turned off before the maximum permissible output voltage swing
is reached. For this reason, it would be inadvisable to replace the constant-current sources
by resistors, as this would cause the current to decrease with a rising output voltage.
The most favorable driver circuit would be one that supplies a larger base current for
an increasing output voltage, and such a circuit is represented in Fig. 15.11. The FETs T3
Vo
Vi Vo
Vi
Fig. 15.10. Generation of a bias voltage that Fig. 15.11. Bias voltage generation by FETs
has an adjustable temperature coefficient
874 15 Power Amplifiers
and T4 operate as source followers. The difference in their source voltages settles, due to
series feedback, at a value of about 1.4 V. FETs that have a large saturation drain current
ID sat are suitable for this purpose.
15.3
Complementary Darlington Circuits
With the circuits described so far, output currents of up to a few hundred milliamperes can
be obtained. For higher output currents, transistors that have higher current gain β must be
used. They can be made up of two transistors if they are operated in a Darlington connection.
Such circuits and their parameters have already been discussed in Sect. 2.4.4. The basic
circuit of a Darlington power amplifier is shown in Fig. 15.12, where the transistor pairs
T1 , T
1 and T2 , T
2 are Darlington connected.
For the implementation of push–pull class AB, the adjustment of the quiescent cur-
rent presents problems, as four temperature-dependent base–emitter voltages must now
be compensated. These difficulties can be avoided by allowing the quiescent current to
flow only through the driver transistors T1 and T2 . The output transistors then become
conducting only for larger output currents. To achieve this, the bias voltage V1 is selected
such that a voltage of approximately 0.4 V appears across each of resistors R1 and R2 ; thus
V1 ≈ 2(0.4 V + 0.7 V) = 2.2 V. Then, for zero input, the output transistors carry virtually
no current, even at higher junction temperatures.
At higher output currents, the base–emitter voltages of the output transistors rise to
about 0.8 V. This limits the current through R1 and R2 to double the quiescent value, and
therefore most of the emitter current of the driver transistors is available as base current
for the output transistors.
Resistors R1 and R2 also discharge the base of the output transistors. The lower their
resistance, the faster the output transistors can be turned off. This is particularly important
when the input voltage changes polarity, as one transistor can become conducting before
the other is turned off. In this way, a large crossover current can flow through both output
transistors, and the resulting secondary breakdown will destroy them immediately. This
effect determines the attainable large-signal bandwidth.
V1 V1
Vi Vo Vi Vo
For integrated circuits it is preferable to use power transistors npn. In such cases, the
Darlington circuit T2 , T
2 shown in Fig. 15.12 is replaced by the complementary Darlington
connection, as described in Sect. 2.4. The resulting circuit, shown in Fig. 15.13, is known as
a quasi-complementary power amplifier. To arrive at the same quiescent current conditions
as for the previous circuit, a voltage of about 0.4 V is impressed across resistor R1 . Voltage
V1 must then be 0.4 V + 2 · 0.7V = 1.8 V. The quiescent current flows via T2 and R2 to the
negative supply. If R1 = R2 , a bias voltage of 0.4 V is also obtained for T
2 ’. The purpose
of resistors R1 and R2 is the same as for the previous circuit; that is, to discharge the base
of the output current transistors. This output structure is found in most integrated power
amplifiers.
15.4
Complementary Source Followers
The major advantage of power MOSFETs over bipolar power transistors is that they can be
turned on and off much more quickly. Whereas bipolar power transistors have switching
times of between 100 ns and 1 ms, the corresponding range for MOSFETs is 10–100 ns. It
is therefore preferable to employ power MOSFETs in output stages for frequencies above
100 kHz up to 1 MHz.
As power MOSFETs possess large drain–gate and gate–source capacitances, possibly
of a few hundreds of picofarads, it is advantageous to operate them as source followers. This
prevents the drain–gate capacitance from being dynamically increased due to the Miller
effect, and the gate–source capacitance is even considerably reduced by bootstrapping.
Figure 15.14 shows the basic circuit diagram for a complementary source follower.
The two auxiliary voltages V1 are used, as in the case of the bipolar transistor shown in
Fig. 15.6, to set the required quiescent current. For V1 = Vth , no quiescent current flows at
all: class-B operation is established. However, in order to minimize the transfer distortion,
a quiescent current is usually desired to flow by selecting V1 > Vth . The magnitude of the
quiescent current is stabilized by current feedback via resistors R1 , R2 . The voltage V1
results from the transfer characteristic of the MOSFETs; that is,
2I1
V1 = ID R1 + Vth +
K1
The resulting voltages are markedly higher than for bipolar transistors, as the pinch-off
voltage Vth of power MOSFETs is between 1 V and 4 V. A simple means of producing the
required bias is to replace the emitter followers T3 , T4 in Fig. 15.9 by source followers.
The resultant circuit is shown in Fig. 15.15. Here, T3 produces a bias voltage of
2I3
V1 = Vth +
K3
If the low-power MOSFETs T3 and T4 are fabricated in the same process as the power
MOSFETs T1 T2 , and therefore have the same pinch-off voltages, the maximum quiescent
current for R1 = R2 = 0 is
K1 A1
I1 = I3 = I3
K3 A3
876 15 Power Amplifiers
V1
Vi Vo
Vi V1 Vo
Thus it depends only on the ratio of the areas of the transistors on the chip. The current can
be reduced using R1 , R2 . The currents I3 , I4 are selected to be sufficiently large to charge
the input capacitor of the source followers T1 , T2 at the highest frequency that occurs.
Operation of the drive circuit generally requires a supply voltage that is up to 10 V
higher than for the output stage. Otherwise, the maximum achievable output voltage may
be as much as 10 V below the supply voltage, which results in unacceptably poor efficiency.
15.5
Current Limitation
Due to their low output resistance, power amplifiers can easily be overloaded and there-
fore easily destroyed. Consequently, it is advisable to limit the output current to a defined
maximum value by means of an additional control circuit. The various possibilities are
exemplified by the simple complementary emitter follower shown in Fig. 15.8. A particu-
larly simple circuit is shown in Fig. 15.16. Limitation takes effect when multiple diode D3
or D4 begins to conduct, because this prevents any further increase in the voltage dropped
across R1 or R2 The maximum output current is therefore
VD 3 − VBE 1 0.7 V
Io+max = = (n3 − 1),
R1 R1
VD 4 − |VBE 2 | 0.7 V
Io−max = − =− (n4 − 1).
R2 R2
where n3 and n4 are the number of diodes used for D3 and D4 respectively.
Another method of current limitation is shown in Fig. 15.17. If the voltage drop across
R1 or R2 exceeds a value of about 0.7 V, transistor T3 or T4 begins to conduct, thereby
preventing any further rise in the base current of T1 or T2 .
15.5 Current Limitation 877
Io
Io
Vi Vo
Vi Vo
Fig. 15.16. Current limitation with Fig. 15.17. Current limitation with
diodes transistors
Io max = ±1.4 V/R1,2 Io max = ±0.7 V/R1,2
Vo
Vi Vo
Fig. 15.18. Voltage-dependent current Fig. 15.19. Current limit and output
limitation current characteristics for a resistive
0.7 V R3.4 Vo load
|Io max | = + ·
R1.2 R5.6 R1.2
For higher positive output voltages, an additional voltage drop of Vo R3 /R5 appears across
R3 , thereby raising the current limit to
0.7 V R3 Vo
Io+max ≈ +
R1 R5 R1
Diode D5 prevents transistor T3 from receiving a positive bias when the output voltages
are negative, which could result in it being turned on unintentionally. Diode D3 prevents
the collector–base diode of T3 from conducting when a larger voltage is dropped across
R2 in the event of negative output voltages. Otherwise, the drive circuit would be subject
to additional loading. Similar considerations apply to the negative current limitation due
to T4 .
These characteristics are graphically illustrated in Fig. 15.19. Using this form of
voltage-dependent current limitation, it is therefore possible to utilize fully the safe op-
erating area (SOA) of the power transistors. Consequently, this method is also known as
SOA current limitation.
15.6
Four-Quadrant Operation
A power output stage is subject to the most severe operating conditions when a constant
current limit Io+max and Io−max is required for any given positive and negative output voltage.
Such requirements invariably arise if there is no resistive load present, but a load that can
feed energy back to the output stage. Loads of this kind include capacitors, inductors, and
electric motors. In this case, it is necessary to use current-limiting arrangements shown
in Figs. 15.16 or 15.17. The critical operating condition for the negative output-stage
15.7 Design of a Power Output Stage 879
Vo
.
Vi Vo Vi
.
Fig. 15.20. Push–pull stage for four-quadrant Fig. 15.21. Characteristics of the output
operation voltage and auxiliary potentials V1 , and V2
transistor T2 occurs when the load feeds the current-limit value Io−max into the circuit while
the output voltage is Vo = Vo max ≈ V + . Current Io−max then flows through T2 at a voltage
of VCE 2 ≈ 2V + , resulting in a power dissipation of P T 2 = 2V + · Io−max . However, for
a voltage of 2 V + , secondary breakdown limits the majority of bipolar transistors such
that they can only be loaded to a fraction of their thermal rating. Therefore, it is generally
necessary to connect several power transistors in parallel or, preferably, to employ power
MOSFETs, which are not subject to secondary breakdown.
One method of halving the voltage across the output-stage transistors is shown in
Fig. 15.20. The basic idea is to control the collector potentials of T1 and T2 together with
the input voltage. For positive input voltages, we obtain
V1 = Vi + 0.7 V + 3 V − 0.7 V − 0.7 V = Vi + 2.3 V.
Transistor T1 is therefore being operated well outside saturation. For negative input volt-
ages, diode D3 takes over the current, and we have V1 = −0.7 V. If the input voltage
falls to Vi = Vi min ≈ V − , a voltage of only VCE 1 max ≈ V − is dropped across T1 .
Likewise, the maximum voltage across T3 is no larger. It occurs when Vi = 0 and is
given by VCE 3 max ≈ V + . The maximum power dissipated in T1 and T3 is therefore
P max = V + · Io+max . Consequently, not only is the maximum collector–emitter voltage
halved, but also the power dissipation. For the negative side, T2 , T4 produce correspond-
ing results due to the symmetry of the circuit. To make this clear, the characteristics of V1
and V2 are plotted in Fig. 15.21.
15.7
Design of a Power Output Stage
To illustrate the design process for a power output stage in more detail, we use the circuit
shown in Fig. 15.22 and determine the rating of its components for an output power of
50 W. The circuit is based on the power amplifier of Fig. 15.12.
880 15 Power Amplifiers
+ 29 V
Vi Vo
that is,
3.1 K
Rth JC = .
W
Instead of the thermal resistance Rth JC , the maximum power dissipation PD25 at a case
temperature of 25 ◦ C is often given in the specifications. It can be calculated from:
ϑJ − 25 ◦ C 150 K
PD25 = = = 48 W.
Rth JC 3.1 K/W
The transistors selected in this way are assumed to have a current transfer ratio of 30 at
the maximum output current. We can therefore determine the data of the driver transistors
T1 and T2 . Their maximum collector current is
4.48 A
= 149 mA ,
30
although this value applies to low frequencies only. For frequencies above fc ≈ 20 kHz,
the current transfer ratio of audio power transistors falls markedly. When the current rises
steeply, the driver transistor must therefore momentarily supply the largest proportion of
the output current. To obtain the largest possible bandwidth, we choose IC max = 1 A.
Transistors within this range of collector currents, which have gain–bandwidth products
in the region of 50 MHz, are still reasonably priced.
We have shown in Sect. 15.3 that it is useful to allow the quiescent current to flow
only through the driver transistors, and to have a voltage of about 400 mV across resistors
R1 and R2 . This is the purpose of the three silicon diodes D3 , which have a total forward
voltage of about 2.1 V. We select a quiescent current of approximately 30 mA to keep the
crossover distortion reasonably small. Thus
400 mV
R1 = R2 = = 13 .
30 mA
The power dissipation of the driver transistors is, at zero input voltage, 30 mA · 29 V ≈
0.9 W, and at maximum input it is still 0.75 W. A small power transistor in a TO-5 case
with cooling fins is obviously sufficient. A value of 100 is usual for the current transfer
ratio of such a transistor. The maximum base current is then:
1 4.48 A 0.8 V
IB max = + ≈ 2 mA.
100 30 13
The current through the constant current sources T3 and T4 must be large compared to this
value, and we select 10 mA.
Emitter followers are prone to unwanted oscillations in the region of the transit fre-
quency of the output transistors. These oscillations can be damped by additionally loading
the output by a series RC element (approximately 1 ; 0.22 mF). However, this also re-
duces the efficiency at higher frequencies. Another method of damping that may also be
used in addition to that described above is to provide series resistors in the base lead
of the driver transistors, in conjunction with an increased collector–base capacitance. If
R7 = R8 = 100 , as shown in Fig. 15.22, the voltage across these resistors remains
below 0.2 V. The achievable output voltage swing is therefore not significantly reduced.
882 15 Power Amplifiers
15.8
Driver Circuits with Voltage Gain
The power amplifiers described above have a certain amount of crossover distortion in
the region of zero output voltage, but this can be largely eliminated by feedback. The
output stage is connected to a preamplifier stage, and negative feedback is applied across
both stages. Figure 15.23 shows a simple possibility. The output stage drive is supplied via
current source T3 which, in conjunction with T7 , forms a current mirror for IC6 . The differ-
ential amplifier T5 T6 provides the required voltage gain. Its effective collector resistance
is relatively high, being produced by the paralleled current-source internal resistances T3 ,
T4 and the input resistances of the emitter followers T1 , T2 .
The whole arrangement has feedback via resistors R7 and R8 , and therefore has a
voltage gain of A = 1 + R8 /R7 . In order to produce sufficient loop gain, A must not be
selected to be too large, practicable values being between 5 and 30.
If we only wish to amplify AC voltages, the zero stability of the circuit can be improved
by connecting a coupling capacitor in series with R7 , thereby reducing the DC voltage gain
to unity. Most power amplifier ICs operate on this principle.
Vi Vo
HF-path
LF-path
OA Vo
Vi
Fig. 15.24. Broadband power amplifier (HF = high frequency, LF = low frequency)
To begin the dimensioning of the circuit, the collector currents of the transistors T3 −T6
are defined. We choose 10 mA. A current of 20 mA must then flow through resistors R3
and R4 . A voltage of 1.4 V is dropped across R3 and R1 . Hence
1.4 V
R3 = R4 = = 70 .
20 mA
The output quiescent potential of the operational amplifier is determined by the offset
voltage of the power output stage and is close to zero. Hence, with no input drive, the
current through resistors R11 , and R12 is virtually zero. The collector currents of T5 and
T6 must therefore flow through resistors R9 and R10 . With supply potentials of ±15 V, it
follows:
15 V
R9 = R10 ≈ = 1.5 k.
10 mA
To attain the full swing of the current sources T3 and T4 , the collector currents of T5 and
T6 must be controlled between zero and 20 mA. These values should be reached for a full
output swing of the operational amplifier. Thus, for resistors R11 and R12 ,
10 V
R11 = R12 ≈ = 1 k.
10 mA
The operational amplifier OA is configured as an integrator. Its gain is defined by the
external circuitry, and is selected such that it is markedly below the open-loop gain of the
operational amplifier. If we select R14 = 10 k and C4 = 160 pF, for instance, the gain is
unity at a frequency of 100 kHz. The lower cutoff frequency of the highpass filter C3 , R13
in the high-frequency path must have a lower value; for example, 1 kHz.
The total gain of the circuit can be set by resistors R15 and R16 to values between
1 and 10. A higher gain is inadvisable, as the loop gain in the high-frequency path then
becomes too low. The open-loop gain of the high-frequency path can be varied by means of
resistors R7 and R8 . They are adjusted so as to obtain the desired transient response for the
whole circuit. For the operational amplifier, the internal standard frequency compensation
is sufficient. To avoid oscillations in the VHF range, it may be necessary to insert resistors
in the base leads of the transistors.
884 15 Power Amplifiers
15.9
Boosting the Output Current
of Integrated Operational Amplifiers
The output current of integrated operational amplifiers is normally limited, the maximum
being about 20 mA. There are many applications for which about ten times this current is
required, but where the number of additional components must be kept to a minimum. In
such cases, the power output stages described here may be used. For low signal frequen-
cies, the number of components can be reduced by the use of push–pull class-B emitter
followers. However, due to the finite slew rate of the operational amplifier, noticeable
crossover distortion occurs even with feedback. It can be reduced considerably by insert-
ing a resistance R1 as in Fig. 15.25, which bypasses the emitter followers in the region of
zero voltage. In this case, the slew rate required of the amplifier is reduced from infinity
to a value that is 1 + R1 /RL times that of the rate of change of the output voltage.
The arrangement shown in Fig. 15.26 has the same characteristics as the previous
circuit. Here, however, the output transistors are controlled by the supply terminals of the
operational amplifier. If we make R2 = 0, this, together with the output transistors of the
operational amplifier, results in two complementary Darlington connections.
At small output currents, the two output transistors T1 and T2 are turned off. The
operational amplifier then supplies the whole output current. At larger output currents,
transistors T1 and T2 alternately become conducting and supply the largest proportion
of the output current. The contribution of the operational amplifier remains limited to
approximately 0.7 V/R1 .
The circuit has an advantage over the previous one in that the quiescent current of
the operational amplifier causes biasing of the base–emitter junctions of the power output
transistors. The values of the resistors R1 are such that the bias is about 400 mV. This
considerably reduces the range of crossover without the need for a quiescent current in the
output transistors, the stabilization of which would require additional measures.
Using the voltage divider R2 , R3 , the output stage can provide an additional voltage gain
of 1 + R2 /R3 . This enables the output voltage swing of the amplifier to be increased nearly
to the supply voltage only limited by the saturation voltage of T1 or T2 . The likelihood of
oscillation within the complementary Darlington circuits is also reduced.
VN VN
VP VP
RL Vo RL Vo
Fig. 15.25. A current booster with Fig. 15.26. A current booster with
complementary emitter followers complementary common-emitter circuits
Chapter 16:
Power Supplies
Every electronic circuit requires a power supply that provides one or more DC voltages.
For larger power requirements, batteries are not economical. The DC voltage is therefore
obtained from the AC line supply by transformation and subsequent rectification. The
DC voltage thus obtained usually has considerable ripple, and changes in response to
variations in the line voltage and the load. Therefore, a voltage regulator is often connected
to the rectifier to keep the DC output voltage constant and counteract these variations. The
following two sections describe ways of providing the unregulated DC voltage; regulator
circuits will be dealt with later.
16.1
Properties of Power Transformers
The internal resistance Ri of the power transformer plays an important part in the design
of rectifier circuits. It can be calculated from the rating of the secondary winding Vsec n ,
Isec n , and from the loss factor L, which is defined as the ratio of the no-load voltage Vsec 0
to the nominal voltage Vsec n :
Vsec 0
L = ; 1 < L < 1.5 (16.1)
Vsec n
Ri = RN (l − 1) (16.3)
The data of the M-core transformers normally used are listed in the table of Fig. 16.1; the
corresponding data for toroidal transformers are given in Fig. 16.2.
As toroidal transformers are more difficult to wind, they are significantly more ex-
pensive, particularly for low powers. However, this is offset by the advantage of minimal
magnetic flux leakage, a higher magnetizing reactance, and therefore a lower magnetizing
current and small no-load losses.
The values in Figs. 16.1 and 16.2 are based on a line voltage Vp rms = 230 V at fl =
50 Hz and a maximum flux density of B̂ = 1.2 T. Should the line voltage deviate slightly
from this value, w1 must be recalculated proportionally to Vp rms and d1 to 1/ Vp rms . If the
line frequency is fl = 60 Hz, B̂ reduces to 1 T, and the parameters w1 and d1 in Figs. 16.1
and 16.2 include a safety margin.
886 16 Power Supplies
16.2
Power Rectifiers
16.2.1
Half-Wave Rectifier
The easiest way to rectify an AC voltage is to charge a capacitor via a diode, as shown
in Fig. 16.3. If the output is not loaded,
√ the capacitor C is charged during the positive
half-cycle to the peak value Vo0 = 2 Vsec 0 rms − VD , where V1 is the forward voltage of
the diode. The peak reverse voltage of the diode occurs√when the transformer voltage is at
its negative maximum, and therefore has the value of 2 Vsec 0 rms .
When a load resistance RL is connected to the DC output, it discharges the capacitor C
for as long as the diode is reverse biased. Only when the no-load voltage of the transformer
exceeds that of the output by the amount VD is the capacitor recharged. The voltage reached
by recharging depends on the internal resistance Ri of the transformer. The shape of the
output voltage at steady state is shown in Fig. 16.4. Owing to an unfavorable ratio of the
16.2 Power Rectifiers 887
Io
Vsec 0 RL Vo
V
Vo Vr pp Vo
Vo ∞
Vo
Vsec 0
Dp
Fig. 16.4. Voltage and current waveforms for a single-phase half-wave rectifier
recharge and discharge times, the output voltage is considerably reduced even for small
load currents, and for this reason the circuit is unsuitable for use in power supplies.
16.2.2
Bridge Rectifier
The ratio of the recharge and discharge times can be greatly improved by charging the
capacitor C during the positive and negative half-cycles. This is achieved by means of the
bridge rectifier shown in Fig. 16.5.
During the recharge period, the diodes connect negative terminal of the transformer to
ground, and the positive terminal to the output. The repetitive peak reverse voltage of the
diodes is identical to the no-load output voltage:
√
Vo 0 = 2Vsec 0 rms − 2VD (16.4)
and is only half that of the half-wave rectifier.
888 16 Power Supplies
Io
Vsec 0
RL Vo
To calculate the voltage reduction at load, we initially assume an infinitely large storage
capacitor C. The output voltage is then a pure DC voltage, which we define as Vo∞ . The
more the output voltage decreases due to the load, the longer is the recharge time. Steady
state is reached when the incoming charge of the capacitor equals the outgoing charge; in
other words, that supplied to the load. Hence,
Ri
Vo ∞ = Vo 0 1 − (16.5)
2RL
where RL = Vo∞ /Io is the load resistance. The deduction of this equation is based on
calculations involving the approximation of sine waves by parabolas and is omitted here
because of its complexity. As comparison with the half-wave circuit in Fig. 16.3 shows, in
this bridge rectifier only half the internal resistance of the transformer is responsible for
the voltage drop at load.
To dimension the rectifier correctly, the currents must be known. As no DC current
flows through the capacitor, the mean forward current of each bridge arm is half of the
output current. As the forward voltage is only slightly dependent on the current, the power
dissipation of a single diode is given by
1
PD = V D Io
2
During every recharge period a peak current IDp flows, the value of which may be
many times that of the output current:
V̂sec 0 − 2VD − Vo ∞ Vo 0 − Vo ∞
IDp = =
Ri Ri
16.2 Power Rectifiers 889
It can be seen from the table in Fig. 16.2 that a toroidal type with D = 80 mm, having a
loss factor of L = 1.15, can be used. We now need to know the internal resistance of the
transformer; however, it is dependent on the rated voltage, the value of which is not yet
known. For its determination, the system of nonlinear equations given in (16.3) – (16.5)
must be solved. This is best done by iteration. We set the initial value of Vsec n rms ≈
Vo min = 30 V. Using (16.3), it follows:
2
Vsec n rms
Ri = Rn (L − 1) = (L − 1)
Pn
(30 V)
= · (1.15 − 1) = 2.65
51 W
Hence, using (16.4) and (16.5),
√ Ri
Vo ∞ = ( 2Vsec n rms · L − 2VD ) 1 −
2RL
√ 2.65
= ( 2 · 30 V · 1.15 − 2 V) 1 − ≈ 37.3 V
2 · 32 V/1 A
The voltage is about 5 V higher than that initially required. For the first iteration, we
decrease the rated transformer voltage by this amount and, correspondingly, obtain
which is already the desired value for the output voltage. The design parameters for the
transformer are therefore
Pn
Vsec n rms ≈ 25 V; Isec n rms = ≈ 2A
Vn
Figure 16.2 gives the winding data for a primary voltage of 220 V/50 Hz:
The no-load output voltage is 39 V. The capacitor must be rated for at least this voltage.
The calculation for transformers that have several secondary windings is the same
as that above. For Pn , the rated power of the corresponding secondary winding must be
16.2 Power Rectifiers 891
inserted. The total power is the sum of the individual powers of the secondary windings.
This determines the choice of the core and therefore the loss factor L.
16.2.3
Center-Tap Rectifier
Full-wave rectification can also be achieved by rectifying two antiphase AC voltages on a
half-wave basis. This principle is illustrated by the center-tap circuit shown in Fig. 16.6.
As we can see from the data given, the advantages of the bridge circuit are retained.
An additional advantage is that the current need only flow through one diode at a time,
rather than through two as in the case of the bridge circuit. As a result, the voltage drop
caused by the forward voltages of the diodes is halved. On the other hand, the internal
resistance of the transformer is doubled, as each part-winding must be rated for half the
output power, thereby further increasing the voltage drop. The ratio of the output voltage
to the forward voltage of the diode will dictate which effect predominates. The center-tap
circuit is better for low output voltages, and the bridge rectifier circuit for high output
voltages.
Io Io
Vsec 0 Vo Vsec 0 Vo
Fig. 16.6. Center-tap circuit Fig. 16.7. Center-tap circuit for symmetrical
output voltages
√
No-load output voltage: Vo 0 = sec 0 −
2V -VD
R
On-load output voltage: Vo ∞ = Vo 0 1 − 2Ri
√ L
Peak reverse voltage: Vpr = 2 2Vsec 0
I¯D
1
Mean forward current: = Io 2
V
Repetitive peak current: IDp = √ o0
2Ri
RL -
Io 4 Ri
Ripple voltage: Vr pp = 2Cf 1 − 2R
l L
2
Minimum output voltage: Vo min ≈ Vo ∞ − 3 Vr pp
892 16 Power Supplies
16.3
Linear Voltage Regulators
Electronic circuits generally require a DC voltage that is accurate to within 5–10% of a
specified value. This tolerance must be maintained over the entire range of line voltage,
load current, and temperature variations. The ripple voltage must not exceed the millivolt
range. For these reasons, the output voltage of the rectifier circuits described is not directly
usable as a supply voltage for electronic circuits, but requires stabilization and smoothing
by a following voltage regulator.
The principal characteristics of a voltage regulator are:
16.3.1
Basic Regulator
Output voltage variations due to supply voltage and load current fluctuations can be reduced
by inserting a controlled series resistance, this method being known as series regulation.
The simplest series regulator is an emitter follower, with its base connected to a ref-
erence voltage source. The reference voltage can, for instance, be obtained from the un-
stabilized input voltage Vi using a Zener diode, as shown in Fig. 16.8. Other possibilities
will be discussed in Sect. 16.4. The output voltage is
Vo = Vref − VBE
The extent to which the voltage varies with load is related to the output resistance:
∂Vo 1 VT
ro = − = =
∂Io gm Io
With VT ≈ 26 mV, we obtain approximately 0.3 for Io = 100 mA.
Io Io
Vi Vo Vi Vo
Vref Vref
Fig. 16.8. Voltage stabilization using an Fig. 16.9. Additional circuitry for output
emitter follower voltage adjustment
Output voltage: Vo = Vref − VBE 0 ≤ Vo ≤ Vref − VBE
16.3 Linear Voltage Regulators 893
Input voltage variations are compensated by the low differential resistance rZ of the
Zener diode. The output voltage variation is given by
rZ rZ
Vo = Vref = Vi ≈ Vi
R1 + r Z R1
This represents a regulation of the output voltage between 1% and 10% of the input voltage
variation, depending on the component values selected.
If an adjustable output voltage is required, a portion of the reference voltage can be
tapped off at a potentiometer, as shown in Fig. 16.9. The potentiometer resistance selected
must be small compared to rBE , so that the circuit output resistance is not increased
appreciably.
16.3.2
Voltage Regulators with a Fixed Output Voltage
The simple circuits shown in Figs. 16.8 and 16.9 are largely inadequate, or fail to meet
the requirements that voltage regulators must satisfy. Consequently, IC voltage regulators
contain a voltage control amplifier, a reference voltage source, and several additional
modules to protect the power transistor. These are shown in the block diagram of Fig. 16.10.
The current-limiting circuit monitors the voltage drop across current-sensing resistor
R. The safe operating area (SOA) of the power transistor is monitored in an additional
block. If the voltage drop across the power transistor increases, the current limit is reduced
accordingly.
A thermal protection device monitors the chip temperature and reduces the output
voltage if hazardous overheating is likely to occur. The diodes realize an analog and-
gate and insure that the output voltage is determined by the lowest of the four correcting
variables. The amplifier holds the output voltage at the nominal value only as long as no
limit value is exceeded.
The practical implementation of a 7800-series IC voltage regulator is shown in
Fig. 16.11. The requirements placed on the amplifier are not particularly stringent, as
an emitter follower alone already constitutes an effective voltage regulator. Consequently,
it is sufficient to have a simple differential amplifier T3 , T4 operating in conjunction with
Darlington circuit T1 , as a power operational amplifier. It acts as a noninverting amplifier
Vi
Vo
Vi
Vref Vo
via voltage divider R1 , R2 in the feedback path and, at the output, produces an amplified
reference voltage of
Vo = (1 + R2 /R1 )Vref
Transistor T2 has a current-limiting function. If the voltage drop across R3 reaches
0.6 V, T2 is turned on, thereby reducing the output voltage. Due to the feedback path
produced, the output voltage is adjusted so that the voltage drop across R3 is stabilized to
the value 0.6 V. This is equivalent to a constant output current of
Io max = 0.6 V/R3
Under these conditions, the output voltage is determined by load resistor RL , in accordance
with Vo = Io max RL .
When the maximum current is reached, the power dissipation in output transistor T1
is given by
P = Io max (Vi − Vo )
In the event of a short-circuit at the output, it is much higher than during normal operation
P = Io max Vi , since the output voltage then falls below the nominal value to zero. In order
to prevent this increased power dissipation, the current limit can be reduced as the output
voltage decreases. This produces the foldback characteristic shown in Fig. 16.12.
A marked increase in power dissipation may also occur if the input voltage V1 is
increased, as in this case the difference Vi − Vo increases likewise. Consequently, the best
Vi
way of protecting output transistor T1 is to match the current limit Io max to the voltage
difference Vi −Vo . Resistor R5 and Zener diode D1 , shown by the dashed line in Fig. 16.11,
are used for this purpose.
If the potential difference Vi − Vo is less than the Zener voltage VZ of diode D1 , no
current flows through resistor R5 . Consequently, the current limit in this case remains
0.6 V/R3 . If the potential difference exceeds the value VZ , voltage divider R5 , R4 causes
a positive base–emitter bias to be applied to transistor T2 . As a result, the latter will be
turned on in response to a correspondingly smaller voltage drop across R3 .
Capacitor Ck provides the frequency compensation required for stability. To provide
additional stabilization, it is generally necessary to connect capacitors with approximately
100 mF to ground at the input and output.
16.3.3
Voltage Regulators with an Adjustable Output Voltage
In addition to the fixed voltage regulators described above, adjustable types are also avail-
able (78 G series). In the latter type, the voltage divider R1 , R2 is omitted and the amplifier
input is brought out as shown in Fig. 16.13. These regulators therefore have four terminals.
By connecting voltage divider R1 , R2 externally, any output voltage between Vref ≈ 5 V
≤ Vo < Vi − 3 V can be selected.
Adjustable voltage regulators with only three terminals can be realized by connecting
the negative supply voltage of the OA to the regulated output instead of ground. In order
to make this difference clear, Fig. 16.13 shows a 78 G series adjustable voltage regulator
with four terminals alongside a 317 series adjustable voltage regulator with three terminals
(Fig. 16.14). Here, the reference voltage source is not connected to ground, but to the
midpoint of the feedback voltage divider. The output voltage therefore increases until
voltage Vref is dropped across R2 . The input voltage difference of the operational amplifier
is then zero.
The output of the voltage regulator in Fig. 16.14 must not be in open-circuit, as this
would prevent the amplifier supply current loop from being closed. It is therefore advisable
to select low values for voltage divider R1 R2 , typically R2 = 240 ; a current of 5 mA
therefore flows for a reference voltage of Vref = 1.25 V. Consequently, the current of
Vi Vi
Vref Vref
Vref Vref
Vref Vo Vref Vo
Fig. 16.13. Adjustable voltage regulator with Fig. 16.14. Adjustable voltage regulator with
four terminals (78 G series) three terminals (317 series)
R2 R1
Vo = 1 + Vref ; Vref = 5 V Vo = 1 + Vref ; Vref = 1.25 V
R1 R2
896 16 Power Supplies
approximately 100 mA flowing out of the reference voltage source cannot appreciably alter
the voltage drop across R1 .
16.3.4
A Voltage Regulator with a Reduced Dropout Voltage
As we can see from Fig. 16.11, the minimum voltage drop between the input and output of
the voltage regulator is made up of the voltage drop of 0.6 V across current-sensing resistor
R3 , the Darlington circuit’s base–emitter voltage of 1.6 V, and the minimum voltage drop
of about 0.3 V across current source I1 . The minimum voltage drop (the dropout voltage)
is therefore 2.5 V. This is particularly troublesome for regulating low output voltages: in
the case of a 5 V regulator, at least 50% of the output power is dissipated. As an additional
voltage drop is required to compensate for line and load changes, an even higher power
dissipation results, which is generally just as large as the output power.
Removal of the heat generated frequently poses problems. Although IC voltage regu-
lators are provided with thermal protection, this means that the maximum output current is
reduced accordingly if cooling is inadequate. It is therefore important to keep the minimum
voltage drop as small as possible. This can be achieved in the circuit shown in Fig. 16.11
by operating the current source I1 from an auxiliary voltage a few volts above the input
voltage.
A method without auxiliary voltage is to use a pnp transistor as the power transistor,
as shown in Fig. 16.15. Here, the minimum voltage drop across the voltage regulator is
equal to the saturation voltage of the power transistor T1 . It can be held below 0.5 V if
an appropriately high base current is applied. However, in order to provide the required
base currents for T1 , a Darlington pair should not be used, as the minimum voltage drop
would be increased by an emitter–base voltage. Transistor T2 is therefore operated in a
common-emitter connection. Current feedback via R3 limits the maximum output cur-
rent and simultaneously improves the stability of the regulation circuit. A drawback of
this principle is the fact that because of the low current gain of the pnp transistors, the
current consumption rises steeply with an increasing output current. This current does
not contribute to the output, but is lost since the base current of T1 flows to ground via
T2 .Particularly disturbing is the rise in the current consumption when reaching the mini-
mum voltage drop, since the current flowing through T2 at this point reaches its maximum.
These problems can be bypassed by replacing the pnp power transistor by a p-channel
power MOSFET, such as the model Max 1658.
Vi
Vref
R3
Vi > 3.8 V 10 mΩ
8 7
57 mV
= OA2
V2 5
IRL
Vref = OA1 3103
1.2 V R1 R2
LP2975 - 3.3 24 kΩ 42 kΩ
4 3 2
C2 C1 Vo
470 pF
220 µF 3.3 V
For high currents it is necessary to use discrete power transistors. The control circuitry
can be constructed from commercially available operational amplifiers. A very simple
solution is achieved when using the voltage regulator LP2975 from National. Its internal
construction and external circuitry are shown in Fig. 16.16. The operational amplifier OA1
together with the reference voltage source forms the voltage regulator circuit. Since the
external power MOSFET represents an inverting amplifier in common-source connection,
the negative feedback signal must be fed to the non-inverted input of the operational
amplifier.
Because of the large capacitance C1 the dominating lowpass filter is located at the
output of the voltage regulator circuit. To still obtain a high stability we suggest using
capacitance C2 to turn the phase back in the critical frequency range.
Operational amplifier OA2 is used for current limitation. When the voltage drop across
R3 reaches the built-in current reference of V2 = 57 mV, the current regulator becomes
active and prevents the output current from rising further.
16.3.5
A Voltage Regulator for Negative Voltages
If a floating input source is available, the voltage regulators described above can also be
used for stabilizing negative output potentials. The resulting circuit is shown in Fig. 16.17.
We can see that it will operate only if the unstabilized voltage source is floating because the
voltage regulator or the output voltage would otherwise be shorted. This problem arises,
for instance, if we use the simplified circuit (Fig. 16.7) for simultaneously generating a
positive and a negative supply voltage. As the center tap is grounded, the negative supply
898 16 Power Supplies
Positive
Voltage-
regulator Vo > 0
Positive
Voltage-
regulator Vo < 0
Negative
Vo < 0 Voltage-
regulator
Fig. 16.17. Stabilizing a negative voltage Fig. 16.18. Stabilizing two voltages balanced
to ground
Vo Vo
Vref Vref
Vi Vi
Fig. 16.19. The 7900 family Fig. 16.20. The 337 family
R2 R1
Vo = − 1 + Vref Vo = − 1 + Vref
R1 R2
potential cannot be stabilized as in Fig. 16.17. In this case voltage regulators for negative
output voltages are required, as shown in Fig. 16.18. If IC types complementary to the 7800
or 317 series are used, the power transistor is operated in common-emitter configuration,
as this results in an easily fabricated npn transistor. The mode of operation of the circuits
shown in Figs. 16.19 and 16.20 thus corresponds to the voltage regulator with reduced
dropout voltage in Fig. 16.15. For this reason, the negative voltage regulator ICs have a
significantly lower dropout voltage than the corresponding positive voltage regulators.
16.3.6
Symmetrical Division of a Floating Voltage
The problem often arises, especially in battery-operated equipment, of obtaining two reg-
ulated balanced-to-ground voltages from a floating unstabilized voltage source. To solve
this problem, the sum of the two voltages can be stabilized to the desired value using one
of the circuits previously described. A second circuit is then required to insure that the
voltage is split in the correct ratio. In principle, we could use a voltage divider with its
tap grounded. The division of the voltage is kept constant if the internal resistance of the
voltage divider is low, but the loss in the voltage divider is then considerably increased. It
is therefore better to use a high resistance voltage divider and two power transistors. Only
the transistor connected to the DC bus carrying the smaller load current is turned on at any
one time. The relevant circuit is shown in Fig. 16.21.
The voltage divider formed by the two resistors R1 halves the voltage Vi . It may have a
high internal resistance, as its only load is the input bias current of the operational amplifier.
16.3 Linear Voltage Regulators 899
Vi /2
Vi
–Vi /2
If the tap of the voltage divider is at zero potential, the voltage Vi is split into a positive
and a negative voltage in the ratio 1:1, as required. The operational amplifier therefore
compares the tap potential with the ground potential and adjusts its output voltage so that
their difference becomes zero. Negative feedback is provided as follows: if, for instance,
the positive output is loaded more than the negative output, the positive output voltage
falls, thereby reducing the potential at the P-input of the operational amplifier. Because of
the high gain, the amplifier output potential reduces even further, so that T1 is turned off
and T2 is turned on. This counteracts the assumed voltage dip at the positive output. Under
steady-state conditions, the current through T2 is just large enough to insure that the two
output voltages share the load equally. Transistors T1 and T2 therefore operate as shunt
regulators, only one of which is conducting at any one time.
If the load is only slightly unbalanced, the output stage of the operational amplifier
can be used directly, instead of transistors T1 and T2 . The amplifier output is then simply
connected to ground.
16.3.7
Voltage Regulator with Sensor Terminals
The resistance Rw of the connecting wires between the voltage regulator and the load,
including possible contact resistances, may cancel out the low output resistance of the
regulator. This effect can be eliminated by incorporating the unwanted resistances in the
feedback loop; that is, by measuring the output voltage as near to the load as possible. This
is the purpose of the sensor terminals S + and S − in Fig. 16.22. The resistors in the sensing
leads produce no errors, as only small currents flow in them.
The four-wire regulation method described can also be implemented using IC voltage
regulators if the ground or voltage sensor terminal is externally accessible. Suitable types
include the 78 G, 79 G, L 200, or LT 1087.
Vi
w
V Vo
16.3.8
Bench Power Supplies
The output voltage of the voltage regulators described so far can be adjusted only within a
certain range Vo ≥ Vref . The current limit only serves to protect the voltage regulator and
is therefore fixed at I max .
A bench power supply must have an output voltage and a current limit, both of which
are linearly adjustable between zero and the maximum value. A suitable circuit is shown in
Fig. 16.23. Voltage regulation is provided by operational amplifier OA1, which is operated
as an inverting amplifier. The output voltage
R2
Vo = − Vref 1
R1
is proportional to the variable resistance R2 . The voltage can also be controlled by varying
Vref 1 . The output current flows from the floating unregulated power voltage source VL via
Darlington transistor T1 through the load and via current-sensing resistor R5 back to the
source.
The voltage across R5 is therefore proportional to the output current Io . It is compared
with a second reference voltage Vref 2 by operational amplifier OA2, which is operated as
an inverting amplifier. As long as
Io R5 Vref 2
<
R4 R3
VP 2 remains positive. The output voltage of OA2 therefore goes to its positive limit, and
diode D2 is reverse biased. In this operating condition, voltage regulation is therefore
unaffected. If the output current reaches the limit
R4
Io max = Vref 2 ,
R5 R3
V
A
Vo
V
V A
Io
Fig. 16.23. A bench power supply with a fully adjustable output voltage and a current limit
R2 R4
Vo = − Vref 1 ; Io max = Vref 2
R1 R5 R3
16.4 Reference Voltage Generation 901
then VP 2 = 0. The output voltage of OA2 falls, and diode D2 becomes forward biased.
This causes the base potential of the Darlington pair to fall; that is, current regulation
comes into effect. Amplifier OA1 tries to prevent the fall in output voltage by raising its
output potential to the maximum, thereby turning off diode D1 , and current regulation is
unimpaired. Therefore the two diodes act as an analog and gate.
In power supplies whose output voltage can be adjusted to zero, exceptionally high
power dissipation may occur. In order to be able to achieve Vo max , the unstabilized volt-
age VL must be greater than Vo max . Maximum power dissipation in T1 occurs when the
maximum output current Io max is allowed to flow at low output voltage levels. It is then ap-
proximately Vo max · Io max ; that is, just as high as the maximum available output power. For
this reason, it is preferable, when comparatively high powers are involved, to use switched-
mode regulators in the output stage. This is because their power dissipation remains small
even if the voltage drop is large.
16.3.9
IC Voltage Regulators
Apart from a small number of voltage regulators for special applications, these devices
can be subdivided into two main families: the 7800 or 317 series (see Fig. 16.24). Both
categories also include negative voltage regulators. Whereas types with adjustable output
voltage are the exception in the 7800 series, all the 317-series types are adjustable and
have only three terminals.
We can see that the dropout voltage for all types is 2 V or more. This is particularly
troublesome for 5 V regulators that have to handle large currents, as the power dissipation
in the voltage regulator is then in excess of 40% of the output power. Consequently, the
power supply efficiency is only 25%, which means that three times the power delivered
is converted into heat. One solution is to use voltage regulators with a reduced dropout
voltage. Nevertheless, by using suitably rated discrete components as – for example, in
Fig. 16.16 – even a 3.3 V power supply can be made to operate at over 50% efficiency.
Another way of minimizing losses is to use switching regulators, as we shall describe
in Sect. 16.5.
16.4
Reference Voltage Generation
Every voltage regulator requires a reference voltage with which the output voltage is
compared. The stability of the output voltage is only as good as that of the reference. In
this section, we shall therefore examine various aspects of reference voltage generation in
greater detail.
16.4.1
Zener Diode References
The simplest method of generating a reference voltage is to apply the unstabilized input
voltage to a Zener diode via a series resistor, as in Fig. 16.25. The quality of the stabilization
is characterized by the suppression of input voltage variations (line regulation) Vi /Vref ,
902 16 Power Supplies
Vi R R
= 1+ ≈ = 10 . . . 100
Vref rZ rZ
where rZ is the dynamic resistance of the Zener diode at the operating point selected. In a
first-order approximation, rZ is inversely proportional to the current flowing in the diode.
Increasing the series resistance R for a given input voltage will not therefore produce an
improvement in stabilization. An important aspect to consider when defining the diode
current is the noise in the Zener voltage, which increases markedly at low currents. The
resistance R is selected such that an adequate diode current will still flow at minimum
input voltage and maximum output current.
Considerably improved stabilization can be achieved if the series resistor R is replaced
by a current source, as shown in Fig. 16.26. The simplest method is to use a FET current
source, as this has only two terminals (see Fig. 4.126). Stabilization factors of up to 10,000
can then be achieved.
16.4 Reference Voltage Generation 903
Vi Vref Vi Vref
Fig. 16.25. Voltage stabilization using a Fig. 16.26. Improving the stabilization using
Zener diode a constant current source
Another way of operating the Zener diode with a constant current is to connect it to
the stabilized output voltage, rather than to the unregulated input voltage. As shown in
Fig. 16.27, we generate an output voltage
R2
Vref = 1+ VZ ,
R1
which is higher than Zener voltage VZ . Constant current IZ = (Vref − VZ )/R3 then flows
through R3 . In this case, line regulation is primarily determined by the supply ripple rejec-
tion D = Vb /VO of the operational amplifier, where VO is the operational amplifier
offset voltage. Using the relations
rZ R1
VO = VP − VN , VP = Vref , VN = Vref
rZ + R 3 R1 + R 2
Values of around 10,000 are achieved. Even if the input voltage ripple amounts 10 V, the
output voltage will then vary only 1 mV.
Considerably larger variations may occur due to changes in temperature. The tempera-
ture coefficient of the Zener voltage is in the order of ±1·10−3 /K. For small Zener voltages,
it is negative and for larger ones, positive. Its typical characteristic is plotted in Fig. 16.28.
We can see that the temperature coefficient is at its smallest for Zener voltages around 6 V.
For larger Zener voltages, it can be reduced by connecting forward-biased diodes in series.
Although discrete components of this kind are available as reference diodes, in most cases
IC reference voltage sources containing reference diodes are used, as shown in Fig. 16.27.
Temperature coefficients up to 10− 6 /K = 7 1 ppm are achieved.
Vi
Vref
Fig. 16.27. Operating the Zener diode from the
regulatedvoltage
R2
Vz Vref = 1 + VZ
R1
904 16 Power Supplies
TC
. VZ 1
TC = ·
VZ ϑ
Vz
.
.
16.4.2
Bandgap Reference
In principle, the forward voltage of a diode or the base–emitter voltage of a bipolar transistor
can also be used as a voltage reference. However, the temperature coefficient of −2 mV/K
at 0.6 V is rather high. It can be compensated for by adding a voltage with a temperature
coefficient of +2 mV/K. The characteristic feature of the circuit in Fig. 16.29 is that this
voltage is generated by a second transistor. Transistors T1 and T2 are driven by different
collector currents; IC2 > IC1 . From the transfer characteristic, we obtain a voltage drop
across R1 of
IC2
VBE = VBE 2 − VBE 1 = VT ln
IC1
VBE
VTemp = R2 (IC1 + IC2 ) = R2 (1 + n)
R1
R2
= VT (1 + n) ln n = AVT
R1
Vref
We now have the possibility of achieving any gain factor A = (1 + n)ln n R2 /R1 by select-
ing suitable values of n and R2 /R1 . Thus, for VTemp we obtain a temperature coefficient
of +2 mV/K if we select A ≈ 23, because
dVTemp dVT k VT
= A· = A = A
dT dT e0 T
26 mV mV
= 23 · = +2
300 K K
According to (2.21) on page 54 the theoretical value for the temperature coefficient of a
bipolar transistor is
dVTemp mV
≈ −2
dT K
where VBG , = Eg /e0 = 1.205 V is the bandgap voltage of silicon, Eg being the bandgap.
The temperature coefficient of the output voltage Vref = Vtemp + VBE 2 is therefore zero if
This is a more precise and at the same time simpler adjustment criterion than setting the
gain A to some calculated value.
For the discrete circuit design shown in Fig. 16.29, we obtain useful component values
if IC2 = 10IC1 . In this case, R1 = R2 is suitable. In order to achieve good matching of T1
and T2 , a double transistor is necessary, such as an LM 394.
Discrete-component bandgap references are only of interest in special cases, as a wide
variety of IC versions is available (see Fig. 16.31). As transistors T1 and T2 in Fig. 16.29
are sometimes operated with identical collector currents, different current densities must
then be achieved by connecting several transistors in parallel for T1 .
The significant advantage as compared to reference diodes is that bandgap references
can be operated at lower voltages, which can be as low as the bandgap voltage VBD ≈
1.2 V. Reference diodes, on the other hand, require voltages of 6.4 V and above. Moreover,
reference voltages of any value can be produced using bandgap references, when only a
portion of the operational amplifier output voltage is fed back to the base terminals, as in
Fig. 16.30. In this case, the actual reference voltage source T1 , T2 is operated from the
regulated output voltage. This provides significantly improved line regulation, as shown
in Fig. 16.27.
Vi
Vref
With many IC bandgap references, it is permissible to connect the output to the sup-
ply voltage, or the corresponding connection is already established internally. Only two
terminals are then brought out of the circuit, and it can be used like a Zener diode.
As voltage VTemp is proportional to the absolute temperature, it can be used for tem-
perature measurement (see Sect. 21.1.5). In many circuits Vtemp = 2 mV/K is brought out
to the terminals for this purpose.
16.4.3
Types
We have listed a number of commonly used reference voltage sources in Fig. 16.31. Note
the tight tolerances and the low temperature coefficients of many types. These are achieved
by laser adjustment of the relevant resistors during manufacture. However, the specified
values only give a rough indication, as all of the circuits are available in various precision
categories.
The two-terminal types behave like Zener diodes. Consequently, their current must
never reach zero. Some three-terminal types incorporate a simple emitter follower at the
output. This enables them to carry high currents at the output, but accept only low input
currents. Other types incorporate complementary emitter followers at the output. Therefore,
they can also accept high currents.
All voltage regulators and many AD und DA converters have built-in reference voltage
source so that often a separate reference voltage source is not required.
16.5
Switched-Mode Power Supplies
The power supplies incorporating linear series regulators described hitherto are subject to
three basic loss factors: the line transformer, the rectifier, and the regulating transistor. The
efficiency η = Poutput /Pinput is in most cases only 25 − 50%. The power dissipation,
1
P loss = Pinput − Poutput = − 1 Poutput
η
can therefore be up to three times as large as the power output. Consequently, there is not
only a large power loss but also an associated cooling problem.
The losses in the series regulator can be substantially reduced by replacing the con-
tinuously controlled transistor by a switch, as shown in Fig. 16.32. In order to obtain the
required DC output voltage, a lowpass filter is additionally required to provide time aver-
aging. The magnitude of the output voltage can be determined in this case by the duty cycle
with which the switch is closed. If an LC lowpass filter is used, the regulator no longer
possesses an inherently lossy element. As the switching regulator described is connected
on the secondary side of the power transformer, these circuits are also known as secondary
switched-mode power supplies.
The losses in the power transformer are, of course, not reduced by the switching reg-
ulator. However, they can be brought down by transforming a high-frequency alternating
voltage instead of the line voltage. For this purpose, the line voltage is directly recti-
fied as shown in Fig. 16.33, and an alternating voltage with a frequency in the range
20 kHz…2 MHz is generated using a switching regulator.
As the number of turns required on the power transformer decreases with the switching
frequency, the copper losses can be greatly reduced. The secondary voltage is rectified,
filtered, and then fed directly to the load. The duty cycle of the switches on the primary
side is varied to regulate the DC voltage.
Circuits of this kind are known as primary switched-mode power supplies. Their effi-
ciency can be 60–80%. An additional advantage is their compactness and the low weight
of the HF transformer.
PWM
controller
If we compare the two basic circuits shown in Figs. 16.32 and 16.33, we can see that
in both cases a switch is used to generate an AC voltage whose duty cycle determines
the output voltage. Whereas, with the secondary regulator, line isolation is provided by a
normal 50 Hz power transformer, with the primary regulator this function is performed by
an HF transformer. Consequently, the switches in the primary regulator are at line potential
and must be qualified for line voltages. In this case, the regulator comprises two sections,
one that is at the line potential and controls the switches, and another that is at the output
potential and measures the output voltage. Both sections must be suitably insulated from
each other and fulfill the public regulations.
In spite of these problems and the associated circuit complexity, primary switching
power supplies are to be preferred due to their high efficiency. Secondary switching power
supplies are mainly used as low-power DC/DC converters.
16.6
Secondary Switching Regulators
The three basic types of DC/DC converters are shown in Figs. 16.34–16.36. Each one
consists of three components: a power switch S, a storage choke L, and a smoothing
capacitor C. However, each of the three circuits delivers a different output voltage. In the
case of the circuit shown in Fig. 16.34, the switch produces an AC voltage whose average
value lies between the input voltage and zero, depending on the duty cycle.
With the circuit in Fig. 16.35, Vo , = Vi if the switch remains permanently in the upper
position. If the switch is moved to the lower position, energy is stored in the choke and
additionally delivered to the output when the switch is returned to the upper position. The
output voltage is therefore higher than the input voltage.
In the circuit shown in Fig. 16.36, energy is stored in the choke as long as the switch
is in the left-hand position. When it switches to the right, the choke current direction does
not change and the capacitor is charged to negative values (if the input voltage is positive).
In the circuit shown in Fig. 16.34, current flows continuously into the reservoir ca-
pacitor. The circuit is therefore also known as a forward converter. This is not true of
Figs. 16.35 and 16.36, because there the capacitor is not recharged as long as the choke is
being charged. These circuits are known as flyback converters.
16.6.1
Step-Down Converters
The toggle switch can be realized by two simple switches or by using only one switch
and a diode for the other path. This results in the step-down converter (the buck regulator)
S S
S
Vi Vo Vi Vo Vi Vo
Fig. 16.34. Step-down Fig. 16.35. Step-up converter Fig. 16.36. Inverting converter
converter 0 ≤ Vo ≤ Vi Vo ≥ Vi Vo < 0 for Vi > 0
16.6 Secondary Switching Regulators 909
V1
Vi
ton toff
Vo
i S o
Vi Vo
V Io
Io
Fig. 16.37. Step-down converter with a Fig. 16.38. Current and voltage waveforms
simple switch ton
ton Ii = Io I L = Io
Vo = Vi for Io ≥ Io min T
T
shown in Fig. 16.37. As long as the switch is closed, V1 = Vi . When it opens, the choke
current retains its direction and V1 falls until the diode is turned on; that is, virtually to
zero potential. This can be seen from the timing diagram shown in Fig. 16.38.
The time characteristic of the reactor current is deduced from the law of induction:
dIL
VL = L · (16.9)
dt
During the on-time ton , a voltage of VL = Vi − Vo is applied across the choke; during the
off-time toff , a voltage of VL = −Vo is present. Using Eq. (16.9), the rate of change of
the current is given by
1 1
IL = (Vi − Vo )ton = Vo toff (16.10)
L L
From this, we can calculate the output voltage:
ton ton
Vo = Vi = Vi = p V1 (16.11)
ton + toff T
where T = ton + toff = 1/f is the period and p = ton /T is the duty cycle1 . We can see
that the output voltage is the arithmetic mean of V1 , as we would expect.
The circuit behaves quite differently if the output current Io becomes smaller than
1 Vo
Io min = IL Vo = 1 − (16.12)
2 Vi
The choke current then reaches zero during the off-time of the switch, the diode is turned
off, and the voltage across the choke becomes zero, as shown in Fig. 16.39 (discontinuous
operation). To calculate the output voltage, we shall assume that the circuit provides lossless
operation. The average input power must therefore be equal to the output power:
Vi I e = Vo Io (16.13)
The current flowing through the choke rises during ton , from zero to the value IL =
VL ton /L. The arithmetic mean of the input current is therefore
ton 1 t2 T
Ii = · IL = on VL = (Vi − Vo )p 2 (16.14)
T 2 2T L 2L
1 Sorry, the commonly used letter for the duty cycle η is spent for the efficiency already.
910 16 Power Supplies
V1
Vi
Vo
ton toff
Io
Io min Io
Fig. 16.39. Current and voltage waveforms Fig. 16.40. Duty cycle p = ton /T as a
in the step-down converter for output function of the output current Io at constant
currents of less than output voltage Vo
T Vo
Io min = Vo 1 −
2L Vi
Substitution into (16.13) gives the output voltage and duty cycle, respectively:
Vi2 p 2 T 2L Vo
Vo = p = Io (16.15)
2LIo + Vi p T 2 T Vi (Vi − Vo )
In order to prevent the output voltage from rising in the event of low currents (Io < Io min ),
p must be reduced accordingly. This is shown schematically in Fig. 16.40. We can see
that very short switching times must be achieved in this region. For currents higher than
Io min , the duty cycle remains constant in accordance with Eq. (16.11). However, this only
applies to a lossless circuit. Otherwise, p must also be increased – albeit by a much smaller
amount – as the output current increases above Io min , in order to keep the output voltage
constant.
Design Considerations
If possible, the inductance of the storage choke is selected to be large enough to prevent
the current from falling below Io min . From (16.12), it therefore follows:
Vo Vo
L = T 1− (16.16)
Vi 2Io min
The maximum current flowing through the storage choke, and consequently through the
switch and diode, is therefore IL max = Io + Io min . This still leaves the parameter with
a period of T = 1/f . In order to allow a small inductance to be used, the frequency f
is selected to be as high as possible. However, the problem then arises that at high fre-
quencies the switching transistor becomes more costly and the drive circuit more complex.
In addition, the dynamic switching losses increase proportionally with the frequency. For
these reasons, switching frequencies between 20 kHz and 2 MHz are preferred.
The smoothing capacitor C determines the output voltage ripple. The charge current
is IC = IL − Io . The charge applied and removed during one cycle therefore corresponds
to the hatched area in Fig. 16.38. For the ripple, we therefore obtain
QC 1 1 ton toff IL T
Vo = = · · + · = IL
C C 2 2 2 2 8C
16.6 Secondary Switching Regulators 911
16.6.2
Generating the Switching Signal
The switching signal is generated using two modules: a pulsewidth modulator and a reg-
ulator with voltage reference. The block diagram is shown in Fig. 16.41.
The pulsewidth modulator comprises a sawtooth generator and a comparator. The
comparator closes the switch as long as voltage VR is greater than the triangular voltage.
The resultant control voltage, Vctl , is shown in Fig. 16.42 for VR ranging from the lower
limit to the upper limit. The resulting duty cycle,
ton VR
p = =
T V̂ST
is therefore proportional to VR .
The subtractor takes the difference between the reference voltage and the weighted
output voltage Vref −kVo . The control amplifier increases VR until this difference becomes
zero. The output then has the value Vo = Vref /k.
An example should serve to illustrate the design process for a switching regulator. Let
us assume that an output voltage of 5 V at a maximum current of 5 A is required.
The minimum output current is to be 0.3 A, and the input voltage is approximately 15 V.
A suitable switching regulator for this application would be the LM 2678-5 from National.
The resulting circuit is shown in Fig. 16.43. Apart from the LC output filter only a few
external resistors and capacitors are required for the operation of the integrated circuit.
The switching regulator is to be operated at a frequency of 250 kHz, resulting in a period
of T = 4ms. From Eq. (16.11), we obtain a switch-on time of
Vo 5V
ton = T = 4 ms = 1.3 ms
Vi 20 V
Vctl V Vo Vo
VST
V
on
Vctl
Vi
VST 2 3 DBoost 8...
40 V
Vref
250 kHz Vctl MOS CBoost
= VR Driver
1.2 V PWM T
L = 22 H
R1 Komp.
15 k
1 Vo
2.5 k C
C1 LM2678-5 5V
7.8 k 220 F
4 6
10 nF
If the output ripple is to have a magnitude of 10 mV the value of the smoothing capacitor
is determined from (16.17)
Io min 0.3A
C = T = 4 ms = 30 mF
4Vo 4 · 10 mV
In the calculation of the filter capacity, the equivalent series resistance (ESR) and the
equivalent series inductance (ESL) have not been taken into account. To still reach an
acceptable level of output ripple requires a distinctly higher capacitance. In the practical
realisations several small capacitances are connected in parallel, since this always gives
lower values of the series resistance and the series inductance than with one large capacitor.
The output voltage remains constant even if the output current is lower than the value
of Io min = 0.3 A. In this case the control amplifier reduces the duty factor through the
comparator as in Fig. 16.40. Problems will arise if the required on-time is shorter than the
minimum achievable on-time of transistor T. In this case the output voltage increases so
greatly in response to a single turn-on pulse of T that the transistor is then turned off for
several cycles. This results in very unsmooth operation.
The desired control response is achieved by connecting the on-chip network R1 C1
to the output of the the high-impedance operational amplifier. It must be considered that
the voltage regulating circuit of switching regulators is prone to instability. This has two
causes: firstly, the switching regulator is a sampling system with an average dead time
equal to half the period; secondly, the output filter represents a second-order lowpass filter
which produces a phase lag of up to 180◦ . For these reasons it is advisable to insure that
the control amplifier produces no phase lag at high frequencies. Resistor R1 in Fig. 16.43
is used for this purpose.
The static circuit losses are mainly due to the voltage drops in the power circuit. The
smoothing choke can easily be dimensioned so that the resistive losses are low. The losses
due to the voltage drop across the power switch formed by transistor T and diode D then
remain.
16.6 Secondary Switching Regulators 913
The output current flows through T during ton and through D during toff . For the case
of a voltage drop of 0.7 V across the transistor or the diode with an output current of 5 A,
a power dissipation of 3.5 W results. The maximum efficiency is therefore
Poutput 25 W
η = = = 88%
Pinput 25 W + 3.5 W
This does not include the switching losses, which are not insignificant at high switching
frequencies. In addition, the efficiency is further diminished by the current consumption
of the switching regulator itself. Since this contribution does not depend on the output
current, the efficiency is diminished especially at low output currents. To operate the n-
channel MOSFET in the resistive region requires a gate potential that is positive compared
with the input voltage. For this purpose the MOS gate driver generates an auxiliary voltage
that is positive compared with the source potential. The boost capacitor transfers this
alternating output voltage to the MOS driver. As an alternative the manufacturer could
have used a p-channel MOSFET which has twice as high an on-resistance for the same
chip area.
16.6.3
Step-Up Converters
Figure 16.44 shows a practical implementation of the step-up converter (boost regulator)
presented in Fig. 16.35, and Fig. 16.45 gives the voltage and current waveforms. Once
again, the relations required for the design parameters of the circuit can be derived from
the rise or fall in the choke current IL during the two states of switch S. These relations are
given below the diagrams. The smallest output voltage is Vo = Vi . For a lossless circuit,
this is obtained when switch S is continuously open.
Here too, the specified output voltage is only achieved if the choke current does not
become zero. If the output value falls below the minimum Io min , the on-time must be
reduced, as shown in Fig. 16.40, in order to prevent a rise in the output voltage. This case
is shown by the dashed lines in Fig. 16.45. The switching signal is generated in precisely
the same way as for the step-down converter.
V1
Vo
Vi Vo
V1 S Vi
S V1
Vi
Vi Vo
V1
Vo
Fig. 16.46. Inverting converter
ton
Vo = Vi for Io > Io min
toff
Vi2 VoT
Io min = ·
(Vi + Vo )22L Fig. 16.47. Voltage and current waveforms
Vi Vo T in the inverting converter. Dashed lines: for
L = · Io < Io min
(Vi2 + Vo )2 2 L
T Io max
C ≈
Vo
16.6.4
Inverting Converter
The inverting converter (inverting regulator) and associated waveforms are shown in
Figs. 16.46 and 16.47.
As we can see, the capacitor is charged to a negative voltage via the diode during the
off-phase. The relations given are again deduced from the fact that the reactor current
changes are equal during the on- and off-times.
If the output current falls below the value Io min , the reactor current occasionally be-
comes zero. In order to keep the output voltage constant in this case, the on-time must again
be reduced, as shown in Fig. 16.40. This is represented by the dashed lines in Fig. 16.47.
16.6.5
Charge Pump Converter
If the current requirement is low, there is a simple method of inverting a voltage. For this
purpose, the input voltage is converted into an AC voltage using switch S1 , as shown in
Fig. 16.48. This alternating voltage is made “floating” by capacitor C1 and then rectified
again, this time by switch S2 . In the switch position shown, C1 is charged up to the input
voltage: V1 = Vi . The two switches then change over, causing voltage −V1 , to be applied
to C1 , and the latter is charged up to the voltage Vo = −V1 = −Vi after several switching
cycles.
Rectification of the output voltage does not necessarily require a controlled switch, but
can also be performed by two diodes, as shown in Fig. 16.49. Depending on which potential
is applied to the rectifier and on the polarity of the diodes, voltage Vi can be added to or
subtracted from that potential. However, the disadvantage of rectification using diodes
is that the output voltage is reduced by the two forward voltage drops. Consequently,
integrated circuit voltage inverters employ CMOS switches for rectification. Due to the on
resistances the voltage drop is proportional to the output current.
16.6 Secondary Switching Regulators 915
S1 S2 S1
Vi Vo Vi Vo
V1 V1
Fig. 16.48. Voltage inverter employing the Fig. 16.49. Simplified arrangement for
charge pump principle rectifying the output voltage
Vo = −Vi Vo = −(Vi − 2VD )
Additional losses are incurred due charge changes of the capacitors. However, these
depend only on the size of the voltage difference produced, which under steady-state
conditions can easily be minimized by selecting suitably high-value capacitors.
16.6.6
Integrated Switching Regulators
In Fig. 16.50 charge-pump voltage converters are collected which only need external
capacitors. They are especially appropriate for low voltages and low currents. Examples
for integrated switching regulators that need a storage choke are listed in Fig. 16.51. Even
types with internal power transistors can provide remarkable output power. For high output
currents types for external power transistors are available. In order to reduce the power
loss in the rectifying diode an internal or external power transistor is used for synchronous
rectification.
Fig. 16.51. Examples for non-isolated (secondary) switching converters using inductors
16.7
Primary Switching Regulators
Primary switching regulators fall into two categories, namely single-ended and push–pull
converters. As single-ended types generally require only one power switch, the number of
components involved is low. However, their use is limited to low-power applications. For
powers in excess of 100 W, push–pull converters are preferred, even though they require
two power switches.
16.7 Primary Switching Regulators 917
16.7.1
Single-Ended Converters
The single-ended converter shown in Fig. 16.52 represents the simplest practical primary
switching regulator. It is similar to the flyback converter in Fig. 16.46, except that the
storage choke has been replaced by a transformer. As long as power switch S is closed,
energy is stored in the transformer. This energy is transferred to smoothing capacitor C
when the switch opens. The resulting relation for the output voltage is the same as for the
circuit shown in Fig. 16.46. The only difference is that the output voltage is now reduced
by the winding ratio W of the transformer, where W = w1 /w2 (see Sect. 18.7.3).
The waveform of the voltage across the switch is plotted in Fig. 16.53. When the switch
opens, the voltage rises until diode D is turned on; that is, to VS max = Vi + W Vo . In order
to prevent it from becoming too high, we make the on-time ton ≤ 0.5T , which means that
VS max ≤ 2Vi . As a DC voltage of
√
Vi = 230 V · 2 = 325 V
is produced when rectifying the 230 V AC line, a voltage of VS max = 650 V is present at
the power switch. The voltages actually present are even higher due to the unavoidable
leakage inductance.
The current characteristic is also shown in Fig. 16.53. As long as the switch is closed,
the rise of the current is I = Vi ton /L. When the switch opens, the diode is turned on and
the current falls accordingly I = W Uo toff /L, producing the output voltage indicated.
However, the transformer inductance must be large enough to insure that the current does
not fall to zero during the off-time.
One disadvantage of the circuit is that the transformer has to provide not only AC line
isolation and the required stepping-down of the voltage, but must simultaneously act as a
storage choke. Due to the DC biasing which occurs, it has to be considerably overrated. A
better solution is to keep the transformer free of any DC component and to use a separate
storage choke. All of the following circuits operate on this principle.
In the case of the single-ended converter in Fig. 16.54, the primary and secondary
windings have the same polarity. Consequently, energy is transferred to the output via
diode D2 , as long as the power switch is closed. The circuit is therefore a forward converter.
The voltage characteristics are shown in Fig. 16.54. As long as the power switch is closed,
VS VS max
W
Vi
w1 w2 Vo
Vi
IS W
V
Fig. 16.52. Single-ended flyback converter Fig. 16.53. Voltage and current waveforms
for Io > Io min
ton Vi
Vo = · for Io > Io min
toff W
ton
VS max = Vi 1 +
toff
W = w1 /w2
918 16 Power Supplies
VS VVgm
W W S max
2V i
V2 Vo Vi
Vi
V2 Vi
VS S W
the input voltage Vi is present at the primary winding and therefore voltage V2 = Vi /W is
present at the secondary winding. When switch S opens, D2 is turned off and the current
through storage choke L is carried by diode D3 . The conditions on the secondary side
are therefore precisely the same as for the forward converter in Fig. 16.37. Consequently,
apart from the factor W , we obtain the same relations for the output voltage, and the same
considerations apply to the design procedure for the storage chokes and the smoothing
capacitor.
At the instant at which the power switch is turned off, diode D2 also becomes reverse
biased. Without further action, the energy stored in the transformer would then generate
an extremely high-amplitude voltage spike. In order to prevent this, the transformer is
provided with a third winding that has the same number of turns as the primary winding,
but a smaller cross-section. For the given polarity, diode D1 then becomes conducting
when the induced voltage equals the input voltage. In this way, the voltage across the
power switch is limited to VS max = 2Vi . In addition, the same energy is fed back to
the input voltage source during the off-time as was stored in the transformer during the
on-time. In this way, the transformer is operated without DC magnetization.
16.7.2
Push–Pull Converters
With circuits of this type, the DC input voltage is converted to AC form by an inverter
that comprises at least two power switches. This AC voltage is stepped down in an HF
transformer and subsequently rectified.
In the circuit shown in Fig. 16.56, the period T is subdivided into four time periods.
Initially, switch S1 is closed, which means that diode D1 is on and voltage V3 = Vi /W is
present at storage choke L. Switch S1 then reopens and all of the voltages at the transformer
fall to zero. Diodes D1 and D2 then each carry half of the choke current.
In the next time period, switch S1 remains open, whereas switch S2 closes. This turns
on D2 , which likewise transfers voltage V3 = Vi /W . When S2 reopens, all of the voltages
at the transformer become zero once more, as during the second time period. The relevant
voltage waveforms are shown in Fig. 16.57.
The secondary side of the circuit therefore operates in basically the same way as the
forward converter in Fig. 16.37. However, energy is now transferred to the storage choke
16.7 Primary Switching Regulators 919
S2 2Vi V
W W VS1
VS2 V3 Vi
Vo
Vi V3 on on
S1
Vi
VS1 W
1
twice during period T , due to full-wave rectification. Consequently, 2 T instead of T must
be substituted in the forward converter equations.
Due to the balanced mode of operation, the transformer operates without direct current.
However, this only applies if the on-times of the power switches are precisely equal; that
is, t1 on = t2 on = ton . This condition must be fulfilled when drive is applied to the switches;
otherwise, the transformer will be driven into saturation, the currents will become high,
and the switches will be destroyed. For the same reason, it is necessary to prevent one
switch from not closing at all during a cycle. However, these conditions are taken into
account in the majority of IC drive circuits for push–pull switching regulators. The drive
arrangement for the power switches is simple here due to the fact that their two negative
terminals are at the same potential.
In the case of the push–pull converter shown in Fig. 16.58, an AC voltage is produced by
connecting one end of the primary winding alternately to the positive or negative terminal of
1
the input voltage, while the other is at 2 Vi . The power switches are again driven alternately.
The voltage waveforms in Fig. 16.59 are the same as for the previous circuit. The only
difference is that the amplitude is halved, a feature that is particularly advantageous for
switch selection.
V
V
Vi
Vi W
Vi
V3 Vo
Vi
V3 on on
Vi
A further advantage of the circuit is that the transformer is always DC-free due to
capacitive coupling, even if the on-times of the two switches are unequal. In this case, only
the DC voltage across capacitors C1 and C2 is slightly displaced. However, a disadvantage
is that the negative terminals of the power switches are at quite different potentials, which
makes the drive arrangement more complex.
16.7.3
High-Frequency Transformers
Storage chokes are commercially available in a wide variety of types. Various manufactur-
ers offer types rated from 1 mH to 10 mH and from 0.1 A to 60 A. There is therefore little
necessity for the user to wind them him- or herself. However, this is not the case with high-
frequency transformers. Here, the user would be lucky to find a ready-made transformer
with the appropriate turns ratio. Consequently, if only small quantities are required, the
user generally has to calculate the transformer data and also wind the transformers him-
or herself.
According to the law of electromagnetic induction, the voltage induced in a transformer
is given by
˙ = w · Ac · Ḃ ,
V = w (16.18)
where is the magnetic flux, B is the magnetic induction, and is Ac the cross-sectional
area of the core between the two coils. If the number of turns on the primary side is w1 , it
follows from Eq. (16.18) that
V1 V1 t
w1 = = ·
Ac · Ḃ Ac B
With B = B̂, the minimum number of turns results from the permissible peak value of
the magnetic induction, B̂ and from the maximum value of
t = ton max = p max · T = p max /f = 1/2f
Hence
V1
w1 = (16.19)
2Ac B̂ · f
We can see that the required number of turns is inversely proportional to the frequency.
Consequently, the power that can be transferred for a given core, and thus for a given
winding area, is proportional to the frequency.
V2 w1
w2 = w1 = (16.20)
V1 W
The magnetizing and copper losses can generally be kept negligibly low. The wire gauge
depends on the currents to be handled. Current densities of up to CD = 5 . . . 7 A/mm2
are permissible in terms of thermal requirements. However, if the copper losses are to be
minimized, lower values should be adopted. The wire diameter is given by
-
I
d = 2 (16.21)
π · CD
16.7 Primary Switching Regulators 921
.
.
.
.
.
Fig. 16.60. Skin effect: skin depth as a function
of frequency
However, due to the skin effect, at higher frequencies the current no longer flows uniformly
through the entire cross-section, but only at the wire surface. For the skin depth (drop to
1/e) of the current, one can use the empirical formula
δ = 2.2 mm/ f/kHz . (16.22)
We can see from Fig. 16.60 how the skin depth reduces with increasing frequency. For this
reason, it is inadvisable to select the wire diameter to be greater than twice the skin depth.
In order that the required cross-sections can still be achieved, litz wire composed of fine,
separately insulated strands can be used. It is also preferable to employ ribbon cable or
correspondingly thin copper foils.
The principal characteristics of a number of ferroxcube EC cores are listed in Fig. 16.61.
The maximum power is only a rough guideline. If the wire diameter is substantially over-
sized in order to minimize the losses, it is possible that the next larger size of core will be
required in order to provide sufficient winding space.
16.7.4
Power Switches
The aspects discussed in this section apply to the power switches of all switching regulators.
The components that we shall consider here are bipolar transistors and power MOSFETs.
The use of IGBTs is only of interest when high powers in the kilowatt range are involved;
consequently, these devices will not be discussed here. If we look at the safe operating
area (SOA) of power transistors, we can see that there are virtually no power transistors
that can handle 100 W at high voltage levels. However, when these devices are used as
high-speed switches, there are a number of exceptions, as shown in Fig. 16.62.
922 16 Power Supplies
. secondary
breakdown
VCE
Fig. 16.62. Safe operating area SOA of a bipolar transistor used as a switch
DC power up to VCE = 50 V : 50 W
DC power up to VCE = 500 V : 5W
Pulse power for 5 ms VCE = 500 V : 2500 W
We can see that the power dissipation and secondary breakdown can be exceeded, albeit
briefly, and in extreme cases (for a few microseconds) it is even permissible for VCE max
and IC max to be applied simultaneously. It is therefore possible to use a 50 W transistor to
switch several kilowatts, a characteristic that is utilized in switched-mode power supplies.
However, there is a second reason for switching the transistors on and off quickly: a
switch only operates in a lossless condition if the transition from the off-state to the on-
state and vice versa is instantaneous. Otherwise, so-called switching losses occur. These
are greater the longer the switching process lasts. As they occur each time the transistor
switches, they are proportional to the switching frequency.
In addition, with most switching regulators it is also preferable to provide short on-
times ton , in order to insure orderly operation even at low load currents Io < Io min . For this
purpose, it is essential to switch the transistor off rapidly. Consequently, we have to avoid
the problem of the storage time of bipolar transistors by preventing them from going into
saturation during the conducting phase (VCE > VCE sat ). These two cases are compared
in Fig. 16.63. We can see that a slight increase in the voltage drop across the conducting
transistor must be tolerated in order to eliminate the storage time.
The basic arrangement for a bipolar transistor operated as a power switch is shown
in Fig. 16.64. In order to turn on the transistor, switch S is moved to the upper position,
allowing a large base current to flow via resistor R1 . This causes the collector current to
rise rapidly, and a short fall time is produced. When the collector potential falls below the
base potential, the Schottky diode is turned on, preventing the transistor from being driven
into saturation. The major portion of the current through R1 is now diverted via the diode
to the collector, and the remaining base current immediately assumes the value required
by the transistor at this operating point. In order to turn the transistor off, it is not sufficient
to cut off the base current. Its direction must be reversed in order to remove the charge of
the base junction. If this is to be effected rapidly, a large negative base current is required,
its magnitude being determined by resistor R2 in Fig. 16.64.
16.7 Primary Switching Regulators 923
unsaturated saturated
L
L
Vi
Fig. 16.64. Base drive Fig. 16.65. Practical base drive Fig. 16.66. Darlington
for short switching times arrangement pair with a speed-up
diode
Q = 500 pF · 10 V + 50 pF · 325 V = 5 nC + 16 nC = 21 nC
924 16 Power Supplies
325
325
Vi Vi
Fig. 16.67. Drive circuit for a power MOSFET Fig. 16.68. Drive circuit for a power MOSFET
with a complementary emitter follower with a totem pole circuit
For the gate potential to rise in 100 ns, a current of I = 21 nC/100 ns = 210 mA is
required. The gate current is therefore of the same order of magnitude as the base current
of the bipolar transistors. The only difference is that the gate current only flows at the
switching instant. In order to switch power MOSFETs on and off rapidly, low-impedance
drivers are therefore required. Figure 16.67 shows a complementary emitter follower and
Fig. 16.68 a totem pole output stage of the type commonly used in TTL gates. Being easier
to implement in monolithic technology, it is therefore preferred in driver ICs. In terms of
their drive circuitry, power MOSFETs have the advantage of not requiring negative voltage
sources, as is the case with bipolar transistors.
16.7.5
Generating the Switching Signals
The switching signals for single-ended converters can be generated using a pulsewidth
modulator, as described earlier in Sect. 16.6.2. However, push–pull converter operation
requires two alternately activated pulsewidth modulated outputs. To generate these signals,
in the pulsewidth modulator in Fig. 16.41 a toggle flip-flop is added, giving the circuit shown
in Fig. 16.69. It changes state on each negative-going edge of the sawtooth signal, thereby
enabling one or the other of the and gates. The waveform diagram is shown in Fig. 16.70.
We can see that two signals from the sawtooth generator are required to produce a complete
pulse cycle at the output. Its frequency must therefore be twice as high as that at which the
VSt
V1
VST
Vi
PWM on
V
V controller on
Fig. 16.69. Pulsewidth modulator for push–pull Fig. 16.70. Signal waveforms
converters
16.7 Primary Switching Regulators 925
Fig. 16.71. Line isolation using an opto-coupler for the analog regulator signal
V2
driver
Vi V2
Fig. 16.72. Line isolation using a pulse transformer Fig. 16.73. Pulse amplitude as a function
of the on-time
16.7.6
Loss Analysis
There are three types of losses that determine the efficiency of a switching regulator. The
static losses are due to the current consumption of the pulsewidth modulator and the drivers,
and are compounded by the on-state power losses of the power switches and the output
rectifier. These losses are independent of the switching frequency. The dynamic losses occur
926 16 Power Supplies
power
dissipation
total
dynamic
static
copper
as switching losses in the power switches and as magnetic losses in the HF transformer
and in the choke. They are approximately proportional to the switching frequency. The
copper losses in the HF transformer and in the choke result from the voltage drop across
the ohmic resistance of the windings. As fewer windings are required at higher frequencies,
as stated in (16.19), these losses are inversely proportional to the frequency as long as the
skin effect can be neglected. Figure 16.74 shows the three loss sources as a function of
frequency. An advisable operating range is between 20 kHz and 200 kHz. Although for
high-frequency operation the magnetic components are lighter and smaller, the dynamic
losses are so predominant in this range that the overall losses increase.
The overshoots that occur when the power switches are turned off are an additional
problem. They are due to the voltages across the leakage inductance of the HF transformer
and other circuit inductances. In order to minimize them, all the leads in the power circuit
must be kept as short as possible. Nevertheless, during rapid switching, high overshoots
may occur even if the leakage inductances are small. To give a numerical example:
I 1A
V = Lleakage = 100 nH = 100 V
t 100 ns
This voltage is added to the regular voltage of the power switch. In order to prevent damage
to the power switches, an additional snubber network is required, although this also causes
additional dynamic losses.
16.7.7
IC Drive Circuits
A number of commonly used switching regulators are listed in Fig. 16.75. The control
devices for push–pull converters have two outputs that switch alternately. However, these
can also be used in single-ended converters if one output is left uncommitted. If limiting
the on-time to 50% causes problems, it is also possible to OR the two outputs together.
16.7 Primary Switching Regulators 927
The integrated circuits for transformer-based switching regulators in Fig. 16.75 can be
divided in 3 groups:
• Flyback controllers with integrated high voltage power MOSFETs. They offer the sim-
plest way to build offline power supplies for low power applications below 50 W. They
are especially suitable to build wall power supplies.
• Flyback controllers for external power transistors. Here an arbitrary power transistor
can be added that is matched the voltage and current requirements.
• Push-pull controller for two power transistors that are alternately switched on. They
are adapted to half- and full-bridge regulators. Normally high voltage floating switch
drivers are here needed; some examples are given in Fig. 16.76.
The high power drivers in Fig. 16.76 are suitable for controlling power MOSFETs
that need high peak and low continuous gate currents. For half- and full-bridge regulators
the upper transistors are floating. They need floating gate drivers which sustain the full
intermediate link voltage. For this application drivers with internal high voltage MOSFETs
for signal transmission are the cheapest solution. However the floating switch must always
be positive with respect to the control input. When full line isolation is needed drivers
based on optocouplers or transformers must be applied.
In Fig. 16.76 some drivers for active rectifiers are also added. They are useful if high
power is needed at low voltages. Here in the rectifier the usual schottky diodes with on
voltages of 0.4–0.7 V can be replaced by power MOSFETs the on voltages of which can be
as low as 0.1–0.3 V if low on resistance types are used. In order to generate the appropriate
928 16 Power Supplies
Fig. 16.76. Examples for MOS-drivers. low = low side driver; high = high side driver; half = half
bridge driver = 1 low side + 1 high side driver
gate signal an active rectifier driver is needed that is synchronized to the timing of the
controller on the primary side.
Chapter 17:
Analog Switches and Sample-and-Hold Circuits
An analog switch is designed to switch a continuous input signal on and off. When the
switch is in the on-state, the output voltage must be as close to the input voltage as possible;
when the switch is off, it must be zero. The principal characteristics of an analog switch
are defined by the following parameters:
17.1
Principle
There are several switch arrangements that fulfill the above requirements. They are repre-
sented in Fig. 17.1 as mechanical switches.
A single-throw series switch is shown in Fig. 17.1a. As long as its contact is closed,
Vo = Vi . On opening the switch, the output voltage becomes zero, although this only
applies to no-load conditions. For capacitive loads, the output voltage will only fall to zero
slowly because of the finite output resistance ro = R.
The single-throw short-circuiting switch shown in Fig. 17.1b overcomes this difficulty.
However, in the on-state – that is, when the contact is open – the circuit possesses a finite
output resistance ro = R.
The double-throw series/short-circuiting switch in Fig. 17.1c combines both advan-
tages, and has a low output resistance in both states. The forward attenuation is low and
the reverse attenuation is high. However, the fact that the output is short-circuited in the
off-state may also cause problems – for example, if the output voltage is to be stored in a
capacitor, as in the sample-and-hold circuits of Sect. 17.4. In this case, switch S3 can be
added, as shown in Fig. 17.2. When the switch is open, the input signal that is capacitively
coupled via S3 is short-circuited by S2 ; however, the output remains at high impedance
due to S3 . This arrangement therefore behaves like the series switch in Fig. 17.1a, but has
a much better reverse attenuation for high frequencies.
Extending this principle to several inputs, we obtain the arrangement shown in Fig. 17.3.
One of the four switches is closed at any one time, which means that the output voltage is
Vi Vo Vi Vo Vi Vo
V1 S1
S1 S3
V2 S2 Vo
Vi Vo V3 S3
S2
V4 S4
equal to the particular input voltage selected. This arrangement is therefore also known as
an analog multiplexer.
By inverting the arrangement, an input voltage can be distributed to several outputs,
thus providing an analog demultiplexer function. The corresponding circuits for digital
signals have already been described in Sect. 8.2 on page 643.
17.2
Electronic Switches
Field effect transistors, diodes, or bipolar transistors are used to implement the switches.
They possess quite different characteristics, and specific advantages and disadvantages.
They do, however, have the same basic arrangement, which is shown in Fig. 17.4. In most
cases, TTL-compatible control signals are required. These are amplified by a power gate
followed by a level converter, which generates the voltages that are required for opening
or closing the switch.
17.2.1
FET Switch
As we saw in Sect. 3.1.3, a FET behaves like an ohmic resistor whose value can be varied
by several orders of magnitude using the gate–source voltage UGS . This behavior makes
it extremely useful as a series switch (see Fig. 17.5). The FET is turned off if we apply a
control voltage that is negative with respect to the input voltage by at least the threshold
VC off ≤ Vi + Vth .
To make the junction-FET conduct, the voltage VGS must he zero. This condition is
not so easy to fulfill, as the source potential is not constant. A solution to this problem is
shown in Fig. 17.6, where diode D becomes reverse biased if VC is made larger than the
most positive input voltage, and therefore VGS = 0, as required.
Vi Vi
Vo Vo
VC VC
Fig. 17.5. FET series switch Fig. 17.6. Simplified drive arrangement
VC on = Vi VC on = Vi max
VC off ≤ Vi + Vth VC off ≤ Vi + Vth
For sufficiently negative control voltages, diode D is forward biased and the FET is
turned off. In this mode, a current flows from the input voltage source via resistor R1 , into
the control circuit. This can usually be tolerated, as the output voltage in this case is zero.
However, this effect becomes troublesome if the input voltage is connected to the switch
via a coupling capacitor, as the latter becomes charged to a negative voltage during the
off phase.
These problems do not arise if a MOSFET is used for switching. An n-channel MOS-
FET can be made to conduct by applying a control voltage that is larger than the most
positive input voltage. No current flows from the gate to the channel, so that diode D and
resistor R1 are no longer necessary. To insure that the input voltage range is as large as
possible, it is better to use, instead of a single MOSFET, a CMOS switch consisting of two
complementary MOSFETs connected in parallel, as shown in Fig. 17.7.
To turn the switch on, V + is applied to the gate of n-channel MOSFET T1 , and that
of p-channel MOSFET T2 is connected to ground. Around the mid-range of voltage Vi ,
both MOSFETs are therefore conducting. If the input voltage increases to higher positive
values, VGS 1 is reduced, making T1 have a higher impedance. However, this has no adverse
effect, as the absolute value of UGS 2 increases simultaneously. This makes T2 go to low
impedance, and vice versa for small input voltages. This is illustrated in Fig. 17.8, in which
we see that the input voltage can assume any value between 0 and V + .
With standard CMOS switches, neither the control voltage nor the analog signals must
be outside this range, because this could result in the destruction of the switches due to
latch-up. In this case, the channel substrate diode becomes conducting and floods the
substrate with charge carriers that could fire the parasitic thyristor shown in Fig. 7.39 on
p-Channel n-Channel
Vi
Vo
VC
Vi
Fig. 17.7. Transmission gate Fig. 17.8. FET resistance versus input voltage for
VC on = V + VC = VC on = V + = 5 V
VC off = 0 V
932 17 Analog Switches and Sample-and-Hold Circuits
Fig. 17.9. Examples for analog switches in CMOS technology. Most other switch configurations
are also available with corresponding factory numbers
page 629, short-circuiting the supply voltage. If it cannot be guaranteed that the relevant
values will stay within the safe input voltage range, a resistor must be connected at the
input to limit the current.
Because of these problems, most integrated CMOS switches are provided with addi-
tional protection – that is, current-limiting – structures or are manufactured with dielectric
insulation. In this case, an oxide layer is used as the insulator to the substrate, instead of
a pn junction. Consequently, CMOS components with dielectric insulation are not subject
to latch-up effects, but their manufacturing process is considerably more expensive.
A number of commonly used CMOS switches and multiplexers are listed in Fig. 17.9
and 17.10. The 74 HC types are normal, extremely low-cost CMOS gates, but they are
prone to latch-up and have only a limited voltage range. The other types are protected
against this effect and can therefore be used without any difficulty. The manufacturers
listed also offer a wide range of other types, of which just a few examples are given.
17.2 Electronic Switches 933
Fig. 17.10. Examples of analog multiplexers in CMOS technology. Most other switch
configurations are also available with corresponding factory numbers
The typical reverse currents of the switches are between 0.1 nA and 1 nA at room
temperature. These values double for every 10 degree increase in temperature and can
therefore be as much as 100 nA. The switching times are between 100 ns and 300 ns.
17.2.2
Diode Switch
Diodes are also suitable for use as switches because of their low forward resistance and
high blocking resistance. If a positive control voltage is applied to the circuit shown in
Fig. 17.11, diodes D5 and D6 become reverse biased. The impressed current I then flows
through branches D1 , D4 and D2 , D3 , from one current source to the other. The potentials
V1 , and V2 thereby assume the values
V1 = Vi + VD , V2 = Vi − VD
Vo = V1 − VD = V2 + VD = Vi
if the forward voltages VD are the same. Should this not be the case, an offset voltage
occurs.
934 17 Analog Switches and Sample-and-Hold Circuits
VC
Vi Vo
–VC
Fig. 17.11. A series switch using diodes
If the control voltage is made negative, the two diodes D5 , D6 become forward biased,
and the diode bridge is turned off. This means that the output is doubly disconnected from
the input and the midpoint is at constant potential. The analog switch therefore has a high
reverse attenuation as shown in Fig. 17.2.
By employing this principle, switching times of less than 1 ns can be achieved if fast-
switching diodes are used. Suitable types include the Schottky diode quartet 5082-2813
from Hewlett-Packard.
Rapid switching naturally requires correspondingly fast drive signals. An example of a
suitable drive circuit is shown in Fig. 17.12. This consists of a bridge circuit made up of four
constant-current sources, T1 . . . T4 . The upper two are switched on alternately by the drive
signal. When T1 is on, a current of magnitude I flows through the diode bridge, causing
it to conduct. When T2 is on, the diode bridge is reverse biased. In order to insure that, in
this case, current sources T2 and T3 are not driven into saturation, the reverse voltages are
limited by transistors T5 and T6 . These also insure that the diode bridge driver is at low
impedance during reverse-biased operation. Good reverse attenuation is then achieved due
to the reduction of capacitive feedthrough.
.
Vi Vo
The amplitude of the analog signal must be smaller than the maximum control voltage
across the diode bridge. For the values specified, it must be limited to ± 2.7 V.
17.2.3
Bipolar Transistor Switch
To investigate the suitability of a bipolar junction transistor for use as a switch, we examine
its output characteristic curves around the origin, an expanded view of which is given in
Fig. 17.13 for small positive and negative collector–emitter voltages.
The first quadrant contains the familiar output characteristics shown in Fig. 2.3. If the
voltage VCE is made negative without changing the base current, the output characteristics
of the third quadrant are obtained. In this reverse-mode polarity, the current gain of the
1
transistor is considerably reduced and is about 30 β. The maximum permissible collector–
emitter voltage in this mode is the breakdown voltage VEB 0 , since the base–collector
junction is forward biased and the base–emitter junction is reverse biased. This type of
operation is known as reverse-region operation, and the accompanying current gain is
termed the reverse current gain ratio βr . The collector current is zero for a collector–
emitter voltage of about 10…50 mV. If the base current exceeds a few millamperes, this
offset voltage increases steeply; for small base currents, it remains constant over a large
range.
The offset voltage can be reduced considerably by insuring that the transistor is in
reverse-region operation when the output current crosses zero. To achieve this, the collector
and the emitter must be interchanged. The resulting output characteristics are shown in
Fig. 17.14. At larger output currents, virtually the same curves are obtained as for the
normal operation in Fig. 17.13, if VCE is still measured at the correct polarity (collector-
. .
. .
VCE,sat
VCE,sat
VCE VCE
VCE
VCE
Fig. 17.13. Complete output characteristics Fig. 17.14. Complete output characteristics
of a transistor in common-emitter for an interchanged emitter and collector,
connection, with the associated test circuit with the associated test circuit
936 17 Analog Switches and Sample-and-Hold Circuits
Vi Vi
Vo Vo
VC VC
to-emitter). The reason for this is that the emitter current, which is now the output current,
is very nearly the same as the collector current.
Near the origin, however, a major difference arises in that the base current can no
longer be neglected with respect to the output current. If, for normal operation, the output
current is made zero, the emitter current is identical to the base current – that is, is not
zero – and an offset voltage of 10 − 50 mV appears at the output. If the collector and the
emitter are interchanged and the output current is made zero, the collector current becomes
the base current. The collector–base junction is then forward biased (reverse operation).
The offset voltage for this mode of operation is usually about one-tenth of that in normal
operation, but it is also positive since, for the circuit shown in Fig. 17.14, Vo = −VCE .
Typical values for the offset voltage are between 1 and 5 mV, and it is therefore desirable
to operate transistor switches with an interchanged collector and emitter. If the emitter
current is kept small, the transistor operates almost exclusively in the reverse mode.
Short-Circuiting Switch
Figures 17.15 and 17.16 show how a transistor can be used as a short-circuiting switch. In
the circuit of Fig. 17.15, the transistor is operating in normal mode, whereas in Fig. 17.16
it is in reverse-region mode. To obtain a sufficiently low transistor resistance, the base
current must be in the milliampere range. The collector current in Fig. 17.15, and the
emitter current in Fig. 17.16, should not be much larger, in order to insure that the offset
voltage remains small.
Series Switch
A bipolar transistor used as a series switch is shown in Fig. 17.17. A negative control
voltage must be applied to turn off the transistor. It must be more negative than the most
negative value of the input voltage, but it also has a limit, since the control voltage may
not be more negative than −VEB 0 ≈ − 6 V.
To render the transistor conducting, a positive control voltage is applied which is
larger than the input voltage by a value of V = IB RB . The collector–base junction is
then forward biased and the transistor operates as a switch in reverse-region mode. The
disadvantage is that the base current flows into the input voltage source, and unless the
internal resistance of the source is kept very small, large errors may occur.
If this condition can be fulfilled, the circuit is particularly suitable for positive input
voltages, as the on-state emitter current is positive. The offset voltage is therefore reduced
and even becomes zero for a particular emitter current, as can be seen in Fig. 17.14. In this
mode of operation, the circuit is known as a saturated emitter follower, since for control
17.2 Electronic Switches 937
Vo
Vi
Vi
Vo Vi
VC Vo VC V
Vo
Vo
Vi VC
Fig. 17.17. Saturated emitter follower as a Fig. 17.18. Transfer characteristics for
series switch positive input voltages
Vi
VC Vo
voltages between zero and Vi , it operates as an emitter follower for VC . This is illustrated
in Fig. 17.18, by the transfer characteristics for positive input voltages.
Series/Short-Circuiting Switch
If the saturated emitter follower shown in Fig. 17.17 is combined with the short-circuiting
switch shown in Fig. 17.16, a series/short-circuiting switch is obtained, which has a low
offset voltage for both modes of operation. It has the disadvantage that complementary
control signals are required. The control arrangement is particularly simple if a comple-
mentary emitter follower is used, as shown in Fig. 17.19. It is saturated in both the on and
off states if VC max > Vi and VC min < 0. Due to the low output resistance, a fast switchover
of the output voltage, between zero and Vi , is possible. A practical implementation of this
was shown in the case of the function generator in Fig. 14.36 on page 862.
17.2.4
Differential Amplifier Switch
The gain of a differential amplifier is proportional to the transconductance, which is in turn
proportional to the collector current. Consequently, the differential gain can be made zero
by cutting off the emitter current. Figure 17.20 shows how this principle can be applied to
a differential amplifier used as an analog switch.
If the control voltage is made negative, diode D is turned off and the differential am-
plifier carries the emitter current Ik = I . If the output voltage is taken from the collectors,
we obtain:
Ic 1
Vo = gm Rc Vi = R c Vi = Ik R c V i
VT 2VT
If the control voltage is made positive, the diode takes over the current I and the
transistors are turned off; Ik = 0. Although this causes the two output potentials to increase
to V + , the output voltage difference Vo becomes zero.
938 17 Analog Switches and Sample-and-Hold Circuits
Vo
Vi
Fig. 17.20. Differential amplifier used as a
switch
VC 0 for VC = +1 V
Vo =
gm RC Vi for VC = − 1 V
Figure 17.21 shows how this principle can be employed to design an analog switch for
low frequencies. As long as the input voltage VD = 0, the control current IC is equally
divided between the two transistors of the differential amplifier, and current I flows in all
the current mirrors. The output current becomes zero. If a positive input voltage is applied,
the collector current of T2 increases by I = 21 gm VD and that of T1 decreases by the
same amount. The output current is therefore:
Ic IC
Io = 2I = gm VD = VD = VD
VT 2VT
If the control current IC is made zero, all the transistors are turned off and the output
current also becomes zero.
Amplifiers that operate on this principle are known as transconductance amplifiers.
They are available in IC form; for example, the CA 3060 or CA 3280 from Intersil. They
can also be used as operational amplifiers if the control current remains constant. If the
control current is made proportional to a second input voltage, they can also be used as
analog multipliers.
Figure 17.22 shows how the principle illustrated in Fig. 17.20 can be used to design a
switch for high frequencies. Here, the two differential amplifiers T1 , T2 and T3 , T4 employ
common-collector resistors R1 . However, only one pair is in operation at a time: when the
VD
= Io
C
C
Vo2
V1 V2 V3 V4 Vo1
VC
control voltage is positive, the differential amplifier T1 /T2 receives current I ; for negative
control voltages, current I flows in the pair T3 /T4 . This arrangement has an advantage over
that in Fig. 17.20 in that the output potentials remain constant during switching.
We therefore have a device that can be used to switch from one input voltage, Vi1 =
V1 − V2 , to another, Vi2 = V3 − V4 . If we make V3 = V2 and V4 = V1 by connecting the
relevant inputs, then Vi2 = −Vi1 and we have a polarity changer.
The circuit can be designed as a wideband amplifier, like the complementary cascode
differential amplifier shown in Fig. 5.25. The current feedback resistors RE and cascode
circuits T7 , T8 are used for this purpose. By selecting suitable component values, band-
widths of 100 MHz or more can he achieved. The circuit can therefore be used, for instance,
as a modulator, demodulator, or phase detector for telecommunications, and as a channel
(beam) chopper in wideband oscilloscopes.
Integrated circuits that employ this principle include the OPA 676 from Burr Brown
or the AD 539 from Analog Devices for high-bandwidth applications, or the AD 630 from
Analog Devices for high-precision circuits.
17.3
Analog Switches Using Amplifiers
If analog switches are combined with operational amplifiers, a number of special charac-
teristics can be obtained. In the following sections, the switches themselves will only be
shown symbolically. The CMOS types listed in Fig. 17.9 and 17.10 are the most suitable
for practical implementation.
940 17 Analog Switches and Sample-and-Hold Circuits
Vi Vi
S
Vo Vo
Fig. 17.23. Switching high voltages at low Fig. 17.24. Switching high voltages with
switch voltages high precision
0 0
Vo = Vo =
−Vi R2 /(R1 + rDS on ) −Vi R2 /R1
17.3.1
Analog Switches for High Voltages
In the circuit shown in Fig. 17.23, the operational amplifier operates as an inverting am-
plifier. When the switch is open, the voltage across it is limited by diodes D1 and D2 to
± 0.7 V. When the switch is closed, both terminals are at ground potential, as they are con-
nected to the summing point. In this case, the circuit operates as an inverting amplifier. The
diodes have no effect, as virtually no voltage is dropped across them. The circuit gain can
therefore be selected using R1 and R2 such that the operational amplifier is not overdriven
even at the highest input voltages.
Another method of switching large voltages is illustrated in Fig. 17.24. In the switch
position shown, the operational amplifier again operates as an inverting amplifier. The
advantage in this case is that the switch is inserted in the feedback loop and, as a result, its
on-state resistance has no effect on the gain. However, it is necessary to use switches whose
analog voltage range is identical to the maximum output voltage swing of the operational
amplifier.
When the switch is changed over, the output is connected via R2 to the summing point;
that is, to zero potential.
17.3.2
Amplifier with Switchable Gain
In the circuit shown in Fig. 17.25, the gain of a noninverting amplifier can be switched using
an analog multiplexer. Depending on which switch of the multiplexer is closed, any gain
factor A ≥ 1 can be realized by selecting suitable component values for the voltage divider
chain. The main advantage of this circuit is that the switches of the analog multiplexer can
be operated without current. This means that their on-state resistance does not affect the
output voltage. An IC amplifier that operates on this principle is the AD 526 from Analog
Devices. Its gain can be switched between values of 1 and 16.
With the circuit shown in Fig. 17.26, the sign of the gain can be reversed using switch
S. When the switch is in the lower position, the circuit operates as an inverting amplifier
that provides a gain of A = − 1.
When the switch is in the upper position, Vp = Vi . The output voltage therefore
assumes a value such that no voltage is dropped across R1 . This occurs when Vo = Vi .
The amplifier therefore operates as a noninverting amplifier. The circuit is then very similar
to the bipolar coefficient network shown in Fig. 11.5.
17.4 Sample-and-Hold Circuits 941
Vi Vo
S1
S2 Vi
S
Vo
17.4
Sample-and-Hold Circuits
17.4.1
Basic Principles
The output voltage of a sample-and-hold circuit should follow the input voltage when it is
in the on-state. In this mode, it therefore behaves like an analog switch. In the off-state,
however, the output voltage must not become zero, but the voltage at the instant of turn-off
must be stored. This is why sample-and-holds are also known as track-and-hold circuits.
The basic arrangement for a sample-and-hold circuit is shown in Fig. 17.27. The central
component is the capacitor C, which performs the storage function. When switch S is
closed, the capacitor is charged up to the input voltage. In order to insure that the input
voltage source is not loaded, an impedance converter is employed. This is implemented
by voltage follower OA1. It must be capable of delivering high output currents in order to
be able to charge and discharge the storage capacitor quickly.
When switch S is open, the voltage across capacitor C must be kept constant for as long
as possible. Consequently, a voltage follower is connected after it, to eliminate loading of
the capacitor. In addition, the switch must have a high off-state resistance and the capacitor
a low leakage current.
The main nonideal characteristics of a sample-and-hold circuit are given in Fig. 17.28.
When the switch is closed by the sample command, the output voltage does not instan-
taneously increase to the value of the input voltage, but only at a defined maximum slew
rate. It is primarily determined by the maximum current of the impedance converter OA1.
This is followed by a settling time, whose duration is determined by the damping due to
the impedance converter and the on-state resistance of the switch. The acquisition time
Vi
A Vo
A
.
Vi
Vo
track hold
Fig. 17.28. Specifications of a sample-and-hold circuit, showing typical values for an LF 398 with
1 nF capacitor. The duration of the tracking phase must he at least equal to the acquisition time
t Ac is defined as the time that elapses after the start of the track command until the output
voltage is equal to the input voltage within the specified tolerance. If the charging of the
storage capacitor is determined solely by the on-state resistance RS of the switch, the
acquisition time can be calculated from the charging function of an RC network and the
required accuracy of acquisition. Thus
4.6 for 1%
tAC = RS · C ·
6.9 for 0.1%
It is therefore shorter the smaller the value of C selected.
During transition to the hold state, it takes a while for the switch to open. This is known
as the aperture delay t Ap . It is not usually constant, but tends to vary, often as a function
of the particular value of the input voltage. These fluctuations are termed aperture jitter
tAp .
In general, the output voltage does not now remain at the stored value, but there is a
small voltage change Vo (the hold step) with subsequent settling. This is due to the fact
that, when the circuit is switched off, a small charge is coupled by the drive signal via the
switch capacitance CS into storage capacitor C. The resultant hold step is given by
Cs
Vo = VC
C
where VC is the amplitude of the drive signal. This effect is smaller the larger is the value
selected for C.
Another nonideal characteristic is the feedthrough. This results from the fact that the
input voltage has an effect on the output even though the switch is open. This effect is
mainly caused by the capacitive voltage divider formed by the capacitance of the open
switch with the storage capacitor.
The most important variable in the store condition is the droop (the hold decay). This
is mainly determined by the input current of the impedance converter at the output and the
reverse current of the switch. For a discharge current IL , we have
Vo IL
=
t C
In order to minimize the discharge current, an FET-input amplifier is used for OA2.
17.4 Sample-and-Hold Circuits 943
As we can see, all of the characteristics in the hold state are better the larger the value
selected for C, whereas during the tracking operation small values of C are desirable.
Consequently, a compromise has to be found, depending on the application.
We have hitherto assumed that the hold capacitor possesses ideal characteristics. It
is also possible to find capacitors that have virtually no leakage current. Nevertheless, a
voltage change can occur in the hold state due to charge storage in the dielectric. This effect
may be explained by reference to the equivalent circuit shown in Fig. 17.29. Capacitor C1
represents the charge stored in the dielectric. It initially remains unchanged in the event
of a hold step and only varies slowly. If the sampling time is short, the charge required for
this purpose is taken from capacitor C during the hold phase (dielectric absorption). In the
case of a hold step of magnitude V , this produces a subsequent voltage change of
C1
V = V
C
that is, 0.6% in the example shown in Fig. 17.29. The size of this effect depends on the
dielectric used. Teflon, polystyrene, and polypropylene are good in this respect; on the
other hand, polycarbonate, Mylar, and most ceramic dielectric materials are poor.
17.4.2
Practical Implementation
The fastest sample-and-hold circuits can be designed on the principle illustrated in
Fig. 17.27, if the diode bridge of Fig. 17.12 is used as a switch and the circuits described
in Figs. 4.111 and 4.112 are used as voltage followers.
A higher degree of accuracy can be achieved using an overall feedback arrangement,
as shown in Fig. 17.30. When the switch is closed, the output potential V1 of amplifier
OA1 assumes a value such that Vo = Vi . This eliminates offset errors due to OA2 or the
switch. Diodes D2 and D3 are nonconducting in this operating condition, as only a small
voltage V1 − Vo is dropped across them, which is precisely equal to the offset voltage.
If the switch is opened, the output voltage remains constant. Resistor R2 and diodes
D2 , D3 prevent amplifier OA1 from being overdriven in this operating condition. This is an
important consideration, because any overdriving is followed by a considerable recovery
time, which is added to the settling time.
A
Vi Vo
A
Vi
A A Vo
This is the principle employed by type LF 398 which, being inexpensive, represents
the most commonly used sample-and-hold circuit for general applications.
Fig. 17.32. Examples for integrated sample and hold circuits. The S&H circuits needed for
analog-digital converters are today normally included on the ADC chips (Sampling ADC).
Chapter 18:
Digital-Analog and Analog-Digital Converters
To display or process a voltage digitally, the analog signal must be translated into numeric
form. This task is performed by an analog-to-digital converter (A/D converter, or ADC).
The resultant number Z will generally be proportional to the input voltage Vi :
Z = Vi /VLSB
where VLSB is the voltage unit for the least significant bit; that is, the voltage for Z = 1.
To convert a number back into a voltage, a digital-to-analog converter (D/A converter,
or DAC) is used, whose output voltage is proportional to the numeric input; that is,
Vo = VLSB · Z
18.1
Sampling Theorem
A continuous input signal can be converted into a series of discrete values by using a
sample-and-hold circuit for sampling the signal at equidistant instants tµ = µTs , where
fs = 1/Ts is the sampling rate. It is obvious from Fig. 18.1 that a staircase function
arises, and that the approximation to the continuous input function is better the higher the
sampling rate. However, as circuit complexity increases markedly with higher sampling
rates, it is essential to keep the latter as low as possible. The question is now: What is the
lowest sampling rate at which the original signal can still be reconstructed error-free; that
is, without loss of information. This theoretical limit is defined by the sampling theorem
(the Nyquist criterion), which we shall now discuss.
In order to obtain a simpler mathematical description, the staircase function shown in
Fig. 18.1 is replaced by a series of Dirac impulse functions, as illustrated in Fig. 18.2:
∞
0
Ṽi (t) = Vi (tµ )Ts δ(t − tµ ) (18.1)
µ=0
Their impulse area Vi (tµ ) · Ts is represented by an arrow. The arrow must not be mistaken
for the height of the impulse, as a Dirac function is, by definition, an impulse with infinite
height but zero width, although its area has a finite value. This area is often misleadingly
Vi (t) Vi (t)
Vi (tµ)
Vi (tµ)
s s
Fig. 18.1. Example of an input signal Vi (t) Fig. 18.2. Representation of the input
and sampled values Vi (tµ ) signal by a Dirac impulse sequence
946 18 Digital-Analog and Analog-Digital Converters
Vi t
s
Vi t
known as the impulse amplitude. The characteristics of the impulse function are shown by
Fig. 18.3, where the Dirac impulse function is approximated by a rectangular pulse rε ; the
limit of the approximation is
It can be seen that this spectrum is a periodic function, the period being identical to the
sampling frequency fs . When this periodic function is Fourier analyzed, it can be shown
that the spectrum |X̃(jf )| is, for − 21 fs ≤ f ≤ 21 fs , identical to the spectrum |X̃(jf )| of
the original waveform. Thus it still contains all of the information, although only a few
values of the function were sampled.
There is only one restriction, and this is explained with the help of Fig. 18.4. The original
spectrum reappears unchanged only if the sampling rate is chosen such that consecutive
bands do not overlap. According to Fig. 18.4, this is the case for
s s s s s s
Fig. 18.4. Spectrum of the input voltage before (upper diagram) and after (lower diagram)
sampling
18.1 Sampling Theorem 947
Fig. 18.5. Overlapping of spectra if the Fig. 18.6. Aliasing due to an excessively low
sampling frequency is too low sampling frequency for f < ≈ fs
18.1.1
Practical Aspects
For a practical realization, the problem arises that a real system is unable to generate Dirac
impulse functions. The impulses must thus be approximated, as shown in Fig. 18.3, by a
finite amplitude and a finite time interval, thereby abandoning the limit concept of (18.2).
By inserting (18.2) into (18.1), we obtain, for finite ε, the approximated impulse sequence:
∞
0
Ṽi
(t) = Vi (tµ )rε (t − tµ ) (18.5)
µ=0
<
(jf ) = sin πεTs f · X(jf
X < ) (18.6)
πεTs f
948 18 Digital-Analog and Analog-Digital Converters
Weightfunction
s s s s
Fig. 18.7. Transition from the spectrum of a Dirac impulse sequence to the spectrum of the
staircase function by means of the weighting function | sin πf/fs )/(πf/fs )|
which is the same as for the Dirac impulse sequence, except for a superposed weighting
function that causes attenuation of the higher-frequency components. The case of the
staircase function is particularly interesting, as the pulse width εTs is here identical to the
sampling interval Ts . The spectrum is then given by:
sin(πf/fs )
X̃
(jf ) = · X̃(jf ) (18.7)
πf/fs
The magnitude of the weighting function is represented in Fig. 18.7, along with the sym-
bolic spectrum of the Dirac impulse functions. At half the sampling rate, an attenuation of
0.64 is obtained.
The example in Fig. 18.8 should serve to illustrate a possible approach for selecting the
sampling rate and the input or output filters. Consider an input spectrum for a music signal
in the range 0 ≤ f ≤ f max = 16 kHz, which is to be sampled and reconstructed with the
utmost fidelity. In this case, it is insignificant whether 16 kHz components actually occur
with full amplitude; rather, the linear frequency response should indicate that constant gain
is required in this range.
Even if it can be insured that no tones above 16 kHz are present, this does not automat-
ically mean that the spectrum at the sampler input is limited to 16 kHz. A typical source of
broadband interference is amplifier noise. For this reason, it is always advisable to provide
the input lowpass filter shown in Fig. 18.8. This is designed to limit the input spectrum
to half the sampling rate in order to prevent aliasing. Its cutoff frequency must be at least
f max in order to maintain the true input signal. On the other hand, it is desirable for it to
1
reject completely a frequency 2 fs and above that is only slightly higher, in order to allow
the lowest possible value to be used for the sampling rate, the point being that the circuit
complexity of the A/D or D/A converters and digital filter increases with the sampling
frequency. On the other hand, the complexity of the lowpass filter increases with greater
filter cutoff sharpness and stop-band attenuation. It is therefore always necessary to find a
compromise between the complexity of the lowpass filter on the one hand and that of the
converters and digital filter on the other. In the example with f max = 16 kHz, one could,
1
for example, select 2 fs = 22 kHz; in other words, using a sampling rate of fs = 44 kHz.
Sampling causes the band-limited input signal to be continued periodically to fs , as
1
shown in Fig. 18.8. Consequently, the baseband 0 ≤ f ≤ 2 fs must be extracted again
following D/A conversion. As a staircase function is obtained at the output of the D/A
converter, the (sin x)/x weighting as expressed in (18.7) must also be taken into account.
18.1 Sampling Theorem 949
s s
band-limited
signal Vi
weighting
sampling Vi
output lowpass
filter
Vo Vi
output
spectrum Vo
s s
Fig. 18.8. Reconstruction of the input spectrum in a digital system with AD-conversion and
following DA-conversion
The equalization required for this purpose can either be provided in the digital domain
or performed in the output lowpass filter. The latter possibility is illustrated in Fig. 18.8.
However, the main purpose of the output filter is to extract the baseband 0 ≤ f ≤ 21 fs
from the spectrum: at f max it must still exhibit full passband characteristics, whereas the
higher frequencies above 21 fs should attenuate completely. We can see that, in terms of
filter steepness, the same problems arise as with the input filter. Consequently, in order to
implement the filter, it is again necessary to provide an adequate margin between f max
and 21 fs .
The problems of implementing the input or output filter can be mitigated by employing
a markedly higher sampling rate – raising it by a factor of two or four, for example.Although
this oversampling naturally (see page 979) increases the complexity of the A/D and D/A
converters, the sampling rate can be reduced again to the value specified by the sampling
theorem by inserting a digital lowpass filter after the A/D converter. This avoids high
data rates for transmission or storage. Prior to D/A conversion, intermediate values can
950 18 Digital-Analog and Analog-Digital Converters
18.2
Resolution
When an analog signal is converted into a digital quantity that has a finite number of bits, a
systematic error is incurred due to the limited resolution; this is known as the quantization
error. If the number Z is reconverted to a voltage by a D/A converter, the quantization error
gives rise to superimposed noise. Figure 18.9 shows the test setup. According to Fig. 18.10,
it is ± 21 VLSB ; that is, it corresponds to half the input voltage step required to change the
least significant bit.
One can show that the noise voltage has the value
VLSB
Vn rms = √ (18.8)
12
For a full sinusoidal swing, the rms output signal voltage for an N -bit converter is deter-
mined by
1 1
Vs rms = √ · · 2N · VLSB
2 2
Hence, the signal-to-noise ratio is
Vs rms
SNR = 20 dB lg = N · 6 dB + 1.8 dB ≈ N · 6 dB (18.9)
Vn rms
Inversely you can measure the SNR and calculate the resolution:
SNR Vs,rms
N= or n = 2N − 1 =
6 db Vn,rms
Vi
V
V Vi
Vi
Vi –V
V
Vi
V
Fig. 18.9. Model for evaluating the Fig. 18.10. Quantization noise of an ideal A/D
quantization error converter
18.3 Principles of D/A Conversion 951
18.3
Principles of D/A Conversion
The purpose of a DAC is to convert a digital number into a proportional voltage. There are
basically three methods of conversion:
These three methods are shown schematically in Fig. 18.11. With the parallel method
(Fig. 18.11a), a voltage divider is used to provide all the possible levels of output voltage.
The switch to which the required output voltage level is assigned is then closed by a 1-of-n
decoder.
With the weighting method in Fig. 18.11b, a switch is assigned to each bit. The output
voltage is then added up via appropriately weighted resistors.
The counting method in Fig. 18.11c requires just a single switch that is opened and
closed periodically. Its duty cycle is set using a presettable counter, in such a way that the
arithmetic mean of the output voltage assumes the desired value.
A comparison of the three methods shows that the parallel method requires Z max
switches, the weighting method ld Z max switches and the counting method a single switch.
Due to the large number of switches, the parallel method is rarely used. Likewise, the
counting method is seldom used, its main disadvantage being that the output voltage can
only change slowly due to the lowpass filter required.
On the other hand, DACs employing the weighting method are widely used, and we
shall now describe the various ways in which they can be implemented. Two methods of
realizing the switches have become standard: CMOS circuits use the transmission gates
shown in Fig. 17.7; while in bipolar circuits, constant currents are generated and switched
using diodes or differential amplifiers, as in Fig. 17.20.
Vo
Vo
S
Vo
Vref
RFB=R
Vo
18.4
D/A Converters in CMOS Technology
18.4.1
Summation of Weighted Currents
A simple circuit for converting a straight binary number to a voltage that is proportional to
it is shown in Fig. 18.12. The resistors are selected such that, when the appropriate switch
is closed, a current flows through them, which is equivalent to the relevant binary weight.
The switches must therefore always be closed if a logical “1” appears in the relevant bit
position. Due to the operational-amplifier feedback via resistor RFB , the summing point
remains at zero potential. The current components are therefore added together without
affecting one another. If the switch controlled by z0 is closed, the output voltage is
RFB 1
Vo = VLSB = −Vref = − Vref
16 R 16
In general,
1 1 1 1
Vo = − Vref z3 − Vref z2 − Vref z1 − Vref z0
2 4 8 16
giving
1 Z
Vo = − Vref (8 z3 + 4 z2 + 2 z1 + z0 ) = −Vref (18.10)
16 Z max + 1
18.4.2
D/A Converters with Double-Throw Switches
A disadvantage of the above D/A converter is that the voltages across the switches depends
on their state. As long as the switches are open, they are at Vref potential; when closed,
they are at zero potential. As a result, the charges of the stray capacitances of the switch
must be reversed every time the switch is operated. This disadvantage can be avoided if
double-throw switches are used, as shown in Fig. 18.13, to connect the resistors either
to the summing point or to ground. The current through each resistor therefore remains
constant. This has a further advantage over the previous circuit in that the load of the
reference voltage source is constant and its internal resistance need not be zero. The input
18.4 D/A Converters in CMOS Technology 953
Vref
Vo
resistance of the network, and thus the load resistance of the reference voltage source, is
given by
16
Ri = 2R 4R 8R 16R = R
15
18.4.3
Ladder Network
When fabricating integrated D/A converters, the implementation of accurate resistances of
widely differing values is extremely difficult. The weighting of the bits is therefore often
effected by successive voltage division, using a ladder network as shown in Fig. 18.14
consisting of the series resistances Rs and the parallel resistances Rp . The basic element
of such a ladder network is the loaded voltage divider in Fig. 18.15, which is required to
have the following characteristics: if it is loaded by a resistor R2 , its input resistance R1
Vref Rp = 2 R R2 = 2R
Vo
Fig. 18.14. D/A converter with a ladder network. This is the commonly used CMOS circuit
Z
Vo = −Vref
Z max + 1
954 18 Digital-Analog and Analog-Digital Converters
must also assume the value R2 . At this load, the attenuation α = V2 /V1 along the ladder
element must have a predetermined value. With these two conditions, we obtain
(1 − α)2 (1 − α)
Rs = Rp and R2 = Rp (18.11)
α α
In the case of straight binary code, α = 0.5. We choose Rp = 2R, and obtain
Rs = R and R2 = 2R (18.12)
in accordance with Fig. 18.14.
The reference voltage source in Fig. 18.14 is loaded by the constant resistance
Ri = 2R || 2R = R
The output voltage of the summing amplifier is
Vo = −RFB Ik
RFB Z
= −Vref (8 z3 + 4 z2 + 2 z1 + z0 ) = −Vref (18.13)
16R Z max + 1
The D/A converter in Fig. 18.14 only requires resistors of size R, if the 2R resistors
are realized by two resistors connected in series. This arrangement is therefore ideally
suitable for fabrication in monolithic IC form. Although the required matching tolerances
for the resistors can be easily achieved, their absolute values cannot be precisely specified.
Consequently, tolerances up to ± 50% are common. Of course, the currents Ik or Ik
may
also deviate by correspondingly large amounts. In order to obtain tight output voltage
tolerances despite this tolerance, feedback resistor RFB is also integrated. This cancels out
the absolute value of R from (18.13) for the output voltage. For this reason, the internal
feedback resistor should always be used for current-to-voltage conversion, and never an
external one.
18.4.4
Inverse Operation of a Ladder Network
Sometimes the ladder network is also operated with exchanged input and output
(Fig. 18.16), as no summing amplifier is then required. However, one must then accept the
drawbacks, mentioned earlier, of a high voltage swing across the switches and a variable
loading of the reference voltage source.
To calculate the output voltage, we need to know the relationship between the applied
voltages Vi and the associated node voltages Vi
. For this purpose, we use the superposition
principle; in other words, we set all the injected voltages, apart from the voltage Vi in
question, equal to zero and add the individual components. If we terminate the network
on the right and on the left with resistance RL = R2 = 2R, we obtain, as required, a load
of R2 = 2R at each node to the right and the left. This gives us the voltage components
18.5 A Ladder Network for Decade Weighting 955
Fig. 18.16. Inversely operated ladder network. This circuit is used in converters with voltage
output without operational amplifier
RL Z RL Z
Vo = Vref · = Vref ·
R + RL Z max + 1 R + RL 16
Io
Vo open Vo load
Fig. 18.17. Equivalent circuit for calculating the
no-load voltage and short-circuit current
Vi
= 13 Vi , and by adding the correspondingly weighted components we obtain the
output voltage:
1 1 1 1 2Vref Z
Vo = V3 + V2 + V1 + V0 = · (18.14)
3 2 4 8 3 16
As the internal resistance of the network, irrespective of the set number Z, has a constant
value of
Ri = R2 Rp = (1 − α)Rp = R (18.15)
the weighting is retained even if the load resistance RL does not possess the initially
specified value R2 = 2R. From the equivalent circuit diagram shown in Fig. 18.17, we
can calculate the no-load voltage and the short-circuit current directly using (18.14):
Z Z Vref Z Vref Z
Vo open = Vref = Vref ; Io short = · = · (18.16)
16 Z max + 1 R 16 R Z max + 1
18.5
A Ladder Network for Decade Weighting
The ladder network in Fig. 18.14 can be extended to any length if longer straight-binary
numbers are to be converted. For the conversion of BCD numbers, the method is modified
somewhat, as shown in Fig. 18.18. Each decimal place (decade) is converted by a 4-bit D/A
converter, as shown in Fig. 18.13 or 18.14, and the individual converters are connected to a
ladder network. This introduces, from stage to stage, an attenuation of α = 10 1
. In (18.11),
resistance Rp must then be replaced by the input resistance Ri of the D/A converter stages,
so that the coupling resistors are Rs = 8.1Ri and the terminating resistor is R2 = 9Ri ,
as indicated in Fig. 18.18. In this manner, each input voltage for a D/A converter stage
956 18 Digital-Analog and Analog-Digital Converters
. i . i s . i 2 i
V
thousands hundreds tens units
Vo
is one-tenth of that for the previous stage. For the example of four decades, the output
voltage
Vref 1 1 1
Vo = − Z 3 + Z2 + Z1 + Z0
16 10 100 1000
is obtained if, for each decade, a ladder network as shown in Fig. 18.14 is used.
18.6
D/A Converters in Bipolar Technology
With D/A converters employing bipolar technology, it is easy to implement constant-
current sources that individually contribute to the total output current. The principle is
illustrated in Fig. 18.19. The currents are weighted according to the significance of the
associated bit position. Depending on whether the relevant binary digit is 1 or 0, the
associated current flows to the output or is diverted to ground. The busbar for current
Ik need not necessarily be at ground potential, as the current of the current sources is
not a function of the voltage. However, this only applies within the output voltage range
for constant-current operation (compliance voltage range). Consequently, an ohmic load
resistance can be used, which need not be connected to ground or to virtual ground.
Simple transistor current sources of the type shown in Fig. 4.18 are used to generate
the constant currents. If all of the base potentials are made equal and all of the emitter
resistors are connected to V − , the latter must be inversely proportional to the significance
of the associated binary digit. This causes tolerancing problems, even in the bipolar pro-
Vo
Vref
V1 V1 V1 V1 V1 V1
Fig. 18.20. Generation of weighted constant currents. This is the commonly used circuit
employing bipolar technology
V 2R
Iref = Rref = 8ILSB ; V1 = Iref · 2R = R Vref
ref ref
cess. Consequently, a ladder network is again used for current flow division (Fig. 18.20).
The current source bank T1 − T6 is at equal base potential, which is established via the
operational amplifier in such a way that current Iref = Vref /Rref flows via reference tran-
sistor T1 . This is the case when V1 = 2R · Iref . If the emitter–base voltages of the other
transistors are identical to that of T1 , we obtain the same voltage drops across the emitter
resistors and therefore the required weighting of the currents.
However, identical emitter–base voltages do not occur, even if the transistors are
completely identical, as the currents are not the same. From the transfer characteristic
in Eq. (2.2), we obtain
IC
VBE = VT ln
IC0
Consequently, the voltage increases by 18 mV if the collector current doubles. To avoid any
resultant error, all of the transistors are operated using the same collector current density.
For this purpose, a sufficient number of transistors are connected in parallel to insure that
only current ILSB flows through each one. In integrated circuits, this is taken into account
by using transistors with correspondingly larger areas for the higher currents.
The 2R termination of the ladder network shown in Fig. 18.20 must not be connected
here to ground, but a point must be selected that is at emitter potential. This is generated
by the otherwise unused transistor T6 . For simplicity, its emitter can also be connected in
parallel with T5 and the two emitter resistors combined to form a single resistor of value R.
Another method of D/A conversion using switched current sources is shown in
Fig. 18.21. Here, identical currents are generated, which appear at the output after be-
ing weighted by a ladder network. The arrangement corresponds to the inversely operated
ladder network shown in Fig. 18.16. The resistors 2R providing the attenuation within the
chain must be connected to ground, as they would have no effect connected in series with
the constant current sources. On the other hand, the attenuation in the chain is unchanged
by connecting a current source, as – theoretically at least – the latter has an infinitely high
internal resistance.
958 18 Digital-Analog and Analog-Digital Converters
Vo
Fig. 18.21. D/A converter with an inversely operated ladder network. This circuit is used in video
converters
IZ Z
Short-circuit current (RL = 0): Ik short = = 2I
8 Zmax + 1
Output voltage: Vo = Ik short (2R || RL )
18.7
D/A Converters for Special Applications
18.7.1
Processing Signed Numbers
When describing D/A converters, we have hitherto assumed that positive numbers are
involved which have to be converted into positive or negative voltages, depending on the
circuit. We shall now examine ways of producing bipolar output voltages using the D/A
converters described. The conventional representation of binary numbers of either sign is
in two’s-complement notation (see Sect. 8.1.3). In this way, the range from − 128 to +127
can be represented with 8 bits, as shown again in Fig. 18.22.
To enter data into the D/A converter, the number range is shifted to 0–255 by adding
128. Numbers above 128 are deemed to be positive, and those below negative. In this
case, the mid-scale number 128 denotes zero. This characterization of signed numbers by
purely positive numbers is known as offset binary representation. The addition of 128 can
be performed simply by negation of the sign bit (see Fig. 18.22).
Fig. 18.22. Processing negative numbers in D/A converters. VLSB = Vref /256
18.7 D/A Converters for Special Applications 959
V
D/A converter
Vo
V
V
A A
In order to obtain an output voltage of correct sign, the addition of the offset is achieved
by subtracting 128VLSB = 21 Vref . The summing operational amplifier OA 2 in Fig. 18.23
is used for this purpose, providing the output voltage:
1 Z + 128 1 Z
Vo = −V1 − Vref = Vref − Vref = Vref (18.17)
2 256 2 256
Its magnitude is listed together with voltage V1 in Fig. 18.22.
The zero stability of the circuit shown in Fig. 18.23 can be improved by using the com-
plementary output current Ik , instead of using the fixed reference voltage for subtraction
of the offset. In the case of two’s-complement number 0, which actually corresponds to
128 in offset binary, we have
Ik = 128ILSB and Ik
= 127ILSB
Therefore, if we add an ILSB to Ik and subtract the result from Ik , we obtain the correct
zero point. This method is illustrated in Fig. 18.24. As before, operational amplifier OA 1
converts current Ik into the output voltage. To eliminate errors, the amplifier is fed back
via DAC internal resistor RFB . Operational amplifier OA 2 inverts the sum of ILSB and
Ik , and adds this current into the summing point of OA 1. The absolute values of the two
V A
D/A converter
V A Vo
Fig. 18.24. Bipolar D/A converter with improved zero stability. By using both output currents the
output voltage is doubled
Z
Vo = Vref for − 128 ≤ Z ≤ 127
128
960 18 Digital-Analog and Analog-Digital Converters
resistors R1 are irrelevant, they must only have the same value. Current ILSB is added via
resistor R2 . If ILSB = Vref /(256R), it follows:
Vref
R2 = = 256R
ILSB
To calculate the output voltage, we only need to add the currents at the summing point
of OA 1 and obtain
Vref Z + 128 Vref 255 − (Z + 128) Vref 1 Z
Vo = R − − = Vref (18.18)
3 R 45256 6 3 R 45256 6 3 R 452566 128
Ik Ik
ILSB
18.7.2
Multiplying D/A Converters
As we have seen, D/A converters provide an output voltage that is proportional to the input
number Z and the reference voltage Vref ; in other words, they form the product Z · Vref .
For this reason, types that allow the reference voltage to be varied are also known as
multiplying D/A converters.
With circuits realized using bipolar technology, the reference voltage can only assume
positive values, as the current sources in Fig. 18.20 would otherwise be turned off. With
CMOS types, on the other hand, positive and negative reference voltages are permissi-
ble. If circuits such as those shown in Figs. 18.23 and 18.24, which allow positive and
negative numbers to be converted with their correct signs, are used, even four-quadrant
multiplication is possible.
18.7.3
Dividing D/A Converters
A D/A converter can also be operated in such a way that it divides by the input number.
To achieve this, it is inserted in the feedback loop of an operational amplifier, as shown
in Fig. 18.25. This means that the reference voltage Vref is set such that Ik = −Vi /RFB .
Using the converter equation
Vref Z
Ik = ·
R Z max + 1
we obtain the output voltage:
Z max + 1 R Z max + 1 Z max + 1
Vo = Vref = Ik R = −Vi · · = −Vi · (18.19)
Z RFB Z Z
Multiplying
D/A converter Vo
This simple method of performing division frequently obviates the need for analog or
digital division, with its associated cost and complexity, if some accuracy is required.
18.7.4
D/A Converter as Function Generator
The output voltage Vo of the usual D/A converter is proportional to the applied number
Z according to Vo = aZ. If, instead of the proportional function, any other relationship
Vo = f (Z) is to be realized, the function X = f (Z) must be generated by a digital
function network and then applied to a D/A converter.
If no stringent requirements are imposed on the accuracy, there is a much simpler
solution: the binary number Z is used to control an analog multiplexer. We apply analog
input values, each of which is assigned to the appropriate binary number. For each analog
value, a separate switch is needed, and the attainable resolution is therefore limited to
about 16 steps.
A possible implementation is shown in Fig. 18.26. In contrast to the usual D/A con-
verters, only one of the switches S0 , to S7 , is closed at any time. The values of the output
voltage function are thus given by the expression
⎧ RN
⎪
⎪+Vref for Z = 0 . . . 3
⎪
⎨ RZ
Vo (Z) =
⎪
⎪
⎪
⎩−Vref RN for Z = 4 . . . 7
RZ
An important application of this principle is the digital generation of sine waves (e.g., in
modems). A simple and widely used method of generating signals of different frequencies,
all synchronized to a common time base, is to employ frequency division. However, a
serious drawback for use in analog systems is that the signals obtained are square waves.
Sine waves can he produced by filtering the fundamental with a lowpass or bandpass filter,
but these filters must always be tuned to the appropriate frequency.
The D/A converter described avoids these problems in that it allows the frequency-
independent generation of sine waves. According to Fig. 18.27, we require a digital input
signal representing a rising and falling sequence of equidistant numbers. This input signal
corresponds to the triangular wave-shape for sine-wave generation by an analog function
network, as described in Sect. 11.7.4 on page 745 and Sect. 14.5 on page 859.
Vref
Vo
–Vref
fi
A
B Analog-Multiplex
for example CD 4051
Vo
C
Since in a complete period each step occurs twice, the sine wave is approximated by a
total of 16 steps. Correspondingly, the input frequency fi of the counter must be 16 times
that of the sine wave.
The staircase in the output signal in Fig. 18.27 contains some harmonics. To get a poor
sinus a lowpass filter can be added. With variable frequencies a switched capacitor filter
is advantageous (Sect. 13.12 on page 836) because its cutoff frequency follows the clock
frequency.
18.8
Accuracy of DA Converters
18.8.1
Static Errors
The zero point error of a DA converter is determined by the leakage currents flowing
through the open switches.
The full-scale error is determined by the on-state resistances of the switches and the
accuracy of the feedback resistor RFB . Both errors can be largely eliminated by trimming.
Nonlinearity, on the other hand, cannot be eliminated by adjustment. It is defined as
the amount by which a step is larger or smaller than 1 LSB, under worst-case conditions.
Figure 18.30 illustrates a nonlinearity of ± 21 LSB. The critical case occurs at mid-scale:
if only the most significant bit is a 1, the current flows via a single switch. If the number
is reduced by 1, the total current of all the lower-order switches is reduced by only ILSB .
If the linearity error is greater than 1 LSB, the trend is reversed. The output voltage
then falls at the mid-scale point although the number is increased by 1. A serious error of
this kind is termed a monotony error, an example of which is shown in Fig. 18.31. Most
D/A converters are designed such that their nonlinearity does not exceed ± 21 LSB, as the
least significant bit would otherwise be meaningless.
964 18 Digital-Analog and Analog-Digital Converters
V V
VLSB VLSB
Fig. 18.30. D/A converter with a Fig. 18.31. D/A converter with a
nonlinearity of ± 21 LSB nonlinearity of 1 21 LSB and an associated
monotony error
18.8.2
Dynamic Characteristics
The settling time is defined as the time that it takes for the output signal to reach the steady-
state value with an accuracy of ± 21 LSB after the number Z has changed from 0 to Z max .
Only then is the analog signal available with the accuracy provided by the resolution of
the D/A converter. Defining the settling time with reference to ± 21 LSB means, of course,
that D/A converters with the same time constant but higher resolution settle more slowly
than those with lower resolution.
With many D/A converters, a current is initially formed which can be converted into
a voltage, as required, in a following operational amplifier. In this case, the settling time
of the operational amplifier, which is usually much greater than that of the D/A converter,
is also added. In order to achieve short settling times for the voltage it is advantageous
to use converters that need no operational amplifier. For ladder networks, the only option
is to use the inversely operated ladder network shown in Fig. 18.16. Types with current
sources like in Fig. 18.21 can all generate a voltage across an ohmic load resistance. In
order to achieve bandwidths in the 100 MHz range, it is advisable to use D/A converters
whose output currents are so large that they can produce the required amplitudes across
load resistors of 50 or 75 .
Unwanted interference pulses (glitches) may also occur at the transition from one
input number to the other. In most cases, these are due only to a small extent to capacitive
feedthrough of the binary drive signals to the output. Large glitches occur if the switches in
V
VLSB
Fig. 18.33. Examples for low-frequency DACs with voltage outputs. Interface: SPI = serial
peripheral interface. I2 C = I2 C-bus interface (serial too). Output: RRO = rail to rail output
the D/A converter do not change state simultaneously. The most critical point again occurs
at mid-scale: if the most significant bit (MSB) is a 1, the current flows via one switch only.
If the number is reduced by 1, the switch for the MSB opens and all of the others close. If
the MSB switch opens before the other switches have closed, the output signal briefly goes
to zero. However, if the MSB switch opens slightly late, the output signal momentarily
assumes its full-scale value. In this way, unwanted pulses with an amplitude equal to half
the range may occur. An example of the switches closing more rapidly than they open is
shown in Fig. 18.32.
As the glitches are short pulses, they can be reduced using a lowpass filter at the output.
As a result of this, however, they become correspondingly longer, the voltage–time area –
in other words, the glitch energy – remaining constant.
966 18 Digital-Analog and Analog-Digital Converters
18.9
Principles of A/D Conversion
The purpose of an A/D converter (ADC) is to transform an analog input voltage into a
proportional digital number. There are three basically different conversion methods that
we have distinguished at the D/A converters already:
the parallel method (a word at a time),
the weighting method (a digit at a time),
the counting method (a level at a time).
The parallel (flash) method compares the input voltage with n reference voltages si-
multaneously and determines between which two reference levels the value of the input
voltage lies. The resulting number is thus obtained in a single operation. However, the
circuitry involved is very extensive, as a separate comparator is required for each possible
number. For a range of measurement from 0 to 255, n = 255 comparators are needed.
18.10 Design of A/D Converters 967
With the weighting (successive approximation) method, the end result is not obtained
in a single operation; instead, one bit of the corresponding straight-binary number is de-
termined at a time. The input voltage is first checked against the most significant bit and
the most significant bit is determined. Then the second bit is evaluated: This process is
repeated until all bits are processed. For a result with 8 bit accuracy 8 steps are required.
The simplest method is the counting method. It involves counting how often the ref-
erence voltage of the least significant bit must be added to arrive at the input voltage. The
number of operations is the required result. If the maximum number to be represented is
256, a maximum of 256 operational steps must be performed to obtain the result.
18.10
Design of A/D Converters
18.10.1
Parallel Converter
The construction of a parallel converter for 3 bits is illustrated in Fig. 18.35. With three bits,
eight different numbers including zero can be represented, for which seven comparators
Vi
V
V
Priority decoder
are required. The seven associated equally spaced reference voltages can be generated by
means of a voltage divider from a single reference voltage source.
For an input voltage that has a value 3 VLSB , the comparators 1–3 produce ones and the
comparators 4–7 zeros. A logic circuit is therefore required to convert these comparator
states to the number Z = 3. The comparator output states and the corresponding straight-
binary numbers are listed in Fig. 18.36. A comparison with Fig. 8.18 on page 646 shows that
the required conversion can be carried out by the priority decoder described in Sect. 8.3.
However, the priority decoder must not be connected directly to the outputs of the
comparators, since totally erroneous straight-binary numbers may arise if the input voltage
is not constant. The example of a change from 3 to 4 – that is, in straight-binary code from
011 to 100 – illustrates this. If the most significant bit changes before the two other bits
because of a shorter propagation delay in the priority decoder, the number 111 (i.e., 7dec
occurs temporarily. This is equivalent to an error of half the range. The result of an A/D
conversion is usually transferred to a memory, and there is therefore a certain probability
that this erroneous number will be stored. The use of a sample-and-hold circuit can prevent
this effect, as it holds the input voltage constant during the conversion process. However,
high speed sample and holds are expensive. Therefore a digital sample and hold is usual
in parallel converters.
The D flip-flops at the output of each comparator in Fig. 18.35 are used for this purpose,
thus insuring that the priority decoder receives constant input signals for one complete clock
period. Steady-state data is therefore available at the priority decoder output prior to the
arrival of the next clock edge.
The sampling instant is essentially determined by the trigger edge of the clock signal,
although the actual sampling is performed a little earlier because of the comparator delay.
The delay differences is therefore determine as the aperture jitter. In order to achieve the
low values the signal delay from the analog input to the storage devices should be as low as
possible. On most types, therefore, the storage element is incorporated in the comparator
and inserted directly after the analog input. The resulting input circuit of a comparator of
this type is shown in Fig. 18.37.
If switch S is set to the left, transistors T1 , T2 operate as a comparator. When the
switch is changed over, comparator T1 , T2 is deactivated and flip-flop T3 , T4 is activated
instead. The flip-flop then stores the state of the comparator. For this purpose, it is not
even necessary for the comparator to have already changed state completely. Since the
flip-flop is likewise designed as a differential amplifier, differences of a few millivolts
decide whether the flip-flop assumes one state or the other. In this way, the aperture jitter
can be reduced to picoseconds.
Figure 18.38 lists a number ofA/D converters that employ flash converter. Today mainly
8 bit converters are offered. If higher accuracy is needed additional bits can be generated
by averaging. For one additional bit 4 samples must be averaged. This technique is used
in high speed oscilloscopes to get up to 11 bits from an 8 bit quantizer if the maximum
sample rate is not needed.
The linearity of A/D converters at low signal frequencies is equal to a resolution of
± 21 LSB, or even ± 41 LSB in some cases. However, at high signal frequencies, nonlin-
earity increases, with the result that the least significant, or even the two least significant
bits, becomes unusable. The quantizing noise increases correspondingly by 6 or 12 dB, in
accordance with (18.9).
18.10.2
Two Step Converters
A disadvantage of flash converters is that the number of comparators required rises
exponentially with the word length. For a 10-bit converter, for example, a total of
210 − 1 = 1, 023 comparators are needed. This number can be considerably reduced
if at first the five most significant bits are parallel-converted, as can be seen in the block
diagram of Fig. 18.39. The result is the coarsely quantized value of the input voltage. A
D/A converter is used to produce the appropriate analog voltage, which is then subtracted
from the input voltage. The remainder is digitized by a second 5-bit A/D converter.
970 18 Digital-Analog and Analog-Digital Converters
Vi
Fig. 18.39. Two step ADC. It
requires only 2 × 32 = 64
comparators for a 10 bit converter
9 ... 4 ... instead of 1023
If the difference between the coarse value and the input voltage is amplified by a factor
of 32, two A/D converters with the same input voltage range can be employed. There are,
however, different requirements for the accuracy of the two converters: the accuracy of the
first 5-bit converter must be as high as that of a 10-bit converter; otherwise, the calculated
difference is meaningless.
However, parallel A/D converters that provide such a high degree of linearity are not
obtainable; nor can they be implemented for higher signal frequencies. Consequently, the
difference signal exceeds the fine range and overdrives the second A/D converter, resulting
in serious errors in the output signal (missing codes).
This problem can be overcome by reducing the gain for the difference signal from
32 to 16, as shown in Fig. 18.40. Bit z5 is then formed both by the coarse and the fine
quantizer. If the fine signal now exceeds the range due to errors of the coarse quantizer,
the coarse value can be increased or lowered by 1 using z5
. This allows coarse quantizer
linearity errors to be corrected to within ± 21 LSB. The coarse quantizer linearity need not
be better than the resolution, a requirement which is not as stringent as that for the circuit
in Fig. 18.40. It is merely necessary for the D/A converter to have full 10-bit accuracy. For
successful error correction, the coarse and fine quantizer ranges must overlap by at least
one bit. So as not to reduce the resolution of the circuit as a whole, the fine quantizer must
therefore have one additional bit.
The coarse and fine values must, of course, be formed by the same input voltage Vi (tj )
in each case. However, due to the transit through the first stage, a time delay is introduced.
Consequently, with this method the input voltage must be kept constant using an analog
sample-and-hold circuit until the complete number is formed. This constitutes a serious
disadvantage as compared with the purely parallel method.
The method can be extended to multi-step converters. In Fig. 18.41 a 3-step converter
is presented. Here in each step 5 bit converters are used. In order to get a range overlap for
error correction the intermediate amplifiers have a gain of only 16 as in Fig. 18.40.
Here in each step a sample and hold circuit is added. The consequence is that the most
significant bits of the first input sample Vi1 will be quantized in step 1 after the first sample.
With the second sample the remainder stored in the second S&H for step 2 that forms the in-
termediate bits. At the same time step 1 quantizes the most significant bits of the next input
Vi
9 ... z ’5 4 ...
Vi
sample Vi2 . With the third sample the remainder of Vi1 stored in the third S&H for step 3
that forms the least bits.At the same time step 1 quantizes the most significant bits of Vi3 and
step 2 quantizes the intermediate bits if Vi2 . The process is illustrated in Fig. 18.42. The sam-
pling frequency of the converter is only determined by the conversion time of one step. But
it needs 3 samples until the first input voltage Vi1 is quantized and available at the output.
Then with each sample a further conversion is available. This latency is typical for pipelined
circuits. Without the intermediate S&H circuits the maximum sampling frequency would be
only 1/3 because all signals up to the last step must settle before a new sample can be taken.
The two-step converter was introduced to reduce the number of comparators in flash
converters. The ultimate reduction of comparators is achieved if only 1 bit is used in
one stage because then only one comparator is needed per stage. Then the number of
comparators is equal to the number of bits N. The first conversion takes N samples,
all following only one. Some examples for commercially available chips are given in
Fig. 18.43.
Fig. 18.43. Examples for pipeline AD-converters. All types have an on-chip reference
972 18 Digital-Analog and Analog-Digital Converters
Fig. 18.43. (cont.) Examples for pipeline AD-converters. All types have an on-chip reference
18.10.3
Successive Approximation
The basic design of an A/D converter employing successive approximation is shown in
Fig. 18.44. The comparator compares the sampled input value with the output voltage
of the D/A converter. When measurement starts, the number Z is set to zero. The most
significant bit (MSB) is then set to 1 and the comparator checks to ascertain whether the
sample- successive
and-hold approximation
Vi
register
V D/A
converter
yes no
step 1
100
yes no yes no
step 2
100
Fig. 18.45. Flowchart for the weighing sequence. The thick line represents the example.
input voltage is greater than V (Z). If so, the bit remains set. If not, it is reset again. Thus
the MSB is “weighed.” The process is then repeated for each additional bit until, finally,
the least significant bit (LSB) is established. In this way, a number is produced in the
register. This number is converted in the DAC into a voltage corresponding to Vi within
the resolution VLSB , giving
Z ! Vi
V (Z) = Vref = Vi that is, Z = (Z max + 1) (18.20)
Z max + 1 Vref
If the input voltage changes during the conversion time, a sample-and-hold circuit is
required to buffer the sampled values, so that all the digits are formed from the same input
voltage value Vi (tj ). If no sample-and-hold is present, an error may occur that is equal to
the input voltage change during the conversion period.
The flowchart for the first three weighing steps is shown in Fig. 18.45. It can be seen
that, in each step, a decision is made as to whether the relevant bit is 1 or 0. The previously
determined bits remain unchanged.
The timing diagram for the weighing process is shown in Fig. 18.46a for the voltage
V (Z), and in Fig. 18.46b for the number Z. Each bit is set on a trial-and-error basis. If, as
V (Z)
V (Z max )
Vi
Fig. 18.46. Timing diagram for a successive approximation converter according to the path in
Fig. 18.45 for a 3 bit converter
974 18 Digital-Analog and Analog-Digital Converters
a result, the input voltage is exceeded, the bit is reset again. In this example, conversion is
therefore complete after 3 weighing steps.
The conversion is controlled by the successive approximation register (SAR). The basic
mode of operation will now be discussed, with reference to Fig. 18.47. At the start of the
conversion, the reset signal R is used to clear all of the flip-flops. In the shift register F7
to
F0
a single 1 is shifted one position to the right on each clock pulse, causing bits z7 to z0
to be set in turn on a trial-and-error basis. The particular weighing result is stored in latch
flip-flops F7 − F0 by reading the relevant comparator state D. Only the flip-flop whose
associated bit is currently being tested is enabled via the C-input at any one time.
When the least significant bit z0 has also been established, the last flip-flop F of the
shift register is set. This indicates end of conversion eoc. Because of the or gate at the
D-input, it retains this state even if further clock pulses are applied. It, together with the
result, is not cleared until the next conversion begins.
The truth table for the successive approximation register is shown in Fig. 18.48. As we
can see, all of the outputs are cleared with the reset signal. At each step the decision D
of the comparator is stored in the relevant position and the next less significant bit is
weighed. The truth table illustrates the operation of the shift register. After eight steps, “1”
has arrived at the end of conversion output and the conversion is done. The result Z is then
R D z7 z6 z5 z4 z3 z2 z1 z0 eoc
0 1 0 0 0 0 0 0 0 0 0 0
1 0 D7 1 0 0 0 0 0 0 0 0
2 0 D7 D7 1 0 0 0 0 0 0 0
3 0 D6 D7 D6 1 0 0 0 0 0 0
4 0 D5 D7 D6 D5 1 0 0 0 0 0
5 0 D4 D7 D6 D5 D4 1 0 0 0 0
6 0 D3 D7 D6 D5 D4 D3 1 0 0 0
7 0 D2 D7 D6 D5 D4 D3 D2 1 0 0
8 0 D1 D7 D6 D5 D4 D3 D2 D1 1 0
9 0 D0 D7 D6 D5 D4 D3 D2 D1 D0 1
Fig. 18.48. Truth table for the successive approximation register
18.10 Design of A/D Converters 975
available in parallel form. However, it can also be obtained in serial form at the comparator
output.
Some examples for successive approximation converters are given in Fig. 18.49.
18.10.4
Counting Method
A/D conversion using the counting method requires the least circuit complexity, but the
conversion time is considerably longer than with the other methods – generally between 1
ms and 1 s. However, this is adequate for slowly changing signals, such as those involved
in temperature measurement, and it is also fast enough for digital voltmeters, as there is a
limit to how quickly the result can be read off. The counting method can be implemented in
various ways. The principal techniques are discussed below, the most important being the
dual-slope and method, as it allows maximum accuracy to be achieved with minimum
circuit complexity.
976 18 Digital-Analog and Analog-Digital Converters
up-down
Vi counter
V
DAC
Compensation Converters
The compensating A/D converter shown in Fig. 18.50 is closely related to the successive
approximation (SA) type shown in Fig. 18.44. The basic difference is that an up–down
counter is used instead of the SA register.
The comparator compares the input voltage Vi , with the compensating voltage V (Z).
If the difference is positive, it causes the counter to count upward, and vice versa. This
means that the compensating voltage rises or falls until it has reached the level of the input
voltage, and then follows the latter as it changes. For this reason, the circuit is also known
as a tracking A/D converter.
One drawback with the simple circuit in Fig. 18.50 is that, as the clock is never switched
off, the counter never stops, but always oscillates by 1 LSB around the input voltage. If this
causes problems, the simple comparator can be expanded to form a window comparator.
This allows the clock to be inhibited when the compensating voltage V (Z) is within
± 21 VLSB of the input voltage Vi .
The significant reduction in control logic as compared to the weighting method is
achieved at the expense of considerably longer conversion times, as the compensating
voltage only changes in steps of size VLSB . However, if the input voltage only changes
slowly, a short settling time can again be achieved, as approximation is performed contin-
uously because of the tracking characteristic, rather than always starting at zero as with
the weighting method.
Single-Slope Converter
The single slope A/D converter shown in Fig. 18.51 does not require a DAC. The principle
here is that the input voltage is initially converted into a proportional time interval using
the sawtooth generator in conjunction with the window comparator K1 , K2 , and G1 .
The sawtooth voltage can be raised from negative to positive values in accordance with
the relation
Vref
VS = t − V0
τ
Logic y = 1 is present at the equivalence gate G1 only for as long as the sawtooth voltage
is between the limits 0 and Vi . The corresponding time interval is t = τ Vi /Vref . This is
measured by counting the oscillations of the crystal-controlled oscillator. If the counter is
zeroed at the start of the measurement, the count after the upper comparator threshold has
been exceeded is
t Vi
Z = = τf (18.21)
T Vref
18.10 Design of A/D Converters 977
Vi
counter
sawtooth crystal
generator oscillator
V
If a negative measurement voltage is applied, the sawtooth voltage will first cross the
measurement voltage and then pass through zero. This sequence therefore enables the sign
of the measured voltage to be determined. The measuring time is the same; it is purely a
function of the magnitude of the measurement voltage. After each conversion the counter
must be reset to zero and the sawtooth voltage adjusted to its negative initial value. In order
to retain the result, the old count is normally stored until a new one is available.
As (18.21) indicates, the tolerance of time constant τ directly affects the measuring
accuracy. Since it is determined by an RC network, it is subject to the temperature and
long-term drift of the capacitor. Consequently, an accuracy of better than 1% is difficult to
achieve.
Dual-Slope Converter
With this method, not only the reference voltage but also the input voltage is integrated.
In the inactive state, switches S1 and S2 in Fig. 18.52 are open, while switch S3 is closed.
As a result, the integrator output voltage is zero.
At the start of the conversion cycle, the counter is cleared, switch S3 is opened, and S1
is closed, causing the input voltage VI to be integrated. If it is positive, the integrator output
becomes negative, as shown in Fig. 18.53, and the comparator enables the clock generator.
The end of the cycle is reached when the counter overflows after Z max +1 clock pulses and
is thus again at zero. Then the reference voltage is integrated by opening S1 and closing S2 .
switch
Vi controller
A V
V counter
As it is negative the integrator voltage now rises. The second integration interval is finished
if VI has reached zero. Then the comparator goes to zero and the counter is stopped. The
counter reading equals the number of clock pulses during the time t2 , therefore the reading
is proportional to the input voltage.
The relation between the input voltage VI and the result Z can be calculated if one
keeps in mind that the integration starts at and ends at VI = 0. From
t1 t2
1 1 !
VI = − Vi dt − Vref dt = 0 (18.22)
RC 0 RC 0
follows, if Vi is taken to be constant
1 1
− Vi t1 − Vref t2 = 0
RC RC
With
t1 = (Z max + 1)T and t2 = ZT (18.23)
results
1 1
− V1 (Z max + 1)T − Vref ZT = 0
RC RC
One recognizes that the time constant RC and the clock period T can be reduced from the
equation:
Vi (Zmax + 1) + Vref Z = 0
Solving for Z renders the result:
Vi
Z = (Z max + 1) (18.24)
Vref
This equation reveals the salient feature of the dual-slope method; namely, that neither
the clock frequency 1/T nor the integration time constant τ = RC have any effect on
the result. The only requirement is that the clock frequency must be constant during the
time t1 + t2 . This short-term stability can be achieved using simple clock generators. For
these reasons, accuracies of 0.01% = 100 ppm can be achieved by this method with cheap
components.
As we have seen, it is not the instantaneous value of the measurement voltage that
affects the result, but its average value over measuring time t1 . Consequently, alternating
voltages are attenuated more heavily the higher their frequency. Alternating voltages whose
18.10 Design of A/D Converters 979
frequencies are equal to an integral multiple of 1/t1 are rejected completely. It is therefore
advisable to adjust the clock generator frequency such that t1 is equal to the period of the
AC supply voltage or a multiple thereof. This will eliminate any hum.
As the dual-slope method allows a high degree of accuracy and noise rejection to be
achieved with minimum circuit complexity, it is preferable to use this type of converter in
digital voltmeters. In this application, the relatively long conversion times do not present
any problem.
The counter in Fig. 18.52 need not be a straight-binary counter. The mode of operation
is identical if a BCD counter is employed. This possibility is utilized in digital voltmeters,
as no binary-to-decimal conversion of the measured value is then required.
18.10.5
Oversampling
In any AD converter the effective resolution can be raised by using a higher sampling
frequency than needed and a following digital lowpass filter that limits the bandwidth to the
used range as shown in Fig. 18.54. By oversampling the quantizing noise Vn2 = VLSB 2 /12
is distributed over a larger frequency range so that the quantizing noise density
2
VLSB 2
VLSB
Vn
2 = = (18.25)
12 · OSR · (fs /2) 6 · fOSR
is lowered as shown in Fig. 18.55. If the frequency range is subsequently limited to the re-
1
quired signal bandwidths fc = 2 fS the quantizing noise (double hatched area) is reduced.
But this method is not very efficient because a fourfold oversampling is needed to halve
the quantizing noise Vn
corresponding to additional 1 bit of resolution as can be seen in
Fig. 18.57.
Vi
c S S c S S S c S S S
Fig. 18.55. Effect of oversampling on quantizing noise density. Hatched area: quantizing noise that
is the same in all three cases. Double hatched area: noise within signal bandwidth. fc = signal
bandwidth, fS = sampling frequency, fOSR = oversampling frequency
980 18 Digital-Analog and Analog-Digital Converters
Delta-Sigma Converter
The advantage of a delta-sigma converter is that the modulator in Fig. 18.56 not only
performs oversampling but also noise shaping. This is shown in Fig. 18.55 also. Here the
quantizing noise is not equally distributed but shifted to higher frequencies where it is
effectively removed by the following digital lowpass filter.
The profit of additional bits with first order noise shaping is shown in Fig. 18.57 in
comparison to oversampling without noise shaping. At fourfold oversampling only 1 bit
is obtained by conventional oversampling whereas 3 bits are obtained by noise shaping
oversampling.
The internal construction of a converter is shown in Fig. 18.58. The core is the
integrator that ensures that the difference between the input signal and the reconstructed
Vi
VI f OSR fS
∫
ADC
Z
N bit N N+x
Vi
DAC
V ADC
Vr N bit
f OSR
Vi + V Z 1 VI
– τs
Vi VADC
Vn
Fig. 18.59. Noise modell
for an -converter
signal from the DAC will be near zero. The noise transfer function can be calculated from
the model in Fig. 18.59. The input signal has lowpass characteristic
VADC 1
=−
Vi 1 + τs
with a cutoff frequency fc = 1/(2πτ ). The integration time constant τ must be chosen
not to limit the signal bandwidth. The noise transfer function has highpass characteristic
VADC τs
= ≈ τs for τs 1
Vn 1 + τs
that results in the noise shaping characteristic shown in Fig. 18.55. You can even use a
chain of two or three integrators to improve the noise shaping and to get additional bits.
But additional feedback paths are needed to stabilize the loop.
The most simple modulator with an N = 1 bit ADC and DAC is shown in
Fig. 18.60. The 1 bit ADC consists of a comparator and a clocked flip-flop. If the inte-
grator output voltage VI is negative the flip-flop will be set (Q = 1) and the negative
refernce voltage will be switched on by the 1 bit DAC causing the integrator to move in
positive direction. Like the dual-slope converter it consists of an integrator followed by
a comparator. Here the input voltage Vi is always connected to the integrator. In order to
measure the input voltage, reference pulses of opposite sign are applied to the integrator
so that its output remains near zero. This process is shown in Fig. 18.61. A compensation
R C
Vi R
1 bit ADC
Q fOSR fS
VI 1D Z
1 1+x
C1
–Vref fOSR
Q
VQ 1 bit DAC
Vi
max
5
Fig. 18.61. Signals in the simple -modulator. Example for Vi = V and N = 3 bit. The
8 ref
flip-flop is set for one clock if the integrator output voltage is negative at the triggering clock edge
pulse is issued if the integrator output is negative at the triggering clock edge. Its duration
is one clock period.
From the number of the compensation pulses required to balance the input charge the
reading can be calculated. The current trough of the input resistance must be as high as
the average current from the reference source:
Vi Vref Z !
+ =0 (18.26)
R R Z max
Vi
Z= Z max (18.27)
Vref
The maximum number Z max is the number of clock pulses in which the number of com-
pensation pulses is counted. In the example in Fig. 18.61 we have chosen Z max = 8 in
5
order to get a 3 bit conversion. With the supposed input voltage Vi = 8 Vref a number of
5 compensation pulses are required during 8 clock cycles.
The main difference to the dual-slope converter is that the charge is not balanced once
at a measurement but with a frequency that is higher by a factor of 2N . The advantages of
the converter are:
– only a simple analog antialiasing filter is required at the input because of the high
oversampling frequency
– high increase of resolution by noise shaping oversampling
– high linearity
– good for integration
Some common converters are collected in Fig. 18.62. Some converters with 7-segment
interface follow in Fig. 18.63.
18.11 Errors in AD-Converters 983
18.11
Errors in AD-Converters
18.11.1
Static Errors
In addition to the systematic quantization noise, errors arise from nonideal circuitry. When
the mid-values of the steps are joined, as shown in Fig. 18.64, a straight line is obtained
which passes through the origin and has a slope of 1. For a real A/D converter, this line
misses the origin (offset error) and its slope is different from 1 (gain error). The gain error
gives rise to a relative deviation of the output quantity from the desired value, which is
constant over the whole range of operation. The offset error gives rise to a constant absolute
deviation. Both errors can usually be eliminated by adjusting the zero-point and the gain.
The remaining deviation is then due to drift and nonlinearity only.
A linearity error that exceeds the quantization error is incurred whenever the steps
are of different heights. To determine this linearity error, the offset and gain are adjusted
and the maximum deviation of the input voltage from the ideal straight line is measured.
984 18 Digital-Analog and Analog-Digital Converters
V V total
nonlinearity
(negative)
total
V V nonlinearity
(positive)
V Vi V Vi
Vi –V Vi – V V
V
Vi Vi
V
V
Fig. 18.64. Quantization error of an ideal A/D Fig. 18.65. Transfer characteristic of an A/D
converter converter that has linearity errors
This value, which is reduced by the systematic quantization error of 21 VLSB , is the total
nonlinearity. It is usually quoted as a percentage of the LSB voltage unit. For the example
shown in Fig. 18.65, the total nonlinearity is ± 21 VLSB .
A further measure of the linearity error is the differential nonlinearity, which indicates
by how much the widths of the individual steps deviate from the desired value VLSB . If
this deviation is larger than VLSB , a number is skipped (missing code). For even larger
deviations, the number Z may even decrease for an increasing input voltage (monotonicity
error).
18.11.2
Dynamic Errors
A/D converter applications fall into two categories: those in digital voltmeters and those in
signal-processing circuits. Their use in digital voltmeters is based on the assumption that
the input voltage remains constant during the conversion process. For signal-processing
applications, however, the input voltage changes continually. For digital processing, sam-
ples must be taken from the alternating voltage by means of a sample-and-hold circuit.
The samples are then A/D converted. It has been shown in Sect. 18.1 that the resulting
number sequence {Z} represents the continuous input signal without loss of information
only if the sampling theorem (the Nyquist theorem) is satisfied. The sampling frequency,
fs , must therefore be at least twice the highest signal frequency f max . Consequently, the
conversion time of the A/D converter and the settling time of the sample-and-hold circuit
must together be less than 1/(2f max ). In order to meet this requirement without introduc-
ing excessive complexity, the signal bandwidth is limited to the lowest value possible. It
is therefore necessary to insert a preceding lowpass filter.
In order to judge the attainable accuracy, the properties of the A/D converter and of the
sample-and-hold circuit must, in this application, be considered together. For example, it
is pointless to operate a 12-bit A/D converter in conjunction with a sample-and-hold circuit
that does not settle to 1/4,096 ≈ 0.025% of full scale within the allowed time.
18.12 Comparison of AD-Converters 985
Vi
Vi
Another dynamic error is incurred due to aperture jitter. Due to the aperture time tA of
the sample-and-hold circuit, the measured value is with a certain delay. If the aperture time
is constant, every measured value will be delayed by the same time, and so equidistant
sampling will still be insured. However, if the aperture time varies by the aperture jitter
tA , as shown in Fig. 18.66, a measurement error is introduced which is equal to the
voltage change V in this time. To calculate the maximum error V , the input signal is
taken to be a sine wave with the highest frequency f max for which the system is designed.
The maximum slope is at the origin:
dV
= V̂ ω max
dt
t=0
Hence, the amplitude error is:
V = V̂ ω max tA
If the error is to be smaller than the smallest conversion voltage VLSB of the A/D converter,
the condition for the aperture jitter must be:
VLSB VLSB
tA < = (18.28)
V̂ ω max 1
V ω
2 max max
It is difficult to fulfill this condition for high signal frequencies, as the following example
illustrates. For an 8 bit converter, VLSB /V max = 1/255. For a maximum signal frequency
of 10 MHz, the aperture jitter must, according to (18.28), be smaller than 125 ps.
18.12
Comparison of AD-Converters
In order to compare the different methods, we have listed their main characteristics in
Fig. 18.67. Figure 18.68 shows the resolution and frequency range in which these methods
are implemented.
986 18 Digital-Analog and Analog-Digital Converters
24
22
dual slope
20
delta sigma
18
16
resolution/bit
counting converters
14
pipeline
12 successive converters
10 approximation
converters
8
flash converters
6
4
0
1 10 100 1k 10k 100k 1M 10M 100M 1G 10G
f/HZ
In Chapter 13, several transfer functions are discussed and their realization by active filters
is described. The processed signals are voltages which, in turn, are continuous functions
of time. The circuits are made up of resistors, capacitors, and amplifiers.
Recently, the trend has been toward signal processing by digital rather than analog
circuits. The advantages are high accuracy and consistency in the results, and a lower sen-
sitivity to disturbances. The high number of digital components required is a disadvantage,
however, but in view of the increase in integration of digital circuitry, it is becoming less
important.
Sequences of discrete numbers are processed instead of continuous signals, and the
circuit elements are memories and arithmetic circuits. The transition from an analog to a
digital filter raises three questions:
1) How can a sequence of discrete numeric values be derived from the continuous input
voltage without loss of information?
2) How must this numeric sequence be processed in order to obtain the required transfer
function?
3) How can the output values be converted back into a continuous voltage?
The embedding of a digital filter in an analog environment is shown schematically in
Fig. 19.1.At sampling instants tµ , the sample-and-hold circuit extracts voltages Vi (tµ ) from
the input signal Vi (t) and holds them constant for one sampling interval. In order to prevent
any information loss from occurring during sampling, the input signal must be band-
limited to half the sampling rate, in accordance with the sampling theorem. Consequently,
a lowpass filter is generally required at the input.
The ADC converts the time-discrete voltage sequence Vi (tµ ) into a numeric sequence
x(tµ ). The x values are usually N bit binary numbers, where N determines the magnitude
of the quantizing noise (see (18.9)). Typical sampling frequencies and resolutions are listed
in Fig. 19.2.
The digital filter shown in Fig. 19.1 produces the filtered number sequence y(tµ ).
In order to convert it back into a voltage, a DAC is used, which delivers an amplitude-
and time-discrete staircase signal at its output. To convert this to a continuous voltage, a
following lowpass smoothing filter is required.
value-discrete
amplitude-discrete
time-discrete
19.1
Digital Transfer Function
As we have seen in Chap. 13, analog filters can be implemented using integrators, adders,
and coefficient networks. A digital filter is obtained by replacing the integrators by delay
elements. The latter can be implemented by, for example, shift registers that are used to
shift the input function samples through at the sampling rate fs . The simplest case is that
of delay by one time interval Ts ; a delay element of this kind is shown schematically in
Fig. 19.3.
19.1.1
Time Domain Analysis
The number sequence {x(tµ )} = {xµ } is given and may be used as sampled values with
a word length of 8, 16 or 32 bits. These are shifted into a register using a corresponding
number of parallel-clocked flip-flops. The output sequence in Fig. 19.3 {y(tµ )} = {yµ }
represents the input sequence shifted by one clock period Ts . We therefore have
y(tµ ) = x(tµ− 1 ) (19.1)
19.1.2
Frequency Domain Analysis
To examine the frequency response, the sinusoidal sequence x(tµ ) = x̂ sin ωtµ is applied
to the input. If the system is linear, a sinusoidal sequence appears at the output. As for
analog filters, the ratio of the amplitudes is equivalent to the magnitude of the transfer
function for s = j ω. The linearity of a digital filter is indicated by the linearity of the
difference equation. According to (19.1), the filter in Fig. 19.3 is therefore linear.
The transfer function may be inferred from the circuit with the help of complex calculus,
as for analog filters. This requires that the frequency response of a delay element be known.
With the harmonic input sequence
x(tµ ) = x̂ej ωtµ
the harmonic output sequence
y(tµ ) = x̂ej ω(tµ −Ts ) = x̂ej ωtµ · e−j ωTs = x(tµ )e−j ωTs
is obtained, and with jω = s, the transfer function
L{y(tµ )}
A(s) = = e−j ωTs = e−sTs (19.2)
L{x(tµ )}
It is a periodic function, the period being f = fs = 1/Ts , where fs is the sampling – that
is, the clock – frequency. To abbreviate this,
Ã(z) = z− 1 (19.4)
This is the frequency-domain description of the delay element shown in Fig. 19.3.
It was mentioned in Chap. 13 that the transfer function A(s) describes the relation-
ship between the output signal and any desired time-dependent input signal if Laplace
transforms according to
L{y(t)} = A(s) · L{x(t)} (19.5)
are used. This relationship also holds for a digital system. Using the converted transfer
function of (19.4), the relation for number sequences can be simplified, since
Z{y(tµ )} = Ã(z) · Z{x(tµ )} (19.6)
where
∞
0
Z{x(tµ )} = X(z) = x(tµ )z−µ (19.7)
µ=0
is the Z-transform of the input sequence. The output sequence is obtained by the corre-
sponding reverse transform. Because of this property, Ã(z) is called the digital transfer
function.
From this, we can calculate the analog transfer function or the quantities derived from
it, such as the magnitude, phase, and group delay. For the delay element, it follows from
Y (z)
Ã(z) = = z− 1 with z− 1 = e−j ωTs ,
X(z)
that
A(j ω) = z− 1 = e−j ωTs = cos ωTs − j sin ωTs .
The magnitude is therefore
'
|A(j ω)| = cos2 ωTs + sin2 ωTs = 1
the phase
− sin ωTs f
ϕ = arctan = arctan(− tan ωTs ) = −ωTs = − 2π
cos ωTs fs
990 19 Digital Filters
z−1
y(tµ+1 ) = x(tµ ) − β1 y(tµ ) Y (z) = X(z)
1 + β1 z−1
Time Domain Frequency Domain
Fig. 19.4. Example of a recursive first-order digital filter
s s s s
Fig. 19.6. Amplitude–frequency response of the digital filter in Fig. 19.4 for β1 = −0.75
19.2
Basic Structures
Lattice filters apart, there are three arrangements for implementing digital filters. These
are illustrated in Figs. 19.7–19.9. All three possess the same transfer functions if the filter
coefficients αk and βk are used at the locations shown.
We can see from Figs. 19.7–19.9 that, in addition to the delay elements, the filters
require multipliers that multiply the variables by the constant filter coefficients, and adders
that add two or three numbers. The structure shown in Fig. 19.7 is the most commonly
used, as in this case each multiplier–accumulator stage (MAC) is separated from the next
by a delay element. As a result, a complete clock period is available for these operations.
992 19 Digital Filters
Fig. 19.8. Digital filter with a global adder at the output and input
Fig. 19.9. Digital filter with a single global adder at the output
The delay elements here produce a “pipeline” structure. In the other two circuits, many
variables have to be added in a single clock period. Although this does not require more
adders, it does require more computing time.
In the circuit shown in Fig. 19.8, it can be seen that the input signal for the delay
chain is derived from the input signal X and all of the weighted intermediate values.
Accordingly, the output signal is the weighted sum of all the intermediate values. The
adders can therefore be combined to form two global adders: one at the input and one at
the output.
The circuit in Fig. 19.9 has a single global adder at the output. It adds both the delayed
and weighted input signal as well as the delayed and weighted output signal. For this
purpose, an additional delay chain is required, which is connected to the output. However,
the additional complexity involved is minimal.
19.2 Basic Structures 993
The number of filter sections determines the order N of the filter. One delay element
(two in Fig. 19.9) and two coefficient multipliers are required for each section, and three
numbers must be added. Only the first and last sections are somewhat simpler.
We shall analyze the circuits taking Fig. 19.7 as an example. The difference equation
is of the form
0
N 0
N
y(tN ) = αk xN−k − βk yN−k (19.8)
k=0 k− 1
f
fn = or ωTs = 2πfn (19.10)
fs
In order not to violate the sampling theorem, the following must apply:
1 1
0≤f ≤ fs or 0 ≤ fn ≤
2 2
Thus, for the magnitude of the complex frequency response, it follows from (19.9) that
2 N 2
8 N 8
+
α k cos 2πkfn α k sin 2π kfn
k=0 k=0
|A(j ω)| = 2 (19.11)
8 2
8
N N
βk cos 2πkfn + βk sin 2π kfn
k=0 k=0
The coefficient β0 is always 1. The multiplier for β0 can therefore be omitted in all of the
filter structures.
994 19 Digital Filters
19.3
Design Analysis of FIR Filters
The coefficients βk in the digital filters (see Figs. 19.7–19.9) determine the amount of
feedback. If they are all made zero, there is no feedback and the output signal is merely
the weighted sum of the input signal and its delays. Filters of this kind are known as
nonrecursive filters, transversal filters, or finite impulse response filters(FIR). The term
“FIR” means that the impulse response is of finite length (N +1 values). The circuits shown
in Figs. 19.7–19.9 then become the simplified circuits shown in Figs. 19.11 and 19.12.
19.3.1
Basic Equations
The elimination of the coefficients βk also simplifies the transfer equations. The difference
equation is
0
N
yN = α0 xN + α1 xN− 1 + . . . + αN− 1 x1 + αN x0 , yN = αk xN−k (19.12)
k=0
For the transfer function, we obtain
1 2
Y (z) = α0 + α1 z− 1 + α2 z− 2 + . . . + αN− 1 z−(N− 1) + αN z−N X(z)
Y (z) 0
N
Ã(z) = = αk z−k . (19.13)
X(z)
k=0
Inserting Euler’s relation,
z− 1 = e−j 2πfn = cos 2πfn − j sin 2πfn , (19.14)
the complex frequency response
0
N
A(j ω) = αk e−j 2πkfn (19.15)
k=0
is obtained. This relation can be simplified if the coefficients are symmetrical:
αN−k = αk even symmetry (19.16)
αN−k = −αk odd symmetry (19.17)
Two terms with coefficients that are equal in absolute value can then be combined and a
common phase factor can be factored out. Equation (19.15) is then simplified:
for even symmetry,
0
N
A(j ω) = e−j πNfn αk cos π(N − 2k)fn (19.18a)
k=0
for odd symmetry,
0
N
A(j ω) = e−j πNfn αk sin π(N − 2k)fn . (19.18b)
k=0
For odd symmetry, the middle coefficient must disappear in even-order filters; in other
words, α 1 N = 0. We thus obtain an expression in terms of the magnitude B(ω) and the
2
phase ej ϕ , of the form
996 19 Digital Filters
B(ω)e−j πNfn for even symmetry
A(j ω) =
B(ω)j e−j πNfn for odd symmetry
In order to calculate the magnitude, we only need to take account of the sum in (19.18).
The phase shift follows from the exponential function
−πNfn for even symmetry
ϕ = (19.19)
−πNfn + π/2 for odd symmetry
In both cases, we can see the linear phase behavior that is exactly fulfilled for any sym-
metrical coefficients.
The group delay is obtained from the definition:
dϕ dϕ dfn Ts dϕ
tgr = − = − · = − · (19.20)
dω dfn dω 2π dfn
Hence, by differentiating (19.19),
1
tgr = N Ts (19.21)
2
Consequently, it is frequency-invariant. Delay distortion cannot therefore occur with sym-
metrical FIR filters. This is one of their chief advantages, and it is the reason why FIR
filters are only designed with symmetrical coefficients. The design procedures and exam-
ples given in this chapter all produce FIR filters with a constant group delay. Choosing
symmetrical coefficients results in another advantage: the calculation of the filter algorithm
is simplified because each coefficient multiplier can be used for 2 coefficients.
19.3.2
Simple Examples
In order to become familiar with the behavior and design analysis of FIR filters, it is useful
to examine a few simple examples.
The DC voltage gain of an FIR filter is equal to the sum of all of the filter
coefficients.
At the highest signal frequency permitted by the sampling theorem – that is,
1 1
f = fs bzw. fn =
2 2
a unit input sequence is produced in which the values +1 and −1 occur alternately:
{Xµ } = {+1, −1, +1, −1, . . .}. Consequently, in Fig. 19.12 the output signal
Y = +0.5 − 0.5 = −0.5 + 0.5 = 0; that is, it is constantly zero. This is also appar-
ent from the amplitude–frequency response. This characteristic can also be generalized:
19.3 Design Analysis of FIR Filters 997
The gain of an FIR filter at half the sampling frequency is equal to the sum of
the coefficients alternately weighted with +1 and −1.
If all of the filter coefficients are multiplied by the same factor, the effect is the same as if
the input signal were multiplied by that factor. From this, we can derive the general rule:
If all of the coefficients of an FIR filter are multiplied by the same factor, only
the basic gain of the filter will be modified, but its filter characteristic will remain
unchanged.
A first-order highpass filter is shown in Fig. 19.14. We can see from the coefficients
α0 = +0.5 and α1 = −0.5 that their sum is zero. This also results in zero DC voltage gain.
With an input sequence of +1, −1, . . . (the highest signal frequency), the same sequence
will appear at the output. The gain is therefore unity. The cutoff frequency of the highpass
filter, like that of the lowpass filter, is fc = 41 fs .
The two examples described also reveal the linear phase behavior and the resulting
constant group delay. The lowpass filter can also be used for averaging, as may be seen
from the coefficients. Similarly, the highpass filter can be used as a differentiator, since for
low frequencies,
Fig. 19.17. Log–log plot of the frequency responses of the lowpass filters used as examples
N = 1 Fig. 19.13 : α0 = α1 = +0.5
N = 2 Fig. 19.15 : α0 = α2 = +0.25 α1 = +0.5
Fig. 19.18. Log–log plot of the frequency responses of the highpass filters used as examples
N = 1 Fig. 19.14 : α0 = +0.5 α1 = − 0.5
N = 2 Fig. 19.16 : α0 = α2 = +0.25 α1 = − 0.5
1000 19 Digital Filters
19.3.3
Calculating the Filter Coefficients
Two methods are commonly used to calculate the coefficients of FIR filters: the window
method and the Remez Exchange Algorithm. The latter is a numeric method of Chebyshev
approximation of a given gain tolerance scheme. It provides a minimal number of coeffi-
cients and therefore produces particularly efficient circuits. The advantage of the window
method is that it allows a clear understanding of the mode of operation, while at the same
time being less computationally intensive. We shall now describe how this method is used
to calculate the filter coefficients.
FIR filters possess a particularly distinctive impulse response. If a unit impulse con-
forming to the sequence
1 for k = 0
{x(kTs )} = (19.22)
0 otherwise
is applied to the input, we obtain, in accordance with Figs. 19.11 and 19.12 or (19.12), the
unit-impulse response
{y(kTs )} = α0 , α1 , α2 , . . . αN = {αk } ; (19.23)
that is, the sequence of the filter coefficients.
On the other hand, it can be shown that the impulse response of a system represents
the inverse Fourier transform of its frequency response Aw (j ω), in accordance with
+∞
y(t) = Aw (j ω)ej ωt dω . (19.24)
−∞
With discrete-time systems, the frequency response is periodic with fs = 1/Ts , and the
time can be specified as multiples of the sampling period: t = kT s . This simplifies (19.24)
to
+1/2f
s
y(kTs ) = Aw (jf )ej 2πf kTs df . (19.25)
−1/2fs
The required filter coefficients are obtained by equating (19.23) and (19.25) and specifying
the desired frequency response Aw (j ω).
19.3 Design Analysis of FIR Filters 1001
Fig. 19.20. Desired frequency response of an ideal lowpass filter and its periodic continuation
Of particular interest is the case, shown in Fig. 19.20, of the ideal lowpass filter with
cutoff frequency fc,n = fc /fs and a gain of 1 in the passband and 0 in the stop band. If a
1
constant group delay tgr = 2 N Ts is additionally required, Aw (jf ) can also be expressed
as a delay function:
−j πf NT
e s for −fc ≤ f ≤ fc
Aw (jf ) = (19.26)
0 otherwise
Inserting this ideal frequency response in (19.25), we obtain
fc fc,n
αkr = e−j πf NTs ej 2πf kTs df = ej πF (2k−N) dfn
−fc −fc,n
sin(2k − N )πfc,n
αkr = 2fc,n for k = 0, 1, 2 . . . N (19.27)
(2k − N )πfc,n
These are the filter coefficients that are being sought, but we have only their raw values –
hence the subscript “r.” They must be modified such that the desired cutoff frequency
or gain is achieved precisely. For this reason, we shall simplify by omitting factor 2fc,n ,
which is common to all coefficients of (19.27). This simplification is allowed since factor
2fc,n will eventually be replaced during the necessary normalization of the gain. To the
value (sin 0)/0, which occurs for even order – in other words, of odd coefficient numbers –
we assign the following limit value:
sin x
lim = 1.
x→0 x
Since, in practice, only finite orders N can be realized, the sequence αkr , must be termi-
nated. This can be interpreted – as in Fig. 19.21 – as multiplication by a square window.
This, of course, means that we have an incomplete approximation to the desired frequency
response. In Fig. 19.22 we can see a marked deviation from the desired ideal frequency
response and a poor stop-band attenuation. This situation can be greatly improved by using,
instead of the square window, a window that gradually reduces the coefficients toward the
edge. Commonly used window functions are
the Hamming window, the Hanning window,
the Blackman window, and the Kaiser window.
We shall use the Hamming window, as it provides good results with minimal computing
effort. The Hamming function
2πk
Wk = 0.54 − 0.46 cos for k = 0, 1, 2 . . . N (19.28)
N
1002 19 Digital Filters
normalizing
normalized
the gain
normalizing normalized
the cutoff
frequency
weighting with
Hamming window
window function
normalizing normalized
the gain
Fig. 19.21. Steps in calculating the filter coefficients based on the example of a fifth-order lowpass
filter that has a cutoff frequency Fc,n = 0.25
Result: α0 = α5 = − 0.00979, α1 = α4 = +0.00979, α2 = α3 = 0.5000
19.3 Design Analysis of FIR Filters 1003
. . fc,n . . fn .
Fig. 19.22. The effect of the individual steps for coefficient calculation on the frequency response
of a filter, based on the example of a fifth-order lowpass filter with fc,n = 0.25
fc,n = 0.25
fc,n = 0.1
fc,n = 0.025
. . fn
In order to arrive at the simplest solution for realizing lowpass filters, we can ask the
question: How must the cutoff frequency fc,n be chosen to insure that, in (19.27), as many
filter coefficients as possible become zero? Two special cases of this kind are shown in
Fig. 19.25. If we make fc,n = 21 , all of the coefficients apart from the middle one with the
value α( 21 N ) = 1 vanish in even-order filters (odd number of coefficients). The resulting
filter is an allpass filter, and therefore cannot be used as a lowpass filter.
If the cutoff frequency is halved, half-band filters with fc,n = 41 are produced. If this
condition is inserted into (19.27), we obtain:
sin(2k − N )π/4
αkr = (19.29)
(2k − N )π/4
As we can see from Fig. 19.25, an appreciable simplification is again obtained for even
order (an odd number of coefficients), as every second coefficient becomes zero. In order
1006 19 Digital Filters
to arrive at a practicable filter design, the coefficients have still to be evaluated with a
window. Using the Hamming window in (19.28), we obtain, with αkw = αkr · Wk ,
sin(2k − N )π/4 2πk
αkw = 0.54 − 0.46 cos for k = 0, 1, 2 . . . N (19.30)
(2k − N )π/4 N
If filter coefficients for all k have been calculated, it is now merely necessary to normal-
ize them by dividing them by their sum in order to obtain the final coefficients. An iterative
process is no longer necessary. Consequently, these filter coefficients may be calculated on
a pocket calculator. The resulting cutoff frequencies are of course not precisely fc,n = 41 ,
as they have not been normalized; but a normalization is ruled out, as the advantage of
every second coefficient vanishing would be lost. The frequency responses of a number of
half-band filters are shown in Fig. 19.26 and a coefficient table is given in Fig. 19.27. It
may be seen from Fig. 19.26 that the −6 dB cutoff frequencies approach fc,n = 0.25 more
accurately the higher the order; that is, they approach the “half band.” The −3 dB cutoff
frequencies normally specified are therefore lower; their precise values are additionally
given in Fig. 19.27. The imprecise values for fc,n nevertheless allow any cutoff frequencies
to be realized by selecting the sampling frequency accordingly: fs = fc /fc,n
19.3 Design Analysis of FIR Filters 1007
Half-band Filters
We can see in Fig. 19.25 that half of the values become zero only for an odd number of
coefficients. Consequently, only half-band filters with an even order are used. It is also
apparent that the boundary coefficients (α0 and α8 in the example) vanish for all orders
that are divisible by four. They are therefore particularly useful, as we can obtain two
additional filter orders without an additional multiplication. Even the two delay elements
allpass filter
fc,n
C
.
half-band filter
fc,n
C
.
S C S C S C S C S
1. filter 2. filter 3. filter 4. filter
fc,n
C = 1/4 fc,nC = 1/4 f c,nC = 1/4 fc,nC = 1/4
S S S S
Highpass Filters
Calculation of the filter coefficients of highpass filters can be related to the design of
lowpass filters. To do this, we use the addition theorem of Fourier transformation, which
states that an addition in the frequency domain corresponds to an addition in the time
domain. Figure 19.29 shows how this statement can be used to design highpass filters.
We can see that a highpass filter is produced in the frequency domain by subtracting a
lowpass filter from an allpass filter. The associated filter coefficients are therefore obtained
by subtracting the coefficients of the lowpass filter from those of the allpass filter, as shown
on the right-hand side of the figure. Of course, the coefficients must once again be weighted
√
using a window, and the magnitude of the gain normalized to 1 at fn = 0.5 and to 1 2 at
fn = fn,c .
However, it is apparent that odd-order highpass filters designed using this method have
a zero at fn = 0.5, which makes their performance unsatisfactory. Therefore, only even-
order filters – that is, those with an odd number of coefficients – have been taken into
account in the coefficients tables in Fig. 19.30 and the frequency responses in Fig. 19.31.
allpass filter
fc,nC .
fn
lowpass filter
fc,nC .
fn
highpass filter
fc,nC .
fc,n
C . fn
Fig. 19.30a. Filter coefficients of FIR highpass filters with a cutoff frequency of fc,n = 0.25;
that is, fc = 0.25fs
19.3 Design Analysis of FIR Filters 1011
Another method of obtaining the filter coefficients of bandpass and band-stop filters
consists of multiplying together the transfer functions Ã(z) of a corresponding highpass
and lowpass filter. The implementation can then be performed either from the individual
filters in a cascade arrangement or, after multiplying out, in a continuous arrangement.
1012 19 Digital Filters
fc,n = 0.25
fc,n = 0.1
fc,n = 0.025
. . fn
fc1 fc2
19.4
Realization of FIR Filters
To implement FIR filters, (19.12) on page 995 must be used to calculate the output values,
0
N
y(tN ) = αk x(tN−k )
k=0
as the sum of the N last input values weighted with the coefficients. This operation can be
performed either in parallel (i.e., in one step) or serially (i.e., in N steps). In the former case,
considerable hardware complexity is involved, and in the latter a considerable amount of
time, as shown in Fig. 19.33. If we assume, for example, 10 ns for the basic operation – that
is, multiplication and addition (MAC-operation) – sampling frequencies of 100 MHz can
be achieved with parallel processing; otherwise, only the N th part thereof is considered.
To calculate (19.12), all the coefficients and the last N sampling values must of course
be available in memory. In both cases, this requires a memory for 2N + 1 values.
The required word length w of the data x is determined by the signal-to-quantizing-
noise ratio, which is in the order of w · 6 dB. The word length available for the coefficients
determines the accuracy to which the calculated coefficients can be realized. They are
normally selected to be at least as large as the data word. After multiplication, this produces
words of double the word length; that is, 2w. When calculating the sum, the word length
can increase by one bit in each step; that is, to 2w + N . However, the actual increase
is smaller, as the majority of the coefficients αk 1. Nevertheless, a rounding down to
smaller word length is generally unavoidable if circuit complexity is to be kept within
acceptable limits.
ycas
s s s
y
Fig. 19.34. Design of integrated FIR filters employing the parallel method
19.4.1
Realization of FIR Filters Using the Parallel Method
The structure shown in Fig. 19.11 is particularly suitable for realizing FIR filters by the
parallel method, as an entire clock period is available for an MAC operation. Multiplication
of the input sequence by the filter coefficients can, in principle, be carried out by parallel
multipliers. Their multiplicand is defined by connecting bit by bit to 0 or 1, depending on
the value of the coefficients. It would also be possible to work out the multiplication table
for each coefficient and store it in a look-up table.
However, both methods are now outdated, as FPGAs with signal processing blocks are
used. They are highly complex circuits that contain a large number of parallel multipliers,
adders, and memories. An additional shift register for storing the coefficients is shown
in Fig. 19.34. The coefficients are read into the shift register once the supply voltage has
been switched on; the filter is then configured. The coefficients may also be exchanged
during operation in order to make the filter characteristic adaptive. This facility is used in
echo cancellation, for example. The coefficient input can also be used as a second signal
input. In this case, the arrangement calculates the cross-correlation function of the input
signals. The additional input ycas allows similar devices to be cascaded so as to raise the
filter order.
19.4.2
Realization of FIR Filters Using the Serial Method
The serial realization of FIR filters is derived from the basic structure shown in Fig. 19.12,
with a global adder at the output. A shift register is used to store the coefficients, as shown
in Fig. 19.35. It is then possible to replace all of the multipliers and adders by a single
MAC, as shown in Fig. 19.36. To calculate the output value, the multiplier inputs are shifted
S S S
S S S
Input
Fig. 19.37. Serial calculation of y. To calculate the output value, the two output pointers are
rotated once and all of the sub-products are added. Then the next value x is read in
once through all the stages and the resultant sub-products are added together. The two shift
registers are implemented as FIFOs (see Sect. 10.2.3). It is not necessary to shift the data
physically; rather, it is only the relevant input and output pointers that are stepped forward.
This is illustrated in Fig.19.37. A number of freely programmable signal processors are
listed in Fig. 19.50 on page 1028.
19.5
Design of IIR Filters
Recursive filters are also known as infinite impulse response (IIR) filters, as – at least
theoretically – their impulse response possesses an infinite number of nonzero sampling
values. Their basic structure and transfer functions, which have already been discussed in
Sect. 19.2, apply to digital filters in general.
1016 19 Digital Filters
19.5.1
Calculating the Filter Coefficients
Two methods in particular are commonly used to calculate the filter coefficients, the Yule-
walk Algorithm and the bilinear transformation. The Yulewalk Algorithm approximates
a given tolerance scheme in the frequency domain by means of a minimum number of
filter coefficients. It thus provides coefficients for a minimized IIR filter, and therefore
represents the analogon to the Remez Exchange Algorithm for FIR filters. We shall now
describe the bilinear transformation in greater detail, because it is less computationally
intensive and therefore facilitates understanding of the principles involved.
The bilinear transformation is based on the frequency response of an analog filter
and attempts to model it as accurately as possible using an IIR filter. However, this is
not directly possible, as the transfer function of a digital filter can only be utilized up to
half the sampling frequency 21 fs and must be periodic beyond that. For this reason, the
amplitude–frequency response of the analog filter in the range 0 ≤ f ≤ ∞ is mapped into
the range 0 ≤ f
≤ 21 fs of the digital filter and continued periodically. A transformation
that possesses this characteristic is
fs πf
f = tan (19.31)
π fs
For f → ∞, f
tends to 21 fs , as required. For f
fs , we have f ≈ f
. The compression
of the frequency axis is therefore smaller the higher is the clock frequency fs with respect
to the frequency range of interest.
In order to be able to employ normalized frequencies as with the analog filters, we
normalize all the frequencies to the sampling frequency:
fn = f/fs or fc,n = fc /fs (19.32)
Equation (19.31) therefore becomes:
1
fn = tan πfn
. (19.33)
π
To illustrate the transformation of the frequency axis, in Fig. 19.38 we have plotted the
amplitude–frequency response of a second-order Chebyshev lowpass filter. We can see that
the typical passband characteristic is retained, although the cutoff frequency is shifted. In
order to avoid this effect, we introduce a shift-factor l into (19.33) for frequency mapping.
We select this factor such that the cutoff frequency is retained in the transformation; that
:
is, fc,n = fc,n
f = fc,n cot πfc,n tan πfn
(19.34)
3 45 6
l
The resultant frequency response curve is shown in Fig. 19.39. We interpret the for-
mally introduced quantity fn
as a new frequency variable fn and denote the transformed
frequency response by A
(j ωn ). We can see that this provides a good approximation to
the analog filter characteristic.
The transformed frequency response A
(jfn ) now possesses a shape which can be
implemented using a digital filter. To calculate the digital transfer function Ã(z), we now
require the transformation equation for the complex frequency variable sn . With sn =
j ωn = jfn /fc,n , it follows from (19.34) that
s = l · j tan πfn
19.5 Design of IIR Filters 1017
.
| f )|
| f )´ |
analog
.
. bilinear
´ f c,n
f c,n . . f n , f n´
c s
analog
bilinear
. . . fn
c s
Fig. 19.39. Matching the cutoff frequency. Second-order Chebyshev characteristic with a 3 dB
ripple as an example. Cutoff frequency: fc,n = 0.3, i.e., fc = 0.3fs . Logarithmic plot
1 − e− 2πjfn 1 − z− 1
s = l = l with l = cot πfc,n (19.35)
1 + e− 2πjfn 1 + z− 1
19.5.2
IIR Filters in a Cascade Structure
As in the case of analog filters, the simplest method of implementing IIR filters is to cascade
first- and second-order blocks. In this case, the values tabulated in Fig. 13.14 for analog
filters can also be used to calculate the filter coefficients. The recalculation of the filter
coefficients is therefore given below in more detail.
A0 1 + z− 1
A(sn ) = ⇒ Ã(z) = α0 (19.40)
1 + a 1 sn 1 + β1 z − 1
A0 1 − a1 l
α0 = α1 = ; β1 = (19.41)
1 + a1 l 1 + a1 l
Similarly, for a highpass filter,
A∞ A∞ sn 1 − z− 1
A(sn ) = = ⇒ Ã(z) = α0
1 a1 + sn 1 + β1 z − 1
1 + a1
sn
A∞ l a1 − l
α0 = −α1 = ; β1 = (19.42)
a1 + l a1 + l
By way of example, we shall calculate the coefficients for a first-order highpass filter for
phone applications. Its cutoff frequency will be fc = 100 Hz, and the bandwidth of the
input signal will be 3.4 kHz. We select fs = 10 kHz and obtain the normalized cutoff
frequency
α0 + α1 z− 1 0.9695 − 0.9695z− 1
Ã(z) = =
1 + β1 z − 1 1 − 0.9391z− 1
The ratio of the sampling frequency to the cutoff frequency is determined by the
parameters selected and has the value 100. The cutoff frequency is therefore proportional to
the sampling frequency. It can therefore be simply controlled using the sampling frequency.
This characteristic is peculiar to all digital filters. The only other filters exhibiting this
property are the switched-capacitor types described in Sect. 13.12.
1020 19 Digital Filters
Y α0 + α1 z − 1 + α2 z − 1
Fig. 19.41. Second-order IIR filter Ã(z) = =
X 1 + β 1 z − 1 + β2 z − 2
Second-Order IIR
A second-order IIR filter obtained by particularizing Fig. 19.7 on page 992 is shown in
Fig. 19.41. Into the linear transfer function,
d0 + d1 sn + d2 sn2
A(sn ) =
c0 + c1 sn + c2 sn2
we insert the bilinear transformation as defined in (19.35) and obtain
α0 + α1 z− 1 + α2 z− 2
Ã(z) = (19.43)
1 + β 1 z − 1 + β2 z − 2
with the coefficients
d0 + d1 l + d2 l 2 2(d0 − d2 l 2 ) d0 − d1 l + d2 l 2
α0 = ; α1 = ; α2 = ;
c0 + c 1 l + c 2 l 2 c0 + c 1 l + c 2 l 2 c0 + c 1 l + c 2 l 2
2(c0 − c2 l 2 ) c0 − c1 l + c2 l 2
β1 = ; β2 =
c0 + c 1 l + c 2 l 2 c0 + c 1 l + c 2 l 2
From this, we can calculate the following second-order filters with l = cot πfc,n :
Lowpass filter (19.44):
A0 1 + 2z− 1 + z− 2
A(sn ) = ⇒ Ã(z) = α0
1 + a1 sn + b1 sn2 1 + β 1 z − 1 + β2 z − 2
A0 2(1 − b1 l 2 ) 1 − a1 l + b 1 l 2
α0 = ; β1 = ; β2 =
1 + a1 l + b1 l 2 1 + a1 l + b1 l 2 1 + a1 l + b1 l 2
Highpass filter (19.45):
A∞ sn2 1 − 2z− 1 + z− 2
A(sn ) = ⇒ Ã(z) = α0
b1 + a1 sn + sn2 1 + β 1 z − 1 + β2 z − 2
A∞ l 2 2(b1 − l 2 ) b1 − a1 l + l 2
α0 = ; β1 = ; β2 =
b1 + a 1 l + l 2 b1 + a 1 l + l 2 b1 + a 1 l + l 2
Bandpass filter (19.46):
Ar sn /Q 1 − z− 2
A(sn ) = ⇒ Ã(z) = α0
1 + sn /Q + sn2 1 + β 1 z − 1 + β2 z − 2
lAr /Q 2(1 − l 2 ) 1 − l/Q + l 2
α0 = ; β1 = ; β2 =
1 + l/Q + l 2 1 + l/Q + l 2 1 + l/Q + l 2
19.5 Design of IIR Filters 1021
A0 (1 + sn2 ) α0 + A0 β1 − z− 1 + α0 z− 2
A(sn ) = ⇒ Ã(z) =
1 + sn /Q + sn2 1 + β 1 z − 1 + β2 z − 2
A0 (1 + l 2 ) 2(1 − l 2 ) 1 − l/Q + l 2
α0 = ; β1 = ; β2 =
1 + l/Q + l 2 1 + l/Q + l 2 1 + l/Q + l 2
We shall now discuss the design procedure with the aid of an example. We require a
second-order Chebyshev lowpass filter with a 0.5 dB ripple and a 3 dB cutoff frequency of
fc = 100 Hz. The analog signal will have a bandwidth of 3.4 kHz and be sampled at fs =
10 kHz. This gives a normalized cutoff frequency of fc,n = 0.01 and a normalizing factor
of l = 31.82. From the table in Fig. 13.14, we can obtain a1 = 1.3614 and b1 = 1.3827.
This produces the continuous transfer function
1
A(sn ) =
1 + 1.3614sn + 1.3827sn2
Using (19.44), we obtain from this the digital transfer function
1 + 2z− 1 + z− 2
Ã(z) = 6.923 · 10− 4
1 − 1.937z− 1 + 0.9400z− 2
As a second example, we will design a bandpass filter. The sampling frequency will be
10 kHz as before. The resonant frequency will be fr = 1 kHz. Hence fc,n = 1 kHz/10 kHz
= 0.1. For Q = 10, the continuous transfer function, in accordance with (13.24), for
Ar = 1 is of the form
0.1sn
A(sn ) =
1 + 0.1sn + sn2
Using l = cot πfr,n = 3.078 and (19.46), we obtain the digital transfer function for
Q = 10 and fr,n = 0.1
1 − z− 2
Ã(z) = − 2.855 · 10− 2
1 − 1.572z− 1 + 0.9429z− 2
Similarly, for Q = 100 and fr,n = 0.1 we obtain
1 − z− 2
Ã(z) = − 2.930 · 10− 3
1 − 1.613z− 1 + 0.9941z− 2
We will now consider the case Q = 10 and fr,n = 0.01. This gives
1 − z− 2
Ã(z) = − 3.130 · 10− 3
1 − 1.990z− 1 + 0.9937z− 2
We can see that as Q increases or the resonant frequency fc,n decreases, the coefficient
α0 → 0, whereas β2 → 1 and β1 → 2. The information on the filter characteristic is
therefore to be found in the very small deviation with respect to 1 or −2. This means that
there is an increasing accuracy requirement on the filter coefficients, which results in a
correspondingly large word length in the filter. In order to minimize the circuit complexity,
the sampling rate must therefore not be selected to be any higher than necessary.
1022 19 Digital Filters
19.6
Realization of IIR Filters
19.6.1
Construction from Simple Building Blocks
We shall now demonstrate the procedure for arriving at the simplest possible circuit, using
the example of the first-order highpass filter from Sect. 19.5.2. There, we have already
calculated the digital transfer function for a highpass filter with a cutoff frequency of
fc = 100 Hz at a sampling rate of fs = 10 kHz; that is, fc,n = 0.01:
α0 + α1 z− 1 0.9695 − 0.9695z− 1
Ã(z) = −
=
1 + β1 z 1 1 − 0.9391z− 1
The corresponding circuit is shown in Fig. 19.42. We can see that the three coefficients
are close to 1. The numerator coefficients α0 and α1 can be rounded to 1 without any
appreciable error, as they only determine the gain. This does not apply to coefficient
β1 , whose deviation from 1 determines the filter cutoff frequency. However, in this case
simplification is possible by the transformation
−β1 = 1 − β1
= 0.9361 = 1 − 0.0609
where β1
= 1 + β1 is the deviation from unity. This coefficient possesses substantially
fewer significant digits than β1 . The nearest power of two is 2−4 = 0.0625. The binary
arithmetic can be greatly reduced by rounding β1
to this value, as a multiplication by 2−4
only represents a shift by four digits, which can be implemented by appropriate wiring.
From (19.42), the resulting shift in the cutoff frequency is given by
1 − β1 2 − β1
2 − 2−4
l = = = = 31 .
1 + β1 β1
2−4
Thus with (19.35) we get fc,n = 0.0103; that is, the cutoff frequency increases to fc =
103 Hz.
We simplify further by rounding the numerator coefficients to α0 = −α1 = 1, and use
(19.42) to obtain, for high frequencies (f ≈ 21 fs ), the gain
1+l 1 + 31
A∞ = α0 = 1· = 1.032
l 31
This small deviation is also acceptable. The resulting simplified arrangement is shown in
Fig. 19.43. We can see that in simple filters it is possible to reduce the circuit considerably
by slightly modifying the design objective.
The practical implementation is shown in Fig. 19.44 for an input word length of 4 bits.
In order to be able to represent positive and negative numbers, we have selected the two’s-
complement notation introduced in Sect. 8.1.3. The highest-order bit is therefore the sign
bit. As we can perform the multiplication by shifting, only adding circuits are required.
For this purpose, we use 4-bit arithmetic circuits of type SN 74 LS 382. These can also be
operated as subtractors by controlling appropriate inputs. In this way, the computation of
the two’s-complement of the coefficients α1 = −1 and −β1 = 1 − 2−4 can be carried out
in the adder.
The two arithmetic circuits IC 8 and IC 9 form the expression
r = −β1 y = y − 2− 4 y
19.6 Realization of IIR Filters 1023
. .
. .
Fig. 19.42. First-order IIR highpass filter Fig. 19.43. Highpass filter with simplified
coefficients
0.9695 − 0.9695z− 1
Ã(z) = 1 − z− 1
1 − 0.9391z− 1 Ã(z) =
fc,n = fc /fs = 0.01 1 − (1 − 2− 4 )z− 1
fc,n = fc /fs = 0.0103
A(f = 0.5fs ) = 1
A(f = 0.5fs ) = 1.032
Fig. 19.44. Circuit implementation of a first order digital IIR highpass filter with a word length of 8 bit internal and 4 bit external
19.6 Realization of IIR Filters 1025
19.6.2
Design Using LSI Devices
There are three methods of implementing IIR filters using LSI circuitry:
1) using CPLDs or FPGAs programmed as filter,
2) using FIR filters,
3) using programmable signal processors.
Programmable digital hardware circuits like CPLDs and FPGAs can be configured
as digital filters. For this task the hardware description language VHDL is advantageous
because the adders and multipliers required can be defined with a single command line.
FPGAs with predefined DSP-block are especially efficient for high speed processing and
little usage of resources. Examples for such devices are given in Fig. 10.48 on page 721.
An IIR filter can be constructed from two FIR filters. It is possible to start with the
basic structure shown in Fig. 19.8, with global adders at the input and output, and to double
the delay chain. This results in the circuit shown in Fig. 19.47, in which we can identify
the two FIR filters. Whether the FIR filters employ a global adder at the output, as shown
here, or distributed adders, as shown in Fig. 19.11, is unimportant. The result in both cases
is the same if the coefficients are arranged accordingly.
The basic structure shown in Fig. 19.9 can also be broken down into two FIR filters
if the global adder at the output is split into two sections. In Fig. 19.48 we can see that
this produces two FIR filters whose partial effects (i.e., results) can be combined using an
additional adder.
Single-chip signal processors are the most suitable devices for the serial method of
implementation of digital filters, as they possess the required data memories in addition
to a parallel multiplier with accumulator. The calculation of the filter output sequence can
be programmed at machine level in Assembler, as on a microprocessor. Additionally, the
programming is now being supported in higher-level languages such as “C.” For program-
1026 19 Digital Filters
–ˇ0
Fig. 19.47. Implementation of an IIR filter with a global adder at the input and the output, from
two FIR filters and one additional adder
–ˇ0
Fig. 19.48. Implementation of an IIR filter using a single global adder at the output, from two FIR
filters and one additional adder
ming, it is best to start from the basic structure with a global adder at the output, according
to Fig. 19.48. As shown in Fig. 19.49, the new value of the output sequence can then be
calculated in accordance with (19.8),
0
N 0
N
YN = αk xN−k − βk yN−k
k=0 k=1
by weighting all input and feedback signals with the relevant coefficients and subsequently
accumulating them. To do this, the tap is shifted along the delay chain and the particular
coefficient αk or βk is selected. When the entire chain has been run through once, the new
function value yN is available. Then the contents of the two shift registers can be shifted
forward by one clock, in order to calculate a further function value y in the next pass. The
data are, of course, not shifted physically: only the pointers addressing the values xk , yk ,
αk , and βk rotate.
19.7 Comparison of FIR and IIR Filters 1027
Fig. 19.49. Serial implementation of an IIR filter with a single global adder at the output,
preferably using a programmable signal processor. Data X is shifted with the sample clock Ts .
Coefficients are shifted in once before start of filtering.
Using a signal processor, it is just as easy to make design calculations for IIR filters
in cascade form. For this purpose, the order of the filter in Fig. 19.49 is reduced to N = 2
and we calculate
y2 = α0 x2 + α1 x1 + α2 x0 − β1 y1 − β2 y0
using a small subroutine. The entire filter is therefore obtained by repeatedly calling up
the program for a second-order filter and exchanging the relevant data or coefficient sets.
Figure 19.50 provides an overview of a number of more recent signal processors. The
preferred number notations are 16-bit fixed-point numbers for universal applications or 32-
bit floating-point numbers for high accuracy and dynamic range. The data word length of
the accumulator is generally more than twice that size, in order to insure that the rounding
errors have no effect on the result. The majority of signal processors have high-speed data
and program memories on the chip. These should be used where possible, because each
external memory access results in the insertion of wait states even if the access times are
short.
The time taken to perform a multiplication and accumulation (MAC) operation deter-
mines how rapidly a signal processor can process fitter algorithms, as these consist virtually
exclusively of MAC operations. For an N th-order FIR filter, N + 1 MAC operations are
required for each sampled value, whereas an IIR filter requires 2N + 1. In most of the
more recent signal processors, a MAC operation is performed in a single machine cycle.
19.7
Comparison of FIR and IIR Filters
If we compare the structure of the IIR filters shown in Figs. 19.7–19.9 with that of the
FIR filters shown in Figs. 19.11 and 19.12, we can see that IIR filters of the same order
require approximately twice as many MAC operations as FIR filters. However, they possess
a higher selectivity than FIR filters with the same number of MAC operations. This is
illustrated by the example shown in Fig. 19.51. Generally speaking, the required order for
1028 19 Digital Filters
an FIR filter is, for the same performance, more than twice as high as for an IIR filter. In
Sect. 19.5.2 we have shown that a lowpass filter with a low cutoff frequency of fc,n = 0.01
can be implemented using a first-order IIR filter. For an FIR filter with this cutoff frequency,
an order of at least N = 65 would have been required since, to a first approximation, a
complete cycle at the cutoff frequency fc,n is to be weighted with the coefficients; that is,
N ≥ 1/fc,n = fs /fc .
However, this characteristic must be set against a number of important advantages of
FIR filters. We have seen that FIR filters can easily be used to implement linear phase
behavior – that is, a constant group delay – with precision. All of the FIR filters mentioned
in this section possess this characteristic; that is, they do not produce any phase distortion.
As FIR filters possess no feedback path, they are also stable for any coefficient values.
IIR filters, like analog filters, tend to oscillate more the higher is their pole Q-factor or the
lower is their cutoff frequency compared to the sampling frequency (see Sect. 19.5.2). In
order to insure that no significant deviations from the calculated frequency response are
incurred, the coefficients of IIR filters must be realized with considerably more precision
than for FIR filters; this requires a larger word length. In addition, with IIR filters the
rounding errors due to limited computational accuracy frequently result in limit cycles.
These are periodic oscillations in the lowest-order bits, which are particularly disturbing
in the case of small input signals. The various advantages and disadvantages are listed in
Fig. 19.52.
19.7 Comparison of FIR and IIR Filters 1029
FIR N = 8
fc,n = 0.1
IIR N = 4
Butterworth
. . fn
Fig. 19.51. Comparison of an eighth-order FIR lowpass filter with a fourth-order IIR lowpass filter
In the previous chapters, a number of methods for processing analog and digital signals have
been described. Many applications, however, require that even electrical signals must be
conditioned before they can be processed in analog computing circuits or A/D converters.
In such cases, measurement circuits are needed which have a low-resistance single-ended
output; that is, produce a ground-referenced output voltage.
20.1
Measurement of Voltage
20.1.1
Impedance Converter
If the signal voltage of a high-impedance source is to be measured without affecting the load
conditions, the noninverting amplifier (follower-with-gain, electrometer amplifier) shown
in Fig. 5.58 can be employed for impedance conversion. It must be noted, however, that
the high-impedance input is very sensitive to noise arising from capacitive stray currents
in the input lead. Therefore the lead is usually screened, but this results in considerable
capacitive loading of the source to ground with 30 . . . 100 pF/m. For an internal source
resistance of 1 G, for instance, and a capacitance of the coax cable of 100 pF, an upper
cutoff frequency of only 1.6 Hz is obtained.
The capacitance is not constant but can change; for example, when the lead is moved.
This gives rise to an additional problem in that this variation produces very large noise
voltages. If, for instance, the lead is charged up to 10 V, a change in capacitance of 1%
results in a voltage step of 100 mV.
These disadvantages can be avoided if a noninverting amplifier is employed to keep
the voltage low between the inner conductor and the shield. The shield is then connected
not to ground but to the buffer output, as shown in Fig. 20.1. In this manner, the effect of
the lead capacitance is reduced by the open-loop gain of the operational amplifier. Since
now only the offset voltage of the amplifier appears across the lead capacitance, the noise
also is eliminated to a large extent.
Vi
Vo
follow the input voltage. This is achieved by using the bootstrap circuit shown in Fig. 20.2,
that has two emitter followers to stabilize the potential differences V1 − Vo and Vo − V2 to
a value VZ − 0.7 V. The maximum output voltage swing is thus no longer determined by
the operational amplifier, but by the permissible voltage across the emitter followers and
the current sources.
20.1.2
Measurement of Potential Difference
The measurement of potential differences involves amplifying the voltage difference
VD = V2 − V1
while minimizing the effect of the superimposed common-mode voltage
1
VCM = (V2 + V1 )
2
It is frequently the case that differential voltages in the millivolt range have common-mode
voltages of 10 V or more superimposed on them. The quality of a subtractor is therefore
characterized by its common-mode rejection:
AD Vo /VD VCM
G = = =
ACM Vo /VCM VD
In the example given, G must be greater than 10 V/1 mV = 104 . Particular problems arise if
the superimposed common-mode voltage exhibits high values or high frequencies. There
are three different methods of amplifying voltage differences:
– operational amplifiers configured as subtractors;
– differential amplifiers with feedback, and subtraction using switched; capacitors
– subtractor with operational amplifier circuitry.
V1
V1
A
A Vo
V2 Fig. 20.3. Subtractor with preceding
impedance converters
A R2
V2 Vo = (V2 − V1 )
R1
voltage followers shown in Fig. 20.3 renders the operation of the subtractor independent
of the internal resistances of the potentials to be measured.
A higher common-mode rejection can be achieved by shifting the voltage gain to the
impedance converters and giving the subtractor unit gain. This variant is shown in Fig. 20.4.
For R1 = ∞, OA1 and OA2 operate as voltage followers; in this case, the circuit is virtually
identical to the previous one.
The circuit has the further advantage that the differential gain can be adjusted by varying
a single resistor. It can be seen in Fig. 20.4 that the potential difference V2 − V1 appears
across resistor R1 . Hence,
2R2
V2
− V1
= 1 + (V2 − V1 )
R1
This difference is transferred to the single-ended output by the subtracting amplifier OA3.
For a purely common-mode input drive (V1 = V2 = VCM ), V1
= V2
= VCM . The
common-mode gain of OA1 and OA2 is therefore unity, irrespective of the differential
gain chosen. Using Eq. (11.4), we thus obtain the common-mode rejection ratio in the
following form:
2R2 2α
G = 1+
R1 α
where α/α is the relative matching tolerance of the resistors R3 .
With the instrumentation amplifier in Fig. 20.4, one operational amplifier can be elim-
inated at the expense of circuit symmetry. Electrometer amplifier OA2 in Fig. 20.5 has
the gain 1 + R1 /R2 . OA1 amplifies potential V2 by 1 + R2 /R1 and simultaneously adds
V1
A
V1
V2 A Vo
V2 V2
A Vo A Vo
V1 V1
A A
V2
Vo
Fig. 20.7. Subtractor with a single high-impedance input
V1 RN RN RN
Vo = 1 + + V2 − V1
R1 R2 R1
voltage V1
injected at the base point with weighting −R2 /R1 . As a result, the two input
potentials are amplified by 1 + R2 /R1 . If the circuit is modified as shown in Fig. 20.6, the
gain can again be set by a single resistor.
For some applications, it is acceptable to use a subtractor with only one high-impedance
input. In this case, only a single operational amplifier is required, as shown in Fig. 20.7.
However, the transfer equation reveals the limitation that the gain for V2 is always less
than that for V1 in absolute-value terms (although this is no disadvantage, for example, in
the case of the gain and zero offset of sensor signals). An interesting special case arises
for RN = R1 = R and R2 = ∞; we then obtain the output voltage Vo = 2V2 − V1 .
V1 V1
Vo Vo
V2 V2
Fig. 20.8. Subtraction of high voltages Fig. 20.9. Subtraction of high voltages
with a freely selectable gain
R2 R2
Vo = (V2 − V1 ) = 0.05(V2 − V1 ) Vo = (V2 − V1 ) = (V2 − V1 )
R1 R1
VCM =
R2
= 0.045V2 R2 |R3
R1 + R2 VCM = V2 = 0.045V2
R1 + R2 |R3
be designed independently. Resistors R1 and R2 again determine the gain; the additional
resistors R3 merely reduce the common-mode input voltage. For the component values
selected, we obtain unit gain, whereas the common-mode input voltage range is virtually
unchanged in comparison with the example in Fig. 20.8. An IC subtractor that employs
this principle is the INA 148 from TI.
However, increasing the common-mode input voltage range with resistors R3 in
Fig. 20.9 also presents problems that must be taken into account when selecting the opera-
tional amplifiers. Resistors R3 actually operate as attenuators for the operational amplifier
input signals. They therefore reduce the loop gain and, consequently, in most cases also
the bandwidth. They simultaneously increase, to the same extent, the unwanted amplifi-
cation of the offset voltage and offset voltage drift. As a result, a high-grade operational
amplifier is required. Resistors R3 must, of course, introduce the same attenuation on both
sides, and so it is particularly important to use tightly matched resistors. In order to insure
tight matching tolerances at both operational amplifier inputs, resistors R2 and R3 at the
noninverting input should not generally be combined to form a single resistor.
V2 − V1 V4 − V3
IC = =
RG RS
1036 20 Measurement Circuits
Vo
V2 V1
Fig. 20.10.
Instrumentation
amplifier with negative-feedback differential amplifiers
R2 RS
Vo = 1 + (V2 − V1 )
R1 RG
Resistors R1 and R2 are already included in ICs based on this principle. The user is
therefore able to fix the gain at the desired value using RS and RG . The advantage of this
circuit compared to the operational amplifier subtractor is that the degree of common-mode
rejection is not dependent on the matching tolerance of the voltage dividers R1 and R2 .
For this reason, the circuit shown in Fig. 20.10 can be manufactured entirely in monolithic
IC form, whereas the critical resistors would otherwise have to be implemented separately
in thin-film technology.
VH = VS = VD = V2 − V1
20.1 Measurement of Voltage 1037
Vo
VD VS VH
1 nF 1 nF
Fig. 20.11.
Subtractor
in switched-capacitor technology. Example: LTC 2053
R2
Vo = 1 + (V2 − V1 )
R1
20.1.3
Isolation Amplifiers
Using the instrumentation amplifiers described above, voltages in the range 10–200 V can
be processed, depending on the circuit principle employed. There are many applications,
however, in which the voltage to be measured is superimposed on a considerably higher
common-mode voltage of perhaps 1000 V. To deal with such high potentials, the measuring
circuit is split, as shown in Fig. 20.13, into two electrically isolated units. Electrical isolation
may also be required for safety reasons; for example, in most medical applications. The
transmitter unit operates at the measuring potential (floating ground), and the receiver unit
at system ground. The transmitter must have its own floating supply, which is connected
to the floating ground. Although the floating ground is electrically isolated from system
ground, it must not be forgotten that there is still some capacitive coupling, which is due
mainly to the capacitance CS of the supply transformer. This is indicated in Fig. 20.13.
To keep the coupling low, it is advisable not to use a mains supply transformer but a
1038 20 Measurement Circuits
Fig. 20.12. Examples for instrumentation amplifiers. All amplifiers can be operated from a single
supply voltage
signal coupling
transmitter receiver
Vo
V V
system
floating ground
ground floating oscillator grounded
supply supply
stray
capacitance
high-frequency transformer for about 100 kHz, fed from a separate oscillator. In this way,
coupling capacitances CS < 10 pF can be achieved.
If both test points have a high internal resistance, even the reduced stray capacitance
current may produce a considerable voltage drop in the floating ground loop. In such a case,
it is advisable to connect the floating ground terminal to a third point, and to determine the
potential difference between the two measuring points using the instrumentation amplifier
20.1 Measurement of Voltage 1039
A A Vo
measuring
coupler
reference
coupler
Fig. 20.14. Opto-electronic transmission of an analog signal. A suitable matched dual optocoupler
is the HCNR 200 from Agilent.
shown in Fig. 20.4. The measuring signals then carry no current; and the stray current has
no influence because it flows through the floating ground connection which is no longer
a signal path. The remaining common-mode voltage with respect to floating ground can
usually be kept low if the floating ground is connected to a suitable potential within the
circuit that is being tested.
The question is how to transmit the analog signal from the transmitter to the receiver
with electrical isolation. Three methods exist: transformers, opto-couplers, and capaci-
tors. For transmission by transformers or capacitors, the signal must be modulated by a
carrier that offers enough bandwidth (amplitude modulation or pulse-width modulation),
whereas opto-couplers enable the direct transmission of DC signals. When high accuracy
is required, the analog signal can be digitized at floating-ground potential, and the dig-
ital values subsequently transmitted by opto-couplers or capacitors to the receiver unit.
Nonlinearities of the transmission path are eleminated this way.
A method of optical analog transmission is presented in Fig. 20.14. In order to com-
pensate for the linearity error of the opto-coupler, the operational amplifier OA1 controls
the current though the light-emitting diodes in such a way that the photoelectric current
in the reference receiver T1 equals a desired value. The feedback loop is closed via the
reference coupler, so that
Vf+ V1 − V2
IF 1 = +
R2 R1
+
As the photoelectric current cannot change polarity, a constant current Vf /R2 is super-
imposed on both sides to enable the transmission of bipolar input signals. If the two
opto-couplers are well matched, we obtain on the receiver side the current IF 2 = IF 1 , and
thus the output voltage
R1
Vf+ V+
Vo = (V1 − V2 ) for =
R1 R2 R2
Isolation amplifiers with transformer, opto-, or capacitor coupling are available as ready-
made modules. A number of such types are listed in Fig. 20.15. Among the most user-
1040 20 Measurement Circuits
Fig. 20.15. Examples for isolation amplifiers. Examples for isolated power supplies are the
DCP 02-series from Texas Instruments or the HPR100-series from Power Convertibles
friendly are the types that already incorporate the required DC converter. An external
voltage converter is only worthwhile if it can be used to drive several isolation amplifiers
whose floating ground is at the same potential. In the types with a built-in voltage converter,
the floating power supply is also available to the user; for example, for driving a preceding
instrumentation amplifier or a sensor. An all-purpose device is the Analog Devices AD 210,
in which the receiver circuit also operates from a floating supply. Consequently, the receiver
signal ground can in this case be isolated from the power supply ground and, as there are
therefore three mutually isolated ground terminals, this arrangement is termed “three-port
isolation.”
20.2
Measurement of Current
20.2.1
Floating Zero-Resistance Ammeter
Section 12.2 describes a current-to-voltage converter that is almost ideally suited for the
measurement of currents because of its extremely low input resistance. However, as the
input represents virtual ground, only currents to ground can be measured.
Floating ammeters can be realized by the instrumentation amplifier in shown Fig. 20.4
if its two inputs are connected to a measuring shunt. The advantage of the low input
resistance is then lost, but if the shunt is incorporated in the feedback loop of the input
amplifiers, as illustrated in Fig. 20.16, a floating ammeter is obtained that has virtually
zero input resistance.
Since the input voltage differences of both operational amplifiers is zero the potential
difference across the inputs 1 and 2 becomes zero, too. If a current I flows into terminal 1,
the feedback causes the output potential of OA2 to have the value
V2 = Vi − I R1 (20.1)
With VN = V1 , we obtain
R2 R 1 R2
V1 = V2 + 1 +
(Vi − V2 ) = Vi + I (20.2)
R2 R2
Vi
VD Vi
A Vo
Vi A
V1 − Vi R1 R2
I
= =
I (20.3)
R1
R1 R2
If both inputs are to behave like those of a floating circuit, there must be I
= I . Hence
we have the condition
R1 R2
= (20.4)
R1
R2
The subtracting amplifier OA3 computes the difference V1 − V2 . Its output voltage, using
(20.1) and (20.2), is therefore
R2
Vo = R1 1+
I (20.5)
R2
20.2.2
Measurement of Current at High Potentials
The permissible common-mode voltage of the previous circuit is limited to values between
the supply voltages. For the measurement of currents at higher potentials, the simple circuit
of Fig. 12.5 can be employed if it is connected to the floating ground terminal of an isolation
amplifier rather than to system ground. Its output voltage is referenced to system ground
with the help of the isolation amplifier.
The required circuitry may be reduced quite considerably if a voltage drop of 1…2 V
can be tolerated for the measurement of current (e.g., in the anode circuit of high-voltage
tubes). In such cases, the current to be measured is made to flow through the light-emitting
diode of an opto-coupler, so that a floating supply is no longer needed. For linearization
of the transfer characteristic, a reference opto-coupler may be used on the receiver side, as
demonstrated in Fig. 20.17. Its input current I2 is controlled by the operational amplifier in
1042 20 Measurement Circuits
measuring reference
coupler coupler Vo Fig. 20.17. Simple isolation
amplifier for current measurement
Vo = RI
such a manner that the photoelectric currents of the reference and the measuring coupler
cancel each other out. If the couplers are well matched,
I2 = I
20.3
AC/DC Converters
Various quantities are used to characterize alternating voltages: the arithmetic mean abso-
lute value, the root-mean-square (rms) value and the positive and negative peak value.
20.3.1
Measurement of the Mean Absolute Value
To obtain the absolute values of an alternating voltage, a circuit is required in which the
sign of the gain changes with the polarity of the input voltage; in other words, its transfer
characteristic must have the shape represented in Fig. 20.18.
A full-wave rectifier of this kind can be realized by use of a diode bridge. However,
the accuracy of such a circuit is limited due to the forward voltages of the diodes. This
effect can be avoided if the bridge rectifier is operated from a controlled current source; a
simple solution is shown in Fig. 20.19. The operational amplifier is employed as a voltage-
controlled current source, in accordance with Fig. 12.8. Hence, independently of the diode
forward voltage,
|Vi |
IA =
R
Vo
full-wave half-wave
Fig. 20.18. Characteristics of a half-wave and a
Vi full-wave rectifier
20.3 AC/DC Converters 1043
Vo
Vi
To display the mean value of this current, a moving-coil ammeter can be used, for instance,
and this method is therefore often employed in analog multimeters.
For output potentials in the range − 2VD < Vo < 2VD , the amplifier has no feedback,
as none of the diodes conduct. While Vo changes from 2VD to − 2VD , VN remains constant,
this causing a delay time within the control loop. Because of this, any phase shift can be
incurred in the control loop, depending on the frequencies involved, so that stabilization of
the operational amplifier is particularly difficult. To reduce the delay time, amplifiers must
be chosen that have a fast slew rate, and the frequency compensation must be stronger than
for linear feedback.
Vi
A Vo Vi
A
V
Vi
V
Vo
Vi
Vi
A Vo Vi
Vi
Vi Vo Vi
CA CA
of opposite polarity are produced. Depending on which input amplifier is selected by the
comparator, the output voltage +Vi or −Vi is obtained.
This method of rectification is practicable because there are ICs that operate on this
principle, such as theAnalog DevicesAD 630, which also contains the required comparator.
However, at high frequencies the delay due to the comparator introduces appreciable errors,
as the delayed changeover then becomes a critical factor.
Vi Vo Vi
Vi Vo Vo
Vi
20.3.2
Measurement of the rms Value
Whereas the arithmetic mean absolute value (the mean modulus) is defined as
T
1
|V | = |V |dt (20.9)
T
0
the definition of the root-mean-square value (the rms value) is given as
'
T
1
Vrms = (V 2 ) = V 2 dt (20.10)
T
0
where T is the measuring interval, which must be large compared to the longest period
contained in the signal spectrum. In this manner, a reading is obtained that is independent
of T . For strictly periodic functions, averaging over one period is sufficient to yield the
correct result.
For sinusoidal voltages,
√
Vrms = V 7/ 2
so that the rms value can be determined simply by measuring the peak value. For other
wave-shapes, this method would produce errors, particularly for highly peaked voltages;
7/Vrms .
that is, for wave-shapes that have large crest factors V
The errors become smaller if the measurement of rms values is reduced to that of the
mean absolute values. For a sinusoidal voltage,
7
T
V 27
|V | = | sin ωt|dt = V (20.11)
T π
0
√
7/ 2,
and, with Vrms = V
π
Vrms = √ |V | ≈ 1.11 · |V | (20.12)
2 2
20.3 AC/DC Converters 1047
π
Sinusoidal Vrms = √ |V | = 1.11 |V | exact
2 2
Fig. 20.27. Errors in the rms value for different waveforms if the mean absolute value is calibrated
for sinusoidal signals
The relative magnitude of the individual values is demonstrated in Fig. 20.26. The form
factor of 1.11 is incorporated in the calibration of most available mean-absolute value
meters; for sinusoidal signals, therefore, they show the rms value although they actually
measure the mean-absolute value. For other wave-shapes, this modified reading produces
varying deviations from the true rms value as shown in Fig. 20.27.
Vi
U V1
U
Vi Vo = Vi rms
V1
Vi2
Vo
Vi Vo = Vi rms
Fig. 20.29. Circuit for true rms measurement, having an improved input voltage range
In this respect, the circuit shown in Fig. 20.29 is preferable, since the square-rooting
operation at the output is replaced by division at the input. The voltage at the output of the
lowpass filter is thus
Vi2
Vo = (20.13)
Vo
At steady-state, Vo is constant; hence,
'
(Vi2 ) 2
Vo = that is Vo = (V i ) = Vrms
Vo
An advantage of this method is that the input voltage Vi is not multiplied by the factor
Vi /U which, for low input voltages, is small with respect to unity, but by the factor Vi /Vo ,
which is close to unity. The available input range is therefore considerably larger. However,
the precondition for this is that the division Vi /Vo can be carried out sufficiently accurately
even for small signals. Divider circuits based on logarithmic operations are best suited for
this purpose, as described in Sect. 11.8.1.
The implicit equation (20.13) can therefore be solved using the principle illustrated in
Fig. 20.30. Before taking the logarithm, we must first form the absolute value of the input
voltage. Squaring is performed simply by multiplying the logarithm by two. To divide by
Vo , the logarithm of the output voltage is subtracted.
The practical implementation of this principle is shown in Fig. 20.31. The full-wave
rectified input signal is produced at the summing point of OA2. The latter forms the
logarithm of the input voltage. The voltage doubling required for squaring is effected by
the two transistors T1 and T2 connected in series:
Vi Vi 2
V2 = − 2VT ln = −VT ln
IC0 R IC0 R
OA4 takes the logarithm of the output voltage:
Vo
V4 = −VT ln
IC0 R
Vi
Vi Vo
Vi average
Vi Vo Vo
Vi
A Vo
A A
A
The voltage V4 −V2 across T3 , implementing the exponential function, produces the output
voltage
V4 − V2 V2
Vo = ICS R exp = i (20.14)
VT Vo
Using capacitor C for averaging, the same output voltage is therefore produced as expressed
by (20.13)
Transistors T1 –T4 must be of monolithic IC form to insure that they possess identical
characteristics – as was assumed in the above calculation. It is even possible to integrate
the operational amplifiers and resistors on the same chip, as shown in Fig. 20.32.
Thermal Conversion
The rms value of an AC voltage is defined as that DC voltage which will produce the same
average power in a resistor; that is,
Vi2 /R = Vrms
2
/R
The rms value of an AC voltage Vi can therefore be determined by increasing a DC voltage
Vrms across a resistor R until the latter is exactly as hot as that heated by Vi . This is
the principle on which thermal rms measurement is based. Essentially, any method of
temperature measurement can be employed (see Sect. 21.1). A particularly useful one is to
use temperature sensors, which can be fabricated together with the heating resistors as an
IC. Consequently, diodes are mainly used nowadays as temperature sensors, as illustrated
in Fig. 20.33.
Resistor R1 is heated by the input voltage, and resistor R2 by the output voltage. The
latter increases until the difference between the two diode voltages is zero; in other words,
1050 20 Measurement Circuits
Vi Vo=Vi rms
both temperatures are the same. In this case, the operational amplifier configured as a
subtractor with a lowpass filter is used as the control amplifier. The capacitors C1 keep
high-frequency signals away from the operational amplifier.
The diode at the amplifier output prevents resistor R2 being heated by a negative
voltage, as this would result in latch-up of the circuit due to positive thermal feedback.
Since the thermal power is proportional to the square of Vo , the loop gain is also propor-
tional to Vo2 , giving a nonlinear step response: the switch-off time constant is considerably
larger than the switch-on time constant. A considerable improvement is achieved with an
additional square-function AC feedback circuit.
As resistors R1 and R2 are usually low-value components (50 ) in order to achieve a
large bandwidth, correspondingly high currents are required to provide the drive. Conse-
quently, an emitter follower is generally inserted at the output of the control amplifier. A
preamplifier or impedance converter at the input would have to be of considerably more
complex design, since it would not only have to provide high current peaks of several
100 mA, but also be capable of handling the bandwidth of the AC input signal. Therefore
the input signal is directly connected to R1 and preamplifiers are not applied.
In order to obtain accurate measurements, the two measuring couples must have good
matching characteristics. An IC that meets these requirements is the LT 1088 from Linear
Technology, with which accuracies of 1% can be achieved up to 100 MHz. In power meters
frequencies up to 40 GHz are attained by this method.
20.3.3
Measurement of the Peak Value
The peak value can be measured very simply by charging a capacitor via a diode. To
eliminate its forward voltage, the diode is inserted into the feedback loop of a voltage
follower, as shown in Fig. 20.34. For input voltages Vi < VC , the diode is reverse biased.
For Vi > VC the diode conducts and, due to the feedback, VC = Vi . The capacitor therefore
charges to the peak input voltage. The voltage follower OA2 draws only very little current
from the capacitor, so that the peak value can be stored over a long period. The push-button
switch P discharges the capacitor to prepare it for the next measurement.
The capacitive load of amplifier OA1 may give rise to oscillations; this effect is elim-
inated by resistor R1 . However, the charging time is increased as the capacitor voltage
20.3 AC/DC Converters 1051
Vi
A
P A Vo = Vi
Vi
A
P A Vo = –Vi
Vi
A ^
A Vo = Vi
CA
P
Vo
Vi
on
off
Fig. 20.37. Signal waveforms in the peak voltmeter with sample-and-hold circuit
it falls again. The output voltage only rises further when the input voltage exceeds the
last maximum stored. A typical mode of operation is shown in Fig. 20.37. This circuit
can be implemented using the sample-and-hold circuits of Fig. 17.32 on page 944 and the
comparators from Fig. 6.39 on page 601.
V1 (t) = −RC
dVi (t) 7i RC d sin ωt = −V
= −V 7i ωRC cos ωt (20.16)
dt dt
If the frequency is known, the coefficient ωRC can be adjusted to 1, so that the required
term for (20.15) is obtained. By squaring and adding Vi (t) and V1 (t), a continuous voltage
for the amplitude is obtained which requires no filtering.
For variable frequencies, the circuit must be expanded as shown in Fig. 20.38, by
inserting an integrator to enable provision of the cos2 ωt term with a frequency-indepen-
dent amplitude. The output voltage of the integrator is
7
1 1 7i sin ωdt = Vi cos ωt
V2 (t) = − Vi (t)dt = − V (20.17)
RC RC ωRC
20.3 AC/DC Converters 1053
Vi
Vi Vi
U
U
Vi U
Vo = Vi
Vi
A
U Vi
U
Vi
A
20.3.4
Synchronous Demodulator
In a synchronous demodulator (synchronous detector, phase-sensitive rectifier), a rectifier
is controlled by an external signal that has the same origin as the input signal.
A synchronous demodulator can be used in the arrangement shown in Fig. 20.39, to
separate a sine wave from a noisy signal and to determine its amplitude. The sine wave
selected has a frequency equal to that of the control signal VCS , and a phase shift ϕ that is
square wave
shaping
V
Vi
VCS
Vo
Fig. 20.40. Operation of a
synchronous demodulator
constant with respect to the control signal. The special case in which fi = fCS and ϕ = 0
is illustrated in Fig. 20.40. It can be seen that, under these conditions, the synchronous
demodulator operates as a full-wave rectifier. If ϕ = 0 or fi = fCS , negative voltage–time
areas occur as well as the positive areas, and reduce the mean value of the output voltage
so that it is always lower than that of the example shown.
The output voltage will now be determined as a function of frequency and phase. The
input voltage Vi is multiplied by +1 or by − 1 in time with the control frequency fCS , this
effect being represented mathematically as
Vo = Vi (t) · VCS (t) (20.18)
where
1 for VCS > 0
VCS (t) =
−1 for VCS < 0
By rewriting this in Fourier series form, we obtain
∞
40 1
VCS (t) = sin(2n + 1)ωCS t (20.19)
π 2n + 1
n=0
Let us now assume the input voltage to be a sinusoidal voltage that has a frequency
fi = m · fCS and a phase angle ϕm . Using (20.18) and (20.19), we then have the output
voltage:
∞
0
7i sin(mωCS t + ϕm ) · 4
Vo (t) = V
1
sin(2n + 1)ωCS t (20.20)
π 2n + 1
n=0
The arithmetic mean value of this voltage is evaluated by the subsequent lowpass filter.
Using the auxiliary equation
T
1
sin(mωCS t + ϕm ) = 0
T
0
Vo
Vi
0.6
0.4
0.2
fi
=m
fCS
Fig. 20.41. Filter characteristic of a synchronous demodulator. A suitable chip for frequencies up
to 2 MHz is the AD 630 from Analog Devices. The gain and phase detector AD 8302 is even
capable up to 2.7 GHz.
Vi lowpass
U filter
U CS vector
oscillator sum
Vi U CS Vo
Vi lowpass
U filter
voltage is chosen to be equal to the computing unit voltage U of the multiplier, we obtain,
instead of (20.21), the result
,
17
V cos ϕ for fi = fCS
2 i
Vo = (20.22)
0 for fi = fCS
According to (20.22), the synchronous demodulator does not produce the amplitude
7i , directly, but gives the real part V
V 7i cos ϕ of the complex amplitude V7i . To determine the
magnitude |Vi | = V 7i , the phase angle of the control voltage can be adjusted by a suitable
phase-shifting network so that the output voltage of the demodulator is at a maximum. The
signal Vi (t) and the control voltage VCS (t) are then in phase, and we obtain, from (20.22),
17 1
Vo = Vi = |Vi | for fi = fCS
2 2
If a calibrated phase-shifter is employed, the phase shift ϕ of the tested circuit can be read
directly.
Often, we are only interested in the amplitude of a spectral input component and not in
its phase angle. If one wants nevertheless to use the selectivity of the synchronous rectifier
two synchronous demodulators can be used in parallel that are operated by two quadrature
control voltages as shown in Fig. 20.42:
where U is the computing unit voltage of the demodulating multipliers. The oscillator
shown in Fig. 14.30 on page 858 is particularly suitable for generating these two voltages.
Only the spectral component of the input voltage which has the frequency fCS con-
tributes to the output voltages of the two demodulators. If it has the phase angle ϕ with
respect to V1 , it is of the form
7i sin(ωCS t + ϕ)
Vi = V
This chapter deals with circuits for measuring nonelectrical quantities. For this purpose,
it must first be detected by a sensor and then converted into an electrical quantity. The
interface circuit for the sensor normally converts this quantity into a voltage which, after
conditioning, is then displayed or employed for control purposes.
The individual stages are shown in Fig. 21.1. A concrete example is then given in
Fig. 21.2, for a humidity sensor. In this example, the sensor has a capacitance that is de-
pendent on the relative humidity. In order to measure it, the sensor must be incorporated in
a capacitance-measuring circuit, the output of which delivers a voltage that is proportional
to the capacitance but in no way proportional to the humidity. Another circuit is therefore
required for linearization and calibration of the sensor. There are a wide variety of sen-
sors for the most diverse measurands and measurement ranges. An overview of the types
available is provided in Fig. 21.3.
21.1
Temperature Measurement
The following subsections describe various ways in which temperature can be measured.
It is evident from the overview table that metallic sensors, such as the thermocouple
and the resistance thermometer, can be employed for a very large range of temperatures.
Semiconductor-based temperature sensors (PTC and NTC thermistors, transistors) produce
larger output signals but at a smaller temperature range.
analog-to-
interfacing
sensor calibration digital
circuit
conversion
analog digital
display, display,
control control
measured value
V
Fig. 21.2. A humidity sensor, as an example of how the measured value is obtained
1060 21 Sensors and Measurement Systems
21.1.1
Metals as PTC Thermistors
The resistance of metals increases with temperature; that is, metals possess a positive
temperature coefficient. The metals most widely used for temperature measurement are
platinum and nickel–iron. To a first approximation, the resistance increases linearly by
some 0.4% per degree of temperature. Thus, if the temperature increases by 100 K, the
resistance increases by a factor of 1.4.
With platinum temperature detectors, the resistance R0 is specified at 0◦ C. A usual
value is 100 (Pt 100), but it can also be 200 (Pt 200), 500 (Pt 500), or 1,000
(Pt 1,000 ). In the range 0◦ C ≤ ϑ ≤ 850◦ C, the resistance is given by the equation
(DIN4̇3760 and IEC 570):
1 2
Rϑ = R0 1 + 3.90802 · 10−3 ϑ/◦ C − 0.580195 · 10−6 (ϑ/◦ C)2
The usable temperature range of − 200◦ C to +850◦ C is very wide. For higher tempera-
tures, thermocouples are employed (see Sect. 21.1.6). The nonlinearity of the equation is
relatively small. Within a limited range of temperatures, linearization can therefore often
be dispensed with. Examples of interfacing circuits are given in Sect. 21.1.4.
With nickel–iron temperature detectors, the nominal resistance R0 is specified at 20◦ C.
The temperature curve within the range − 50◦ C ≤ ϑ ≤ 150◦ C is then given by
It can be seen that in addition to the linear term there is a quadratic component that causes
a deviation of approximately 25◦ at 150◦ C. Linearization is therefore invariably necessary.
Section 21.1.4 describes the operation and design of the relevant interfacing circuits.
21.1.2
Silicon-Based PTC Thermistors
The resistance of uniformly doped silicon increases with temperature. The temperature
coefficient is approximately twice as large as for metals. The resistance approximately
doubles for an increase in temperature of 100 K. The relevant equation takes the form
This applies only for sensors of KTS-series from Infineon or Philips and is only approximate
for other manufacturers. In the equation, R25 is the nominal resistance at 25◦ C, being
mainly between 1 and 2 k. ϑ is the difference between the actual temperature and the
nominal temperature: ϑ = ϑ− 25◦ C.As with nickel–iron sensors, the usable temperature
range is between 50◦ C and +150◦ C. Section 21.1.4 will show how silicon PTC thermistors
are used and how their characteristics are linearized.
21.1 Temperature Measurement 1063
21.1.3
NTC Thermistors
NTC thermistors are temperature-dependent resistors with a negative temperature coeffi-
cient. They are made of metal-oxide ceramic material and their temperature coefficients
are very large, being between − 3% and −5% per degree. NTC power thermistors are
used for inrush current limiting. With these devices, heating due to the flow of current
is desirable. When hot, they must possess a low resistance and a high current-carrying
capacity. In contrast, self-heating in NTC measurement thermistors is kept to a minimum.
What matters here is a resistance curve that is specified as precisely as possible. If the
temperature of interest T is close to the nominal temperature TN , the relationship between
temperature and resistance can be approximated by the relation
1 1
RT = RN · exp B − (21.1)
T TN
Temperatures must be inserted in Kelvin (T = ϑ + 273◦ ). Depending on the type of
thermistor, the constant B is between 1,500 K and 7,000 K. To enable the resistance char-
acteristic to be described precisely, even if the temperature differences are large, it is
preferable to use the equation
1 1 1 RT 1 RT 3
= + ln + ln
T TN B RN C RN
This additionally includes the term with the coefficient 1/C, allowing an accuracy of 0.1 K
to be achieved even over a temperature range of 100 K. This naturally requires that the
coefficients or the resistance characteristics are specified with sufficient accuracy by the
manufacturer. Interfacing circuits for NTC thermistors are dealt with in Sect. 21.1.4.
21.1.4
Operation of Resistive Temperature Detectors
With the resistive temperature detectors (RTD) described here, resistance is a function of
temperature; the relationship is described by the relevant equations R = f (ϑ). The mag-
nitude of the change in resistance with temperature is given by the temperature coefficient:
1 dR
TK = · (21.2)
R dϑ
in percent per degree. This also allows the resulting temperature tolerance to be calculated
from a resistance tolerance:
1 R
ϑ = · (21.3)
TK R
3 45 6 3 45 6
Temperature tolerance Resistance tolerance
For a temperature coefficient of 0.3% per degree, a resistance tolerance of ± 1% conse-
quently produces a temperature tolerance of ± 3 K. The larger the temperature coefficient,
the smaller is the temperature tolerance for a given resistance tolerance.
The resistance of resistive temperature detectors can be measured by making a constant
current flow through the sensor. This current must be small enough to insure that no
appreciable self-heating occurs, if possible keeping the heat dissipation below 1 mW. A
1064 21 Sensors and Measurement Systems
Fig. 21.4. A four-wire resistance measuring circuit that provides independence from lead
resistances
voltage is then obtained at the sensor that is proportional to its resistance. Where there are
long leads between the current source and the sensor, it may be useful to employ a four-wire
resistance measuring circuit, such as the one shown in Fig. 21.4. Here, the lead resistances
do not falsify the result when Vϑ is measured using high- resistance instruments.
Although the voltage Vϑ is proportional to the resistance Rϑ , it is not necessarily a linear
function of temperature due to nonlinear characteristics. However, if the measured values
are digitized, the corresponding temperature can be calculated by solving the relevant
characteristic curve equation for ϑ. For analog linearization, a function network of the
type described in Sect. 11.7.5 can be connected at the output.
For most applications, however, adequate linearization is obtained by connecting a
suitable fixed resistor Rlin in parallel with the sensor, as shown in Fig. 21.5a. Figure 21.6
shows the effect of Rlin on a silicon PTC thermistor. As the value of Rϑ increases, the
linearization resistor causes the value of the parallel circuit to increase more slowly. This
largely compensates for the quadratic term in the characteristic curve equation. The quality
of the linearization is basically dependent on optimizing the linearization resistance for
the required measurement range. In the simplest case, this value can be obtained from the
data sheet.
The question that remains, however, is how to proceed when no information is available
in the data sheet for the required measurement range. The usual requirement is for a constant
error limit that is as low as possible throughout the range. The linearizing resistance allows
the error for three temperatures (ϑL , ϑM , and ϑV ) to be reduced to zero. These three
temperatures are now shifted, selecting an appropriate value of Rlin , until the maximum
error between them and at either end of the range is of the same magnitude. This process
is illustrated in Fig. 21.6.
Vref
V V
Rϑ Rϑ
Fig. 21.5a. Vϑ = Iref · Rlin Fig. 21.5b. Vϑ = Vref
Rϑ + Rlin Rϑ + Rlin
Fig. 21.5a,b Linearization of an NTC thermistor characteristic using Rlin .
For Vref = Iref · Rlin , both circuits produce the same output signal
21.1 Temperature Measurement 1065
measuring range
U
real
ideal
adjustment points
L U
Vref
Vo
.
V
KTY83
Fig. 21.8. Linearization, zero shift, and gain for a silicon PTC thermistor
Vo = 20 mV ϑ/◦ C for 0◦ ≤ ϑ ≤ 100◦ C
1066 21 Sensors and Measurement Systems
ϑ Rϑ Vϑ Vo
ϑL = 0◦ C RϑL = 820 VϑL = 0.558 V VoL = 0.00 V
ϑM = 50◦ C RϑM = 1202 VϑM = 0.741 V VoM = 1.00 V
ϑU = 100◦ C RϑU = 1706 VϑU = 0.936 V VoU = 2.00 V
. .
Vref
.
V V
.
.
can be operated from a single 5 V source. In order to achieve almost zero output, the output
should be loaded with an additional 1 k.
To linearize platinum temperature sensors, a negative linearizing resistance is required
because of the negative quadratic term in the characteristic. For a Pt 100 sensor operated
in the temperature range between 0◦ C and 400◦ C, a linearization resistor Rlin = − 2.5 k
is required in accordance with (21.4). Linearization of the type shown in Fig. 21.8 is
therefore impossible. In this case, a current source with a negative internal resistance must
be used. The equivalent circuit diagram is shown in Fig. 21.10. The current source shown
in Fig. 12.9 is particularly suitable for implementing this circuit. If R3 is given a slightly
lower value than would be necessary for a constant-current source, a negative resistance
is produced:
Vϑ R1 R3
ro = − = = Rlin
I R3 − R 2
Vref
Vo
desired temperature range. Here the linearizing resistance can be calculated according:
B − TM
Rlin = RT M ∼ RT M
B + 2TM
where B is the B-value of the NTC thermistor in (21.1). Once again, the temperature sensor
can be connected in series with the same linearizing resistor Rlin , giving a linearized voltage
curve. In order to obtain a voltage that increases with temperature, it is advisable to take
off the voltage across the linearizing resistor. This is shown in Fig. 21.13. The circuit and
its design procedure are otherwise identical to the interfacing circuit for PTC thermistors
shown in Fig. 21.8.
21.1.5
Transistors as Temperature Sensors
Because of its internal structure, a bipolar transistor is a heavily temperature-dependent
component. Its reverse current doubles for an approximately 10 K increase in tempera-
ture, and its base–emitter voltage falls by some 2 mV/K (see Sect. 2.21). These otherwise
undesirable side-effects can be utilized for temperature measurement. In Fig. 21.14, a
transistor configured as a diode is operated with a constant current. This produces the
temperature dependence of the base–emitter voltage shown in Fig. 21.15. At room tem-
perature (T ≈ 300 K), its normal value is about 600 mV. With a temperature increase of
100 K, VBE falls by 200 mV, and it increases accordingly if the temperature is reduced.
The temperature coefficient is therefore
V
= 0.3%/K
V · T
Unfortunately, the dispersion of the forward voltage and of the temperature coefficient is
large. For this reason, individual transistors are nowadays only used for temperature mea-
surement where the degree of measuring accuracy required is not too great for instance
in CPUs. Here the type MAX 6699 is usefull. Improved calibration can be achieved for
circuits based on the difference in the base–emitter voltages of two bipolar transistors op-
erated at different current densities. The principle is shown in Fig. 21.16. This is effectively
a bandgap reference, as described in Sect. 16.4.2. The difference between the base–emitter
voltages is given by
IC2 IC1 IC2 A1
VBE = VT ln − VT ln = VT ln
IC0 A2 IC0 A1 IC1 A2
VBE
VBE
Fig. 21.14. The use of the Fig. 21.15. Base-emitter voltage as a function of
base–emitter voltage for (absolute) temperature (typical)
temperature measurement
21.1 Temperature Measurement 1069
VTemp
Vo
Fig. 21.16. Use of a bandgap reference for Fig. 21.17. A supplementary circuit for
temperature measurement (e.g., the implementing a Celsius zero point
LT 1019 from Linear Technology)
Since the two collector currents are the same and the surface ratio of the transistors is
A1 /A2 = 10, it follows:
kT mV
VBE = ln 10 = 200 ·T
e K
To implement a bandgap reference, this voltage is amplified with R2 so that a voltage
VTemp ≈ (2 mV/ K) · T is produced, which compensates for the temperature coefficient of
T2 (see Sect. 16.4.2).
Voltage VTemp can be used directly for temperature measurement: it is proportional to
the absolute temperature T (PTAT). For ϑ = 0◦ C,
mV
VTemp = 2 · 273 K = 546 mV
K
To obtain a Celsius zero point, a constant voltage of this magnitude can be subtracted
from VTemp . For this purpose, the subtractor in Fig. 21.17 uses the appropriately weighted
voltage VBG .
The principle employed in Fig. 21.16 can be modified by connecting the emitters to the
same potential. The output voltage of the operational amplifier in Fig. 21.18 again assumes
a value such that the two collector currents have the same magnitude. This produces the
same value for VBE , but in this case between the base terminals. The voltage across
R1 is therefore proportional to T , (PTAT). It can be increased to any value by the series
connection of additional resistors. In the example shown in Fig. 21.18, it is amplified by a
factor of 50:
mV
VTemp = 50VBE = 10 ·T
K
At room temperature (T ≈ 300 K) this produces a voltage of VTemp ≈ 3 V. The advantage
of this variant is that VTemp occurs at the output of the operational amplifier and a load can
therefore be applied to it.
Temperature sensors that employ the principle shown in Fig. 21.18 are manufactured
as ICs by National (LM 335). They do not have a separate supply voltage terminal and
therefore behave like Zener diodes.
1070 21 Sensors and Measurement Systems
VTemp
Vref
.
.
Vo
VTemp
. .
Fig. 21.18. A modified bandgap reference for Fig. 21.19. An example of Celsius zero
direct temperature measurement (e.g., shift for a two-terminal temperature sensor
LM 335 from National)
The operation of this type of temperature sensor with Celsius zero shift will be illus-
trated by the example shown in Fig. 21.19. Since the sensor behaves like a Zener diode
with a low internal resistance (approximately 0.5 at 1 mA), the current flowing has vir-
tually no effect on the voltage and the sensor can be operated from any supply voltage. It
is merely necessary to insure that the minimum operating current (here, 0.4 mA) is main-
tained. On the other hand, an unnecessarily large operating current should not be chosen,
so that self-heating is kept to a minimum. If a series resistance of 7.5 k is selected in
Fig. 21.19, a current of about 1 mA flows through the sensor at 0◦ C; at 150◦ C it is still
higher than 0.4 mA. To obtain a Celsius zero point, a current of
2.73 V 2.5 V
= = 273 mA
10 k 9.16 k
must be subtracted. Since the operational amplifier inverts in this case, a positive Celsius
scale is obtained at the output, as required.
A sensor that incorporates Celsius zero shift is also available. The LM 35 from National,
for example, produces a voltage of 10 mV/◦ C. It represents a significantly simpler solution
if only positive temperatures are to be measured.
Voltage VBE , which is proportional to temperature, can also be used to generate a
current that is proportional to the absolute temperature. In both Fig. 21.16 and Fig. 21.18,
the collector current IC is proportional to T . In order to obtain the required current, it
is therefore sufficient to replace the operational amplifier in Fig. 21.16 with the current
balancing circuit shown in Fig. 21.20. The condition IC1 = IC2 is then still fulfilled. The
voltage
A1 k A1 mV A1
VBE = VT ln = ln · T = 86 ln ·T
A2 e A2 K A2
VBE
VTemp
21.1.6
Thermocouple
At the contact point of two different metals or alloys, the Seebeek effect gives rise to a
voltage, in the millivolt range, which is known as the thermoelectric voltage. The principle
of temperature measurement illustrated in Fig. 21.22 shows that even if one of the two
metals is copper, we always obtain two thermocouples with opposite polarities. At identical
temperatures ϑM = ϑR , their thermoelectric voltages therefore compensate one another.
Copper
V
Fig. 21.22. Principle of temperature measurement with
R Copper thermocouples, using a copper–constantan thermocouple as
an example
1072 21 Sensors and Measurement Systems
isothermal V
block 1
isothermal
Fig. 21.23. Compensation of the
block 2
reference junction temperature ϑR
isothermal V
R
block
Fig. 21.25. Overview of thermocouples. The most widely used types, J and K, are shown in bold
type. Types B and G are so nonlinear that no average temperature coefficient can be specified
Constantan = copper-nickel; Chromel = nickel-chromium; Alumel = aluminum-nickel
V/mV
plotted in Fig. 21.26. As we can see, none of the curves is precisely linear. Types T, J, E,
and K, however, do possess a decent linearity and also deliver relatively high voltages.
They are therefore preferred if the temperature range permits their use. For the other types,
if it is not possible to restrict the application to a narrow temperature range, linearization
is required.
To evaluate the thermoelectric voltage, a voltage corresponding to reference tempera-
ture ϑR must be added in accordance with Fig. 21.24, in order to refer the measurement to
the “ice point”; that is, 0◦ C. This correction can either be made at thermocouple level or
after amplification. Figure 21.27 gives a schematic representation of the second situation,
using an iron–constantan thermocouple as an example. To amplify its voltage to 10 mV/K,
a gain of
10 mV/K
A = = 193
51.7 mV/K
is required. Then the reference temperature must be added with the same sensitivity; that
is, also with 10 mV/K. Figure 21.28 shows one way of implementing this principle. Since
1074 21 Sensors and Measurement Systems
. R R
Vo
R
R
Fig. 21.27. Amplification and reference point compensation for thermocouples, using an
iron-constantan thermocouple as an example
Vo
R
R
e.g.
or
Fig. 21.28. A practical arrangement for the interfacing circuit for thermocouples, using an
iron-constantan type as an example
the thermoelectric voltages are in the microvolt range, a low-drift operational amplifier is
necessary. In order to obtain adequate loop gain despite the high voltage gain of 193, the
operational amplifier must possess a high open loop gain AD . Measurement of the reference
junction temperature can be greatly simplified by using a ready-made temperature sensor
with a Celsius zero point, such as the LM 35 from National or the LT 1025 from Linear
Technology. However, any other circuit described in this chapter that provides an output
signal of 10 mV/K could also be used.
Figure 21.29 shows the alternative principle whereby the ice point correction value
is added to the thermocouple voltage before the signal is amplified. For this purpose, a
voltage of 51.7 mV/K must be added in the case of an iron–constantan thermocouple. The
circuit becomes very simple if we make use of the fact that the thermocouple is electrically
. R .
Vo
R
. R
Fig. 21.29. Reference junction compensation prior to amplification of thermocouple signals, using
an iron-constantan type as an example
21.1 Temperature Measurement 1075
Vo
R . R
Fig. 21.30. A practical design for reference junction compensation prior to amplification, using
iron-constantan thermocouples as an example
Vo
. R
floating and can therefore simply be connected in series with the correction voltage source,
as shown in Fig. 21.30.
The simplest solution is to use specific ICs for operating thermocouples, such as the AD
594–597 series from Analog Devices. Types AD 594 and 596 of this series are calibrated for
the operation of iron-constantan thermocouples (type J) and types AD 595 and AD 597 for
Chromel-Alumel (type K). Here, the wires of the thermocouple are connected directly to the
integrated circuit, as shown in Fig. 21.31. The latter constitutes the isothermal block with
reference temperature ϑR . It is assumed that the silicon crystal is at the same temperature
as the IC pins. The ice point correction is generated for the chip temperature, added to
the thermoelectric voltage, and amplified. An internal zero point and gain calibration to
within 1◦ C are available. If the inputs are shorted and the thermocouple omitted, only the
ice point correction voltage of
mV mV
Vo = 51.7 ◦ · ϑR · 193 = 10 ◦ ϑR
C C
is produced at the output. The circuit then operates as a transistor temperature sensor with
a Celsius zero point.
21.1.7
An Overview of Types
A number of representative manufacturers and products for temperature measurement are
listed in Fig. 21.32. There are large variations in price. It is therefore worthwhile comparing
different principles and types. In general, it can be said, however, that the more expensive
a sensor is, the more accurately it has been adjusted (laser-trimmed) by the manufacturer.
1076 21 Sensors and Measurement Systems
21.2
Pressure Measurement
Pressure is defined as force per unit area
p = F /A
1 Newton 1N
1 Pascal = ; 1 Pa =
1 Square meter 1 m2
21.2.1
Design of Pressure Sensors
Pressure sensors record the pressure-induced deflection of a diaphragm. For this purpose, a
number of strain gauges are mounted on the diaphragm and form a Wheatstone bridge. They
vary their resistance as a result of the piezo-resistive effect of deflection, pressure, or tensile
force. They are mainly constructed from vapor-deposited constantan or platinum–iridium
layers. Nowadays, resistors implanted in silicon are generally used, with the silicon sub-
strate simultaneously acting as the diaphragm. They have the advantage of being cheaper
to manufacture and are over ten times more sensitive. Their disadvantage, however, is their
higher temperature coefficient.
Figure 21.34 gives a schematic representation of a pressure sensor. In the differential
pressure sensor shown in Fig. 21.34a, a pressure p1 is exerted on one side of the diaphragm
1078 21 Sensors and Measurement Systems
diaphragm
casing
Fig. 21.35a. Expansion and compression of the Fig. 21.35b. Arrangement of strain gauges on
diaphragm of pressure sensors the diaphragm
and a pressure of p2 on the other. Therefore, only the pressure difference p1 − p2 causes
the deflection of the diaphragm. With the absolute pressure sensor shown in Fig. 21.34b,
one side of the diaphragm takes the form of a vacuum chamber.
Figure 21.35 shows a typical arrangement of the strain gauges on the diaphragm.
The left-hand diagram is intended to show that when the diaphragm deflects, zones are
produced which are elongated and others which are compressed. It is in these areas (see the
right-hand diagram) that the four bridge resistors are arranged. They are interconnected in
such a way that the resistances in the bridge arms change inversely. As can be seen from
Fig. 21.36, this arrangement produces a particularly large output signal, while concurring
effects, such as the absolute value of the resistances and their temperature coefficient,
compensate for each other. In spite of this, the output signal is low because of the very
small changes in resistance R. At maximum pressure and at an operating voltage of
Vref = 5 V, it is between 25 and 250 mV, depending on the sensor. The relative change in
resistance is therefore between 0.5% and 5%.
The output signal of a real pressure sensor is made up of a component that is proportional
to the pressure and an undesirable offset component:
Vref
VD
VD R
S = =
p Vref p · R
is the sensitivity, and O is the offset. Both terms provide a contribution that is proportional
to the reference voltage. In order not to obtain too small a signal, the reference voltage
should be as large as possible. However, constraints exist due to the self-heating of the
sensor. Reference voltages between 2 V and 12 V are therefore used.
21.2.2
The Operation of Temperature-Compensated Pressure Sensors
Silicon-based pressure sensors have such high temperature coefficients that some form
of temperature compensation is generally required. The easiest solution is for the user to
employ pressure sensors which already have temperature compensation, provided by the
manufacturer. However, in some cases cost considerations may force users to implement
their own temperature compensation. We shall now describe a possible approach.
There are a number of basic considerations relating to the conditioning of pressure
sensor signals:
1) Although the four bridge resistors in Fig. 21.36 are well matched among each other, their
absolute value exhibits a large tolerance and is also heavily temperature-dependent. For
this reason, the output signals must not be loaded, and an instrumentation amplifier is
normally used to provide gain.
2) Pressure sensors usually have a zero error, which is quite small in absolute terms (e.g.,
± 50 mV); however, comparison with the wanted signal shows that it is usually of
the order of magnitude of the measurement range. A zero adjuster covering the entire
measurement range is therefore required.
3) In general, the sensitivity of a pressure sensor also exhibits significant tolerances (e.g.,
± 30%), so that gain adjustment is additionally required.
4) Zero and gain adjustment must be possible without having to use iteration procedures.
5) Since the wanted signals of a pressure sensor are small, a large amount of amplifica-
tion is required. This results in appreciable amplifier noise. The pressure sensor itself
also produces considerable circuit noise. The bandwidth at the amplifier output must
therefore be limited to the required frequency range of the pressure variations.
1080 21 Sensors and Measurement Systems
Vref
VD A
Vo
A
gain
adj.
Vref
zero
adj.
VZ A
Vref
VD A
Vref Vo
A
A
gain
Vref adj.
zero VZ A 2R2 2 R2
adj.
VZ
Vref
To allow zeroing to be performed before the subtractor containing the gain adjuster,
operational amplifier OA4 has been added in Fig. 21.38. For potential V3 , this gives:
R3 1
V3 = V2 + Vref − VZ
R3 + R 4 2
Thus a voltage of up to ± 21 Vref R3 /(R3 + R4 ) can be added to potential V2 , depending on
the size of VZ . The amplifier zero has not been fixed at 0 V but at Vref by connecting the foot
of voltage divider chain R2 to Vref instead of to ground. This shifts the zero balance point
from zero pressure and zero output voltage to the pressure that corresponds to Vo = Vref
(see Sect. 21.5.1).
An example will serve to illustrate component value selection for the circuit. An air
pressure meter is to deliver an output voltage of 2.5 mV/hPa. An uncalibrated pressure
sensor is to be used. At a supply voltage of Vref = 5 V, it delivers a signal of 10…40 mV/hPa;
its zero error can be as much as ± 50 mV. To calculate the zero adjustment, we choose
R3 = 1 k. At a reference voltage of Vref = 5 V, this then produces the required adjustment
range with R4 = 49 k. The gain must be adjustable between 62.5 and 250, depending on
the sensitivity of the sensor. If R2 = 10 k is specified, a minimum value of 80 (fixed
resistor) and a maximum value of 330 is obtained for R1 . The variable resistor must
therefore have a value of 250 .
To calibrate the circuit, it is first necessary to adjust the zero point. To avoid an iterative
process, the pressure at which no voltage is present at the gain adjuster is selected; that is,
V1 − V3 = 0. This is the case when Vo = Vref /2 = 2.5 V and corresponds to a pressure
of 1,000 hPa. The zero adjuster is therefore set so that Vo is actually 2.5 V at 1,000 hPa.
1082 21 Sensors and Measurement Systems
Vref
V2 V A
Vo
Vref
A
zero Z V2 –Vref
gain
adj. adj.
–Vref
Fig. 21.39. Transfer of the bridge signal to the right-hand bridge arm
R2
Vo = 1 + (VD + Vn ) = A(VD − VO )
R1
Since no voltage is dropped across R1 after the adjustment, its value has no effect on the
adjustment. To calibrate the gain, we select the pressure that is as far away as possible
from the zero point at 1,000 hPa – at the upper or lower end of the desired measurement
range, for example – and adjust the output voltage to the nominal value using R1 . A more
detailed description of sensor calibration is given in Sect. 21.5.
Since the sensor signals are in the microvolt range, it is advisable to use operational
amplifiers with low offset voltages and offset voltage drifts. However, as the bandwidth
requirements are small, operational amplifiers with a low power consumption can he used.
A ready-made instrumentation amplifier can, of course, be used for the subtractor, although
the advantage is not significant: only the four resistors R2 are saved.
The circuit for conditioning the sensor signals can be greatly simplified if a negative
voltage is additionally available or can be generated with a voltage converter. In the circuit
shown in Fig. 21.39, one bridge arm of the pressure sensor is in the negative feedback
path of amplifier OA1. If we imagine V2 = 0, the entire bridge signal VD will therefore
be transferred to the right-hand bridge output and subtraction will no longer be required.
Therefore the simple noninverting amplifier OA2 is all that is required here to provide
gain. For nulling the offset voltage VO of the sensor, voltage V2 = −VO is applied
A lowpass filter can easily be implemented using capacitor C to limit the noise band-
width of the circuit. A second-order lowpass filter can also be implemented by connecting
a second capacitor to ground directly at the bridge output.
21.2.3
Temperature Compensation for Pressure Sensors
By their nature, the doped silicon resistors of a pressure sensor are temperature-dependent.
They are even used for temperature measurement themselves (see Sect. 21.1). The typical
21.2 Pressure Measurement 1083
Fig. 21.40. Resistance and sensitivity of silicon pressure sensors as a function of temperature
R ppm S ppm
TKR = ≈ +1350 , TKS = ≈ −2350
Rϑ K Sϑ K
resistance curve is shown in Fig. 21.40. At room temperature, the resistor’s temperature
coefficient is
R ppm %
TKR = ≈ 1350 = 0.135
R · ϑ K K
In a bridge arrangement of the type employed in pressure sensors, the temperature-induced
resistance variation has no adverse effect, provided that it is the same in all of the resistors
and no load is placed on the output signal. However, a problem arises due to the fact
that the pressure sensitivity of the sensor is also temperature-dependent; its temperature
coefficient is
S ppm %
TKS = ≈ − 2350 = − 0.235
S · ϑ K K
Thus, for a temperature increase of 40◦ C, it has already dropped by 10%, as can
be seen from Fig. 21.40. To prevent this from invalidating the measurement, the gain
must be increased accordingly with temperature. Naturally, this must not be based on the
temperature of the amplifier, but on that of the pressure sensor. The temperature detector
must therefore be incorporated into the pressure sensor. Hence, the obvious solution would
be to perform the temperature compensation in the sensor itself. This can be done by
increasing the reference voltage Vref with temperature in such a way that the reduction in
sensitivity is just compensated:
VD = S · P · Vref + O · Vref = VP + VO
It is usually accepted that the zero point Vo = O · Vref will shift slightly.
Temperature-compensated pressure sensors differ only in the method of tempera-
ture compensation employed. Three widely used methods are illustrated in Fig. 21.41.
In Fig. 21.41a, an NTC thermistor is used to increase the bridge voltage with temperature.
In Fig. 21.41b, the negative temperature coefficient of a diode, of − 2 mV/K, is used. The
arrangement of the transistor in the circuit produces the effect of three diodes. A tem-
perature sensor using the bandgap principle can also be incorporated; the type used in
Fig. 21.41c is the LM 335 (see Fig. 21.18). It operates as shown in Fig. 21.18 and delivers
a voltage of 10 mV/K; that is, about 3 V at room temperature. The interesting feature of
this solution is that the temperature compensation circuit can be simultaneously used for
temperature measurement.
1084 21 Sensors and Measurement Systems
Vref
Vref
.
a An NTC thermistor; for example, b Approximately three diodes; c A bandgap temperature sensor;
in the SDX-series from SenSym for example, in the KP 100 A 1 for example in the LM 335
from Philips from National
. .
Vref
.
.
VB VB
Fig. 21.42. Operation of a pressure Fig. 21.43. Practical implementation of the current
sensor from a current source that source
has a negative internal resistance
Ik = 1 mA Ri = − 7.05 k
21.2 Pressure Measurement 1085
|TKS |
Ri = RB = − 2.35RB
TKR − |TKS |
The circuit shown in Fig. 12.9 is again ideally suited for use as a current source. Figure 21.43
shows how it can be used for temperature compensation in a pressure sensor with a bridge
resistance of RB = 3 k. We select a short-circuit current of Ik = 1 mA. This gives
R1 = Vref /Ik = 2.5 k. The rated operating voltage of the bridge is therefore
VB = |TKS /TKR |RB Ik = (2350/1350) · 3 k · 1 mA = 5.22 V
In order to obtain the component values for the circuit, the required internal resistance
must first be determined:
|TKS |
Ri = RB = − 2.35 · 3 k = − 7.05 k
TKR − |TKS |
If R2 = 10 k is then specified, we obtain
R1
R3 = R2 1 + = 6.45 k
Ri
21.2.4
Commercially Available Pressure Sensors
Figure 21.44 gives some idea of the wide range of pressure sensors available. There are
not only a large number of other manufacturers, but most of the types listed are merely
representative of whole families of sensors. It can be seen from Fig. 21.44 that in addition to
the pressure sensors with a range of 1–2 bar, which are primarily designed for barometers,
there are also types with very much smaller and greater measurement ranges. There are
two designs of pressure sensors: one type measures pressure against atmospheric pressure,
while the other measures the difference in pressure between two ports.
The sensitivity of the sensors appears to vary considerably. The reason for this is the
great diversity of measurement ranges. At full pressure and a nominal supply voltage, they
all deliver a difference signal of 50 – 150 mV. The only exception to this rule concerns those
types that incorporate amplifiers. These provide an amplified, temperature-compensated,
and calibrated output signal. With many types, the zero error is of the order of the entire
measurement range. The types with internal compensation (Fig. 21.41a) perform much
better in this respect, since not only the sensitivity but also the zero point is calibrated by the
manufacturer. Amplifiers with built-in calibration for offset, gain, temperature coefficient
and linearity are found in the family MAX 1450–1458 Maxim.
21.3
Humidity Measurement
Humidity specifies water content. Of particular interest is the water content of air. The
absolute humidity Habs is defined as the amount of water contained in a unit volume of air:
Mass of water g
Habs = ; [Habs ] = 3
Volume of air m
The maximum amount of water that can be dissolved in air is given by the saturation
humidity Hsat :
Hsat = Habs max = f (ϑ)
Its magnitude is heavily dependent on temperature, as shown in Fig. 21.45. When saturation
humidity is reached or exceeded, water condenses: the dew point is reached. Determining
the dew point thus allows Fig. 21.45 to be directly employed to specify the absolute
humidity.
Most of the reactions due to atmospheric humidity, which include such things as phys-
ical well-being, are dependent on the relatire humidity Hrel :
Habs
Habs
Habs
Hrel =
Hsat
which indicates the percentage of the saturation humidity attained. The amount of the
relative humidity can be determined with the aid of Fig. 21.45. If, for example, a dew point
of 25◦ C is established by cooling the air, the absolute humidity Habs = 20 g/m3 . However,
at a temperature of, for example, 55◦ C, the air could take up Hsat = 100 g/m3 of water.
Therefore, at 55◦ C the relative humidity is
Habs 20 g/m3
Hrel = = = 20%
Hsat 100 g/m3
The relationship between relative air humidity and temperature can be obtained directly
from Fig. 21.46.
21.3.1
Humidity Sensors
The example given above shows that the relative humidity can be determined by measur-
ing the ambient temperature and the dew point. Although the dew point can be measured
precisely and no further calibration is required, the cooling equipment required is com-
plex. The sensors commonly used for determining humidity simplify the measurement
by providing an output that is a direct function of the relative humidity – the quantity
generally of interest. They consist of a capacitor with a dielectric whose permittivity is
humidity-dependent.
Figure 21.47 shows this type of device. The dielectric is made from aluminum oxide or
a special plastic foil. One or both electrodes consist of a metal that is permeable to water
vapor. The capacitance characteristic for a typical device is shown in Fig. 21.48. It can
be seen that a specific basic capacitance C0 occurs and that the increase in capacitance
is nonlinear. Within a restricted range, this nonlinearity can be largely eliminated using
a series capacitor. A humidity sensor with a voltage output is the HIH 4000 family from
Honeywell.
porous electrode
CS H rel 1.4
= 1 + 0.4
C0 100%
porous electrode
water-adsorbent
dielectric Hrel
Fig. 21.47. Internal design Fig. 21.48. Sensor capacitance versus relative
principle of a capacitive humidity humidity. Example: No. 2322691 90001 from
sensor Philips
1088 21 Sensors and Measurement Systems
21.3.2
Interfacing Circuits for Capacitive Humidity Sensors
In order to determine humidity, it is necessary to measure the capacitance of the humidity
sensor. This means that any circuit for measuring capacitance could be employed here.
For example, an AC voltage can be applied to the sensor and the flow of current measured,
as shown schematically in Fig. 21.49. Although this method appears simple, it is actually
quite complex, as it requires not only a calibrated AC meter, but also an AC voltage source
with constant amplitude and frequency.
A simpler solution is to incorporate the sensor in an astable circuit, with the sensor
determining its operating frequency or duty factor. Figure 21.50 shows a circuit of this
type. It consists of two multivibrators of the type shown in Fig. 6.29. Multivibrator M1
oscillates at a constant frequency of about 10 kHz if CMOS gates are used. It synchronizes
multivibrator M2, whose on time is determined by humidity sensor CS . The on times of
both multivibrators are of equal duration at zero humidity; but as the humidity increases,
the on time of M2 becomes longer, as shown in Fig. 21.51. From the difference in the
on times, a signal V3 is obtained which is proportional to C and thus approximately
proportional to the humidity. The lowpass filter at the output averages the signal.
A circuit that provides a considerably higher degree of accuracy is shown in Fig. 21.52.
Here, the capacitance of the humidity sensor is determined in accordance with the definition
of capacitance CS = Q/V . Capacitor CS is initially charged to Vref and then discharged
via the summing point. The average current flowing is
I S = Vref · f · CS
V2
V1
V3
V2
V1
Vo V3
Fig. 21.50. Determining the increase in Fig. 21.51. Output signal resulting
capacitance by measuring the increase in oscillation from the difference in switching
period Gartes: CMOS; for example, the CD 4001 times
21.3 Humidity Measurement 1089
Is
Vref Vo
zero
gain
Vref Vo
VZ
VG
zero
gain
Vref Vo
Fig. 21.54. Nulling and gain adjustment for a humidity sensor using potentiometers
the result. This benefit is lost if resistors are used, as in normal SC techniques; this is also
evident from the circuit in Fig. 21.52. The LTC 1043 from Linear Technology is particularly
suitable for implementing the switches, since it not only contains four changeover switches
but also a clock generator that drives the switches.
It is desirable to adjust the zero point and full-scale span with potentiometers rather
than trimming capacitors. A method that can be employed without losing the benefits of
pure SC technology is shown in Fig. 21.54. Here, the current flowing through CT or CG
is varied by the voltage tapped off via the potentiometers. In order to insure that this does
not increase the charging time, capacitors C2 and C3 are additionally employed and are
selected to be large in relation to CT or CG .
21.4
The Transmission of Sensor Signals
There is often a considerable distance, or an environment with high levels of interference,
between the sensor and the point at which the signal is to be evaluated. For this reason,
special action must be taken in such cases to insure that the measured values are not
impaired by external effects. Depending on the field of application and the protection class
required, a distinction is drawn between electrical signal transmission and a more complex
method that employs electrically isolated transmission.
21.4.1
Electrical (Direct-Coupled) Signal Transmission
For long-distance transmission the line resistance RL cannot be ignored. Even small cur-
rents necessary for the operation of the sensor result in such large voltage drops that impair
the measured value. This problem can be solved by transmitting the measurement signal
for evaluation via two additional lines in which no current flows. The measured quan-
tity is then obtained by employing an instrumentation amplifier such as the one shown in
Fig. 21.55. The voltage drop in the measurement circuit merely causes a common-mode
signal VCM = I0 RL , which disappears after subtraction.
21.4 The Transmission of Sensor Signals 1091
VS Vo
Vo=VS
VS Vo =VS
One wire can be dispensed with by specifying that the resistance in all the circuits must
be identical, giving the three-wire method shown in Fig. 21.56. The voltage drop across
RL can be eliminated in this case by formulating the expression
Vo = 2V1 − V2 = 2VS + 2I0 RL − VS − 2I0 RL = VS
If the sensor signals are small as, for example, in pressure sensors and thermocouples,
they must be preamplified near the sensor before the signal is transmitted over a long
circuit. This principle is illustrated in Fig. 21.57. Although the output signal is affected by
the voltage drop across RL , if gain A is selected to be sufficiently large, this error becomes
insignificant. It can be eliminated altogether if the four-wire method of Fig. 21.55 is
additionally employed. However, one more instrumentation amplifier on the receiver side
would then be required.
It is simpler in this case to convert the sensor signal into a current proportional to
it. A current is unaffected by the line resistances. The principle is shown in Fig. 21.58.
sensor VS Vb
Vo VS
Fig. 21.57. A preamplifier for small sensor signals reduces signal transmission errors
1092 21 Sensors and Measurement Systems
gf VS
sensor VS Vb
Vo
sup
Fig. 21.58. A preamplifier with a current output at the sensor eliminates errors in signal
transmission. An example of a voltage-controlled IC current source: XTR 110 from Texas
Instruments
Vo = IS R1 = gf VS R1 = AVS
sensor VS gf VS
Vb
IS gf VS
Isup Isup IS
Vo
sup o Z
Vref
A
Z
A
VS A
A V1 VS
sensor current
loop
Fig. 21.60. Internal design of a current loop transmitter, using the AD 693 from Analog Devices
with a resistance bridge as anexample for the sensor
R3 Vref R 2 VS
Io = IZ + IS = + 1+
R4 RI R1 RI
end. For unipolar signals, the zero point is set to 4 mA. For bipolar signals it is set to
12 mA, giving a control range of ± 8 mA. If R1 = 250 is selected, as is usual, both cases
produce voltages of Vo =1–5 V at the receiver end.
Figure 21.60 shows the internal structure of a sensor interfacing circuit with a current
loop output. The core of the circuit is a precision current source consisting of transistor
T, operational amplifier OA1, and shunt resistor RI . Current Io assumes a value such
that the input voltage difference of OA1 becomes zero. If R4 is omitted for the sake of
simplicity, this will occur when the voltage drop is Io RI = V1 . Resistor R4 is merely
used to add the zero current of IZ = 4 mA or 12 mA. The sensor signal is conditioned
by the instrumentation amplifier and then controls the current source. The clever feature
of the arrangement shown in Fig. 21.60 is that the load currents for the four operational
amplifiers, the reference voltage source, and any sensors connected to it also flow through
shunt resistor RI . Their sum is thus taken into account for current measurement. Transistor
T then carries only the current lacking in the wanted output current. To insure that the
arrangement operates even with the smallest loop current of Io = 4 mA, the sum of the
load currents Isup must be less than 4 mA. In commercially available ICs, the internal
current drain is less than 1 mA, so that up to 3 mA is still available for operating the sensor.
A positive side-effect of the method described above is that faults can easily be detected.
If the loop current is less than 4 mA, a fault has occurred; for example, a bypass or an open
circuit.
21.4.2
Electrically Isolated Signal Transmission
For transmission over long distances, and in environments with heavy electrical interfer-
ence, noise signals of such magnitude can occur that the methods of signal transmission
described above will not provide an adequate signal-to-noise ratio. In such cases, there is
1094 21 Sensors and Measurement Systems
interfacing Vo
sensor transmitter receiver
circuit
local central
supply supply
V
upper end of range
only one viable solution: the use of fiber-optic transmission. This is affected neither by
electrostatic nor electromagnetic fields, and can handle almost any potential difference.
Figure 21.61 shows the principle involved in the optical transmission of sensor signals.
However, analog signals are not usually transmitted over optical fibers, since the atten-
uation of the optical transmission paths is not well defined and is also subject to temperature
variations and aging. The sensor signal is therefore converted into a serial digital signal
in the transmitter. There are various ways of doing this. In the case of voltage–frequency
conversion, the frequency is a linear function of the voltage; the duty factor of the output
signal is a constant 1 : 1. With voltage to duty factor conversion, the frequency is constant
but the duty factor is a linear function of the voltage. Figure 21.62 shows the principles of
the two methods. They are particularly useful in cases in which an analog signal is to be
recovered at the receiver end.
Digital processing of the received signals is also possible by measuring the frequency
or the duty factor digitally. However, if high accuracy is required, it is preferable to digitize
the signal using a commercially available A/D converter at the sensor end and to transmit
the result serially word by word.
21.5
Calibration of Sensor Signals
Some sensors are manufactured to such tight tolerances that no calibration is required,
provided that the interfacing circuit also employs components with sufficiently tight toler-
21.5 Calibration of Sensor Signals 1095
ances. In this case, the sensor can be replaced without recalibration becoming necessary.
However, this unfortunately applies only to a few temperature sensors. The general rule is
that recalibration is always necessary when a sensor is replaced. Where a high degree of
accuracy is required, regular recalibration may even be necessary.
21.5.1
Calibration of the Analog Signal
For an explanation of the calibration process without reference to the specific characteristics
of the sensor, we shall consider the calibration circuit to be separated from the interfacing
circuit of the sensor, as shown in Fig. 21.63. We shall assume that the sensor signal is a
linear function of the physical quantity G or is linearized by the interfacing circuit. The
input voltage of the calibration circuit can then be expressed in the form
Vi = a
+ m
G (21.6)
The calibrated signal should generally be proportional to the measured quantity, in accor-
dance with
Vo = mG (21.7)
Figure 21.64 shows the voltage characteristic for a temperature measurement as an ex-
ample. The calibration circuit must allow correction of the zero point and the gain. An
important constraint is that calibration should be possible without iteration; in other words,
there must be a procedure whereby one setting does not affect the other. This is possible
with the arrangement shown in Fig. 21.63. Its output is:
Vo = A(Vi + VZ ) (21.8)
Using (21.6) and (21.7), comparison of the coefficients produces the following calibration
conditions:
Zero point: VZ = −a
Gain: A = m/m
For zero adjustment, the physical quantity G = G0 , associated with output value Vo = 0
is applied to the sensor. The output voltage is then adjusted to Vo = 0 by varying VZ . This
adjustment is independent of any setting of gain A: the only requirement is that A = 0. In
Fig. 21.64, zeroing causes a parallel shift of the input characteristic through the origin.
For gain adjustment, we apply a physical quantity G1 and calibrate the gain A so
that the nominal value of the output voltage Vo1 = mG1 is produced. In Fig. 21.64, this
corresponds to a rotation of the shifted input characteristic until it corresponds to the
VZ
interfacing Vi Vi VZ
sensor
circuit Vo Vi VZ
Fig. 21.63. Basic arrangement for the calibration of sensor signals by adjusting the zero point VZ
and the gain A
1096 21 Sensors and Measurement Systems
V
Vo1 Vo
Vi
. .
Fig. 21.64. Illustration of the calibration process – first nulling, and then gain adjustment – using
the example of a clinical thermometer
required function. Nulling is not affected by this, since gain adjustment merely involves
varying factor A in Eq. (21.8).
It can be seen that the reverse sequence does not result in noniterative adjustment. It
is therefore absolutely essential for the zero adjuster to precede the gain adjuster in the
signal path. The circuit arrangement in Fig. 21.63 is therefore invariable.
Calibration will now be further explained using the clinical thermometer example from
Fig. 21.64. For zeroing, the sensor is set to temperature ϑ = 0◦ C and the output voltage
adjusted to Vo = 0 by varying VZ . This is the case for voltage
VZ = −a
= +0.5 V
To calibrate the gain, the second calibration value is applied to the sensor – for example,
G1 = ϑ1 = 40◦ C – and gain A is adjusted until the wanted value of the output voltage
100 mV
Vo1 = mG1 = ◦C
· 40◦ = 4 V
V
Vo1
zero shift Vd
Vo2
Vi VZ
gain adjustment
Vo
zero adjustment VZ
Vi VZ
Vi
Fig. 21.65. Iteration-free adjustment procedure with two nonzero calibration points G1 , G2
VZ Vd
interfacing Vi Vi VZ Vi VZ
sensor
circuit
Vo
Vi VZ VdV
Fig. 21.66. Arrangement for noniterative calibration of sensor signals if none of the calibration
points is zero
the adjustment at the output. Calibration value Vo2 = Vd must then be obtained here. Since
the output voltage of the amplifier after adjustment is precisely zero, it is independent of
the value of A.
The second calibration value is then applied and gain A is adjusted as previously
described. The shifted input characteristic in Fig. 21.65 will rotate until it possesses the
correct slope. The calibrated output signal is then obtained by voltage addition at the output
side.
An example of the practical implementation of a calibration circuit is shown in
Fig. 21.67. The input voltage and the voltage of the zero adjuster are added at the summing
point of OA1. The gain is set at the feedback resistor. The fixed-value resistor is used to
limit the adjustment range; it also prevents the gain being set to zero. Amplifier OA2 effects
the output-side zero shift for the first calibration point. Since the amount of shift can be
selected by a suitable choice of R3 , no adjustment is required here.
The adjustment procedure will now be further explained using the clinical thermometer
example. Let the input and output characteristics be of the form
50 mV 100 mV
Vi = − 0.5 V + ◦C
ϑ; Vo = ◦C
ϑ
+Vref
VZ –Vref
VdvV
–Vref
Vi
A Vo
A
apply a temperature of ϑ2 = 30◦ to the sensor and adjust the output voltage to Vo2 = 3 V.
The voltage required for this purpose is
50 mV
VZ = −Vi1 = +0.5 V − ◦C
· 30◦ C = − 1 V
The output voltage of OA1 is then zero and the value that happens to be set for A has no
effect on the zero adjustment. To calibrate the gain, the other calibration point of ϑ1 = 40◦ C
is set and the output voltage adjusted to Vo1 = 4 V. This is obtained for a gain of
m 100 mV/◦ C
A = = = 2
m
50 mV/◦ C
With R1 = 10 k, the calibrated condition produces a value of R2 = 20 k.
21.5.2
Computer-Aided Calibration
If we intend to undertake further processing of a sensor signal by a microcomputer, it
is advantageous to calibrate the sensor with the microcomputer as well. As can be seen
from Fig. 21.68, this not only obviates the need for an analog calibration circuit, but also
allows calibration to be performed more easily and improves accuracy and stability. For
calibration, let us assume that the number N at the output of the A/D converter shown in
Fig. 21.69 is a linear function of measured quantity G:
N = a + bG (21.9)
N
N
N
The calibration coefficients a and b are determined from two calibration points
(G1 , N1 ) and (G2 , N2 )
by solving the system of equations
N1 = a + bG1 and N2 = a + bG2
for a and b:
N2 − N 1
b = (21.10)
G2 − G 1
and
a = N1 − bG1 (21.11)
To compute the appropriate physical quantity from a measured value N , (21.9) must be
solved for G
G = (N − a)/b (21.12)
For practical calibration, the intended calibration values – for example, G1 = 30◦ C and
G2 = 40◦ C – are stored in a table. They are then applied in turn to the sensor and the
microcomputer is instructed – for example, via push-buttons – to read in the appropriate
measured values – for example, N1 = 1,000 and N2 = 3,000 – and store them in the table.
Using these entries, a program in the microcomputer can compute the calibration values
in accordance with (21.10) and (21.11 and store them in the table as well:
b = 200/◦ C and a = − 5,000
Calibration is now complete. The analysis program can then compute values Gi in accor-
dance with (21.12). For a measured value of N = 2,360, the example gives a temperature of
N −a 2360 + 5000
G = = = 36.8◦ C
b 200/◦ C
For computer calibration, the hardware characteristic is therefore taken as given, its equa-
tion is formulated, and it is then used to map measured values Ni onto physical quantities
Gi . It is therefore unnecessary to shift or rotate characteristic curves as in analog cali-
bration. Any calibration point can be chosen: calibration is always noniterative, since the
calibration points are determined by solving a system of equations.
A particularly difficult problem is posed by the calibration of sensors whose signals
are not merely a function of the quantity sought, but of a second value as well. The most
prevalent form of such unwanted dual dependence occurs in the temperature-dependence
1100 21 Sensors and Measurement Systems
of sensor signals. Pressure sensors are an example of this, and they will be used here to
illustrate the procedure involved. Measured value N comprises four components:
N = a + bp + cϑ + dϑp (21.13)
where p is the pressure, ϑ is the temperature, a is the zero error, b is the pressure sensitivity,
c is the temperature coefficient of the zero point, and d is the temperature coefficient of
the sensitivity.
To determine the four coefficients a, b, c, and d, four calibration measurements are per-
formed, each of which differs in one quantity,
N11 = a + bp1 + cϑ1 + dp1 ϑ1 N21 = a + bp2 + cϑ1 + dp2 ϑ1
N12 = a + bp1 + cϑ2 + dp1 ϑ2 N22 = a + bp2 + cϑ2 + dp2 ϑ2
and we obtain
N22 + N11 − N12 − N21 N22 − N12
d = b = − dϑ2
(p2 − p1 )(ϑ2 − ϑ1 ) (p2 − p1 )
(21.14)
N22 − N21
c = − dp2 a = N22 − bp2 − cϑ2 − dp2 ϑ2
ϑ2 − ϑ 1
Calibration is now complete and the pressure can then be computed from (21.13):
N − a − cϑ
p = (21.15)
b + dϑ
To give an example of how calibration is performed, let us assume that the four calibration
values required are to be obtained at pressures of p1 = 900 mbar and p2 = 1, 035 mbar,
and at temperatures of ϑ1 = 25◦ C and ϑ2 = 50◦ C. This produces the measured values
shown in Fig. 21.70. Using Eq. (21.14), we obtain from the calibration coefficients:
1
a = − 1375 b = 5.18
mbar
1 1
c = 1.71 ◦ d = − 0.0119
C mbar · ◦ C
This calibration is very precise, since it not only calibrates the zero point and gain but also
takes account of the temperature coefficients of the sensitivity and zero point. This method
allows low-cost, uncalibrated sensors to be used for performing precision measurements.
For pressure measurement, we use (21.15). If, for example, a measured value of N =
3351 is obtained for a temperature of ϑ = 15◦ C, this gives a pressure of
N − a − cϑ 3351 + 1375 − 1.71 · 15
p = = mbar = 940 mbar
b + dϑ 5.18 − 0.0119 · 15
Obviously, a calibrated temperature measurement is required to enable proper account to
be taken of the influence of temperature. The temperature measurement will, of course,
ϑ1 = 25◦ C ϑ2 = 50◦ C
p1 = 900 mbar N11 = 3061 N12 = 2837
p2 =1035 mbar N21 = 3720 N22 = 3456
temperature interfacing
sensor circuit
ADC and micro-
display
MUX computer
pressure interfacing
sensor circuit
Fig. 21.71. An arrangement for computer-aided temperature and pressure calibration and
measurement
also be calibrated by computer, in the same way as described above. The resulting block
diagram is shown in Fig. 21.71. The pressure and temperature sensor signals are condi-
tioned by the interfacing circuits and are fed to an analog-to-digital converter with a built-in
multiplexer. The microcomputer receives the measured values N and computes from them
the calibration coefficients during calibration and the measured quantities during normal
operation. To enable sufficient accuracy to be attained, the A/D converter must have an
accuracy of at least 12 bits. As A/D converters possessing this degree of accuracy are not
available in most single-chip microcomputers, separate A/D converters which also contain
an input multiplexer, have to be employed.
The sensor signal processor MSP 430 family from Texas Instruments is specifically
tailored for the evaluation of sensor signals. In addition to the 14 bit AD converter with
multiplexer, it contains a driver for a ten-digit liquid crystal display (LCD). It represent a
particularly simple solution in cases where the measured values are to be displayed only.
Chapter 22:
Electronic Controllers
22.1
Underlying Principles
The purpose of a controller is to bring a physical quantity (the controlled variable X) to
a predetermined value (the reference variable W ) and to hold it at this value. To achieve
this, the controller must counteract the effect of disturbances in a suitable way.
The basic arrangement of a simple control circuit is shown in Fig. 22.1. The controller
influences the controlled variable X by means of the correcting variable Y , so that the error
signal W − X is as small as possible. The disturbances acting on the controlled system
(plant) are represented formally by the disturbance variable Z, which is superimposed on
the correcting variable. In what follows, we shall assume that the controlled variable is a
voltage and that the system is electrically controlled. Electronic controllers can then be
employed.
In the simplest case, such a controller is a circuit that amplifies the error signal W − X.
If the controlled variable X rises above the reference signal W , the difference W − X
becomes negative. The correcting variable Y is thereby reduced by a factor defined by the
amplifier gain. This reduction counteracts the increase in the controlled variable; that is,
there is negative feedback. At steady-state, the remaining error signal is smaller the larger
is the gain AC of the controller. It can be seen from Fig. 22.1 that, for linear systems,
Y = AC (W − X) and Y = AS (Y + Z) , (22.1)
where AS is the gain of the controlled system. Hence, the controlled variable X is:
AC AS AS
X = W+ Z. (22.2)
1 + A C AS 1 + A C AS
It is now obvious that the response of the control system to a reference input, ∂X/∂W ,
approaches unity more closely the greater is the loop gain:
∂X
g = AC AS = . (22.3)
∂(W − X)
The response to a disturbance, ∂X/∂Z, approaches zero more closely the larger is the gain
AC of the controller.
controlled variable
disturbance controlled system = actual value
correcting
variable
controller reference signal
AC = desired value
error signal
It must be pointed out, however, that there is a limit to the value of the loop gain g
since, if the gain is too large, the unavoidable phase shifts within the control loop give rise
to oscillations. This problem has already been discussed in connection with the frequency
compensation of operational amplifiers. The objective of control engineering is to obtain,
despite this restriction, the smallest possible error signal and good transient behavior at the
same time. For this reason, an integrator and a differentiator are added to the proportional
amplifier, and the P-controller is thus turned into one that exhibits PI, or even PID, action.
The electronic realization of such circuits is dealt with below.
22.2
Controller Types
22.2.1
P-controller
A P-controller (a controller with proportional action) is a linear amplifier. Its phase shift
must be negligibly small within the frequency range in which the loop gain g of the control
system is larger than unity. For example, an operational amplifier with resistive feedback
is a P-controller of this kind.
To determine the maximum possible proportional gain AP , we consider the Bode plot of
a typical controlled system, represented in Fig. 22.2. The phase lag is 180◦ at the frequency
f = 3.3 kHz. The negative feedback then becomes a positive feedback. In other words,
the phase condition (14.3) for a self-sustaining oscillation is fulfilled. The value of the
proportional gain AP determines whether or not the amplitude condition of Eq. (14.2) is
also fulfilled. For the example in Fig. 22.2, the gain |AS | of the system at 3.3 kHz is about
0.01 =7 −40 dB. If we select AP = 100 = 7 40 dB, the loop gain at this frequency would
P-controller
system
+ controller
phase margin
If the controller gain is increased to obtain a smaller deviation, the transient response
suffers. A proportional gain of any magnitude can only be set for systems that behave
like first-order lowpass filters, because the phase margin is then greater than 90◦ at any
frequency.
22.2.2
PI-Controller
The previous section has shown that, for reasons of stability, the gain of a P-controller must
not be too large. One way of improving the control accuracy is to increase the loop gain
at low frequencies, as illustrated in Fig. 22.4. In the vicinity of the critical frequency fc ,
the frequency response of the loop gain thereby remains unchanged, so that the transient
behavior is not affected. However, the remaining error signal is now zero, since
lim |g| = ∞ .
f →0
PI-controller
fc fc
system
system + controller
(loop gain |g|)
system
system + controller
proportional Vi
amplifier
Vi Vo
integrating
amplifier
Vi
Fig. 22.5. Block diagram of the Fig. 22.6. Bode plot of the PI-controller
PI-controller
To determine the cutoff frequency f1 , we calculate from Fig. 22.5 the complex con-
troller gain
1 1
AC = AP + = AP 1 + .
j ωτI j ωτI AP
Hence,
ω1 1
AC = AP 1 + where ω1 = 2πfI = (22.6)
jω τI AP
A PI-controller can also be realized using a single operational amplifier, the appropriate
circuit being shown in Fig. 22.7. Its complex gain is
R2 + 1/j ωCI R2 1
AC = − = − 1+ . (22.7)
R1 R1 j ωCI R2
By comparing coefficients with Eq. (22.6), we obtain the controller parameters as
R2 1
AP = − and fI = . (22.8)
R1 2πCI R2
It is quite simple to design a PI-controller if we make use of the fact that the I-component
does not change the phase margin. The size of the P-component (the gain) is therefore
retained; that is, in the example given, fc = 700 Hz and AP = 7.
In order to insure that the I-component does not reduce the phase margin, it is necessary
to select fI fc . However, it is not advisable to select it to be unnecessarily low, as it
will then take longer for the integrator to bring the deviation to zero. An optimum value is
fI = 0.1 fc . The I-component then reduces the phase margin by less than 6◦ . Figure 22.4
has been plotted using these parameters. The appropriate transient behavior of the error
signal is shown by the oscillogram of Fig. 22.8. It is obvious from the lower trace that,
with these optimum parameters, the PI-controller settles to a zero error signal in the same
time interval that a purely P-action controller requires to adjust to a relative error signal of
1/(1 + g) = 1/8 = 12.5%.
Fig. 22.8. Error signal for the P-controller Fig. 22.9. Error signal of the PI-controller: fI
(upper) and the PI-controller with an optimum too small (upper) and fI too large (lower)
value of fI (lower)
The effect of a less than optimum value of fI is demonstrated by the oscillogram shown
in Fig. 22.9. For the upper trace, fI is too small: the settling time is increased. For the
lower trace, fI is too large: the phase margin is reduced.
22.2.3
PID-Controller
By connecting a differentiator in parallel as shown in Fig. 22.10, a PI-controller can be
extended to become a PID-controller (a controller with proportional-integral-derivative
action). Above the differentiation cutoff frequency fD , the circuit behaves like a differen-
tiator. The phase shift rises to +90◦ , as can be seen from the Bode plot in Fig. 22.11. This
phase lead at high frequencies can be used to partially compensate for the phase lag of the
controlled system in the vicinity of fc , allowing a higher proportional gain, so that a higher
gain crossover frequency fc is obtained. The transient behavior is thereby speeded up.
The selection of the controller parameters will again be illustrated for our example
system. In the first step, we raise the proportional gain AP until the phase margin is
only about 15◦ . In this case, we can infer from Fig. 22.12 that AP = 50 = 7 34 dB and
fc ≈ 2.2 kHz, as against 700 Hz for the PI-controller. If the differentiation cutoff frequency
is now chosen to be fD ≈ fc , the phase lag of the controller at frequency fc is approx.
Vi
differentiating
amplifier
proportional Vi
amplifier
Vi Voo
integrating
amplifier
Vi
Fig. 22.10. Block diagram of the Fig. 22.11. Bode plot of the
PID-controller PID-controller
22.2 Controller Types 1109
PID-controller
fc
fc
system + controller
system (loop gain)
system
system + controller
+45◦ ; in other words, the phase margin is increased from 15◦ to 60◦ and we obtain the
desired transient behavior.
To determine the integration cutoff frequency fI , the same principles apply as for the
PI-controller; that is, fI ≈ 10 1
fc . This results in the frequency response of the loop gain
shown in Fig. 22.12.
The reduction in settling time is illustrated by a comparison of the oscillograms for a
PI- and a PID-controller, shown in Fig. 22.13.
In order to implement a PID-controller, we must first determine the complex gain from
the block diagram shown in Fig. 22.10:
1 ω ωI
AC = AP + j ωτD + = AP 1 + j − (22.9)
j ωτI ωD ω
where
AP 1
fD = and fI = . (22.10)
π τD 2πAP τI
A circuit that has the frequency response of (22.9) can also be realized using a single
operational amplifier; this is shown in Fig. 22.14. Its complex gain is given by
R2 CD 1
AC = − + = j ωCD R2 + .
R1 CI j ωCI R1
CD R2
Hence, with ,
CI R1
R2 1
AC = − 1 + j ωCD R1 − . (22.11)
R1 ωCI R2
1110 22 Electronic Controllers
Vi Vo
A comparison of the coefficients with those in (22.9) yields the controller parameters
R2 1 1
AP = − , fI = , fD = . (22.12)
R1 2πCI R2 2π CD R1
22.2.4
The PID-Controller with Adjustable Parameters
To determine the different controller parameters, we assumed that the parameters of the
controlled system were known. However, these data are often difficult to measure, particu-
larly for very slow systems. It is therefore usually better to establish the optimum controller
parameters by experiment. A circuit is then required which allows independent adjustment
of the controller parameters AP , fI , and fD . It can be seen from Eq. (22.12) and Eq. (22.10)
that this condition can be fulfilled neither by the circuit shown in Fig. 22.14 nor by that
in Fig. 22.10 since, for any variation in AP , the cutoff frequencies fI and fD will also
change.
For the circuit shown in Fig. 22.15, however, the parameters are decoupled and can
therefore be adjusted independently. The complex gain of the circuit is
RP 1
AC = 1 + j ωCD RD − . (22.13)
RI ωCI R1
Comparing coefficients with Eq. (22.9) gives the controller parameters:
RP 1 1
AP = , fI = , fD = . (22.14)
RI 2πCI RI 2π CD RD
Controller optimization is again illustrated with reference to our example system. In
the first step, switch S is closed to render the integrator inactive. Resistor RD is made zero;
that is, the differentiator does not contribute to the output signal. The circuit is therefore a
purely P-action controller.
We now apply a square wave to the reference signal input and record the transient
behavior of the controlled variable X. Starting from zero, AP is increased until the step
response is only slightly damped, as in the upper trace of Fig. 22.16. The step response
22.2 Controller Types 1111
Vi
A
A Vo
is then that obtained for the system shown in Fig. 22.12, for a phase margin of 15◦ and
without derivative or integral controller action.
In the second step, the differentiation cutoff frequency fD is lowered from infinity by
increasing RD , and is adjusted to a value for which the desired damping is achieved (see
the lower trace in Fig. 22.16).
In the third step, we consider the transient behavior of the error signal W − X. After
opening switch S, the integration cutoff frequency fI is increased until the settling time
is at a minimum. The appropriate oscillograms have already been shown in Figs. 22.8
and 22.9.
The great advantage of this optimization method is that the optimum controller param-
eters represented by Fig. 22.12 are obtained immediately and without iteration. Using the
controller parameters found in this way, the simple PID-controller shown in Fig. 22.14 can
be designed.
The oscillation test – that is, the occurrence of a just slightly damped oscillation
(Fig. 22.16) – also allows calculation of all the data required for designing the PID con-
troller: the oscillation frequency is the critical frequency fosc = 1/Tosc = fc . The gain
for which oscillation occurs yields the P-gain ACosc = AP . The differentiation cutoff fre-
quency is selected to be equal to the oscillation frequency fD = fosc , and the integration
cutoff frequency to be equal to one-tenth of the oscillation frequency: fI = 10
1
fosc . Taken
together, these parameters produce the design parameters for a PID controller:
22.3
Control of Nonlinear Systems
22.3.1
Static Nonlinearity
We have assumed so far that the equation for the controlled system in Fig. 27.1 is
X = AS Y
in other words, that the controlled system is linear. For many systems, however, this
condition is not fulfilled, so that in general
X = f (Y ) .
For small changes about a given operating point X0 , each system can be considered to
be linear as long as its transfer characteristic is continuous and differentiable in the vicinity
of this point. In such cases, the derivative
dX
aS =
dY
is used, so that for small-signal operation
x ≈ aS y ,
where x = (X − X0 ) and y = (Y − Y0 ). For a fixed point of operation, the controller
can now be optimized as described above. However, a problem arises if larger changes in
the reference variable are allowed: since the incremental system gain aS is dependent on
the actual point of operation, the transient behavior varies as a function of the magnitude
of W .
This can be avoided by providing linearity; that is, by connecting a function network
of Sect. 11.7.5 on p. 750 in front of the controlled system. The corresponding block
diagram is shown in Fig. 22.17. If the function network is used to implement the function
Y = f −1 (Y
), we obtain the required linear system equation
X = f (Y ) = f [f −1 (Y
)] = Y .
If, for instance, the system exhibits exponential behavior,
X = AeY ,
the function network must be a logarithmic one, that has the characteristic
Y
Y = f −1 (Y
) = ln .
A
22.3 Control of Nonlinear Systems 1113
linearized system
linearization non-linear
system controlled
correcting
variable variable
controller
linearized AC error signal reference
correcting variable variable
22.3.2
Dynamic Nonlinearity
A different kind of nonlinearity of a controlled system may arise as the result of the
rate of change of some quantity in the system being limited to a value that cannot be
raised by increasing the correcting variable. We have encountered this effect in operational
amplifiers, in the form of limitation of the slew rate. For large input steps and controllers
with integral action, the above effect leads to large overshoots that decay only slowly.
The reason for this is as follows. For optimized integral action of the controller and
for a small voltage step, the integrator reaches its steady-state output voltage at the precise
instant at which the error signal becomes zero. For double the input step of a linear system,
the rate of change in the system, as well as that of the integrator, would double. The
increased reference value would therefore be reached within exactly the same settling
time.
For a system that has a limited slew rate, only the rate of change of the integrator
is doubled, whereas that of the system remains unchanged. This results in the controlled
system reaching the reference value very much later and the integrator overshooting. The
controlled variable therefore greatly exceeds the reference value and the decay takes longer
the further the integrator output voltage is from the steady-state value. The decay time
constant for this nonlinear operation therefore becomes larger for increasing input steps.
The effect is avoided by increasing the integration time constant (i.e., reducing fI )
until no overshoot is incurred for the largest possible input step. However, this results in
considerably prolonged settling times for small-signal operation (see the lower trace in
Fig. 22.9).
Vi
A
A Vo
Fig. 22.19. Slew rate limiter for the reference variable. Resistors R2 , R3 limit the gain of OA1 and
provide additional frequency compensation
Steady-state output voltage: Vo = −Vi
dVo V max
Maximum slope of output voltage: =
dt RC
A much more effective measure is to limit the slew rate of the reference variable to
the maximum slew rate of the controlled system. This insures linear operation throughout,
and the overshoot effect is thus avoided. This will not increase the settling time for large
reference signals, since the controlled variable cannot change any faster anyway. This is
illustrated by the oscillogram shown in Fig. 22.18.
In principle, a lowpass filter could be used to limit the slew rate, but this would also
reduce the small-signal bandwidth. A better solution is shown in Fig. 22.19. If a voltage
step is applied to the circuit input, amplifier OA 1 saturates at the output limit V max . The
output voltage of OA 2 therefore rises at the rate
dVo V max
=
dt RC
until it reaches the values –Vi determined by the overall feedback. A square-wave volt-
age would therefore be shaped into the required trapezoidal voltage. The signal remains
unchanged if the rate of change of the input voltage is smaller than the predetermined
maximum. The small-signal bandwidth is therefore not affected.
22.4
Phase-Locked Loop
A particularly important application of feedback control in communications systems is
the phase-locked loop (PLL). Its purpose is to control the frequency f2 of an oscillator in
such a manner that it is the same as the frequency f1 of a reference oscillator, and to do
this so accurately that the phase shift between the two signals remains constant. The basic
arrangement of such a circuit is illustrated in Fig. 22.20.
The frequency of the voltage-controlled oscillator (VCO) can be varied by means of
the control voltage Vf , according to the relationship
f2 = f0 + kf Vf . (22.15)
Such voltage-controlled oscillators are described in Chap. 14. For low frequencies, the
second-order differential equation circuit of Sect. 14.4 or the function generators of
Sect. 14.5 can be employed. For higher frequencies, the emitter-coupled multi-vibrator
of Fig. 6.21 is more suitable, or any LC oscillator if a varactor diode is connected in paral-
22.4 Phase-Locked Loop 1115
lel with the capacitor of the oscillating circuit. However, the linear relationship of (22.15)
then holds only for small variations around the point of operation, f0 , as the incremental
control constant (the VCO “gain”) kf = df2 /dVf is dependent on the operating point.
The phase detector produces an output voltage that is defined by the phase angle ϕ
between the tracking oscillator voltage V2 and the reference oscillator voltage V1 :
Vϕ = kϕ · ϕ .
The integrating property of the controlled system is of particular interest. If frequency f2
deviates from the reference frequency f1 , the phase shift ϕ will increase proportionally
with time, and without limit. The error signal in the closed loop therefore rises, even for
a finite controller gain, until both frequencies are exactly the same. The remaining error
signal of the frequency is thus zero.
The remaining error signal of the phase shift, however, does not usually become zero.
From Fig. 22.20, we deduce that Vα − Vϕ = Vf /AC ; hence
f1 − f0
α−ϕ = , (22.16)
A C kf kϕ
where f0 is the VCO frequency for Vf = 0. If it is important not only to keep the phase
shift constant but also to hold it precisely at a predetermined value of
α = Vα /kϕ = −ϕ ,
a PI-controller must be used for which AC (f = 0) = ∞. In many applications it is
sufficient to control for identical frequencies (f1 = f2 ) – that is, for a constant phase shift
(the angle α being unimportant) – so that the control input Vα can be omitted. Voltage Vϕ
is then the error signal.
To determine the controller parameters, the frequency response of the system must
be known. As mentioned before, the phase detector exhibits integral behavior, so that the
phase shift is given by
t t t
ϕ = ω2 dt˜ − ω1 dt˜ = ωdt˜ , (22.17)
0 0 0
where t˜ is a dummy time variable of integration. To determine the frequency response of the
controlled system, we modulate frequency ω2 sinusoidally with a modulating frequency
ωm around the center frequency ω1 . Hence
>
ω(t) = ω cos ωm t .
By inserting this in (22.17), we obtain:
1116 22 Electronic Controllers
>
ω
ϕ(t) = · sin ωm t .
ωm
Taking the phase lag of 90◦ into account, we obtain, in complex notation,
ϕ 1
= , (22.18)
ω j ωm
which is the equation for an integrator. With the constants kf and kϕ , the complex gain of
the controlled system is then
V ϕ 2πkf kϕ kf kϕ
AS = = = . (22.19)
Vf jωm jfm
As will be seen later, the measurement of the phase shift involves a certain delay and the
factor kϕ is therefore complex; in other words, the order of the system is raised.
The behavior of a phase-locked loop generally depends on the type of phase detector
used. The most important circuits will now be discussed.
22.4.1
Sample-and-Hold Circuit as a Phase Detector
The phase angle ϕ between two voltages V1 and V2 can, for example, be measured with
a sample-and-hold circuit by sampling the instantaneous value of V1 at the rising edge of
V2 . For this purpose, V2 in Fig. 22.21 is used to activate an edge-triggered one-shot that
produces the sampling pulse for the sample-and-hold circuit. It can be seen in Fig. 22.22
that the output voltage of the circuit is given by
Vϕ = V̂1 sin ϕ . (22.20)
Around the point of operation (ϕ = 0), the detector characteristic is approximately linear:
Vϕ ≈ V̂1 ϕ/rad .
Hence, the factor of the phase detector is
dVϕ
kϕ = = V̂1 /rad . (22.21)
dϕ
V
V1 V
V1
V V2
V2
V2
ts
Fig. 22.21. A sample-and-hold circuit Fig. 22.22. Voltage waveform in the phase
used as a phase detector detector. The dips in Vϕ disappear to a great
extent if the sampling time ts is not much
larger than the time constant of the
sample-and-hold circuit
22.4 Phase-Locked Loop 1117
V
V1
It is obvious from Fig. 22.23 that a further possible operating point at which Vϕ = 0
would be ϕ = π. Then, kϕ = −V̂1 /rad. The sign of the controller gain defines which of
the two operating points is stable. Further stable points of operation occur at intervals of
2π. This indicates that the phase detector does not recognize a displacement by a whole
number of periods.
If a triangular wave-shape V1 is employed rather than the sinusoidal one, a triangular
detector characteristic is obtained. For rectangular input voltages V1 , the circuit is obviously
unusable.
Controller Parameters
For the controller, it is best to choose a circuit without derivative action, since the output
voltage of the sample-and-hold element changes only in steps. According to (22.23), the
phase shift ϕm between Vϕ and Vf at frequency fm = 41 f2 , is − 135◦ . The phase margin
is thus 45◦ if we adjust the proportional gain AP in such a way that the gain crossover
frequency fc = 41 f2 . By definition, at fm = fc ,
|g| = |AS | · |AC | = 1 .
Using AC = AP and (22.23), this results in
fc f2
AP = = .
kf kϕ 4kf V̂1
1118 22 Electronic Controllers
PI-controller
sample-and- V V
VCO
hold circuit
V V
one-shot
Pull-In
After switch-on there is usually a frequency offset f = f1 −f2 . The phase shift therefore
increases proportionally with time. According to Fig. 22.23, this produces an alternating
voltage at the output of the phase detector, with a frequency f and an amplitude V̂ϕ = V̂1 ,
so that the tracking oscillator is thus frequency-modulated by the voltage:
Vf = AP V̂1 sin ωt .
There will therefore be an instant at which the frequencies are identical, and the loop will
pull in and acquire lock. The precondition for this is that the frequency offset f = f1 −f2
is smaller than the sweep width:
f2 max = ±kf AP V̂1 . (22.24)
This maximum permissible offset is known as the capture range and represents the normal
range of operation of the loop. For our example, it is ±2.5 kHz; that is, 25% of f1 = 10 kHz.
22.4.2
Synchronous Demodulator as a Phase Detector
Section 20.3.4 on page 1053 describes the application of the multiplier as a phase-sensitive
rectifier. If two sinusoidally alternating voltages, V1 = U cos ω1 t and V2 = U cos(ω2 t+ϕ),
are applied to the inputs, the output voltage will be (U = multiplier unit)
V1 V2 1 1
Vo = = U cos[(ω1 + ω2 )t + ϕ] + U cos[(ω1 − ω2 )t − ϕ] . (22.25)
U 2 2
22.4 Phase-Locked Loop 1119
V
U
V
VCO
V
P-controller
V V
lowpass filter
U V
V f1 Vo
Fig. 22.26. PLL with a multiplier as phase detector for frequency demodulation
because of its very slow response. In principle, it could be speeded up by derivative action
of the controller, but this would nullify the effect of the lowpass filter; that is, increase the
ripple.
An increase in the control system bandwidth at the expense of the ripple of Vf can
be attained very simply by using a proportional controller and omitting the lowpass filter,
as shown in Fig. 22.26. A phase margin of 90◦ is then available for any proportional gain
chosen; that is, the control loop is aperiodically damped.
Because of the feedback, the ripple of Vf causes the tracking oscillator to be frequency-
modulated with twice the signal frequency. This results in a distortion of the output sine
wave which is equivalent to a phase noise of the oscillator. For square waves, the mark-
space ratio is changed. The proportional gain must not be too high if the distortion is to be
kept within tolerable limits. The condition flp ≤ 13 f1 can be used as a rule of thumb.
The resulting arrangement from Fig. 22.26 is available as an integrated circuit PLL.
Usually, the multiplier is simplified and reduced to a modulator, as shown in Fig. 17.22.
The 74 HC 4046 from Philips and National, for instance, is based on this principle.
When operated without a lowpass filter, the circuit is usable only in those applications
where it is important to have frequency f2 identical to f1 , and where the shape and phase
shift of the output signal are not significant, for example, as a discriminator for frequency
demodulation. The FM-signal is used as input signal. If the VCO frequency f2 is propor-
tional to Vf , this voltage is also proportional to the input frequency f1 . The superimposed
ripple can be filtered out by a lowpass filter outside the control loop.
22.4.3
The Frequency-Sensitive Phase Detector
One drawback of the phase detectors described above is that they possess only a limited
capture range; in other words, they cannot pull in if the initial frequency offset exceeds
a certain limit. The reason for this is that the phase-equivalent signal Vϕ for a frequency
deviation is an alternating voltage that is symmetrical about zero. The voltage Vf therefore
effects a periodic frequency modulation of the tracking oscillator, but never a systematic
tuning in the right direction; that is, toward the lock-in frequency.
The phase detector in Fig. 22.27, however, is different in this respect, in that it pro-
duces a signal that has the correct sign information for any given frequency offset. The
circuit basically comprises two edge-triggered D flip-flops. For control purposes, the input
voltages V1 (t) and V2 (t) are converted into the rectangular signals x1 and x2 .
22.4 Phase-Locked Loop 1121
V1
Vy1
Vy2 – Vy1
Lowpass
V
Vy2
V2
Fig. 22.27. Phase detector with memory for the sign of the phase shift
We now assume that both flip-flops are reset. If voltage V2 leads voltage V1 (i.e.,
ϕ > 0), we first obtain a positive edge of x2 which sets flip-flop F2 . It remains set until
the following positive edge of x1 sets flip-flop F1 . The state in which both flip-flops are set
exists only during the propagation delay time, since they are both reset subsequently by
gate G. It can be seen from Fig. 22.28 that the output of the subtractor shows a sequence of
positive rectangular pulses. Correspondingly, a sequence of negative pulses is obtained if
the positive-going edge of x2 occurs after the positive-going edge of x1 ; that is, if ϕ < 0.
This behavior is summarized in the state diagram of Fig. 22.29.
The duration of the output pulses is equal to the time interval between the positive-
going zero crossings of V1 (t) and those of V2 (t). Hence, the mean value of the output
voltage is:
t ϕ/rad
Vϕ = V̂ = V̂ · . (22.30)
T 2π
As the value of the time interval increases proportionally with ϕ until the limits ± 360◦
are reached, a range of linear phase measurement of ± 360◦ is obtained. When this limit is
exceeded, the output voltage jumps to zero and increases again, still with the same polarity.
The result is the sawtooth characteristic shown in Fig. 22.30.
Vy2 – Vy1 V
V
V
V
The basic difference between this characteristic and all previous ones is that, for ϕ > 0,
Vϕ is always positive and, for ϕ < 0, always negative. This is the reason for the frequency
sensitivity of the detector. If frequency f2 is, for instance, larger than f1 , the phase shift
increases continuously and proportionally with time. As shown in Fig. 22.30, we then
obtain for Vϕ a sawtooth voltage that has a positive mean value. If this detector is used in a
phase-locked loop, it always indicates a leading phase. For a controller with integral action,
the tracking freqticncy f2 is therefore reduced until it coincides with f1 . The capture range
is thus theoretically infinite, and in practice limited only by the input voltage range of the
VCO.
We described in Sect. 22.4.2 how the averaging lowpass filter has a very unfavorable
effect on the transient behavior. For this reason, it is usually also omitted in this circuit. If
we wish to have ϕ = 0 (with the aid of a PI-controller), no phase distortions are incurred,
since in this case Vϕ = 0 even without filtering. The flip-flops in the phase detector then
produce no output pulses.
One drawback of the circuit is that very small deviations in phase are not detected,
since the flip-flops would then have to produce extremely short output pulses that would
be lost due to the limited rise times within the circuit. This is the reason why the phase
jitter (the phase noise) is somewhat larger than with the sample-and-hold detector.
If a PLL with a large capture range and a small phase jitter is required, this circuit
can be combined with a sample-and-hold detector. After lock-in has been accomplished,
the sample-and-hold detector is switched into the loop instead of the frequency-sensitive
phase detector.
22.4.4
The Phase Detector with an Extensible Measuring Range
With the phase detectors described so far, it is not possible to detect a phase shift of more
than one oscillation period, as the phase-measuring range is limited to values between π/2
and 2π , depending on the detector used. There are applications, however, for which a phase
delay of several oscillations must be recovered. The phase detector shown in Fig. 22.31 is
suitable for this purpose. It is based on the up–down counter as illustrated in Fig. 9.36 on
page 672, which is insensitive to coincident clock pulses.
Near zero phase shift, the detector behaves in the same way as the previous circuit. If
x2 leads x1 , positive pulses of the magnitude VLSB arise, the duration of which is equal
to the interval between the zero crossings of the input voltages. For a phase lag, negative
pulses occur. The mean value of the pulses is:
t ϕ/rad
Vϕ = VD = VLSB = VLSB · .
T 2π
22.4 Phase-Locked Loop 1123
two’s
comple- V lowpass
ment filter
DAC V
V
VLSB
If the phase displacement reaches the value 2π, the value for the interval t jumps from T
to 0. In contrast to the previous circuit, however, the output does not assume zero voltage,
but remains at VLSB , as the difference D simultaneously increases by 1. To generalize, the
resultant output voltage is given by:
t ϕ/rad
Vϕ = VLSB D + = VLSB · .
T 2π
The expression D + t/T indicates the number of periods by which the two signals
are displaced. The resulting detector characteristic is shown in Fig. 22.32 for 4 bits. The
measurement range can he increased as required by extending the counting range.
22.4.5
The PLL as a Frequency Multiplier
A particularly important application of the PLL is that of frequency multiplication. A
frequency divider is connected to each of the two inputs of the phase detector, as illustrated
in Fig. 22.33. The frequency of the VCO assumes such a value that
f1 f2
= .
n1 n2
In this manner, the frequency of the VCO,
n2
f2 = f1
n1
can be adjusted to any rational multiple of the reference frequency f1 .
1124 22 Electronic Controllers
In this application, the phase detector may operate at a frequency that is considerably
lower than that of the VCO. It must therefore be insured that the control voltage Vϕ
contains no ripple. An undesired frequency modulation and phase noise would otherwise
occur instead of simply the distortion of the output waveform, as described in Sect. 22.4.2.
The frequency multiplier circuit can be used to generate frequencies above 50 MHz
with crystal accuracy for which practically no crystals are available. For this purpose, we
use a crystal oscillator that operates at, for example, f1 = 10 MHz and select n2 > n1 .
If it is merely a question of obtaining an integral multiple of the crystal frequency, we
can select n1 = 1, thereby dispensing with the input divider. However, if we would like,
for instance, to step through the frequencies from 90 to 100 MHz in 100 kHz increments,
the crystal frequency must first be divided down to 100 kHz with n1 = 100. We can then
generate all of the desired frequencies using a divider factor of n2 = 900 . . . 1000. This is
the principle on which the digital tuners widely used in today’s radio and TV receivers are
based. A number of integrated PLL components are listed in Fig. 22.34.
V
controller VCO
23.1
Basic Photometric Terms
The human eye perceives electromagnetic waves in the range 400 to 700 nm as light. The
wavelength produces the sensation of color, and the intensity that of brightness. In order
to quantify brightness, it is necessary to define a number of photometric quantities. The
luminous flux Φ is a measure of the number of quanta of light (photons) passing through
a cross-sectional area of observation area A per time. It is expressed in lumen (lm). The
luminous flux Φ is unsuitable for characterizing the brightness of a light source, as it is
generally a function of the cross-sectional area A and the distance r from the light source.
In the case of a spherically symmetrical point source, the luminous flux is proportional to
the solid angle Ω. This is defined as Ω = surface/(radius)2 and is actually dimensionless.
However, it is generally assigned the unit steradian (sr). The solid angle that encloses the
entire surrounding sphere is given by
4π r 2
Ω0 = sr = 4π sr .
r2
A circular cone of aperture angle ±ϕ encloses the solid angle
Ω = 2π(1 − cos ϕ) sr . (23.1)
At ±33◦ , is approximately unity. For small solid angles, we can, as an approximation,
replace the spherical surface by a flat surface, obtaining
An
Ω = 2 sr (23.2)
r
where r is the distance of the surface from the center. As the luminous flux of a point
source of light is proportional to the solid angle Ω, the brightness of the light source
can be characterized by the quantity I = dΦ/dΩ, the luminous intensity. The unit of
luminous intensity is the candela (cd). The relationship between the above units is given
by 1 cd = 1 lm/sr. A light source therefore possesses a luminous intensity of 1 cd if it
emits a luminous flux of 1 lm into a solid angle of 1 sr. In the case of spherical symmetry,
the total emitted luminous flux is therefore Φtot = I Ω0 = 1 cd 4π sr = 4π lm. 1 cd is
defined as the luminous intensity of a black body with a surface area of 1.6667 mm2 at
the temperature of solidifying platinum (1769◦ C). A large candle flame has a luminous
cd
intensity of approximately 1 cd. For incandescent lamps, the relation I = 1 W P can be
used as an approximation, where P is the rated power of the incandescent lamp.
In the case of extended light sources, the luminance L = dI /dAn is generally specified,
An being the projection of the light source area onto the plane perpendicular to the direction
of observation. If the angle between the surface normal and the specified direction is ε,
then dAn = dA · cos ε. The unit of luminance is the stilb (sb): 1 sb = 1 cd/cm2 .
A measure of how bright an illuminated area A appears to the observer is the illuminance
E = dΦ/dAn . The unit is the lux (lx): 1 lx = 1 lm/m2 . A full moon gives an illuminance
of 0.1 to 0.2 lx. A newspaper is just readable at an illuminance of 0.5 to 2.0 lx. At a writing
1128 23 Optoelectronic Components
desk, there should be an illuminance of 500 to 1,000 lx. Daylight can produce illuminances
of up to 50,000 lx.
We shall now calculate the illuminance produced by a point source of light that has a
given luminance at a specified distance r (Fig. 23.1). In order to calculate the illuminance,
we assume that the area term dA is small compared to r 2 and is perpendicular to the
connecting line LM. From (23.2), the solid angle d subtended by dA at point L is
therefore given by
dA
dΩ = sr .
r2
dA
dΦ = I d = I sr .
r2
dΦ I
E= = 2 sr . (23.3)
dA r
1.47 mW
PL = Φ.
lm
lm 1.47 mW
1 lx = 1 =
7
m2 m2
When giving approximate values for various luminous intensifies, we stated that an incan-
descent lamp of rated power P = 10 W possesses a luminous intensity of about 10 cd.
It therefore radiates a luminous flux Φtot = 4 sr · 10 cd = 126 lm into the full solid
angle; at a wavelength λ = 555 nm, this corresponds to a light power PL = 0.185 W .
An incandescent lamp consequently has an efficiency η = PL /P ≈ 2%. In addition to
the photometric units given above, other units are often used, particularly in the American
literature. They are listed in Fig. 23.2. An overview of the efficiency of different light
sources is given in Fig. 23.3. It is remarkable that even the best high-efficiency LEDs do
not reach the efficiency of flourescent lamps.
23.2 Photoconductive Cells 1129
23.2
Photoconductive Cells
Photoconductive cells are junctionless semiconductor devices whose resistance is a func-
tion of the illuminance. Figure 23.4 shows the circuit symbol and Fig. 23.5 the character-
istic.
A photoconductive cell behaves like an ohmic resistor; that is, its resistance is neither a
function of the voltage applied nor of its sign. With moderate illuminance the relationship
R ∼ E −γ applies, where γ is a constant between 0.5 and 1. With higher illuminance,
the resistance tends to a minimum value. At low illuminance the value of γ increases,
and at very low illuminance the resistance tends to the dark resistance. The dark-to-light
resistance ratio may exceed 106 . The resistance is markedly temperature-dependent at low
illuminance. This is shown by Fig. 23.6.
When the cell is illuminated, a steady-state resistance value is not established immedi-
ately. The photoconductive cell requires a certain settling time. This is in the millisecond
range at illuminances of a few thousands of lux, but may exceed several seconds at values
below 1 lx. The steady-state value at which the resistance settles depends not only on the
illuminance but also on the preceding optical history. After comparatively long exposure
to high illuminance, higher resistance values are obtained than if the photoconductive cell
had been kept in the dark.
Photoconductive cells are generally made of cadmium sulfide, for which the figures
quoted above apply. Photoconductive cells made of cadmium selenide are characterized
by shorter settling times and a higher dark-to-light resistance ratio. However, they possess
higher temperature coefficients and exhibit greater dependence on the past optical history.
Cadmium-based photoconductive cells are sensitive in the spectral range from 400 to
800 nm. Some types can be used over the entire range, while others possess a quite specific
color sensitivity. Photoconductive cells with high infrared sensitivity are fabricated from
lead sulfide or indium antimonide. They are suitable for wavelengths up to some 3 or 7 mm,
but are considerably less sensitive than cadmium-based cells.
Photoconductive cells have a sensitivity comparable to that of photomultipliers. They
are therefore suitable for measuring low levels of illumination. They can also be employed
as controllable resistors. As the load may be several watts, components such as relays can
be directly connected without additional amplification.
23.3
Photodiodes
The reverse current of a diode increases on exposure to light. This effect can be used
to measure light. For this purpose, photodiodes are provided with a glass window in the
package. Figure 23.7 shows the circuit symbol, Fig. 23.8 the equivalent circuit diagram, and
Fig. 23.9 the characteristics. Essentially, a short-circuit current flows that is proportional
to the illuminance. Thus, in contrast to photoconductive cells, no external voltage source
is required. Typical sensitivity values are of the order of 0.1 µA/lx for small photo diodes.
23.3 Photodiodes 1131
A
IA
When a reverse bias is applied, the photoelectric current remains virtually unchanged. This
operating mode is useful if short response times are required, as the junction capacitance
decreases with reverse bias.
As the illuminance increases, the no-load voltage rises to approximately 0.5 V in the
case of silicon photodiodes. As Fig. 23.9 shows, the diode voltage decreases only slightly
on load as long as the current is smaller than the short-circuit current IP determined by
the illuminance. Photodiodes are therefore suitable not only for measurement of light,
but also for the generation of electrical energy. For this purpose, particularly large-area
photodiodes are manufactured, which are known as photovoltaic or solar cells.
The spectral sensitivity range of silicon photodiodes is between 0.6 and 1 mm, while
that of germanium photodiodes is between 0.5 and 1.7 mm. The relative spectral response
is shown in Fig. 23.10.
–IA –IA
E = 800 lx E = 800 lx
600 MPP
600
400 400
200 200
Human
eye
Vo
Vo Vreverse
Fig. 23.11. A current–voltage converter for a Fig. 23.12. A current–voltage converter for a
particularly low dark current particularly large bandwidth
Output voltage: Vo = RN · I Output voltage: Vo = RN · I
ICs: OPT 101 Texas Instr.
ISL 29000 Intersil
fc = 1/2π RN CN = 160 Hz .
23.4
Phototransistors
In a phototransistor, the collector–base junction is designed as a photodiode. Figure 23.13
shows its circuit symbol and Fig. 23.14 its equivalent circuit.
The way in which a phototransistor operates can be easily explained by reference to the
equivalent circuit in Fig. 23.14: the current through the photodiode causes a base current
and thus an amplified collector current to flow. Whether it is preferable to connect the base
or leave it open depends on the particular circuit. Phototransistors in which the base lead
is not brought out are also known as photo-duodiodes.
In order to achieve a particularly high current gain, a Darlington-connected phototran-
sistor can be used. Its equivalent circuit is shown in Fig. 23.15.
The equivalent circuits show that phototransistors have a similar performance to that of
comparable photodiodes in terms of their spectral range. However, their cutoff frequency
23.5 Light-Emitting Diodes 1133
Vo Vo
is considerably lower. For phototransistors it is of the order of 300 kHz and for photo-
Darlington connections some 30 kHz.
Figure 23.16 shows a phototransistor used as a photodetector. Denoting the photocur-
rent through the collector–base diode by IP , we obtain an output voltage
Vo = V + − BR1 IP .
Correspondingly, for the circuit in Fig. 23.17,
Vo = BR1 IP .
A wide range of phototransistors and optoelectronic switches is available from Optek and
Infineon.
23.5
Light-Emitting Diodes
Light-emitting diodes (LEDs) are not made from silicon or germanium, but from gallium
arsenide phosphide (III-V compound). These diodes emit light when a forward current
flows. The spectral range of the luminous flux emitted is quite sharply delimited, its fre-
quencies depending on the basic material used. Figure 23.18 shows the circuit symbol,
and Fig. 23.19 provides an overview of the most important characteristics.
The efficiency of high efficiency LEDs can be as high as 10 % but at older types it is
less than 0.1%.The luminance is proportional to the forward current over a wide range.
Currents of a few milliamperes are sufficient to provide a clearly visible display. LEDs
are therefore particularly suitable as display elements in semiconductor circuits. They are
also available as seven-segment or matrix displays.
23.6
Optocouplers
23.7
Visual Displays
Digital information can be displayed in many ways; for example, using incandescent
lamps, glow lamps, LEDs, or liquid crystals. LED and liquid crystal displays (LCDs) have
contrast
Vrms
Fig. 23.21. Contrast of a
LCD-Display as a function
VOFF VON of the AC voltage applied
assumed great importance, as they can be operated with low voltages and low currents.
The application is simplified by the large number of integrated drivers available.
LCDs are not semiconductor components. Unlike LEDs, they themselves generate
no light, relying instead on external illumination. An optical effect is produced because
a liquid crystal element is transparent when no voltage is applied and therefore appears
bright, whereas it is opaque when a voltage is applied and therefore appears dark. The liquid
crystal element comprises two electrodes with an organic substance sandwiched between
them. This substance contains crystals whose orientation can be varied by an electric field
therefore they are named liquid cristal displays. The state of the element therefore depends
on the electric field strength; it behaves, in effect, like a capacitor.
The device is driven by AC voltages at a frequency high enough to insure that no flicker
occurs. On the other hand, the frequency selected must be low enough to insure that the
alternating current flowing through the capacitor remains small. In practice, the values
selected are between 30 and 100 Hz. The driving AC voltage must not contain any DC
component, as even 50 mV electrolyze the LCD and reduce lifetime.
In Fig. 23.21, the contrast is plotted as a function of the rms value of the alternating
voltage amplitude applied. For AC voltages of less than VOFF rms ≈ 1.5 V, the display is
virtually invisible; voltages of more than VON rms ≈ 2.5 V produce maximum contrast.
As the capacitance of a liquid crystal element is only about 1 nF/cm2 , the currents
required for driving the device are well below 1 mA. This extremely low current requirement
represents a significant advantage over LEDs.
23.7.1
Binary Displays
LEDs require a forward current of 5…20 mA for good visibility in daylight. These currents
can be provided most conveniently using gates, as in Figs. 23.22 and 23.23. In Fig. 23.22,
the LED lights when an H-level appears at the gate output by applying an L-level at the
input. In Fig. 23.23, the reverse is true. Current limiting is provided in each case via
Fig. 23.22. Driving a LED connected to Fig. 23.23. Driving a LED connected to
ground from a logic gate supply voltage from a logic gate
1136 23 Optoelectronic Components
VL
VL
VL
Fig. 23.26. Voltage waveform for a liquid crystal display that is switched on and off
the resistor R. Due to the relatively high load due to the LEDs, the gate outputs have no
specified voltage level and must therefore not be used for logic operations. This is indicated
on the circuit diagram by a cross at the gate output.
In order to control the light output, it is possible to use gates with a second input at
which a square-wave AC voltage is applied. Its duty cycle then enables the average diode
current to be reduced to zero. The frequency must be at least 100 Hz to insure that no
flicker is visible.
Drive signal generation for liquid crystal displays is somewhat more complicated. An
alternating voltage must be generated whose rms value is sufficiently high and whose mean
value is zero. The easiest way to achieve this is to connect the display between two switches
(Fig. 23.24), which are switched back and forth between ground and operating voltage V+
either in phase or in antiphase. For in-phase operation, VL = 0, while for antiphase
operation, VL rms = V + . This is illustrated by the waveform diagram in Fig. 23.26.
The practical implementation of this principle is shown in Fig. 23.25. When x1 = 0,
y1 = y2 = x2 ; the two terminals of the display therefore switch in phase with the square-
wave signal x2 . For x1 = 1, y1 = x̄2 , and the display receives antiphased signals. CMOS
gates are most suitable for this purpose, as their output levels with purely capacitive loading
only differ by a few millivolts from V+ or zero potential. In addition, only the use of CMOS
gates fully exploits the low power requirement of liquid crystal displays.
23.7.2
Analog Displays
A quasi-analog display can be obtained using a row of indicating elements. This provides
a dot-position display if only the element belonging to a particular indication value is
23.7 Visual Displays 1137
Fig. 23.27.
Dot-position display (above)
and the bar-graph display (below)
Fig. 23.28. Binary dot-position display drive Fig. 23.29. Binary bar-graph display drive
Vref Vref
Vi Vi
Fig. 23.30. Analog dot-position display drive Fig. 23.31. Analog bar-graph display drive
turned on. A bar-graph display is obtained if all the lower elements are turned on as well.
Figure 23.27 compares these two alternatives.
A dot-position display can be driven by binary signals using a 1-out-of-n encoder (see
Sect. 8.2.1 on page 643). In this case, only the LED connected to the output selected is
turned on. The bar-graph display in Fig. 23.29 is obtained when all the LEDs below the
selected output are also turned on via the additional output gates.
In order to drive a display row using analog signals, it is advisable to employ an
analog–digital converter in a parallel arrangement, as the signals required for driving a
bar-graph display are then produced directly. As shown in Fig. 23.30, the input voltage is
compared with a reference voltage by means of a comparator chain. As a result, all the
1138 23 Optoelectronic Components
comparators whose reference voltages are smaller than the input voltage are activated. With
this method, additional gates are required in order to implement a dot-position display, as
shown in Fig. 23.31.
In many applications the easiest way of implementing a dot-position or bar-graph
display is to use a simple microcontroller for instance the PIC-family from Microchip.
The input signal can be connected to the AD-input, the LEDs to the port outputs.
23.7.3
Numerical Displays
The simplest means of representing the numerals 0 to 9 is to arrange seven indicating
elements to form a seven-segment display (Fig. 23.32). Depending on which combination
of segments a to g is turned on, all digits can be represented with adequate readability.
In order to drive a seven-segment display, it is necessary to assign each digit, which is
normally present in binary coded form (BCD), the associated combination of segments.
A circuit of this type is known as a BCD seven-segment decoder. Its truth table is shown
in Fig. 23.33. The principles for connecting LED and liquid crystal displays are those of
Figs. 23.23 and 23.25 respectively. The corresponding circuits are given in Figs. 23.34
and 23.35.
Fig. 23.34. Connection of a LED display Fig. 23.35. Connection of a liquid crystal
to a seven-segment decoder display to a seven-segrnent decoder
BCD seven-segment decoders are available as integrated circuits; examples are listed
in Fig. 23.36. Some of the LED-driving types possess current source outputs, in which
case the external current-limiting resistors are not required. In addition to the decoders for
driving common-anode displays, types are also available for a common cathode. In the case
of the liquid-crystal driving decoders, the exclusive-OR gates are already incorporated. The
only external device required is therefore the square-wave generator.
In some cases the numbers 10 to 15 that are not used for numeric display are assigned
to the letters A to F. The usual shape of the additional characters in a hexadecimal display
is also shown in Fig. 23.32.
23.7.4
Multiplex Displays
The LED and liquid crystal displays described here can be used for visual representation of
multi-digit data. However, in order to minimize the number of drivers and lines required,
it is advisable in the case of multi-character displays to connect them together as a matrix
and operate them on a time-division multiplex basis. Figure 23.37 shows an arrangement
of this kind for an 8-character, 7-segment LED display. The corresponding segments of
all the displays are connected in parallel. In order to ensure that the same segments of all
1140 23 Optoelectronic Components
1-of-8
decoder
8-bit
parallel anode
port driver
7-seg-
ment
decoder
cathode
driver
the characters do not now light up simultaneously, only one character is activated via the
1-of-8 decoder at a time.
Therefore, only 15 lines are required to operate an 8-character, 7-segment display.
A single 8-bit parallel interface is suitable as a microprocessor interface. A number of
7-segment decoders are shown in Fig. 23.36. Anode and cathode drivers are listed in
Fig. 23.39.
Multiplex operation is performed by the microprocessor program. To do this, the po-
sition number is specified with four bits and the character to be represented in BCD code
using the other four bits. The output is then repeated for the next character position. In
order to ensure that a flicker-free display is produced, the complete display cycle must be
executed at least 100 times a second. There are many applications, particularly in simple
devices, where the processing time required for driving the display is spare. However, it
may be disturbing to have the display flickering when the microprocessor is required for
long periods for other tasks.
If the display is to operate stand-alone, it must possess an additional display memory
and an internal multiplexing facility. The resulting circuit is shown in Fig. 23.38. The
display data is written by the microprocessor into a two-port memory (see Sect. 10.2 on
page 697) which is connected to the microcomputer bus like a normal RAM. The display
contents are read from the two-port memory, independently of the bus operation. During
this process, the binary counter issues the addresses cyclically and activates the appropriate
character positions via the 1-of-8 decoder.
Display drivers operating on this principle are widely available as fully-integrated
circuits. Some types are listed in Fig. 23.47 on page 1145. As well as the types with
parallel data inputs, there are also versions in which the display data is stored in a shift
register. They only require one serial data line and no addresses to control them.
Liquid crystal displays (LCDs) require an AC voltage of a specific amplitude. The
push-pull methods described in Fig. 23.25 on page 1136 is only used for generating this
voltage for a small number of segment drivers. For larger numbers of segments, liquid
crystal displays are also linked together to form matrices, in order to minimize the number
23.7 Visual Displays 1141
binary 1-of-8
counter decoder
0...7
anode
driver
2-port 7-seg-
memory ment
decoder
cathode
driver
Fig. 23.39. High current drivers for LED displays and other applications requirung high currents
of connecting lines. However, three voltage levels are required (in addition to ground
potential) for driving liquid crystal matrices of this kind, in order to ensure that the selected
segments receive a sufficiently high and the remainder a suitable low AC voltage. This
special type of multiplexing is known as the triplex method.
23.7.5
Alphanumeric Displays
Seven-segment displays only allow a few letters to be represented. In order to display the
entire alphabet, a higher resolution is required. This can be achieved by using 16-segment
displays or 35-dot matrix displays.
1142 23 Optoelectronic Components
Fig. 23.41. Usual character set of a 16-segment display. At the border you find the ASCII
equivalent. The first hex-sign is in the top row, the second sign in the column on the left.
For example: K =7 B4Hex
16-Segment Displays
The arrangement of the segments in a 16-segment display is shown in Fig. 23.40. In
comparison with the seven-segment display in Fig. 23.32, segments a, d, and g are divided
into two sections and segments h–m are added. This enables generation of the character
set shown in Fig. 23.41. It is usually limited to 64 characters, which include the upper-case
letters, the numerals, and the most important special characters. Some display drivers are
listed in Fig. 23.47.
Fig. 23.42. Arrangement of dots in a 35-dot Fig. 23.43. Matrix arrangement of display
matrix, in seven rows and five columns elements, using LEDs as an example
character generator. The latter determines, in accordance with Fig. 23.43, which dots
of the particular row are to be turned on. Character generators are available as mask-
programmed ROMs, providing the symbols shown in Fig. 23.44. If other character sets
are required, it is advisable to program an EEPROM accordingly. Figure 23.46 gives the
character generator contents required for the character “K.” Typical matrix display with
integrated drive electronics are shown in Fig. 23.47.
1144 23 Optoelectronic Components
anode
driver
cathode
driver
character generator
Fig. 23.47. Display decoders and drivers and displays with integrated drivers
Part III
Communication Circuits
Chapter 24:
Basics
24.1
Telecommunication Systems
Today, telecommunication systems are as much a part of everyday life as electrical energy.
Besides the analog telephone as a conventional cable system and analog radio and TV
broadcasting as classical wireless systems, there are countless more modern telecommu-
nication systems including ISDN telephones, cordless and mobile telephones, radio and
TV broadcasting via wideband cable networks or satellite transmission, PC modems, wire-
less PC mouses and keyboards, wireless garage door openers, and remote controlled car
locks with the actuator integrated within the car key. Furthermore, heterogeneous systems
such as the Internet evolve from the combination of several systems and the application of
specific network procedures.
A transmission system is defined as a telecommunication system if a modulation is used
for the interface to the transmission channel.According to this criterion, telecommunication
engineering must be perceived as the theory of modulation methods. In contrast, there are
transmission systems without modulation, e.g. computer interfaces such as V.24 and SCSI,
which only provide specific lines and drivers for direct signal transmission over larger
distances. Telecommunication systems are thus characterized by the use of a modulator
in the transmitter and a corresponding demodulator in the receiver.
Figure 24.1 shows the components of an analog and a digital telecommunication sys-
tem. Ordered top to bottom, the components form a transmitter; from bottom to top they
form a receiver. The transmission medium between the transmitter and the receiver is the
channel, which may be a cable or radio link with sending and receiving antenna.
In analog systems, the useful signal to be transmitted s(t) is supplied directly to the
analog modulator. The modulator output signal is amplified in a transmitter amplifier and
fed to the channel. Most analog modulators generate a signal with the desired transmission
frequency; in such cases the transmitter amplifier consists solely of one or more amplifiers
in series. In other cases, the modulator generates a signal with an intermediate frequency
which then must be converted to the transmission frequency by a mixer incorporated
within the transmitter amplifier. The channel causes a signal attenuation that may be as
much as 150 dB in radio transmission links (e.g. 1 kW = 103 W transmitted power → 1 pW
= 10−12 W received power). In extreme cases the power of the signal is only slightly higher
than the unavoidable thermal noise. In the receiver, the receiving amplifier enhances the
signal enough so that it can be fed to the demodulator. This requires a gain control to ensure
a fixed signal level at the demodulator despite the large variations in the received signal
level due to the varying distances to the transmitter. In radio and cable systems for multiple
use, the receiving amplifier must also perform frequency selection by separating the desired
incoming signal from signals of adjacent frequency ranges. Here, several filters as well
as one or two mixers are used for frequency conversion. Finally, the analog demodulator
generates the received useful signal r(t) from the selected and amplified signal.
The digital system contains all of the same components as the analog system except that
the modulator and demodulator are of digital design and are connected to the amplifiers
1150 24 Basics
A D
D A
Source Source
coding decoding
Channel Error
coding correction
D A
A D
Channel Channel
a Analog b Digital
via D/A or A/D converters. Sometimes these converters are regarded as integral parts of the
modulator or demodulator and are not shown separately; in such cases the digital modulator
comprises a digital input and an analog output, while the digital demodulator has an analog
input and a digital output. These components, which correspond to the analog system, are
then ready for application. The system is complemented by a channel coding circuit in the
transmitter that introduces a redundancy in terms of check bits, parity bits, checksums or
a specific code; this redundancy is used for error correction in the receiver. Some systems
use an additional source coding and source decoding in order to reduce the amount of
transferred data. Source coding is usually not loss-free, i.e. the signal is not reconstructed
exactly in the decoding process. Rather, source coding is based on physiological findings,
asserting that the human receiver is not capable of detecting certain parts of speech or
24.1 Telecommunication Systems 1151
image signals. This is the level at which the digital signal s(n) is sent and the signal r(n) is
received. For the transmission of analog signals, additional converters are required in the
transmitter and the receiver; this is the case, for example, in digital telephones in which the
transmitted useful signal s(t) originates from a microphone and the received useful signal
r(t) is reproduced by a loudspeaker.
Compared to a digital system, an analog system comprises fewer components, which
are often of simpler design than the corresponding components of the digital system. One
disadvantage of the analog system is that noise and other interferences caused during sig-
nal transmission can no longer be separated from the signal; therefore the signal-to-noise
ratio declines rapidly, especially in transmissions over considerable distances. Further-
more, analog modulation methods do not make full use of the available bandwidth and
need a relatively high signal-to-noise ratio at the receiver input in order to achieve a high
transmission performance.
Digital systems use sophisticated modulation methods that provide a substantially
better bandwidth utilization than analog systems. Noise and other interferences are totally
eliminated by threshold evaluation in the demodulator as long as a certain amplitude is not
exceeded. If this amplitude is exceeded, an initial erroneous evaluation is made which can
be remedied by error correction if the erroneous failure rate remains below a certain limit.
Digital systems are thus capable of providing near ideal transmission performance even
with a low signal-to-noise ratio at the receiver input. Better utilization of the bandwidth
by using more complex modulation methods is also of high importance as the continual
introduction of new systems causes an increasing shortage of transmission frequencies.
Figure 24.2 presents a comparison of the most important features of analog and digital
systems; these also contain the common advantages of digital systems such as low drift
and low alignment complexity. Some of the properties are redundant; for instance, better
utilization of bandwidth in digital systems is a consequence of higher complexity of mod-
ulation methods, while a lower required signal-to-noise ratio at the receiver input enables
the transmit power to be reduced.
1152 24 Basics
24.2
Transmission Channels
The transmission channels are discussed in the order of their industrial use: cable, radio
link and fibre-optics. Despite their technological differences, all channels have one thing
in common, that is, the transmission is based on electromagnetic waves.
24.2.1
Cable
Telecommunications predominantly use coaxial cables and two-wire lines. Figure 24.3
shows a cross-section of these cables together with the E and H field lines, as well as their
characteristic dimensions. The coaxial cable is a shielded line since the fields are restricted
to the space between the inner and outer conductor; influence on adjacent components is
thus excluded1 . In contrast, the signal of an unshielded two-wire line may be emitted into
neighboring components or other unshielded cables running in parallel due to capacitive
(E field) or inductive (H field) coupling; this occurrence is called crosstalk.
The space between the internal and external conductor of a coaxial cable is filled by
a dielectric to centre the conductors; the dielectric material is usually Teflon (r = 2.05)
or polystyrene (r = 2.5). The conductors of the two-wire line each have a polyethylene
sheath; the conductor and sheath are then either twisted or strapped together.
di E field
H field a
da
d
1 In many coaxial cables used in practice the outer conductor is not perfectly sealed so that weak
fields also exist outside the cable.
24.2 Transmission Channels 1153
The line characteristic impedance consists of the field characteristic impedance ZF and a
geometric factor kG that describes the line. Insertion of ZF results in:
⎧
⎪ 60 da
⎪
⎪ √ ln Coaxial cable
⎪
⎪ r di
⎨
= ⎛ ⎞
ZW 2 (24.2)
⎪
⎪ 120 a a
⎪
⎪ √ ln ⎝ − 1⎠ Two-wire line
⎪
⎩ r d d
In practice, cables are used with ZW = 50 (e.g. r = 2.05, di = 2.6 mm, da = 8.6 mm)
and ZW = 75 for coaxial cables and with ZW = 110 for twisted pair lines. The
calculation of ZW for two-wire lines is difficult since fields exist inside the cable sheath
(r > 1) and in the surrounding space (r = 1); therefore it is necessary to use the effective
value of r in (24.2) that can be determined only by field simulation or measurement.
The line characteristic impedance is not an ohmic resistance and thus cannot be mea-
sured using an ohmmeter or impedance meter. It describes solely the ratio between the
voltage and the current of one wave. Later we shall see that normally two waves exist
1154 24 Basics
on a line: the incidental (forward) wave with Vf = ZW If and the reflected wave with
Vr = ZW Ir . With the equations V = Vf +Vr and I = If −Ir , we can calculate the voltage
V that can be measured between the conductors and the current I that flows through the
cable.
In practice, the prefix line is usually omitted and the parameter is called characteristic
impedance. Often the symbols ZL or Z0 are used where Z indicates that it is a complex
impedance; but sometimes the symbols RW , RL or R0 are also used.
Transmission Equation
A short piece of cable can be described by an equivalent circuit with four components (see
Fig. 24.4) where the following four parameters per unit length are used [24.1]:
Insertion of
V2 = V1 + dV , I2 = I1 + dI
dz → 0 , V1 → V2 = V , I1 → I2 = I
I1 R 'dz L'dz I2
V1 C 'dz G'dz V2
leads to:
dV % &
= − R
+ j ωL
I (24.3)
dz
dI % &
= − G
+ j ωC
V (24.4)
dz
Differentiation of (24.3) with respect to z and insertion of (24.4) lead to the transmission
equation:
d 2V % &% &
2
= R
+ j ωL
G
+ j ωC
V = γL2 V (24.5)
dz
The general solution is
V (z) = Vf e−γL z + Vr eγL z (24.6)
and the propagation constant is:
γL = (R
+ j ωL
) (G
+ j ωC
) (24.7)
Even at frequencies in the lower kHz range the following conditions apply to low-loss
lines: j ωL
R
and j ωC
G
; consequently the propagation constant [24.1] is:
R
C
G
L
√
γL ≈
+
+j ω LC (24.8)
2 L 2 C 3 45 6
3 45 6 βL
αL
with the attenuation constant αL and the phase constant βL . In a loss-free line (R
= G
=
0) the attenuation constant is zero.
For further clarification we take the function of time:
@ = (24.6) @ =
u(t, z) = Re V (z) ej ωt = Re Vf ej ωt−γL z + Vr ej ωt+γL z
% &
= |Vf | e−αL z cos ωt − βL z + ϕf + |Vr | eαL z cos (ωt + βL z + ϕr )
3 45 6 3 45 6
Forward wave Reflected wave
= uf (t, z) + ur (t, z)
It consists of a forward wave uf (t, z) and a reflected wave ur (t, z). Figure 24.5 shows
these waves at a given time t0 and a quarter of a period length later. The propagation in
the opposite direction and the increasing attenuation in the direction of propagation can
be seen. The velocity of propagation v is determined by the displacement of the maximum
value of the cosine function. For the forward wave the following applies:
dz ω 1
ωt − βL z + ϕf = 0 ⇒ v = = = √ (24.9)
dt βL L
C
For the reflected wave the absolute value of the velocity of propagation is the same, but the
sign is negative; this again indicates that the two waves propagate in opposite directions.
The wave length λ corresponds to the distance between two consecutive maxima; this
1156 24 Basics
u f (t,z) v l – aL z
~e
0
z
t = t0
1
t = t0 +
4f
ur (t,z) aL z
~e
0
z
Fig. 24.5. Forward (top) and reflected (bottom) wave on a line at a given time t0 and a quarter of a
period length later
requires the z-dependant portion of the argument of the cosine function to pass through
the 2π region:
2π 1 v
βL λ = 2π ⇒ λ = = √ = (24.10)
βL f LC
f
To calculate the current I on the line, (24.3) is solved for I with the value of V from
(24.6) inserted:
1 dV 1 % −γL z
&
I = −
= −
−γL Vf e + γL Vr eγL z
R + j ωL dz R + j ωL
G
+ j ωC
% &
= Vf e−γL z − Vr eγL z
R
+ j wL
ZW = (24.11)
G
+ j ωC
Here, too, we find a forward and a reflected wave, although in this case these are subtracted.
The current waves are coupled to the corresponding voltage waves via the line characteristic
impedance. This relationship has already been described in the previous section.
The voltages Vf and Vr , as well as the currents If and Ir , of the forward and reflected
waves cannot be measured directly because the two waves are always superimposed on
the cable, thus, only V and I can be measured. A directional coupler must be used for the
measurement of the waves [24.1].
For low-loss lines the influence of R
and G
on the line characteristic impedance can
be disregarded; then:
L
ZW ≈ (24.13)
C
Attenuation
For lines and cables the attenuation per unit length a
is normally specified in decibel per
meter. For a standard 50 coaxial cable [24.1]:
a
−3 f
≈ 2.35 · 10
dB/m MHz
For the attenuation a of a line with length l it follows:
a = a
l
Figure 24.6 shows that the attenuation is dependent upon length and frequency. Two-wire
lines have an attenuation which is two to five times higher, depending on the line type.
kH Hz MH z
H
G z
a
10 z
H
10 1M 10 M
z
H
G
0
10
dB
1
100
z
0
10
5
2
1
0.1
~
Fig. 24.6. Attenuation a of a standard
~
2 5
1 10 100 1k 10k l 50 coaxial cable with length l for
m several frequencies
1158 24 Basics
L
1 c0 m 1
Velocity of propagation v = √ = √ = 3 · 108 ·√
LC
r s r
Z
L
=
W
Inductance per unit lengthg
v
1
Capacitance per unit length C =
ZW v
a
The attenuation constant αL can be calculated from the attenuation per unit length:
−1 a
−1 a
f
αL = 0.115 m · = 0.115 m ·
dB/m dB/m f0 f0
Parameters of a Cable
A cable is usually specified by the line characteristic impedance ZW , the velocity of
propagation v and the attenuation per unit length a
. The relative dielectric constant r
may also be given instead of the velocity of propagation; the velocity of propagation can
then be calculated with (24.1). As an alternative to ZW and v or r , the inductance per unit
length L
and the capacitance per unit length C
can also be specified but this is uncommon
in practice. An overview of the parameters and relationships is given in Fig. 24.7.
I1 I2
ZW
V1 V2
g L = a L + jb L
V1 = Vf + Vr (24.14)
It follows that the reflected wave is determined by the circuitry at gate 2. No reflected wave
exists for V2 − ZW I2 = 0, i.e. with a resistance R = ZW = V2 /I2 at gate 2; this is known
as the termination of the line with its characteristic impedance. Solving (24.18) for Vf
and Vr and insertion into (24.14) and (24.16) results in:
V2 # γL l $ Z I # $
e + e−γL l + eγL l − e−γL l
W 2
V1 =
2 2
V2 # γL l $ I # $
e − e−γL l + eγL l + e−γL l
2
I1 =
2ZW 2
With
1 # γL l $ 1 # γL l $
cosh(γL l) = e + e−γL l , sinh(γL l) = e − e−γL l
2 2
the four-pole equation of a line is obtained:
⎡ ⎤ ⎡ ⎤⎡ ⎤
V1 cosh(γL l) ZW sinh(γL l) V2
⎣ ⎦ = ⎣ 1 ⎦⎣ ⎦ (24.19)
I1 sinh(γL l) cosh(γL l) I 2
ZW
I1 I2
ZW
V1 V2 Z2
gL = a L + j b L
Z1 Z2
and tanh(jβL l) = j tan(βL l), the input impedance for a loss-free line (αL = 0) is:
2πl
Z2 + j ZW tan
λ
Z1 = (24.21)
Z2 2πl
1+j tan
ZW λ
Equations (24.20) and (24.21) show that the line produces an impedance transformation
Z2 → Z1 . For further clarification let us look at some special cases.
– Termination with the characteristic impedance: With Z2 = ZW it follows that
Z1 = Z2 = ZW , independent of the length of the line. It was already mentioned
in the previous section that this eliminates the reflected wave. Termination with the
characteristic impedance is the preferred operating mode for transmission lines since
it ensures an optimum power transmission from the signal source to the load; this is
further detailed in Sect. 24.3.
– Electrically short line: If a line is very much shorter than the wave length λ the tanh
and the tan terms can be disregarded; then Z1 = Z2 . This corresponds to the normal
connecting line in low-frequency circuits which can be regarded as the ideal connection.
With higher frequencies, the permissible length of an electrically short line decreases
in accordance with the wave length, i.e. inversely proportional to the frequency; in the
GHz range this causes a noticeable impedance transformation even at lengths of a few
millimeters.
– λ/4-line: A loss-free line with a length equal to a quarter of the wave length λ results
in tan(2π l/λ) = tan(π/2) → ∞; from (24.21) it follows:
2
ZW
Z1 = (24.22)
Z2
This relationship is sufficiently accurate for low-loss lines as well. The λ/4-line is often
used instead of a transformer for resistance transformations;
√ the resistance Z2 = R2 is
transformed to Z1 = R1 by a λ/4-line with ZW = R1 R2 . Such lines are also known
as λ/4-transformers.
– Open line: A line with Z2 → ∞ is called an open, open-ended or unloaded line; if it
is loss-free, it follows from (24.21):
l<λ/8
ZW ZW 1 1
Z1 =
≈ = = (24.23)
2πl 2πl j ωC
l j ωC
j tan j
λ λ
24.2 Transmission Channels 1161
An open, loss-free line acts as a reactance with a capacitive response (tan(2π l/λ) > 0)
or an inductive response (tan(2π l/λ) < 0) depending on the length; for l < λ/8 the
line acts as a capacitance with C = C
l.
– Short circuited line: For a short circuited (Z2 = 0) loss-free line, (24.21) leads to:
2πl l<λ/8 2πl
Z1 = j ZW tan ≈ j ZW = j ωL
l = j ωL (24.24)
λ λ
Thus, a short circuited, loss-free line also acts as a reactance with an inductive response
(tan(2π l/λ) > 0) or a capacitive response (tan(2π l/λ) < 0) depending on the length;
for l < λ/8 the line acts as an inductance with L = L
l.
The latter three cases play an important role in the realization of matching circuits in the
upper MHz and the GHz range; these applications no longer use coaxial cables or two-
wire lines but strip lines as described below. Figure 24.10 shows the various transformation
properties of a given line.
2
ZW
ZW ZW ZW ZW Z2
Z2
Any length λ
l =
4
ZW C 'l ZW L'l
λ λ
l< l <
8 8
It follows that C = 114 pF. This means that the signal to be measured is loaded with a
capacitance that is substantially higher than the input capacitance of the oscilloscope. The
line of one meter in length is therefore not an electrically short line.
Strip Line
With increasing frequencies, it becomes necessary on printed circuit boards to incorporate
connections as conductors with defined characteristic impedance in order to ensure the
distortion-free transfer of high-frequency analog signals and fast digital signals; strip lines
of various designs are used for this purpose [24.1].
The simplest type of strip line is the microstrip, as shown in Fig. 24.11. It is almost
indistinguishable from normal PC board conductors and can thus be produced in the normal
etching process. Owing to the uniform ground surface on the rear side of the board, circuit
boards with a copper layer on both sides must be used. Boards made of Pertinax should not
be used because of their high dielectric losses. Epoxy resin boards (r ≈ 4.8) can achieve
acceptable results in low-demand applications with frequencies below 1 GHz; although the
scatter of r with epoxy resin is particularly problematic. In general, however, substrates of
Teflon (r = 2.05) and, especially in the GHz range, aluminum ceramics (Al2 O3 , r = 9.7)
are used.
Calculating the line characteristic impedance and the quantities per unit length is only
possible with very complex mathematical models. In practice, the required parameters
are usually determined by field simulation. However, there are semi-empiric equations for
determining the line characteristic impedance of a microstrip with the dimensions given
in Fig. 24.11. These equations provide an accuracy of about 2% provided that w/d 10,
a condition which is easy to satisfy in practice [24.1]; for w > h:
√
ZW 188.5/ r
≈
w r + 1 w 0.082 (r − 1)
+ 0.441 + ln + 0.94 + 1.451 +
2h 2πr 2h r2
Figure 24.12 shows the curves for Teflon, epoxy resin and Al2 O3 .
w
r =1 d
h
d
Teflon: r = 2.05
Epoxy resin: r = 4.8
Al 2O3 : r = 9.7 Fig. 24.11. Cross-section of a microstrip
24.2 Transmission Channels 1163
ZW
Ω 120
110
100
90
80 r = 2.05
70
60 r = 4.8
50
40 r = 9.7
30
20
~
~
Fig. 24.12. Characteristic impedance of a microstripline for Teflon (r = 2.05), epoxy resin
(r = 4.8) and Al2 O3 (r = 9.7)
24.2.2
Radio Communication
Figure 24.13 shows the components of a radio transmission system. The output signal
of the transmitter amplifier is fed to the transmitting antenna through a cable. As the
antenna input impedance is usually not matched to the characteristic impedance of the
cable, optimum power transfer requires a matching network. The electromagnetic wave
radiated from the transmitting antenna is received by the receiving antenna erected at a
distance r. The received signal passes through another matching network before it is fed
along a cable to the receiver amplifier.
Antennas
There are many different antenna designs; an overview is given in [24.1]. They differ in
terms of frequency range, bandwidth and radiation pattern. Normally, the transmitting
Transmitting Receiving
antenna antenna
ro = 50 Ω ZW = 50 Ω ZW = 50 Ω ri = 50 Ω
Antenna Antenna
fr fs
Fig. 24.14. Operating modes using a common transmit and receive antenna
antennas of radio and TV stations send the signal horizontally in all directions so that it
can be received by any receiver within the transmission range. The receiving antennas
of portable radio and TV sets also have a wide radiation pattern to, if possible, avoid
any alignment to the broadcasting station; this, however, allows only relatively powerful
stations to be received. In contrast, stationary devices use directional antennas which can
pick up weak stations as well but must be exactly adjusted for the given transmitter station;
antenna misalignment blocks reception. An example of this are parabolic reflector antennas
of satellite receivers. In mobile communications the mobile unit cannot be aligned because
the site of the base station is generally unknown and undergoes changes in relation to the
location of the mobile unit and the given propagation conditions; this is the reason that
here, too, antennas with a wide radiation pattern are used. The base stations themselves
use sectoring, i.e. the surroundings are divided into sectors each of which is served by
one antenna of the respective directional characteristic. Directional (line-of-sight) radio
systems use transmitting and receiving antennas with extremely narrow radiation patterns;
this provides benefits including long distances to be covered with relatively low transmit
power, the prevention of unwanted radio monitoring and the same transmit frequency to
be used for transmitting in other directions. In principle, each antenna can be used for
transmitting and receiving; the directional characteristic is the same.
In bi-directional transmission links with a common transmit and receive antenna, the
output signal of the transmitter amplifier must be prevented from reaching the sensitive
input of the receiving amplifier which would otherwise be immediately destroyed. Systems
with alternating transmission and receiving modes use an antenna change-over switch
(see Fig. 24.14a). Simultaneous transmitting and receiving is also possible if separate
frequencies are used for the received and transmitted signals; signal separation is done by
a specific filter (duplexer). Figure 24.14b shows a simple duplexer with parallel resonant
circuitry.
Directivity: The parameter for the directional characteristic is the directivity D; it indi-
cates the factor by which the transmit power is higher in the main direction, as opposed
to a hypothetical antenna which has an omnidirectional radiation pattern in all directions.
The reference antenna is hypothetical since there is no single antenna that features an om-
nidirectional radiation pattern; the directivity value of a real antenna is thus always higher
than one.
24.2 Transmission Channels 1165
The directivity relates to the radiated power; in practice, however, the supplied power,
which is higher than the radiated power due to losses, is of interest.
Equivalent circuit: Figure 24.15 shows the equivalent circuit of an electrically short rod
antenna (length < λ/4) including the connection to the transmitter amplifier; in this case
LA and CA are the reactive antenna elements, RS is the radiation resistance and RV the
ohmic loss resistance [24.1]. The operating frequency is below the resonant frequency, i.e.
the antenna impedance has a capacitive part; the sum of the radiation and loss impedances
is less than 50 . The antenna impedance is transformed to 50 by the matching network.
Figure 24.16 shows the radiation resistance RS of a rod antenna plotted over the relative
length l/λ [24.1]. The resistance becomes very low for l < λ/8; at this point matching
to 50 is only possible for a very narrow bandwidth. Especially favorable are rod an-
tennas with l/λ ≈ 0.26 . . . 27. Including the antenna loss resistance, they have an overall
resistance of 50 and are operated slightly above the resonant frequency; matching is
performed with a series capacitance.
Matching
network Antenna
Lmatch LA CA RV
Transmitter
Cable
amplifier Cmatch RS
ZW = 50Ω
ro = 50Ω
Fig. 24.15. Equivalent circuit of a rod antenna (l < λ/4) including the connection to the
transmitter amplifier
RS
Ω l=
RS = 130Ω
140
120
l = /2
100 RS = 99.5Ω
80
60 l = 3 /4
RS = 52.7
40 l = /4
l = /8 RS = 36.6Ω
20 RS = 3.4Ω
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 l
Fig. 24.16. Radiation resistance of a rod antenna plotted over the relative length l/λ
1166 24 Basics
Antenna radiation efficiency: The radiation efficiency η can be taken directly from
Fig. 24.15:
RS
η = < 1
RS + R V
This shows the ratio of the radiated power to the supplied power. If the antenna is used
as the receiving antenna, the equivalent circuit is primarily the same except that the loss
resistance does not have the same value due to a slightly different current distribution;
it is therefore necessary to differentiate between the transmission efficiency ηS and the
receiving efficiency ηR .
Antenna gain: The product of the directivity and the antenna efficiency is known as the
antenna gain:
G = Dη
The antenna gain compares the transmit power of a real, lossy antenna in the main radiation
direction with the transmit power of a hypothetical, loss-free antenna with an omnidirec-
tional radiation pattern at the same supplied power. Due to the different antenna efficiencies
in the transmission and receiving mode, it is necessary to differentiate between the transmit
gain and the receive gain; in practice, however, these differences are usually so low that
there is no need for such a differentiation.
2 Please note that the transmitting and receiving antennas are now to be considered as loss-free
antennas with an omnidirectional radiation pattern as any deviation is already accounted for in
the antenna gains GS and GE .
24.2 Transmission Channels 1167
a0 z
GH Hz
dB 10 1G Hz
0 0M
140 1
z
MH
10
120
Hz
100 1M z
H
0k
10
80
60
40
20
~
~
2 5
0.1 1 10 100 1000 r Fig. 24.17. Basic attenuation of a radio
km link
Frequency Ranges
Frequencies are divided into ranges; Fig. 24.18 shows the ranges from 30 kHz to 300 GHz
and their respective designations. The range between 200 MHz and 220 GHz is also known
as the microwave range; it is subdivided into 12 bands (see Fig. 24.19). The names of the
Fig. 24.18. Frequency ranges and wave lengths for radio links in the range from 30 kHz to 300 GHz
1168 24 Basics
Designation P L S C X Ku K Ka Q E F G
from (GHz) 0.2 1 2 4 8 12 18 27 40 60 90 140
to (GHz) 1 2 4 8 12 18 27 40 60 90 140 220
Fig. 24.20. Frequency and wave length ranges for radio and TV broadcasting in Germany
Fig. 24.21. Frequency ranges for mobile communication and mobile telephones in Germany (U =
uplink: mobile unit → base station, D = downlink: base station → mobile unit)
ranges and bands are often combined with the type designations of components, e.g. UHF
transistor or S-band FET.
Besides this classification of frequency ranges, or bands, which is independent of the
application, a particular frequency range is also assigned to each special use. Figure 24.20
contains an overview of radio and TV broadcasting ranges, while Fig. 24.21 lists the bands
for mobile communication, including mobile telephones according to the DECT standard.
24.2.3
Fibre Optic Links
In addition to coaxial and two-wire cables and radio links, fibre optic links via optical
waveguides (glass fibres) are also becoming increasingly important. This technology uses a
carrier signal in the infrared range (f = 190 . . . 360 THz, λ = 1.55 . . . 0.85 mm) which can
be modulated by signal frequencies of up to 100 GHz; theoretically this allows transmission
rates of up to 200 Gbit/s. Today’s systems use 10 Gbit/s while systems with up to 40 Gbit/s
are presently being tested. Due to their very small relative modulation bandwidth (signal-
to-carrier frequency ratio 10−3 ), the transmission attenuation is constant across the
transmission band; therefore, the equalization requirements in the receiver are lower than
in cable links despite the substantially higher data rates.
24.2 Transmission Channels 1169
Quiescent Bias
current voltage
Optical waveguide
Transmitter = 0.85 / 1.3 / 1.55 µm Receiving
amplifier amplifier
Optical Waveguides
A high quality waveguide consists of a very thin fibre made of silicate glass; here, the
cross-sections shown in Fig. 24.23 are used. The radiation propagates in the core with the
refractive index nco and the diameter dco . The cladding, with its slightly lower refractive
index ncl and the outer diameter dcl , is required for guidance only while the outer sheathing
serves to protect the optical waveguide. Typical values for a step-index fibre are nco ≈ 1.4
and ncl /nco ≈ 0.99, i.e. the refractive index of the cladding is only 1% lower than that of
the core. Waveguides made of glass are called glass fibres.
Optical waveguides made of plastic are on the market. These are known as plastic
fibres. They are lower in price and, due to their high mechanical flexibility, are easier to
lay than glass fibres but have significantly poorer propagation properties so that they can
only be used for short distances and low data rates. Their diameter is much bigger than
that of waveguides made of glass; typical diameters are dco = 0.98 mm and dcl = 1 mm.
Limiting angle and acceptance angle: The signal propagation can be explained by
means of radiation optics. The ray in the core is totally reflected at the interface to the
cladding, i.e. it is returned to the core if the angle between the ray and the interface is
smaller than the limiting angle βg ; this is described by3 :
3 In radiation optics the angle between the ray and the perpendicular to the interface is often used;
in this case, it is described by sin βg = ncl /nco . We relate the angle to the fibre axis.
1170 24 Basics
Core dco = 50 µm r
dco dcl
n
dco = 50 µm r
dco dcl
n
dcl =125 mm n cl n co
b Gradient fibre
dco =10 µm r
dco dcl
dcl =125 m m n cl n co
c Single-mode fibre
Fig. 24.23. Cross-section, refraction index and propagation characteristics in optical waveguides
made of silicate glass
ncl
cos βg = < 1
nco
The typical step-index fibre values result in the angle βg ≈ 8◦ . To ensure that the angle
within the waveguide remains below the limiting angle, the angle of incidence at the front
end must be smaller than the acceptance angle αA . Figure 24.24 illustrates this. From the
refraction law it follows:
sin αA
= nco
sin βg
n =1 b >b g n cl
bg
n co
bg
aA
b < bg
coupling between the transmitter diode and the waveguide; a high value in combination with
a correspondingly high acceptance angle is advantageous. The velocity of propagation is:
c0 c0
v = √ =
r,co nco
Here, r,co = n2co is the dielectric constant of the core material.
Modes: Applying Maxwell’s equations shows that, due to boundary conditions for the
fields, not all angles within 0 ≤ β < βg can be used for signal propagation; moreover,
according to the condition
√
2λm π dco
sin βm = with m = 0, 1, 2, . . . und m ≤ √
πdco 2λ
only discrete angles βm are available [24.2]. The rays corresponding to these angles are
known as modes; their number becomes higher with an increasing core diameter.
The core diameter of the step-index fibre is so large that it allows several modes to
propagate (see Fig. 24.23a). Since the different modes cover different distances, a pulse
originating from the transmitting diode becomes continuously wider as the length of the
fibre becomes longer. This modal dispersion has a strong limiting effect on the bandwidth,
particularly for large fibre lengths; this is the reason why step-index fibres are no longer used
in long-distance communication. Step-index fibres made of plastic are used in simple sys-
tems with distances of up to 100 m and data rates of up to a maximum of 40 Mbit/s [24.4].
The gradient fibre uses a continuous transition of the refractive index; this causes the
modes to be sent back in the form of continuous total reflection in the direction of the fibre
axis (see Fig. 24.23b). Since in the outer portions of the core the velocity of propagation
increases due to the declining refractive index, the slated modes propagate faster than the
modes along the fibre axis; this strongly reduces the modal dispersion and increases the
bandwidth. Even though the gradient fibre does not reach the bandwidth of the single-
mode fibre described below, it still has the advantage that simpler connection systems,
with higher tolerances in terms of alignment, can be used due to the larger core diameter.
In the single-mode fibre4 , the core diameter is so small that it allows only the basic mode
to propagate (see Fig. 24.23c), thus eliminating the modal dispersion. The permissible core
diameter is derived from the condition that the angle of the mode with m = 1 must be
wider than the limiting angle:
√ nco /ncl ≈0.999
2λ
β1 > βg ⇒ dco < ≈ 10 λ
ncl 2
π 1−
nco
4 This type was formerly known as the mono-mode fibre.
1172 24 Basics
0.5
0.2
~ Fig. 24.25. Attenuation
coefficient of a typical
~
0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 waveguide of silicate glass
µm plotted over the wave length
In this case, the refractive index of the cladding is only 0.1% lower than that of the core in
order to prevent the permissible core diameter from getting too small. This fibre achieves
the highest bandwidth. One drawback is that complex connectors are required.
extremely high, thus restricting the link length to 100 m. For the light spectrum, red light-
emitting diodes (LEDs) are used in the transmitter and photodiodes are used in the receiver.
24.2.4
Comparison of Transmission Channels
The following description is restricted to a comparison of the attenuation because com-
paring data rates is only possible by taking the method of modulation into consideration.
Furthermore, the data rate of radio transmission is limited by the assigned frequency range
and not by the carrier frequency.
Figure 24.27 shows the superiority of optical waveguides as compared to coaxial cables.
Since waveguides have a very narrowband modulation, the attenuation depends solely on
the distance; with a permissible attenuation of 40 dB between transmitter and receiver it
is possible to cover up to 100 km without a repeater. In the case of coaxial cables, the
attenuation also depends on the frequency. The distance coverage is therefore determined
by the maximum permissible attenuation at the upper frequency limit.
In radio links, distance has only a logarithmic effect on attenuation; the semilogarithmic
plot in Fig. 24.27 thus shows straight lines. For the borderline case of very large distances,
the radio link is superior to all other types. However, the technically available bandwidth
must be distributed among a high number of systems. Due to the high sensitivity of nar-
rowband receivers, the permissible attenuation can reach up to 150 dB. Figure 24.27 shows
the basic attenuation only; the reduction in attenuation due to the gains of transmitter and
receiver antennas (usually 10 . . . 20 dB; more than 40 dB in large parabolic antennas) and
the additional attenuation caused by air, rain, fog and absorptions close to ground are not
taken into account. The main advantage of the radio link is, of course, its independence
from wires and cables.
Today, almost all telephone and data traffic uses fibre optic systems with several op-
tical waveguides in parallel. This is the basic technology used for the high transmission
capabilities of public and private wide-area networks such as the Internet.
a Coaxial cable
1 GHz 1 MHz
dB
140
120
Wireless
100 Wireless 1 MHz
1 GHz
80
60
40
Optical waveguide
20 (window 3)
0
~
2 5
0.01 0.1 1 10 100 1000 r Fig. 24.27. Attenuation of
km several transmission channels
1174 24 Basics
24.3
Reflection Coefficient and S Parameters
In Sect. 24.2.1 we saw that voltages and currents along a wire are described by forward and
reflected waves, that the relationship between these waves depends on the circuitry and
that an impedance transformation generally takes place; only in the case of electrically
short lines may an ideal link be assumed. This description is now extended to cover
any two-terminal and four-terminal network, i.e. all voltages and currents in a circuit
are broken down into a forward and a reflected wave, allowing a uniform description of
components and interconnecting lines. In this case, components are no longer characterized
by impedances and admittances but by the ratio between the forward and the reflected wave.
The corresponding parameters are the reflection coefficient and the S parameters.
24.3.1
Wave Parameters
The voltages of the forward (index f ) and reflected (index r) waves on a line are interrelated
by the respective currents via the characteristic impedance ZW of the line:
Vf = ZW If , Vr = ZW Ir
For this reason, one parameter is sufficient to describe each of the two waves. The following
wave parameters are used:
Vf
a = = If ZW Forward wave
ZW
(24.28)
Vr
b = = Ir ZW Reflected wave
ZW
They provide a measure for the power transported by the waves and are given in units of
root of Watt:
√ √
[a] = [b] = VA = W
The transported power is5 :
@ = ZW real
Pf = Re Vf If∗ = |a|2
(24.29)
A B ZW real
Pr = Re Vr Ir∗ = |b|2
The characteristic impedance ZW of the line is real, therefore, Vf and If as well as Vr and
Ir , are always in phase and both waves transport effective power only.
a
V
b Fig. 24.28. Equivalent representations of
the parameters in a circuit
V = ZW (a + b) (24.30)
1
I = (a − b) (24.31)
ZW
1 V
a = + I ZW (24.32)
2 ZW
1 V
b = − I ZW (24.33)
2 ZW
24.3.2
Reflection Coefficient
Converting to wave parameters means that a two-terminal network is no longer described
by the impedance Z but by the forward and reflected waves (see Fig. 24.29). The forward
wave is also called the incident wave. The ratio of the reflected wave to the incident wave
is known as the reflection coefficient r:
Reflected wave Vr b
Reflection coefficient r = = =
Incident wave Vf a
Using Z = V /I it follows from (24.32) and (24.33) that:
Vr b Z − ZW
r = = = (24.34)
Vf a Z + ZW
1176 24 Basics
V a b
ZW V Z= ZW r=
I a
b
Z=Z W r=0
ZW –1 1
Re{ Z } Re{ r }
Z= 0 r = –1
–j
Fig. 24.30. Mapping the impedance plane (Z plane) on the reflection coefficient plane (r plane) in
the case of passive two-terminal networks (Re {Z} ≥ 0)
24.3 Reflection Coefficient and S Parameters 1177
consequently it follows from (24.34) that r = 0. The incident active power Pf is fully
absorbed by the two-terminal network.
– Short circuit: Z = 0 results in r = −1, i.e. the incident and reflected waves are of
the same absolute size, but of opposing phase: b = −a. In this case, the two-terminal
network does not absorb any active power; the incident active power is fully reflected:
Pr = Pf .
– Open circuit: Z → ∞ results in r = 1; the incident and reflected waves are of the
same size and in phase: b = a. Here again, the incident active power is fully reflected:
Pr = Pf .
Figure 24.31 shows these special points and regions in the r plane.
Figure 24.32 shows the absolute value of the reflection coefficient and the power
transmission factor of ohmic resistances for ZW = 50 . The absolute value of the
reflection coefficient increases rapidly with any deviation from the matched condition
Z = R = 50 and asymptotically approaches one. The power transmission factor is
less steep around the matching point; a slight mismatch is thus not critical in terms of
power transmission. In the range 20 < Z = R < 130 , (24.34) yields |r| < 0.45 and
jIm{r}
Ohmic-inductive r = j =⇒ Z = j ZW ,L = Z W /ω
Inductive (Z = jω L ) r = 0 ⇒ Z = ZW
(Matching)
L →0 L→∞
r = –1⇒ Z = 0 r = 1⇒ Z = ∞
(Short circuit) R →0 R→∞ (Open circuit)
Re{ r }
C→∞ C→ 0
Ohmic-capacitive r = – j ⇒ Z = –j ZW ,C = 1 /(ω ZW )
Fig. 24.31. Special points and regions in the reflection coefficient plane (r plane)
1178 24 Basics
2
r kP = 1 – r r
1
0.8
0.6
r r
0.4
2 2
kP = 1 – r kP =1 – r
0.2
0
~
Fig. 24.32. Absolute value of the reflection coefficient and power transmission
factorkP = 1 − |r|2 for ohmic resistances where ZW = 50
(24.36) yields kP = 1 − |r|2 > 0.8. In this case the transmission power loss is below 1 dB
(10 log kP = − 0.97 dB).
Here, Vf (0) and Vr (0) are the voltages of the incident and the reflected waves at the point
z = 0. Equation (24.28) leads to the waves a(z) and b(z) along the line:
Vf (z) Vf (0) − γL z Vr (z) Vr (0) γL z
a(z) = = e , b(z) = = e
ZW ZW ZW ZW
This allows the reflection coefficients r1 and r2 to be calculated:
b1 b(0) Vr (0) b2 b(l) Vr (0) 2γL l
r1 = = = , r2 = = = e
a1 a(0) Vf (0) a2 a(l) Vf (0)
r1 a1 ZW a2 r2
Z2
Z1 b1 gL = a L + j b L b2 Z2
r plane r plane
j j
l= λ Inductive
l= λ
8 l = 3λ
16 16 = – 4πl
λ
–1 l= λ 1 –1 Open circuit 1
l= 0 4
Short circuit
l
l = 5λ
l = 7λ 16
16 l = 3λ = – 4π l
8 λ
Capacitive
–j –j
a Resistance: Z2 = R2 = Z W /3, r2 = –1/2 b Short circuit (r2 = –1) and open circuit (r2 = 1)
This means that the line causes an attenuation of the reflection coefficient by double the
value of the attenuation constant αL and a shift double the phase constant βL .
The case of a loss-free line is of particular importance. If αL = 0 it follows from
(24.37):
4πl
βL =2π/λ −j ϕ=− 4πl/λ
r1 = r2 e − 2jβL l = r2 e λ = r2 e j ϕ (24.38)
In this case, the reflection coefficient is only rotated clockwise by two turns per wave length:
l = λ ⇒ ϕ = − 4π . Figure 24.34a demonstrates this with a resistance Z2 = R2 = ZW /3
and r2 = − 1/2 and a gradual increase in the line length by l = λ/16. First, the reflection
coefficient enters the ohmic inductive region. For l = λ/4 (ϕ = − π ) the condition of
r1 = − r2 = 1/2 is reached with Z1 = ZW 2 /R = 3Z . This property of a λ/4 line
2 W
was already described in (24.22) and Fig. 24.8. With a further increase in line length, the
reflection coefficient passes through the ohmic capacitive region until the starting point
r1 = r2 is reached for l = λ/2 (ϕ = − 2π ). This means that the reflection coefficient r1
is periodic with l = λ/2.
Figure 24.34b shows that a short short-circuited line (r2 = −1) has an inductive effect
and a short open-circuited line (r2 = 1) has a capacitive effect; this, too, has already been
described in (24.23) and (24.24) and in Fig. 24.10. With l = λ/4 the short circuit changes
to an open circuit and the open circuit to a short circuit.
Termination with the characteristic impedance (Z2 = ZW ) results in r2 = 0. In this
case, the rotation is without effect; r1 = 0 and Z1 = ZW , regardless of the length of
the line.
1180 24 Basics
r1 ZW r2
V1 V2 Z2
g L = j 2 π/l
1+ r1 1+ r2
r1 1+ r r2
30° r 30°
r
r
1 1 1 1 1
1+ r
1+ r
V(z) l /4
Vmax
Vf
Vmin
0
0 l/8 l/4 3 l/ 4 l = l /2 z
Fig. 24.35. Standing wave on a loss-free line of length λ/2 for r2 = 0.5 e j 30◦
24.3.3
Wave Source
A signal source with an internal resistance is called a wave source. A wave source emits an
independent wave while the passive, two-terminal networks discussed so far only reflect
incident waves. Figure 24.36 shows a wave source with the related parameters.
1182 24 Basics
Zg Z = ZW r= 0 b= ag = 0
rg bg a
Vg V r
Z
ag b
Fig. 24.36. Wave source
V Vg % &
bg,0 = = 1 − rg (24.45)
ZW 2 ZW
Available Power
For high-frequency amplifiers the available power gain is usually specified. The power
at the amplifier output is not related to the power taken from the source but rather to the
available power PA,g . The available power is the maximum active power that can be taken
from a source with power matching7 :
|Vg |2 Zg =Rg |Vg |2
PA,g = A B = (24.47)
4 Re Zg 4Rg
For calculations with the wave parameters, a notation with bg,0 and rg is required. From
(24.45) it follows:
4 ZW |bg,0 |2
|Vg |2 =
|1 − rg |2
With
. ,
A B 1 + rg (1 + rg )(1 − rg∗ ) 1 − |rg |2
Re Zg = Re ZW = Re ZW = Z W
1 − rg |1 − rg |2 |1 − rg |2
it follows by insertion into (24.47):
|bg,0 |2
PA,g = (24.48)
1 − |rg |2
It should be noted that bg,0 also depends on rg , i.e. |rg | → 1 does not result in PA,g → ∞,
instead PA,g = 0 for rg = 1 (a source with Zg = ∞ means no power output) and
PA,g = ∞ for rg = − 1 (for a source with Zg = 0 there is no power limitation).
24.3.4
S Parameters
The description with the help of the wave parameters is now applied to four-terminal
networks by converting the voltages and currents into the corresponding waves using
(24.32) and (24.33) (see Fig. 24.37):
1 V1 1 V1
a1 = + I 1 ZW , b1 = − I 1 ZW
2 ZW 2 ZW
1 V2 1 V2
a2 = + I 2 ZW , b2 = − I 2 ZW
2 ZW 2 ZW
Here, a1 and a2 are the incident waves and b1 and b2 are the reflected or outgoing waves.
S Matrix
The relationships between the waves are expressed in the form of a matrix equation:
b1 S11 S12 a1
= (24.49)
b2 S21 S22 a2
Parameters S11 . . . S22 are known as scattering parameters or S parameters and form the S
matrix. The description of a four-terminal network with S parameters is equivalent to the
description of other four-terminal parameters, e.g. the Y parameters shown in Fig. 24.30
or the Z or H parameters. However, the S parameters are normalized to the characteristic
impedance ZW and, for this reason, this value must always be specified. Figure 24.38
I1 I2
r1 = S11 a1 a2 = 0 rL = 0
S ZW
RL = Z W
b1 = S11 a1 b2 = S 21 a1
rg = 0 a1 = 0 a2 r2 = S 22
Rg = ZW S ZW
b1 = S12 a 2 b2 = S 22 a 2
Fig. 24.38. External circuitry for determining the S parameters S11 and S21 (top) and S12 and S22
(below)
shows the external circuitry of a four-terminal network for determining the S parameters.
In the paragraphs below we will refer to the left pair of terminals as the input and the right
pair of terminals as the output, but will continue to use the indices 1 and 2.
Input reflection coefficient S11 : Parameter S11 corresponds to the input reflection co-
efficient in a system where the output is terminated by the characteristic impedance:
b1
(24.34)
S11 =
= r 1
= r1
(24.50)
a1 a2 =0 rL =0 RL =ZW
This value represents the input impedance Zi when operated with a load RL = ZW :
V1
(24.35) 1 + r1
1 + S11
Zi
=
= ZW
= ZW
RL =ZW I 1 RL =ZW 1−r 1 RL =ZW 1−S 11
Output reflection coefficient S22 : Parameter S22 corresponds to the output reflection
coefficient in a system where the input is terminated with the characteristic impedance:
b2
(24.34)
S22 =
= r 2
= r2
(24.51)
a 2 a1 =0 rg =0 Rg =ZW
This value represents the output impedance Zo when operated with a source where Rg =
ZW :
V2
(24.35) 1 + r2
1 + S22
Zo
= = ZW = ZW
Rg =ZW I
2 Rg =ZW 1−r
2 Rg =ZW 1−S 22
Forward transmission coefficient S21 : Parameter S21 is called the forward transmission
coefficient in a system where the output is terminated by the characteristic impedance, and
it describes the transmission characteristic from the input to the output:
b2
S21 = (24.52)
a1
a2 =0
For further clarification and determination of the relationship between S21 and the overall
gain AB = V2 /Vg , let us look at the circuit shown in Fig. 24.39 where the input is connected
24.3 Reflection Coefficient and S Parameters 1185
Rg = ZW
a1 a2 = 0
Vg
bg,0 =
Vg V1 S ZW
V2 RL = ZW
2 ZW b1 = S11 a1 b2 = S 21 a1
to a source with Rg = ZW and the output to a load RL = ZW . The output voltage is:
(24.30) a2 =0
V2 = ZW (a2 + b2 ) = ZW b2 = ZW S21 a1 (24.53)
The incident wave a1 corresponds to the independent wave bg,0 of the source since there
is no reflected portion because Rg = ZW :
(24.46) Vg
a1 = bg,0 =
2 ZW
After inserting into (24.53) and solving for S21 :
2V2
S21 = = 2AB
(24.54)
Vg Rg =RL =ZW
Thus, S21 corresponds to double the overall gain if both sides are connected to the charac-
teristic impedance.
Reverse transmission coefficient S12 : Parameter S12 is called the reverse transmission
coefficient in a system where the input is terminated by the characteristic impedance, and
it describes the transmission characteristic from the output to the input:
b1
S12 = (24.55)
a
2 a1 =0
with terminating resistances Rg = RL = ZW which are connected with lines with the
characteristic impedance ZW . In this case, there is no impedance transformation, i.e. the
terminating conditions are complied with for all frequencies independent of the length of
the lead.
Another advantage of the S parameters is the fact that they are measured together
with the terminating resistances that are also present in normal operation. A four-terminal
network such as an amplifier is designed for this purpose so that the measuring conditions
do not cause a nonpermissible load. On the other hand, a short circuit generally creates too
high a current, while an open circuit creates too high a voltage due to undamped resonances
in the matching networks.
2
1 + (Y22 − Y11 ) ZW − Y ZW 1 1 − S11 + S22 − S
S11 = Y11 =
2
1 + (Y11 + Y22 ) ZW + Y ZW ZW 1 + S11 + S22 + S
− 2Y12 ZW 1 − 2S12
S12 = Y12 =
2
1 + (Y11 + Y22 ) ZW + Y ZW ZW 1 + S11 + S22 + S
− 2Y21 ZW 1 − 2S21
S21 = Y21 =
2
1 + (Y11 + Y22 ) ZW + Y ZW ZW 1 + S11 + S22 + S
2
1 + (Y11 − Y22 ) ZW − Y ZW 1 1 + S11 − S22 − S
S22 = Y22 =
2
1 + (Y11 + Y22 ) ZW + Y ZW ZW 1 + S11 + S22 + S
S Parameters of a Transistor
For a better understanding, let us look at the S parameters of a bipolar transistor in common-
emitter configuration. We shall use the small-signal model in Fig. 24.41a which is taken
8 See (4.146)–(4.150) and (4.155)–(4.157).
24.3 Reflection Coefficient and S Parameters 1187
from Fig. 2.41. The results for a FET are almost the same because the small-signal models
differ very slightly (see Fig. 3.49). For practical purposes the S parameters are always
quoted for ZW = 50 .
The low-frequency values for parameters S11 and S22 can easily be determined since at
low frequencies the transistor has no reverse transmission. These parameters correspond to
the reflection coefficient r1 at the input and r2 at the output and can be calculated directly
from the input resistance ri and the output resistance ro of the transistor at low frequencies:
(24.50) (24.34) ri − Z W
S11 = r1 =
ri + Z W
(24.51) (24.34) ro − Z W
S22 = r2 =
ro + Z W
From Fig. 24.41, we obtain for low frequencies ri = RB + rBE and ro = rCE ; conse-
quently:
RB + rBE − ZW 2ZW
S11 = ≈ 1− (24.58)
RB + rBE + ZW rBE
rCE − ZW 2ZW
S22 = ≈ 1− (24.59)
rCE + ZW rCE
For the approximations, it is assumed that RB < ZW rBE , rCE . To determine S21 we
first calculate the overall gain with Rg = RL = ZW :
rBE gm rBE ZW βZW
AB = − gm (ZW || rCE ) ≈ − = −
ZW + RB + rBE ZW + rBE ZW + rBE
Again, we make the assumption that RB < ZW rBE , rCE . From (24.54) it follows:
2βZW
S21 = 2AB ≈ − (24.60)
ZW + rBE
Since there is no reverse transmission, S12 = 0. This allows us to locate the low-frequency
S parameters in the r plane: S11 and S22 are located close to the open-circuit point r = 1,
while S12 is at the origin r = 0 and S21 on the negative real axis outside the unit circle.
Loci: The frequency response of the S parameters is presented by means of loci in the
r plane. Figure 24.42 shows these for a bipolar transistor without case using the small-
signal model of Fig. 24.41a and for a transistor with case using the small-signal model of
RB CC C LB RB CC LC C
B B' B B'
10 Ω 0.5 pF 1.5 nH 10 Ω 0.5 pF 1.5 nH
v B'E rBE CE gmvB'E rCE v B'E' rBE CE gmvB'E' rCE
520 Ω 7 pF 192 mS 5kΩ 520 Ω 7pF 192 mS 5kΩ
E'
E
LE
0.7 nH
E
a Without case b With case (simplified)
j Im {S11 } j Im {S22}
Re {S11 } Re { S22}
f
f
Without case
With case
j Im {S 21} j Im {S12 }
f f
4 8 12 16 20 Re {S12 }
0.1 0.2 0.3 0.4 0.5
Re { S21}
Fig. 24.41b which uses a simplified case model with three lead inductances. The small-
signal parameters of the transistor are determined with the help of Fig. 2.45 on page 83
for IC = 5 mA, β = 100, VA = 25 V, fC = 4 GHz and CC = 0.5 pF. These parameters
are typical of high-frequency discrete transistors of the BFR series. By inserting
gm = 192 mS , rBE = 520 , rCE = 5 k
and ZW = 50 into (24.58)–(24.60), we obtain the low-frequency values of the S parame-
ters:
S11 = 0.83 , S12 = 0 , S21 = − 16.9 , S22 = 0.98
With these values the loci in Fig. 24.42 begin with f = 0 and climb to f = 6 GHz.
Without case, S11 and S22 are in the ohmic-capacitive region (Im {r} < 0). Due to the lead
inductances, with case there is a series resonance at both the input and output which causes
the impedances to become ohmic (Im {r} = 0).Above the resonant frequencies, S11 and S22
are inductive (Im {r} > 0). The case has very little effect on S21; with increasing frequency
the magnitude declines while the phase is shifted by approximately 180◦ . Without case,
the magnitude of S12 remains below 0.07 even at very high frequencies, i.e. the reverse
transmission remains relatively low. With case, the reverse transmission increases markedly
due to the lead inductances. As the reverse transmission is a measure of stability (reverse
transmission → feedback → oscillator), high-frequency discrete transistors with relatively
long leads are particularly susceptible to parasitic oscillations which is the reason why, in
24.3 Reflection Coefficient and S Parameters 1189
j Im {S11 }
Im {Y11 }
mS 25
4 GHz 6 GHz
3 GHz
20
2 GHz
15 500 MH z
0.2 0.4 0.6 0.8 1 1 GHz
10 MH z
10
1 GHz Re {S11 } 200 MH z
5 100 MH z
500 MH z 100 MH z
10 MH z 2 GHz
0
5 10 15 20 25 30
200 MH z –5 3 GHz
Re {Y11}
mS
j Im { S22}
Im {Y22 }
mS 30
3 GHz
25
20
6 GHz 0.2 0.4 0.6 0.8 1
10 MH z
15 2 GHz
Re {S22 }
1 GHz
4 GHz 500 MH z 10
3 GHz 100 MH z
2 GHz 1 GHz
200 MH z 5
500 MH z
10 MH z
–5 0 5 10 15 20 25
Re {Y22 }
mS
the GHz range, SMD cases with low lead inductances are mandatory. In integrated circuits,
this problem occurs only with circuit parts connected to external leads; inside integrated
circuits, the lead inductances are usually negligibly low.
Example: Figures 24.43 and 24.44 show the S and Y parameters of the high-frequency
discrete transistor BRF93 for IC = 5 mA and VCE = 5 V. The loci of the S parameters
are determined with ZW = 50 and, with the exception of S12 , correspond well to the
general curves in Fig. 24.42. The deviation of S12 is a consequence of the simplified case
design in Fig. 24.41b.
According to Fig. 24.43, the series resonance at the input occurs at 1 GHz and at the
output at 5.5 GHz. Comparing the locus of S11 to that of Y11 shows the influence of the
reverse transmission: for S11 , which is measured with a termination with ZW at the output,
the series resonance occurs at 1 GHz, whereas for Y11 , which is measured with an output
short circuit, it occurs at 2 GHz (Im {Y11 } = 0). In this case, the operating conditions,
1190 24 Basics
j Im {S21 }
Im{ Y21 } Re { Y21}
mS 40 mS
200 MH z –40 40 80 120 160 200
100 MH z 10 MH z
500 MH z
3 GHz
–40 100 MH z
1 GHz 200 MH z
4 8 12 16 20 2 GHz
10 MH z 2 GHz – 80
6 GHz Re { S21} 1 GHz
500 MH z
– 120
– 160
– 200
j Im{ S12 }
Im{ –Y12 }
mS 3 GHz
12
10
3 GHz 4 GHz
2 GHz 8 2 GHz
1 GHz 6 GHz
10 MH z 0.2 0.4 0.6 0.8 1
6
Re{ S12}
1 GHz
4
2 500 MH z
200 MH z
10 MH z
0 2 4 6 8 10 12
Re{ –Y12 }
mS
termination with ZW and short circuit, must be seen in terms of small-signal conditions,
i.e. the output is connected to resistance ZW or to ground via a sufficiently high capacitance.
The locus of Y22 has a negative real portion between 230 MHz and 1.09 GHz; in
this region the transistor is potentially unstable. If a load YL with Re {Y22 + YL } < 0
and Im {Y22 + YL } = 0 is connected to the output9 , a parasitic oscillation occurs; here,
this takes place with inductances between 16 nH (YL = 1/(j ωL) = − j 9 mS at f =
1.09 GHz) and 550 nH (YL = 1/(j ωL) = − j 1.25 mS at f = 230 MHz). If the input is
terminated with ZW = 50 , no instability occurs since the locus of S22 remains entirely
within the unit circle of the r plane. Consequently, the output impedance is Re{Zo } > 0
9 A transistor is unstable whenever the input and output admittance of the transistor, in combination
with the admittance of the external circuitry, form a negative resistance; then Re{Y } < 0 and
Im{Y } = 0 must apply. The same applies to impedances; here Re{Z} < 0 and Im{Z} = 0 must
apply. These conditions result from the known oscillating conditions for the loop gain and the
phase of an oscillator; here the condition for the real part corresponds to the condition for the
loop gain, and the condition for the imaginary part corresponds to the condition for the phase.
24.4 Modulation Methods 1191
and the output admittance is Re{Yo } > 0. This performance is typical of high-frequency
transistors. Thus, the S parameters can be measured without stability problems while the
measurement of Y, Z or H parameters causes parasitic oscillations that render an accurate
measurement impossible.
24.4
Modulation Methods
The useful signal to be transferred generally has to be converted into a signal suitable for
transmission; this process is called modulation and the methods used are called modulation
methods. One distinguishes between a transmission in the baseband, which transmits the
useful signal in its original frequency range, and carrier-frequency transmission, which
transposes the useful signal to a higher transmit frequency. Transmission in the baseband
is typical in line transmission systems such as telephones. In its most simple form, the
useful signal can be transmitted without conversion, i.e. without modulation, as is the
case in analog telephone systems. However, there are also line transmission systems that
use carrier-frequency transmission, for example, radio and TV signals transmitted via the
broadband network. Transmission via optical waveguides is a line transmission technique
considered suitable for both methods. Wireless systems, on the other hand, must use carrier-
frequency transmission since the size of the required antennas is inversely proportional
to the transmit frequency and the direct transmission of low-frequency signals would
require extremely large antennas. Furthermore, only one transmission channel is available
for radio transmission so that the various systems have no choice but to use different
frequency ranges.
The following discussion is restricted to carrier-frequency transmission. The parame-
ters amplitude, frequency and phase of a high-frequency carrier signal
sC (t) = aC cos ωC t (24.61)
vary due to the useful signal s(t). In analog modulation methods this variation is achieved
directly by the useful signal:
– Amplitude modulation (AM): sC (t) = [aC + k1AM s(t)] cos ωC t 2
Ct
– Frequency modulation (FM): sC (t) = aC cos ωC t + kFM 0 s(τ )dτ
– Phase modulation (PM): sC (t) = aC cos [ωC t + kPM s(t)]
The parameters kAM , kFM and kPM represent the modulation depth. Figure 24.45 shows
the modulated carrier signals for these methods. Since frequency is the derivative of the
phase by time (ω = dφ/dt), these parameters are interdependent; frequency and phase
modulation are thus combined under the term angle modulation. Amplitude and frequency
modulation are the conventional methods for radio broadcasting where long and medium
wave radio uses the AM method while VHF broadcasting uses FM. These modulation
methods will be described in more detail in the following sections.
The binary transmission of digital signals is best achieved by using a two-level square-
wave signal for s(t), e.g. s(t) = 0 for zero and s(t) = 1 for one. A switch-over takes
place between two amplitudes, two frequencies or two phases. This method is known
as keying and the modulation methods are amplitude shift keying (ASK), frequency shift
keying (FSK) and phase shift keying (PSK); Fig. 24.46 shows the modulated carrier signals.
1192 24 Basics
sC aC + k AM s
aC
– aC
– aC – kAM s
sC s
aC
– aC
b Frequency modulation (FM)
sC s
aC
– aC
c Phase modulation (PM)
It is also possible to use more than two levels. In this case, the corresponding methods are
called n-ASK, n-FSK and n-PSK, where n indicates the number of levels. Therefore, the
two-level method is also called 2-ASK, 2-FSK and 2-PSK. Strictly speaking, the keying
methods are not independent techniques because what we actually have are the common
AM, FM or PM methods utilizing a specific useful signal.
There is also a multitude of other modulation methods which modulate the amplitude
and the phase. In these methods there is no simple relationship between the useful signal
s(t) and the modulated carrier signal sC (t). The notation
sC (t) = a(t) cos [ωC t + ϕ(t)] (24.62)
can be used for the general amplitude modulation a(t) and the general phase modulation
ϕ(t). The relationship between a(t), ϕ(t) and the useful signal s(t) characterizes the given
method. In most cases another notation is used that is based on a trigonometric conversion
of (24.62):
sC (t) = a(t) cos [ωC t + ϕ(t)]
24.4 Modulation Methods 1193
sC s
aC
– aC
a Amplitude shift keying (ASK)
sC s
aC
– aC
b Frequency shift keying (FSK)
sC s
aC
– aC
c Phase shift keying (PSK)
Modulation methods
Figure 24.47 gives an overview of the most important modulation methods. In all
modern carrier-frequency methods the modulator first produces the quadrature components
i(t) and q(t) from the useful signal, then an I/Q mixer generates the modulated carrier
signal according to (24.63). Section 24.4.3 will describe this in more detail. Transmission
in the baseband is performed either directly, i.e. without modulation, or with the use of
pulse modulation methods which digitize the message to be transmitted and recode it in a
suitable pulse sequence (see [24.5]). This will not be elaborated on.
24.4.1
Amplitude Modulation
In amplitude modulation (AM), the useful signal s(t) which is to be transmitted modulates
the amplitude of the carrier signal sC (t) while the phase of the carrier signal remains
constant. A distinction is made between amplitude modulation with carrier and amplitude
modulation without carrier:
,
[aC + kAM s(t)] cos ωC t AM with carrier
sC (t) = (24.66)
kAM s(t) cos ωC t AM without carrier
s(t) = as cos ωs t
kAM as kAM as
= aC cos ωC t + cos (ωC − ωs ) t + cos (ωC + ωs ) t
2 2
3 45 6 3 45 6 3 45 6
Unmodulated Useful signal in Useful signal in
carrier signal the lower sideband the upper sideband
sC,u (t) sLSB (t) sUSB (t)
The modulated carrier signal consists of the unmodulated carrier signal, a useful signal at
the frequency fC − fs in the lower sideband and a useful signal at the frequency fC + fs
in the upper sideband. For AM without carrier the unmodulated carrier signal is missing.
Since the useful signal appears twice, once in each sideband, AM is also called double-
sideband modulation. Figure 24.48 shows the signal components that occur in AM as well
as the modulated carrier signals with and without carrier.
The magnitude of the modulated carrier signal is called the envelope curve sC,E :
,
|aC + kAM s(t)| AM with carrier
sC,E =
|kAM s(t)| AM without carrier
For AM with carrier, the envelope consists of the useful signal and the carrier amplitude
providing that the modulation depth
kAM as
m = (24.68)
aC
remains below one; then:
aC + kAM s(t) > 0
Figure 24.48 shows this for m = 0.8. The useful signal can then be regained by a peak
value rectification of the modulated carrier signal with subsequent separation of the DC
portion. This type of demodulation is known as envelope detection. Due to the existence
of this simple demodulation method, the AM broadcasting system uses exclusively AM
with carrier.
s Useful signal
V 0.8
t
– 0.8
–1
– 0.4 t
– 0.4 t
sC AM without carrier
0.8
V
t
– 0.8
AM with carrier
sC Envelope curve = useful signal + carrier amplitude
1.8
V
0.2
– 0.2 t
–1
– 1.8
Modulation
To produce an amplitude-modulated signal according to (24.66), it is necessary to use a
multiplier and a sinusoidal carrier signal cos ωC t. Figure 24.50 shows this for AM with
carrier.
Instead of the sinusoidal carrier signal cos ωC t, a square wave signal with amplitude
levels 0 and 1 and the period length TC = 1/fC can be used. In this case, the multiplication
uses only the factors 0 and 1 so that the multiplier can be replaced by a switch. From the
24.4 Modulation Methods 1197
S S
useful signal band
as
fs f fg f
SC,u SC,u
aC aC
fC f fC f
SC SC BAM = 2 f g
aC aC
Lower Upper
m aC sideband sideband
2
fC – f s f C fC + f s f fC – f g fC fC + f g f
Fig. 24.49. Amplitude modulation with carrier in the frequency range. In AM without carrier, the
carrier is missing in the modulated signal.
one can see that besides the desired carrier of the frequency fC , other carrier components
also occur at uneven multiples of fC , as does a direct component. Each of these components
is modulated by the useful signal and has the corresponding sidebands. The desired carrier
with its sidebands is extracted from this mixture by means of a bandpass filter. Figure 24.51
shows the amplitude modulator with switch along with the time- and frequency-related
presentations of the signals. If the square wave signal is not symmetrical (pulse duty ratio
= 50%), additional carrier components occur at all even multiples of fC ; at the same
time the amplitude of the desired carrier decreases. The electronic switches described in
Sect. 17.2 and the mixers described in Chap. 28 can be used as the switch.
Figure 24.52 shows the sample circuit of a MOSFET used as a short circuit switch and
a two-circuit bandpass filter for extracting the desired carrier and the related sidebands.
Voltage VS corresponds to the signal aC +kAM s(t) in Fig. 24.51; it must be larger than zero
in order to obtain an AM with carrier. An amplifier is required for decoupling the switch
and the filter; Sect. 28.2 describes the dimensioning process for a two-circuit bandpass
filter.
1198 24 Basics
Multiplier
aC + k AM s( t ) sC( t )
cos ωC t
sC,u (t)
aC + kAM s( t )
t
SC,u
0 fC 3 fC 5 fC f
sC,S (t)
Switch sC,u(t)
sC,S (t) t
SC,S
Passband of the bandpass filter
Bandpass
filter
0 fs fC 3 fC 5 fC f
fC – fs fC + fs 3 fC – fs 3 fC + fs 5 fC – fs 5 fC + fs
sC( t )
t
sC( t )
SC
fC f
fC – fs fC + fs
Two-circuit
bandpass filter
R V STS
VS V ST
VT
Demodulation
Envelope detector: The envelope detector shown in Fig. 24.53 can be used to de-
modulate an AM with carrier; this consists of a peak-type rectifier with a lossy storage
circuit (RS , CS ) and a highpass filter (Cc , RL ) to cancel the DC component. The following
conditions must be met for accurate demodulation:
– The carrier frequency must be significantly higher than the maximum frequency of the
useful signal.
– The minimum of the envelope must be higher than the forward voltage of the diode.
– The time constant TS = CS (RS || RL ) of the storage circuit must be selected such that
the rectified voltage can follow the envelope curve11 .
Cc
VST Cs Rs RL VS
– The useful signal must be a pure alternating voltage signal as the highpass filter sup-
presses both the DC component of the useful signal and the DC component caused by
the carrier.
– The cutoff frequency of the highpass filter must be lower than the minimum frequency
of the useful signal.
The prime advantage of the envelope detector is its simple design. A drawback is the
nonlinearity due to the nonlinear characteristic of the diode, especially with smaller carrier
11 The capacitance C can be regarded as a short circuit in the carrier frequency range; R || R
c S L
thus becomes effective.
1200 24 Basics
amplitudes; this produces a lower modulation limit. The envelope detector is used in simple
AM radio receivers.
sC( t ) sC( t )
Multiplier cos !C t SC
sM ( t )
fC – fs fC fC + fs f
Lowpass
sM ( t )
filter
t
SM
Passband of the lowpass filter
aC + kAM s ( t )
2
0 fs 2 fC f
2 fC – fs 2 fC + fs
sC( t ) s( t )
Phase
Limiter detector Phase-locked loop (PLL)
PD
VCO
Fig. 24.55. Synchronous demodulator with switch and phase-locked loop for recovering the carrier
1202 24 Basics
24.4.2
Frequency Modulation
In frequency modulation (FM), the instant frequency or instant angular frequency
dφ ω(t) 1 dφ
ω(t) = ⇒ f (t) = =
dt 2π 2π dt
is modulated by the useful signal:
ω(t) = ωC + kFM s(t) (24.71)
To produce the modulated carrier signal, the instantaneous phase φ(t) must be generated
by integrating the instantaneous angular frequency ω(t)12 :
t
sC (t) = aC cos φ(t) = aC cos ω(τ ) dτ
0
By inserting (24.71) and performing integration, we obtain:
t
sC (t) = aC cos ωC t + kFM s(τ ) dτ (24.72)
0
This means that the frequency-modulated carrier signal corresponds to a phase-modulated
carrier signal
sC (t) = aC cos [ωC t + ϕ(t)]
with the phase:
t
ϕ(t) = kFM s(τ ) dτ
0
s Useful signal
as
– as
Phase
– aC
– aC
= aC J0 (η) cos ωC t
− aC J1 (η) cos(ωC − ωs )t + aC J1 (η) cos(ωC + ωs )t
+ aC J2 (η) cos(ωC − 2ωs )t + aC J2 (η) cos(ωC + 2ωs )t
− aC J3 (η) cos(ωC − 3ωs )t + aC J3 (η) cos(ωC + 3ωs )t
+ aC J4 (η) cos(ωC − 4ωs )t + aC J4 (η) cos(ωC + 4ωs )t
− ···
= aC J0 (η) cos ωC t Carrier (24.76)
∞
0
+ aC (−1)n Jn (η) cos(ωC − nωs )t Lower sideband
n=1
∞
0
+ aC Jn (η) cos(ωC + nωs )t Upper sideband
n=1
Jn are the Bessel functions of first order shown in Fig. 24.57 while η is the modulation
index according to (24.75). The spectrum thus consists of an infinite number of components
located on both sides of the carrier with a spacing according to the frequency of the
useful signal; they form a lower and an upper sideband. Since the magnitude of the Bessel
functions rapidly declines with a constant argument η and higher orders n, for practical
purposes, the two series in (24.76) can be interrupted after a finite number of elements. For
a better understanding, Fig. 24.58 shows the magnitude of the Bessel functions in decibel
and the spectra in decibel for three values of η. It can be seen that the spectrum widens with
an increase in η. Since the Bessel functions contain zeros, some individual components
may be zero, for example, the carrier component at η = 2.4 is missing since J0 (2.4) = 0.
Ji
1
0.9
J0 ( h )
0.8
0.7
J1 ( h )
0.6
J2 ( h ) J3 ( h ) J4 ( h )
0.5
J5 ( h ) J6 ( h ) J7 ( h ) J8 ( h )
0.4
0.3
0.2
0.1
0
1 2 3 4 5 6 7 8 9 10 h
– 0.1
– 0.2
– 0.3
J9 ( h ) J10 ( h )
– 0.4
Ji
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
dB 0
J0 ( h )
–10
–20
J1 ( h )
–30 J2 ( h )
J3 (h )
–40
J4 (h )
–50 J5 ( h )
–60 J6 ( h )
J7 ( h )
–70
J8 ( h )
–80 J9 ( h )
–90 J10 ( h )
–100
SC h = 0.1 h =1 h = 2.4
dB 0
–10
–20
–30
–40
–50
–60
–70
–80
–90
–100
fC f fC f fC f
fs fs fs
Fig. 24.58. Magnitude of the Bessel functions J0 (η) . . . J10 (η) in decibel and spectra of the
modulated carrier signal for η = 0.1/1/2.4
FM radio uses f = 75 kHz and fs,max = 15 kHz; consequently, ηmin = 5 and BFM =
180 kHz.
FM is a nonlinear modulation method. For this reason, the spectrum of the carrier
signal modulated by a general signal cannot be calculated by summing the spectra of the
individual components. Only in exceptional cases does a general signal have a spectrum
that is symmetric to the carrier. Despite these restrictions, the bandwidth formulas can
also be used for the general case; fs,max then represents the upper cutoff frequency of the
useful signal.
Modulation
A voltage-controlled oscillator (VCO) controlled by the useful signal s(t) is used as the
frequency modulator (see Fig. 24.59a). The constant kFM is then determined by the tuning
characteristic of the oscillator:
dω
kFM =
ds
Figure 24.59b shows a simple FM modulator based on a Colpitts oscillator with a tuning
diode D for modulating the frequency. The slope of tuning depends on the characteristic and
coupling of the diode to the resonant circuit; the latter is adjusted by means of capacitance
Cc . As the oscillator output signal usually contains strong harmonics, the desired signal
must be extracted by means of a bandpass filter.
FM modulators based on high-frequency oscillators are used whenever the carrier
frequency should be identical to the transmit frequency. If, however, the modulator signal is
generated on a lower intermediate frequency and later converted to the transmit frequency,
it is also possible to use low-frequency oscillators such as the emitter-coupled multivibrator
from Sect. 6.3.2.
Demodulation
Discriminator: One method of demodulating an FM signal is to convert it to an
amplitude-modulated signal with subsequent envelope detection as shown in Fig. 24.60.
First, the amplitude of the input signal is held constant and independent of the receiving
Vb Vb
Turnable
resonant circuit
Voltage-
controlled Cc C1
oscillator
L
s( t ) VCO sC( t )
VS > 0 VSC
D C2
dw
kFM =
ds
conditions by means of a limiter and a bandpass filter.At the same time, any amplitude mod-
ulation that may interfere with the demodulation process is eliminated (AM suppression).
A series connection of several differential amplifiers with feedback for the DC voltage for
operating point setting is used as a limiter (see Fig. 24.61); the resistances are selected
such that the transistors are not driven into the saturation region.
A (frequency) discriminator with frequency-dependent gain is used for the conversion
of FM to AM. Since FM frequency deviation is generally much smaller than the carrier
frequency, the relative frequency deviation is very small. Therefore, the frequency depen-
dence of the gain must be very high in the region of the carrier frequency in order to
obtain a sufficiently high sensitivity. For the slope detector (discriminator), a circuit with
resonant frequency slightly above the carrier frequency is used so that the FM-modulated
carrier signal undergoes a frequency-dependent amplification at the slope of the resonance
curve. Figure 24.62 shows the slope detector together with the following envelope detector.
Amplification and
AM suppression
Envelope
Bandpass
Limiter filter Discriminator detector
sC L( t ) sFM ( t ) FM sAM ( t )
sC( t ) s( t )
AM
DC voltage feedback
RFB
VSCL
VSC
CFB
R f Res > f C Cc
VSFM VSAM VS
L C CS RS RL
R1 f Res,1 > f C V1
VSFM
L1 C1 CS1 RS1
VS
L2 C2 CS2 RS2
R2 f Res,2 < f C V2
Fig. 24.63. Differential discriminator
Since the slope of the resonance curve is not constant, this simple circuit does not pro-
duce sufficiently linear characteristics, such that even with low modulations the distortion
factor increases rapidly. For this reason, practical applications always use the differential
discriminator shown in Fig. 24.63 which evaluates the difference between two shifted
resonance curves and produces a region with a linear characteristic (see Fig. 24.64). With
a frequency deviation of f , the linear portion of the characteristic must be 2f wide
and the following must therefore be true:
V dV ∆f Res
V2 V1 df 2∆ f
VS = V1– V2 dV1 dVS
df df
dV2
df
~
f Res,2 f C f Res,1 f fC f
The carrier frequency approximately corresponds to the mean value of both resonance
frequencies:
fRes fRes,1 ,fRes,2
fRes,1 + fRes,2
fC = fRes,1 fRes,2 ≈
2
This determines the selection of the resonant frequencies:
5f 5f
fRes,1 = fC + , fRes,2 = fC −
2 2
The bandwidth B of the two resonant circuits must be 4f ; the qualities are thus:
fRes,1 fC fRes,1 fC
Q1 = ≈ + 0.6 , Q2 = ≈ − 0.6
B 4f B 4f
The following allows the resistances to be determined:
L1 L2
R1 = Q1 , R2 = Q2
C1 C2
In practice, slightly higher resistors must be selected since the envelope detectors place an
additional load on the circuits; for CS1 , CS2 ≤ C1 , C2 and RS1 , RS2 R1 , R2 this load is
small. The time constant of the envelope detectors must be selected such that the detectors
can follow the maximum signal frequency.
PLL demodulator: The PLL demodulator shown in Fig. 24.65 is of high quality and very
easy to integrate; it is used to make the frequency of a voltage-controlled oscillator (VCO)
follow the instantaneous frequency of the modulated carrier by means of a phase-locked
loop (PLL). If the VCO has a linear characteristic and the bandwidth of the loop filter is
larger than the maximum frequency of the useful signal, then the output signal of the loop
filter is proportional to the useful signal. In practice, the PLL demodulator usually operates
on an intermediate frequency which is significantly lower than the receiving frequency;
this allows the use of a VCO with a square-wave output signal and renders the subsequent
limiter unnecessary.
24.4.3
Digital Modulation Methods
Digital modulation methods are used for the transmission of binary data. There are two
types of digital modulation: the simple keying methods derived from the analog methods
and the more sophisticated methods; they differ in both the transmission rate and the
susceptibility to errors as well as in the circuit designs used.
1210 24 Basics
Phase Loop
Limiter detector filter
sC( t ) PD s( t )
VCO
Limiter Voltage-
controlled
oscillator
Amplitude Shift Keying (2-ASK): Amplitude shift keying uses a switch for the modu-
lator in order to turn the carrier signal on and off. An envelope detector with a subsequent
comparator is used as the demodulator where the signal level below the switching threshold
of the comparator is regarded as a binary zero and above the threshold as a binary one. As
the amplitude of the received carrier signal can vary considerably, one must either use a
controlled amplifier to enhance the signal to a defined level or perform a suitable adapta-
tion of the comparator switching threshold. Adaptation of the switching threshold can be
done by a second envelope detector with a considerably larger time constant which deter-
mines the amplitude Vs,max of a binary one and maintains it in accordance with its time
constant. The comparator switching threshold is then adapted to half the carrier amplitude
(see Fig. 24.66).
Amplitude shift keying is only used in very simple systems which have a maximum
transmission rate of up to 1.2 kBit/s. The main advantage is the simple circuit design.
Amplitude shift keying with several levels (n-ASK with n > 2) that allows a higher trans-
mission rate, is not used in practical applications; there are other more suitable methods,
e.g. frequency shift keying.
Frequency Shift Keying (2-FSK): Frequency shift keying uses the same components as
analog frequency modulation. The FM modulator is switched between the two frequencies
24.4 Modulation Methods 1211
VSC C1 R1 VS
VS,binary
D2 R2 / 2
VS,max
Fig. 24.66. Demodulator for
C2 R2 / 2 amplitude shift keying with
2 automatic adaptation of the
switching threshold
f1 and f2 by the desired binary signal. The differential detector can be used as the demod-
ulator while the two resonant circuits are set to the frequencies f1 and f2 , and the output
signals of the envelope detectors are compared by a comparator. A linear discriminator
characteristic is not required in this case.
Integrated receive circuits for 2-FSK usually use the binary frequency discriminator
with an edge triggered D flip-flop shown in Fig. 24.67. The modulated carrier signal
with the frequencies fC − f for the binary zero and fC + f for the binary one is
multiplied by a cosine and a sine trigger signal thus yielding the following components:
1 1
cos (ωC ± ω) t · cos ωC t = cos (±ω) t + cos (2ωC ± ω) t
2 2
1 1
cos (ωC ± ω) t · sin ωC t = − sin (±ω) t + sin (2ωC ± ω) t
2 2
0: cos( C –∆ )t
1: cos( C+∆ )t 0 / 1: cos ∆ t
Lowpass 0: 0
cos Ct Edge 1: 1
filter Limiter
triggered
D flip-flop
D Q s( t )
sC( t )
Clk
Lowpass Limiter
sin Ct 0: D
filter
0: – sin ∆ t Clk
1: sin ∆ t 1: D
Clk
The components at the double carrier frequency are suppressed by means of lowpass filters.
If we neglect the prefactors and consider the symmetry of the cosine and sine functions,
then the output of the lowpass filters yields:
After conversion to square-wave signals by means of limiters, the binary data is obtained
from the time sequence of rising edges; an edge triggered D flip-flop is used for evaluating
the binary data. In practice, the multipliers are replaced by two electronic switches that
are actuated by two staggered square-wave signals; the resultant harmonics at multiples
of the carrier frequency are suppressed by the lowpass filters. The carrier frequency in the
receiver may not correspond to the carrier frequency in the transmitter exactly but must be
between fC − f and fC + f . In practice, the carrier frequencies in the transmitter and
receiver are derived from crystal oscillators of the same nominal frequency; this usually
makes the difference between them much smaller than the frequency deviation f .
The frequency shift keying 2-FSK is often used in simple systems with data rates of
up to several kilobits per second; 4-FSK systems are also used. However, more complex
methods are used for higher data rates because they allow higher rates at the same bandwidth
of the transmitted signal and are less prone to errors.
Modulation and demodulation: Modulation is done in two steps. In the first step, a
digital modulator generates the in-phase signal i(t) and the quadrature signal q(t) from
the binary data signal s(n). In the second step an I/Q mixer forms the modulated carrier
signal sC (t). Figure 24.68 shows the configuration of the modulator. In practice, the I/Q
mixer must be followed by a bandpass filter in order to suppress unwanted components.
This particularly applies to mixers used as switches, which is almost always the case in
practical solutions.
I/Q mixer
cos Ct
Bandpass
i (t) filter
Digital
s(n) modulator sC( t )
q( t )
–sin Ct
I/Q mixer
Lowpass
cos Ct filter
iM ( t ) i(t)
Digital
sC (t) demodulator r(n)
qM (t ) q(t)
–sin Lowpass
Ct
filter
Demodulation is also performed in two steps. In the first step, an I/Q mixer forms the
signals
Complex baseband signal: The quadrature components are combined to form a complex
baseband signal
This signal corresponds to the complex phasors known from AC calculations; the following
is true
1214 24 Basics
j 1t j Ct
SC cos ( C+ 1) t = Re {e e } SB e
j 1t
LSB LSB
USB USB
~
~
fC – fg fC fC + f1 fC + fg f – fg 0 f1 fg f
Fig. 24.70. Spectra of the signals (LSB: lower sideband; USB: upper sideband) using the
single-tone signal with baseband frequency f1 as an example
@ = @ =
v(t) = v̂ cos (ωt + ϕ) = Re v̂ e j ϕ e j ωt = Re V e j ωt
⇒ V = v̂ e j ϕ
with the complex phasor V . Correspondingly, for the modulated carrier signal, the follow-
ing applies:
@ =
sC (t) = a(t) cos [ωC t + ϕ(t)] = Re a(t) e j ϕ(t) e j ωC t
@ =
= i(t) cos ωC t − q(t) sin ωC t = Re [i(t) + j q(t)] e j ωC t
The complex phasor is dependent on time because the amplitude and phase of the real
and imaginary components are time-dependent; the result is a complex signal instead of a
complex phasor. With
@ =
sC (t) = Re sB (t) e j ωC t (24.80)
the modulated carrier signal is derived from the complex baseband signal. For practical
purposes, the term complex is usually omitted and it is simply called the baseband signal.
In the frequency domain, the transition from the modulated carrier signal to the base-
band signal corresponds to a shift in the spectrum by the carrier frequency (see Fig. 24.70).
The lower sideband is then mapped on the negative baseband frequencies and the upper
sideband to the positive baseband frequencies. The unmodulated carrier has the baseband
frequency zero. Since the sidebands are independent of one another, the spectrum is usually
asymmetrical.
The main advantages of the baseband signal are the independence of the carrier fre-
quency and the representation of the carrier status by a signal with an amplitude and phase
that correspond to the amplitude and phase of the carrier. For sinusoidal high-frequency and
intermediate-frequency signals, it is not the absolute frequency that is normally specified,
but the frequency deviation from the carrier. This deviation corresponds to the baseband
frequency.
24.4 Modulation Methods 1215
It follows:
i(t) = aC + kAM s(t) , q(t) = 0
The baseband signal is real. For the frequency-modulated carrier signal
t
sC (t) = aC cos ωC t + kFM s(τ ) dτ
0
the following is true:
@1 Ct 2 =
sC (t) = Re aC e j kFM 0 s(τ ) dτ e j ωC t
C t
⇒ sB (t) = aC e j kFM 0 s(τ ) dτ
It follows:
t t
i(t) = aC cos kFM s(τ ) dτ , q(t) = aC sin kFM s(τ ) dτ
0 0
In this case, the baseband signal is complex.
Bandwidth: The upper cutoff frequency fg,B of the complex baseband signal corre-
sponds to the maximum of the cutoff frequencies of the quadrature components. If fg,i is
the upper cutoff frequency of the in-phase signal i(t) and fg,q is the upper cutoff frequency
of the quadrature signal q(t), then:
A B
fg,B = max fg,i , fg,q
According to (24.69), the two amplitude-modulated signals i(t) cos ωC t and q(t) sin ωC t
have a bandwidth corresponding to double the upper cutoff frequency:
BAM,i = 2fg,i , BAM,q = 2fg,q
This means that the bandwidth of the modulated carrier frequency corresponds to double
the maximum of the cutoff frequencies of the quadrature components:
A B A B
B = max BAM,i , BAM,q = 2fg,B = max 2fg,i , 2fg,q (24.81)
For quadrature components, the double-sided bandwidth is always given in practice;
it corresponds to the single-sided bandwidth of amplitude-modulated signals:
Bi = 2fg,i = BAM,i , Bq = 2fg,q = BAM,q
This eliminates the factor 2, and the (single-sided) bandwidth of the modulated carrier sig-
nal, which is identical to the required transmission bandwidth, corresponds to the maximum
of the (double-sided) bandwidth of the quadrature components. This is simply referred to
as the bandwidth B and is illustrated in Fig. 24.71.
1216 24 Basics
I
Bi = 2 fg,i BAM,i = 2 fg,i
AM
modulation
~
~
– f g,i 0 f g,i f f C – f g,i fC f C + f g,i f
Q
Bq = 2 fg,q BAM,q = 2 fg,q
AM
modulation
~
~
– f g,q 0 f g,q f f C – f g,q fC f C + f g,q f
SB SC
B = 2 fg,B B = 2 fg,B
Quadrature
AM
modulation
(QAM)
~
~
Fig. 24.71. Signal bandwidths: in-phase signal i(t) (top), quadrature signal q(t) (centre) and
complex baseband signal sB (t) (bottom)
Constellation diagrams: For the transmission of a binary data signal s(n), m bit are
combined in one symbol (see Fig. 24.72); the data rate rD (clock frequency fD ) is reduced
to the symbol rate rS = rD /m (symbol clock fS = fD /m). The digital modulator assigns
a certain carrier state to each of the 2m possible symbols and generates the associated
quadrature components i and q. Mapping the 2m carrier states, as described by the respec-
tive baseband pointers sB = i + j q, in the IQ plane results in the constellation diagram of
the modulation method. Figure 24.73 shows the constellation diagrams for 2-PSK (m = 1),
4-PSK (m = 2) and 8-PSK (m = 3) together with the resulting quadrature components
for the data signal from Fig. 24.72. The assignment of the symbols to the carrier states is
carried out according to the Gray code so that adjacent carrier states differ by one bit only.
This results in a minimum bit error rate since, in most cases, erroneous symbol detection
in the demodulator caused by interference supplies an adjacent symbol and thus generates
one bit error only.
The bandwidth of the modulated carrier signal is proportional to the symbol clock and,
in practice, amounts to B ≈ (1.3 . . . 2)fS . This means that, as compared to 2-PSK for a
given bandwidth, double the data rate is achieved for 4-PSK while it is tripled for 8-PSK.
The ratio of the data rate to the bandwidth is known as the bandwidth efficiency [24.6]:
24.4 Modulation Methods 1217
Symbol clock fS = fD /2
Symbol clock fS = fD /3
Symbol clock fS = fD /4
q q q
000 100
01 11
001 101
0 1
i i i
011 111
00 10
010 110
t t t
q (t ) q (t ) q (t )
t t t
rD =mrS
rD B=(1.3...2)·fS m Bit
= = (24.82)
B (1.3 . . . 2) s · Hz
As m increases, the spacing of the carrier states decreases if the power of the modulated
carrier signal remains the same, thus increasing the susceptibility to errors. A measure of
the susceptibility is the power efficiency Eb /N0 [24.6]; this indicates by which factor the
mean energy Eb per received bit must exceed the thermal noise-power density N0 in order
1218 24 Basics
q q q
00 0
b a b a
0010 0110 1110 1010
01
1
0011 0111 1111 1011
10 11 i i i
0001 0101 1101 1001
c d c d
0000 0100 1100 1000
Continuous transitions
0 1 0 0 11 1 0 0 1 1 0 0 1 0 0 11 1 0 0 1 1 0 0 1 0 0 11 1 0 0 1 1 0
s( n) s( n) s( n)
i (t) i (t) i (t)
t t t
a a c d c d abadabcbabcb
q( t ) q( t ) q( t )
t t t
to remain at or below a given bit error rate. The power efficiency corresponds to the required
signal-to-noise ratio at the input of the demodulator multiplied by a given factor. With the
received useful signal power Pr = Eb fD (mean energy per received bit x data rate) and the
noise power Pn = N0 B (noise-power density x bandwidth), the signal-to-noise ratio is:
fD =mfS
Pr Eb fD B=(1.3...2)·fS m Eb
SNR = = = = (24.83)
Pn N0 B (1.3 . . . 2) N0
The demands for a high bandwidth efficiency (high ) and a high power efficiency (small
(Eb /N0 or SNR) are diametrically opposed. A good compromise is achieved with 4-PSK
which is also known as quadrature-phase shift keying (QPSK); this method is used quite
often.
Figure 24.74 shows the constellation diagrams of other common modulation methods.
DQPSK (differential quadri-phase shift keying) is one of the differential modulation meth-
ods in which the symbols are represented by status transitions and not by the carrier states.
In these methods, the demodulator can determine the binary data signal by the successive
comparison of two consecutive symbols without knowing the absolute phase; this makes
the demodulator comparably simple. Another differential method is MSK (minimum shift
keying); this method continuously changes the carrier phase with each data bit by ±90◦ .
The advantage of this method is the constant carrier amplitude which is independent of
the speed of the status transitions. Here, nonlinear amplifiers can be used without caus-
ing intermodulation distortions. Likewise in n-PSK and DQPSK, all states have the same
24.4 Modulation Methods 1219
amplitude, although in practice, transitions cannot occur suddenly as we will see in the sec-
tion below; this causes a change in the carrier amplitude in the transition regions. 16-QAM
(quadrature amplitude modulation) uses a 4 × 4 constellation diagram. QAM methods
feature a high bandwidth efficiency and are used whenever extremely high transmission
rates are required at limited bandwidths; systems with 64-QAM (8 × 8) and 256-QAM
(16 × 16) are also used. However, these methods require a high signal-to-noise ratio at the
input of the demodulator.
Pulse Shaping
For the quadrature components i(t) and q(t), the methods n-PSK, DQPSK and 16-QAM
provide a sequence of square pulses with the symbol duration TS = 1/fS (see Figs. 24.73
and 24.74). In this form they are not suitable for transmission as the spectrum of a square
pulse is relatively wide and decreases very slowly with increasing frequencies; the band-
width required for the transmission would be disproportionately high. A considerable
reduction in the bandwidth can be achieved by pulse shaping using suitable filters; for this
purpose the quadrature components i(t) and q(t) are filtered by pulse filters.
Cosine roll-off pulses: Cosine roll-off pulses have particularly favorable characteristics
sin (πfS t) cos (πrfS t)
s(r) (t) = with 0 < r ≤ 1
πfS t 1 − (2rfS t)2
with the spectrum
⎧
⎪
⎪ fS
⎪
⎪ 1 for |f | < (1 − r)
⎪
⎪ 2
⎪
⎨ 1 π |f | 1 − r fS fS
S (r) (f ) = 1 + cos − for (1 − r) ≤ |f | ≤ (1 + r)
⎪ 2 r fS 2 2 2
⎪
⎪
⎪
⎪
⎪
⎪ fS
⎩ 0 for |f | > (1 + r)
2
Parameter r is known as the roll-off factor and influences the (double-sided) bandwidth of
the pulse:
B = (1 + r) fS ⇒ B TS = 1 + r (24.84)
Figure 24.75 shows the spectrum of the cosine roll-off pulse. A typical value in practice is
r = 0.3 . . . 1; consequently B = (1.3 . . . 2)fS .
B = (1+ r ) f S S(r )
1
Cosine
1/2 edge
fS fS fS fS fS fS f
– (1+ r ) – – (1– r ) (1– r) (1+ r )
2 2 2 2 2 2
sR SR
dB 0
1 – 10
0.8 – 20
0.6 – 30
– 40
0.4
– 50
0.2 – 60
~
0
~
~
–3 –2 –1 0 1 2 3 t / TS –4 –2 0 2 4 f TS
s(0.3) T= 6 TS B TS = 1.3
S (0.3)
dB 0
1 10
0.8 20
0.6 30
40
0.4
50
0.2 60
~
0
~
–3 –2 –1 0 1 2 3 t / TS –2 –1 0 1 2 f TS
s(1) T= 6 TS B TS = 2
S (1)
dB 0
1 10
0.8 20
0.6 30
40
0.4
50
0.2 60
~
0
~
–3 –2 –1 0 1 2 3 t / TS –2 –1 0 1 2 f TS
Fig. 24.76. Pulses and absolute spectra: square pulse (top), cosine roll-off pulse s(0.3) with r = 0.3
(centre) and cosine roll-off pulse s(1) with r = 1 (bottom) with the pulse duration T = 6 TS .
Together with the cosine roll-off pulses the spectrum of the square pulse is shown for comparison.
Figure 24.76 shows the time signals and the spectra of the cosine roll-off pulses with
r = 0.3 and r = 1 compared to a square pulse. The spectra of the cosine roll-off pulses
have a much steeper trailing edge. The bandwidth corresponds to the width of the main
region between the two inner zero points. The components outside the main region result
from the necessary limitation of the infinitely long pulse duration; they can be reduced
to any size by lengthening the duration of the pulse. In Fig. 24.76, the pulse duration is
T = 6 TS (−3 ≤ t/TS ≤ 3). With an increasing roll-off factor the trailing edge of the
pulse becomes steeper so that the limitation has less effect.
As the cosine roll-off pulses are longer than the symbol duration TS , pulse crosstalk
occurs and is known as inter-symbol interference (ISI). A special feature of the cosine
roll-off pulse is the fact that the central maximum has the value 1, and there are zero
points interspaced by TS on both sides (see Fig. 24.76). This eliminates the inter-symbol
24.4 Modulation Methods 1221
s( n) 0 1 0 1 1 0 0 1 s( n ) 0 1 0 1 1 0 0 1 s( n ) 0 1 0 1 1 0 0 1
sR s(1 ) s(0.3 )
t t t
2 2 2
1 1 1
0 0 0
–?1 –1 –1
–2 –2 –2
– TS / 2 – TS / 4 0 TS / 4 TS / 2 – TS / 2 – TS / 4 0 TS / 4 TS / 2 – TS / 2 – TS / 4 0 TS / 4 TS / 2
interference if, in the demodulator, the symbols are sampled in the centre of the symbol
duration. Deviation from the ideal sampling time may falsify the sampled value due to
adjacent pulses to such an extent that the demodulator makes a false decision resulting in a
bit error. Knowledge of the permissible shift in the sampling time and the related reduction
in the signal-to-noise ratio can be gained from the eye pattern; for this diagram, all signal
curves that are possible within one symbol duration are calculated and plotted over one
common time axis −TS /2 < t < TS /2. Figure 24.77 shows the eye patterns for cosine
roll-off pulses with r = 0.3 and r = 1 compared to the ideal eye pattern of square pulses.
It can be seen that sampling in the centre of the pulse duration (t = 0) causes no reduction
in the signal-to-noise ratio. For any deviation from this sampling time the signal-to-noise
ratio declines; the smaller the roll-off factor r, the faster the signal-to-noise ratio declines.
The region between the lowest course of the curve for level 1 and the uppermost course
of the curve for the zero level is called the eye. With square pulses, the eye is opened to
its maximum; with cosine roll-off pulses the eye closes for r → 0. The eye opening is a
measure of the synchronization requirements in the receiver: the smaller the eye, the more
accurate the sampling instant must be.
Furthermore, the eye pattern shows that after pulse shaping the amplitude is no longer
constant. This results in a situation whereby amplitude modulation also occurs for n-PSK
and DQPSK, even though all states in the constellation diagram have the same magni-
tude. The amplitude modulation increases with a decreasing roll-off factor; this causes an
increase in the crest factor (ratio of the peak value to the (rms) effective value).
When selecting the roll-off factor it is necessary to compromise between the required
bandwidth and the opening of the eye. For r → 0 the bandwidth assumes the minimum
value B = fS . For r = 1, the opening of the eye is at its maximum and the bandwidth is
thus B = 2fS .
1222 24 Basics
cos Ct
Cosine roll-off
i S (t) bandpass (SAW)
Square-wave
modulator sC,S (t)
s( n ) with sC(t)
analog
output
q S ( t)
– sin Ct
i S (t) i (t)
Square-wave
modulator
s( n) with sC (t)
analog
output
qS (t) q(t)
– sin Ct
i S (n) i (n ) D i (t)
Square-wave
modulator 1...4 10...14 A
s( n) with sC (t)
digital 1...4 10...14 D
output
qS ( n) q( n) A q( t )
– sin Ct
c Digital cosine roll-off lowpass filters and D/A converters in the baseband
Pulse filter: Linear-phase transversal filters with finite pulse response are used for pulse
shaping. These filters contain delay elements whose output signals are weighted and added.
Normally, they function as digital FIR (finite impulse response) filters. The delay elements
can then be designed as shift registers (see Sect. 19.3). The standard design of a transversal
filter is shown in Fig. 19.12 on page 995. The transversal filter can also function as an
analog filter by using delay lines, sample-and-hold circuits or a charge-coupled device
(CCD) for signal delay. Another possibility is to use surface acoustic wave filters (SAW
filters), which use the transit time of an acoustic wave for signal delay. However, SAW
filters operate in the form of bandpass filters only.
24.4 Modulation Methods 1223
Φ1
Φ2
Vo
Φ1 Φ2 Φ1 Φ2
Vi
The most simple pulse filtering method is via a SAW bandpass filter with cosine roll-
off characteristic in the carrier range, i.e. after the I/Q mixer; the modulator thus has the
configuration shown in Fig. 24.78a. In practice, this requires no extra circuitry since a
bandpass is required after the I/Q mixer in order to suppress unwanted components (see
Fig. 24.68). An intermediate frequency must be used for the carrier frequency (fC ≈
10 . . . 100 MHz) so that the SAW filter with the required bandwidth can function.
Pulse shaping in the baseband requires separate filters for the quadrature components
i(t) and q(t). Figure 24.78b shows a modulator with analog cosine roll-off lowpass filters.
The analog transversal filter with sample-and-hold circuits and an inverting operational
amplifier for the weighted summation can be used as shown in Fig. 24.79. As the quadrature
components in the baseband have a double-sided bandwidth B = (1 + r)fS ≤ 2fS , the
clock frequency of the transversal filter must be higher than the symbol frequency fS by at
least a factor of 2 in order to meet the requirements of the sampling theorem. In practice,
this filter is usually clocked with four times the symbol frequency to increase the distance
to the aliasing components (oversampling). This means that for cosine roll-off pulses with
an impulse duration of 6TS , a filter with 6 · 4 = 24 delay elements or 48 sample-and-
hold elements is required. In modulation methods with binary quadrature components, the
delay elements can be replaced by D flip-flops as is the case in 2-PSK, 4-PSK (QPSK) and
DQPSK. To reduce the circuitry of the filters, in simple systems the cosine roll-off lowpass
filters are often realized in approximation only. If a somewhat higher bandwidth and lower
eye diagram opening are acceptable, a standard lowpass filter can be used instead of the
transversal filter.
In complex systems, pulse shaping is accomplished by digital FIR filters requiring
additional D/A converters to generate the analog quadrature components. Figure 24.78c
shows a modulator with digital cosine roll-off lowpass filters. The word length at the
filter input results from the constellation diagram and is a maximum of 4 bit (256-QAM
→ 16 × 16 constellation diagram → 4 bit each for iS (n) and qS (n)). The word length
at the output corresponds to the resolution of the D/A converter and is to be selected in
accordance with the required signal-to-noise ratio; 10 . . . 14 bit are common in practice.
Filters for modulation methods with binary quadrature components (2-PSK, QPSK and
1224 24 Basics
d0 . . . . . . . . . d11
i S (n )
or D Q D Q D Q D Q D Q D Q
qS ( n)
Clk Clk Clk Clk Clk Clk
2 fS fS
D Q D Q
4 fS Clk Q Clk Q
Fig. 24.80. Digital cosine roll-off filter with ROM for modulation methods of binary quadrature
components (iR (n), qR (n) ∈ [0; 1])
DQPSK) are particularly simple as the input signal assumes the values ±1 only and is
represented by one bit. Since the output word of a filter with a pulse length of 6TS depends
on a maximum of 7 consecutive bits, all of the 4 · 27 = 512 possible output words can be
stored in a ROM with a clock frequency of 4fS . A shift register of length 7 and both the full
and half clock frequency, i.e. 4fS and 2fS , are used for addressing. Figure 24.80 shows
this simple filter. Often the clock frequency is raised to 8fS or 16fS in order to enlarge
the distance to the aliasing components in which case a ROM of 1024 or 2048 words is
necessary.
Pulse shaping in most modern systems is performed by a digital signal processor (DSP)
which also carries out all other digital functions, i.e. all functions that are shown above
the D/A or A/D converters in Fig. 24.1b. Where the calculatory power of a standard DSP
is insufficient or the power loss of a standard DSP is too high for the required computing
power, customized DSPs with special digital components are used to speed up time-critical
functions. Such DSPs may contain, for example, two of the filters shown in Fig. 24.80 and
subsequent D/A converters.
If analog transversal filters or digital filters are used for pulse shaping, additional analog
anti-aliasing filters must be used to remove the aliasing components at multiples of the
clock frequencies; these filters are not shown in Fig. 24.78b/c.
sC
Cosine
roll-off
filter
VO
Vb
I Mq
R1
Vb
IM
I Mi
I0
Q mixer
Vb
Vb
I/Q mixer
I0
Vb
I mixer
Ct)
– sign( sin
Vb
Ct)
VCq
VCi
Vq
Vi
sign(cos
Ct)
2 fC
Ct)
q(t)
i (t)
– sign( sin
sign(cos
D Q
Clk Q
D Q
Clk Q
Generation of carrier signals
FF3
FF4
digital modulator
Q
Clk Q
D Q
Clk Q
FF5
FF6
D Q
Clk Q
FF1
FF2
Clk
D
D
QFF1
QFF2
QFF3
QFF4
s( n)
fD
fD
s( n)
2 fC
TD =1/ fD
s( n) a b c d e f g h i j
TS =1/ f S
Vi a c e g i
TC =1/ f C t
VC i
t
I Mi
t
Vq b d f h j
t
VC q
t
I Mq
t
IM
t
sC
wave signal with double the carrier frequency by the divider flip-flops FF5 and FF6. The
fundamental waves of the square-wave signals correspond to the carrier signals cos ωC t
and − sin ωC t of an ideal I/Q mixer. The current switches of the mixers are activated
with the level-equalized carrier voltages VCi and VCq resulting in square-wave currents
IMi and IMq at the output of the mixers. The summation of the mixer output signals is
done by adding the currents IMi and IMq . The summation current IM is converted into
a voltage by resistance R1 while the common-collector circuit serves as a buffer. The
modulated carrier signal sC (t) is obtained from the output voltage Vo after filtering by a
cosine roll-off bandpass filter (SAW filter). Figure 24.81 presents the modulated carrier
signal sC (t) without the delay caused by the filter, thus illustrating the relationship with
current IM .
Although all points of the QPSK constellation diagram have the same absolute value,
there is also an amplitude modulation, in addition to the phase modulation, which is caused
by the cosine roll-off filter. A diagonal transition in the constellation diagram passes the
origin and, in this case, the amplitude falls briefly back to zero.
24.5 Multiple Use and Grouping of Communication Channels 1227
24.5
Multiple Use and Grouping of Communication Channels
A two-dimensional space defined by the frequency and time axes is available for wireless
transmission of signals. The transmission channels of all data communication systems must
be arranged within this space so that there is multiple utilization. The mode of subdividing
this space is called multiplex operation.
The transmission between two communication parties can be unidirectional or bi-
directional. In unidirectional transmission, one of the parties acts as the data transmitter
while the other is the data receiver; typical examples are radio and TV broadcasting.
Unidirectional systems usually have a distributive characteristic, i.e. one transmitter serves
many receivers with the same information. Such systems are therefore called broadcast
systems and the signal distribution is known as broadcasting. In bi-directional transmission,
the two parties act as both data transmitter and data receiver. They can alternately use one
channel or separate channels for the transmission in both directions. The first case is known
as half duplex operation and the second as duplex or full duplex operation. An example
of half duplex operation is CB radio telephony which allows only one partner to speak
at any given time and requires a special change-over signal (Over!) for the transition to
the other party. Modern systems like cordless or mobile phones use duplex operation for
signal transmission requiring two channels to form one link. The method of grouping is
called duplex mode.
24.5.1
Multiplex Operation
Frequency Division Multiplex
The most important method of dividing transmission space is called frequency division
multiple access (FDMA) or frequency division multiplex. This approach permanently as-
signs a certain frequency range to each transmission channel. All channels of a certain
application are grouped together to cover the frequency range available for this applica-
tion; some examples are listed in Figs. 24.20 and 24.21 on page 1168. All communications
systems use frequency division multiple access at the uppermost level; there is no system
that utilizes the entire frequency range available. Figure 24.83a illustrates the division of
the transmission space in the frequency division multiple access mode. In this context,
the channels are also known as frequency channels. There is a frequency gap between the
channels that is required as a transition region for the filter in the receiver, therefore the
channel separation C is larger than the bandwidth B of the signals.
Frequency division multiple access requires no coordination between the systems in
adjacent channels. Each system can use the assigned channel without limitations.
Time Time
t t
Frequency Frequency Frequency
Frequency Frequency Frequency channel 1/ channel 2/ channel 3/
channel 1 channel 2 channel 3 time time time
slot 3 slot 3 slot 3
C C
Frequency Frequency Frequency
channel 1/ channel 2/ channel 3/ TC
B B B time time time
slot 2 slot 2 slot 2
~ ~
~
~
Frequency f Frequency f
a Frequency division multiple access b Frequency and time division
multiple access
A distinction must be made between time division multiple access on the data level
and time division multiple access on the transmitter level. On the data level, several data
streams are combined into one and sent by a single transmitter. Correspondingly, the
transmitted signal is received by one receiver and the resulting data stream is split up
into the original data streams. One example is the radio relay transmission of telephone
conversations. Here, for instance, 30 digitized speech signals with a data rate of 64 kbit/s
each are combined into one data stream of 1.92 Mbit/s for transmission. In this case, the
division of the transmission time into data slots only refers to the arrangement of the data
and has no influence on the transmitter or the transmit signal13 .
In the time division multiple access mode on the transmitter level, time slots are used
by different transmitters which must be coordinated in order to avoid overlapping of the
transmit times. A time gap between the time slots is needed to switch over from one
transmitter to the next. The interval TC between the beginning of two successive time slots
is therefore slightly larger than the duration TS of a time slot (see Fig. 24.83b).
The time slots are cyclically and consecutively numbered and combined into frames
where all time slots of the same number form one time channel. Figure 24.84 illustrates
this for the case of four time channels. The time channels can be further divided if m
transmitters share a single channel so that each transmitter uses one time slot in every mth
frame. This method is used, for example, in GSM radio communications (global system
for mobile communications).
The time division multiple access operation is used in communication systems where
several participants communicate with one common base station (BS) or base transceiver
station (BTS). Frequency division multiple access operation would make it necessary for
the base station to provide each participant with one transmitter and one receiver. Time
13 The term transmitter is not used here in the wider sense and thus only specifies the components
from the modulator to the antenna. This means that the components for combining the data streams
into one are not part of the transmitter.
24.5 Multiple Use and Grouping of Communication Channels 1229
Time channel 1
Fig. 24.84. Frames and time channels for time divison multiplex with four channels
division multiple access operation, on the other hand, can serve several participants with
one transmitter and one receiver. GSM mobile communication uses a time division multiple
access mode with 8 time slots enabling one GSM base station to serve a maximum of
6 · 8 = 48 participants with 6 transmitter-receiver units. Cordless telephone systems of
the DECT standard use a time division multiple access mode with 24 time slots, 12 of
which are intended for both transmission directions. Thus a DECT base station can serve a
maximum of 12 telephones with one transmitter/receiver unit. Consequently, with respect
to the connection capacity, the number of time slots should be as high as possible. This,
however, is more complex in terms of coordination and reduces the efficiency because the
ratio of time slot length to time gap between the time slots is less favorable.
Spreading
(coding)
d1 ( t ) s1 ( t) Transmitter
=1
1
c1 ( t )
d2 ( t ) s2 ( t) Transmitter Frequency
=1
2 channel
c2 ( t )
dm ( t ) sm ( t)
=1 Transmitter
m
cm ( t )
Despreading
(decoding)
r(t)
Receiver r1 (t) T
1 =1 ʃ0 B d1 ( t )
c1 ( t )
Receiver r(t)
r2 (t) T
2 =1 ʃ0 B d2 ( t )
c2 ( t )
Receiver r(t)
rm (t) T
m =1 ʃ0 B dm ( t )
cm ( t )
Fig. 24.85. Code division multiple access operation in direct sequence (CDMA or DS-CDMA)
TB
SF = (24.85)
TC
The spreading factor in Fig. 24.86 is SF = 8. The bits of the coded data stream and the
code words are called chips to distinguish them from the bits of the uncoded data stream.
Thus TB is the bit duration and TC is the chip duration.
24.5 Multiple Use and Grouping of Communication Channels 1231
c5 ( t )
Receiver
TB
5
r5 ( t ) ʃ r 5 (t) dt = 0 no contribution
0
c6 ( t )
Receiver
TB
6
r6 ( t ) ʃ r 6 (t) dt = TB "1" detected
0 ("0" results in
c7 ( t ) ʃ… = –TB )
Receiver
TB
7
r7 ( t ) ʃ r 7 (t) dt = 0 no contribution
0
In the receivers, the received signal undergoes an Exclusive-Or operation with the
code word and is integrated throughout a bit duration. This decoding operation is called
despreading. Owing to the orthogonality14 of the code words the integration only yields
a component not equal to zero in the receiver that uses the same code word as the trans-
mitter. Figure 24.86 illustrates the situation where the received signal r(t) is equal to the
transmitted signal s6 (t) of transmitter 6. Since the spreading, despreading, and addition of
the transmit signals are linear operations, the separation of a received signal consisting of
several transmit signals is achieved in the same fashion.
T ,
k = 0 for i = j
ci (t) cj (t) dt =
0 0 for i = j
1232 24 Basics
Practical realization: Figures 24.85 and 24.86 show the basic principle of code division
multiplex without the use of a special modulation method. In practice, however, the code
division multiple access mode is always used in combination with one of the known mod-
ulation methods, most commonly QPSK or DQPSK. Figure 24.87 shows the integration
of the components for code division multiple access in a system with QPSK modulation.
Spreading is done after modulation but before the roll-off filtering; despreading is done
before demodulation. The IF and RF components of the transmitter and receiver are not
shown in Fig. 24.87. The transmitter is usually that shown in Fig. 25.6c on page 1243 with
a digital I/Q mixer. In this case the components of the modulator also operate digitally and
are implemented with a digital signal processor (DSP). The preferred receiver is that with
IF sampling as shown in Fig. 25.23c on page 1266 or the direct conversion receiver shown
in Fig. 25.33 on page 1277. The components of the demodulator are also implemented
with a DSP.
When designing a transmission system with code division multiple access some ad-
ditional aspects must be taken into consideration. To illustrate these aspects, let us thus
look at a mobile communications system in which several mobile units communicate with
one common base station (see Fig. 24.88). Here, all downlink channels (base station →
mobile unit) are transmitted synchronously from the transmitter of the base station, while
the uplink channels (mobile unit → base station) operate asynchronously, i.e. without
coordination between the transmitters of the mobile units.
– The Walsh codes that were used in Fig. 24.86 are orthogonal only in synchronous
operation; if the code words are shifted in time, an accurate separation of the channels is
I/Q mixer
Roll-off
Spreading filter cos Ct
i( t )
QPSK
=1
modulator
with
d( t ) c( t )
square-
wave
=1
output
q( t )
– sin Ct
Fig. 24.87. Code division multiple access in combination with QPSK modulation: modulator (top)
and demodulator (bottom)
24.5 Multiple Use and Grouping of Communication Channels 1233
Mobile unit 1
Antenna
du,1 (t ) QPSK/CDMA Transmitter
cu,1 (t ) modulator
DUP
dd,1 (t ) QPSK/CDMA Receiver Duplexer
cd,1 (t ) demodulator
Base station
QPSK/CDMA dd,1 (t )
modulator cd,1 (t )
QPSK/CDMA du,1 (t )
Antenna demodulator cu,1 (t )
Transmitter
DUP
Duplexer
Receiver
QPSK/CDMA dd,m (t )
modulator cd,m (t )
QPSK/CDMA du,m (t )
demodulator cu,m (t )
mobile unit m
du,m (t ) Antenna
QPSK/CDMA
Transmitter
cu,m (t ) modulator
DUP
dd,m (t ) QPSK/CDMA
Receiver Duplexer
cd,m (t ) demodulator
Fig. 24.88. Mobile communications system with QPSK modulation and code division multiple
access with separate code words for uplink (mobile unit → base station, index u) and downlink
(base station → mobile unit, index d)
no longer possible. Therefore, the use of the Walsh codes is found in downlink channels
only. The uplink channels require code words that remain approximately orthogonal even
with a time shift. A measure for this is the cross correlation function, which measures
the similarity of signals in dependence of their time shift15 . The absolute value must be
as small as possible for all code words and all time shifts. In practice, a set of binary
pseudo noise (PN) or a pseudo random binary sequence (PRBS) is usually used [24.7].
– The code words are used for separation of the channels and for spectral spreading of
the transmit signal. Often this leads to the problem of code words with little cross-
correlation causing an unfavorable spectral distribution of the transmission power. One
way of avoiding the problem is to use two codes in order to decouple the properties
regarding channel separation and spectral spreading. First, the channel separation is
performed using long codes and then the spectral spreading using short codes. Both
code words are usually of the same chip duration, where the length of the short code
corresponds to the bit duration of the uncoded data stream and the duration of the long
code covers several bits of the uncoded data stream [24.7].
– Since the code words used in practice are not exactly orthogonal, each transmit signal
generates a noise-like interference signal in all nonrelated receivers which reduces the
signal-to-noise ratio. The connecting capacity of the system is fully exploited if the
number of transmit signals has increased so much that the signal-to-noise ratio has
dropped to the minimum value required for an accurate demodulation. In this case, the
number of transmit signals is usually clearly below the number of code words. Therefore,
the connecting capacity of a practical system is not limited by the number of code words
but by the interference levels which in turn depend on the distribution of the mobile units.
This means that the connecting capacity is variable.
– The connecting capacity attains its maximum if the received signal level intended for
that receiver is higher than all other transmit signals or if all received transmit signal
levels are equal. A power control scheme must be used in order to meet this condition.
The transmission power of mobile units must be adjusted such that all uplink channels
reach the base station with the same level; the signal-to-noise ratio of all channels is
thus equal. The power of the downlink channels must be so small that the respective
mobile units are just capable of receiving, thus reducing the interfering signals in the
receivers of other mobile units.
Despite these demands and the corresponding complexity involved in the functioning of
such systems, code division multiple access systems are superior to systems with time
division multiple access. For this reason, existing systems with time division multiplex
(GSM, DECT) are increasingly being replaced by systems with code division multiplex
(UMTS, IS-95).
24.5.2
Duplex Operation
Duplex operation can be explained by way of a mobile communications system. The
channels for the two transmission directions are called the uplink channel (mobile unit →
base station) and the downlink channel (base station → mobile unit).
24.5 Multiple Use and Grouping of Communication Channels 1235
Duplex
Uplink band band gap Downlink band
Duplex separation
f uplink f downlink f
This chapter describes the design of transmitters and receivers for radio transmission. The
terms used shall have a defined meaning such that the components from the modulator up
to the transmitting antenna form the transmitter, while the components from the receiver
antenna up to the demodulator form the receiver.
The demands placed on the transmitter and receiver are clearly distinct since the trans-
mitter must process only the desired signal while the receiver must separate the desired
signal from the frequency mixture received by the antenna. Furthermore, the transmitter
handles signal levels which are constant or which vary very slightly, while the receiver
copes with extremely large level differences that depend on the distance to the transmitter.
The main challenges for the transmitter include the task of converting the useful signal into
a high-frequency transmission signal with as little interference as possible, to amplify this
signal with the highest possible efficiency and to minimize the transmission of undesirable
interference signals generated by the conversion or amplification. The main challenges for
the receiver are to filter out the desired signal even from very weak levels, while at the
same time receiving very strong signals from adjacent frequency ranges, and producing
a clear signal with a high signal-to-noise ratio and minimum intermodulation distortions.
Thus, the main obstacle for concern in transmitters is efficiency, while receivers face issues
of selection, dynamics and noise.
25.1
Transmitters
First we will look at the construction of transmitters with analog modulation, followed by
a description of transmitters with digital modulation. These descriptions are supported by
simplified block diagrams showing only the essential components.
25.1.1
Transmitters with Analogue Modulation
Transmitters with Direct Modulation
The most simple transmitter is obtained when the carrier frequency fC of the analog mod-
ulator is identical to the transmission frequency fRF . In this case, the modulator output
signal only needs to be amplified and fed to the antenna. In practice, the transmission am-
plifier must be followed by an output filter that reduces the distortion products originating
in the amplifier to an acceptable level. Figure 25.1a shows the construction of a transmitter
with direct modulation. The signal spectra are shown in Fig. 25.2.
1238 25 Transmitters and Receivers
f RF Output
filter
Analog
s( t) modulator sRF ( t )
f RF
f IF RF f RF Output
filter filter
M1
Analog
s( t) sRF ( t )
modulator
fIF f LO
S
s (t ) B
f RF
Analog 0 f
modulator
S RF
B
sRF ( t )
~
~
f RF f
S
s( t ) B
f IF
Analog 0 f
modulator
S IF
B B
sIF ( t )
f LO
M1 – f IF 0 f IF f
S M1
B B
sM1 ( t )
~
~
f LO –f IF f LO f RF = f LO + f IF f
RF filter f IF –B
2
RF filter
S RF
B
sRF ( t )
~
~
f LO f RF f
portion below the local oscillator frequency in Fig. 25.3. Then the frequency sequence
in the transmission signal is inverted; this is known as inverted mode. The receiver must
take the inverted frequency operation into account in order to correctly receive the desired
signal. For this purpose, the receiver uses a mixer operated in inverted mode.
The mixer output signal contains a signal portion at the local oscillator frequency fLO
(see Fig. 25.3). Consequently, the transition region of the RF filter (transition from the
pass band to the cutoff band) must not exceed the width fIF − B/2 to ensure that the
transmission signal lies fully within the pass band and the local oscillator signal is in the
cutoff band. Particularly suitable are surface acoustic wave (SAW) filters with their very
narrow transition region and constant group delay but whose high insertion loss (>20 dB)
is disadvantageous. Where no SAW filters are available for the desired transmission fre-
quency, LC filters or filters with dielectric resonators must be used. As these filters have
unwanted group delay distortion at the borders of the transition region, it is necessary
to select a clearly smaller transit region in order to prevent the transmission signal from
being affected. As an alternative, one may use the entire range between the portions above
and below the local oscillator frequency as the transition region and suppress the local
oscillator frequency by a separate serial or parallel resonant circuit (zero transmission at
fLO ).
With rising transmission frequencies, the ratio of the transmitter frequency to the width
of the transition region increases; hence, the quality of the RF filter must also increase:
fIF =fC B
fRF fRF
QRF ∼ ≈
fIF − B/2 fC
This results in a higher filter order and increased group delay distortions. In practice, the
intermediate frequency is made as high as possible so that the transition region becomes
wider and the RF filter quality becomes correspondingly low.
Figure 25.1c shows the construction of a transmitter with two intermediate frequencies,
while the signal spectra are presented in Fig. 25.4. Mixer M1 converts the modulator’s
output signal from the first to the second intermediate frequency. This requires a local
oscillator with the frequency fLO1 = fIF2 −fIF1 . Subsequently the portion above the local
oscillator frequency is filtered out by an IF filter. The quality of the IF filter is proportional
to the ratio of the second intermediate frequency and the width of the transition region:
fIF1 =fC B
fIF2 fIF2
QIF ∼ ≈
fIF1 − B/2 fC
25.1 Transmitters 1241
S
s( t ) B
f IF1
Analog 0 f
modulator
S IF1
B B
sIF1 ( t )
f LO1
M1 – f IF1 0 f IF1 f
S M1
B B
sM1 ( t )
~
~
IF filter f IF1 – B
2
IF filter
S IF2
B
sIF2 ( t )
f LO2
~
~
M2 f LO1 f IF2 f
S M2
B B
sM2 ( t )
~
~
RF filter f IF2 – B
2
RF filter
S HF
B
sRF ( t )
~
~
f LO2 f RF f
The conversion to the transmission frequency is achieved with a mixer M2, which is fed
by a second local oscillator with the frequency fLO 2 = fRF − fIF2 . An RF filter of the
quality
fIF2 B
fRF fRF
QRF ∼ ≈
fIF2 − B/2 fIF2
Obviously the overall quality is Q ≈ fRF /fC , which, in transmitters with one in-
termediate frequency, has to be generated by the RF filter and in transmitters with two
intermediate frequencies can be distributed to two filters:
fRF
Q = QRF QIF ∼
fC
The relative amounts can be controlled by the value of the second intermediate frequency,
more specifically, if it is relatively high then QIF > QRF , if it is relatively low then
QIF < QRF . In practice, the values selected depend on the transmission frequency and the
available filters. The planned number of units also has an important influence since for high
unit numbers customized dielectric or SAW filters can be used, but for mass applications
such as mobile communication even the design of new filter technologies is warranted. For
small batch production, on the other hand, standard filters are used. The use of LC filters
with discrete components is avoided where possible for reasons of space and calibration.
In transmitters with two intermediate frequencies, one can also operate one or both
mixers in inverted mode by filtering out the portions below the local oscillator frequency.
If both mixers are operated in inverted mode, then the transmission signal is in noninverted
mode again.
f IF ( 2)
S RF
C C C C
RF filter
B
~
~
25.1.2
Transmitters with Digital Modulation
In principle, transmitters with digital modulation are of the same design as transmitters with
analog modulation. The essential difference is that digital modulators primarily generate
the quadrature components i(t) and q(t) that are combined into a modulated carrier signal
by an I/Q mixer.
Figure 25.6a shows a digital transmitter with direct modulation. It corresponds to the
analog transmitter with direct modulation in Fig. 25.1a if the combination of digital mod-
ulator, I/Q mixer (MI and MQ) and the subsequent filter are regarded as being equivalent
to the analog modulator. The same applies to the digital transmitter with one or two in-
MI
i (t)
RF f RF Output
o filter filter
0
Digital
s( n) modulator f RF sRF ( t)
o
90
q( t )
MQ
a With direct modulation
MI
i (t) IF RF
f IF f RF Output
o
filter filter filter
0 M1
Digital
s( n) f IF sRF ( t)
modulator
o
90
f LO
q( t )
MQ
b With one intermediate frequency and an analog I/Q mixer
i (n ) IF IF
f f
filter 1 IF1 filter 2 IF2
M1
Digital Digital sIF ( n) D
s( n)
modulator I/Q mixer A
f LO1
q( n)
RF f RF Output
filter filter
f LO2 sRF ( t)
M2
c With two intermediate frequencies and a digital I/Q mixer
25.1.3
Generating Local Oscillator Frequencies
The required local oscillator frequencies are derived by phase-locked loops (PLL) from a
crystal oscillator with reference frequency fREF . Figure 25.7 depicts this for a transmitter
with one intermediate frequency and variable transmission frequency. The intermediate
frequency is fixed and is determined by the divider factors n1 and n2 :
n2
fIF = fREF
n1
The local oscillator frequency is variable in steps according to the channel spacing C.
For this purpose, the reference frequency is divided to the channel distance by the divider
factor n3 and multiplied by a PLL with the programmable divider factor n4 :
fREF n4
C = , fLO = n4 C = fREF
n3 n3
n2
PLL for the
intermediate 1
frequency
Frequency
(IF-PLL) divider
n3 K n4
PD VCO fLO = f
1 n 3 REF
n4
PLL for the
local oscillator 1
frequency
Programmable
(LO-PLL) frequency divider
The local oscillator frequency and thus the transmission frequency is adjusted by changing
the divider factor n4 . If the local oscillator frequencies are not divisible by C, then the
reference frequency must be divided by means of the divider factor n3 to the largest
common divisor of C and the local oscillator frequencies and this common divisor must
be multiplied by n4 .
Example: In Fig. 24.81 on page 1225, a QPSK modulator with I/Q mixer is to be converted
into a transmitter with one intermediate frequency that is capable of a data rate of 200 kbit/s
at a roll-off factor r = 1. A crystal oscillator with fREF = 10 MHz is to be used as a refer-
ence. The data rate fD = 200 kHz is obtained by division by a factor of 50. The carrier or
intermediate frequency is fC = fIF = 70 MHz since inexpensive SAW filters are available
for this frequency. Since the I/Q mixer in Fig. 24.81 must be driven with the frequency
2fC = 140 MHz, we select n1 = 1 and n2 = 14 for the IF PLL in Fig. 25.7. For QPSK,
the symbol frequency is equal to half the data rate fS = fD /2, resulting in a bandwidth
of B = (1 + r)fS = 200 kHz. We assume that the transmitter can use 4 channels ranging
from 433 to 434 MHz with a channel spacing of C = 250 kHz. From the transmission fre-
quencies fRF = 433.125/433.375/433.625/433.875 MHz we obtain the local oscillator
frequencies fLO = fRF − fIF = 363.125/363.375/363.625/363.875 MHz. Since these
are not multiples of C, we calculate the largest common divisor: lcd{K, fLO } = 125 kHz.
For the LO PLL this leads to n3 = 10 MHz/125 kHz = 80 and n4 = fLO /125 kHz
= 2905/2907/2909/2911. For all channels, the RF filter must allow signal transmission
without major group delay distortion and, at the same time, sufficiently attenuate the highest
local oscillator frequency. The double-tuned-circuit bandpass filter described in Sect. 26.2
can be set up for a center frequency of 434.4 MHz and a bandwidth of 10 MHz. Thus the
desired signal is attenuated by 6 dB, while the local oscillator frequency is reduced by
more than 54 dB and the portion below the local oscillator frequency by more than 70 dB.
25.2
Receivers
The receiver has the task of filtering out the desired signal from the antenna signal and
amplifying it enough to feed it to the demodulator. In most instances, the receive frequency
is variable so that different channels, for example, various radio stations, can be received.
As the signal level may vary widely depending on the distance between transmitter and
receiver, the receiver must be provided with amplifiers of variable gain and gain control
in order to compensate for the different levels of receive signals. Limiting amplifiers that
convert the receive signal into a square wave signal and subsequent filtering can be used
only for signals from transmitters with pure angle modulation.
First we shall describe receivers for analog modulation in which the receive signal is
converted to an intermediate frequency and then demodulated in an analog demodulator
(for example, detector for AM and envelope discriminator for FM). Then we shall discuss
the expansions to enable the reception of digital modulated signals.
1246 25 Transmitters and Receivers
Gain control
Pre- Tuneable
amplifier RF filter VGA
Demo-
r Ant (t) dulator
r (t)
f RF
a Direct-detection receiver
f RF f IF Gain control
Pre-
amplifier RF filter IF filter VGA
M1
rIF (t) Demo-
r Ant (t) r (t)
dulator
f LO
25.2.1
Direct-Detection Receivers
In the pioneer days of radio engineering, the direct-detector receiver shown in Fig. 25.8a
was used. The receive signal was filtered by an RF filter and fed directly to the demodulator
after a fixed or variable amplification. The RF filter needed tuning in order to receive the
signals from different radio stations. The only modulation technology that could be used
was amplitude modulation since the envelope detector was the only demodulator that
worked satisfactorily with a variable carrier frequency fC = fRF . All other demodulators
must be set up for a fixed carrier frequency or require frequency-synchronous tuning
according to the RF filter.
Besides being limited to amplitude modulation, the direct-detection receiver has other
significant draw-backs:
– The transmission frequency must be no more than two orders of magnitude greater than
the bandwidth of the signal to be received; otherwise, the quality of the RF filter becomes
too high. In the early days of broadcasting systems, there were only a few stations with
significantly differing transmission frequencies. A simple resonant circuit was therefore
sufficient to filter out the desired station.
– Tuneable filters of high quality are expensive and can only be tuned to a very limited
frequency range if the bandwidth is to be maintained. On the other hand, the resonant
circuits used in the early days allowed easy tuning by means of a variable capacitor.
– The entire amplification must be done at the transmission frequency, thus high-frequency
transistors with high quiescent currents and relatively low gains must be used.
25.2 Receivers 1247
– With increasing frequencies, the performance of envelope detectors decreases due to the
parasitic capacitance of the rectifier diode.
With the growing density of transmitting stations and the use of higher frequencies, the
direct-detection receiver soon reached its limits.
25.2.2
Superheterodyne Receivers
In the superhet(erodyne) receiver, the tuning of the RF filter is replaced by the frequency
conversion from a mixer with variable local oscillator frequency fLO . This converts the
signal to be received to a fixed intermediate frequency (IF frequency):
fIF = fRF − fLO fRF
An intermediate frequency filter (IF filter) of a substantially lower quality
fIF fRF
fIF fRF
QIF ∼ ∼ QRF
B B
is used to filter out the signal. The variable amplification and the demodulation are also done
at the IF frequency. Thus, all disadvantages of the direct-detection receiver are eliminated.
Figure 25.8b shows the construction of a superhet receiver with one intermediate frequency.
RF Filters
In the process of frequency conversion, not only the desired receive frequency
fRF = fLO + fIF
but also the image frequency
fRF,im = fLO − fIF
are converted to the IF frequency (see Fig. 25.9). This causes a region located at the
opposite side of the local oscillator frequency to be converted to the pass band of the IF
filter. In order to prevent this, the RF filter in front of the mixer must be set up such that
all desired receive frequencies are within the pass band and the related image frequencies
in the cutoff region (see Fig. 25.10). The RF filter is thus also known as the image filter.
In practice, the RF filter is designed such that the local oscillator frequencies are also in
the cutoff region. This prevents the relatively strong signal of the local oscillator from
moving backwards into the pre-amplifier and to the receiving antenna. This characteristic
is of high importance because the undesirable emission of local oscillator signals from the
Desired conversion
RF
filter
Image
frequency
conversion
Fig. 25.9. Image frequency in
f IF f RF,im = f LO – f IF f LO f RF = f LO + f IF f the superhet receiver
1248 25 Transmitters and Receivers
R Ant f IF f IF C RF
filter
~ B
~
f
f RF ,im = f LO – f IF f LO f RF = f LO + f IF
receiving antenna is a major problem in the design of receivers which comply with EMC
regulations.
In practice, the local oscillator signals are not sinusoidal but present strong harmonic
distortions. This results in additional image frequencies of higher order on both sides of
the harmonics of the local oscillator frequency which are also converted to the pass band
region of the IF filter:
fRF,im(n) = nfLO ± fIF
These image frequencies and the corresponding harmonics of the local oscillator frequency
must also be suppressed by the RF filter. The RF filter must therefore provide a high stop-
band attenuation even above the range of reception. LC filters or filters with dielectric
resonators are used in practical applications where two to four resonant circuits are typical.
These filters are called 2-, 3- or 4-pole filters. The number of poles refers to the equivalent
lowpass filter and is thus equal to the number of resonant circuits1 .
With an increasing receive frequency and a constant IF frequency, the relative difference
between the receive frequency and the image frequency becomes smaller and smaller; thus
causing the quality
fRF
QRF ∼
fIF
of the RF filter to increase. Where the separation of the receive and image frequencies
can no longer be achieved by reasonable means, it is necessary to either increase the IF
frequency in order to reduce the quality of the RF filter or to use a superhet receiver with
two intermediate frequencies.
It is also possible to configure the RF filter such that the frequency fLO − fIF below the
local oscillator frequency is used as the receive frequency fRF , while the corresponding
image frequency fRF,im = fLO + fIF is suppressed. In this case, the mixer M1 operates in
the inverted mode as the frequency sequence is inverted due to the relation fIF = fLO −fRF ;
but, with fIF = fRF − fLO , the mixer operates in the noninverted mode and the frequency
sequence remains the same.
In noninverted mode, the image frequency is below the receive frequency, while in
inverted mode, it is above. Therefore, the inverted mode is always used in cases where the
frequency range above the receive frequency has clearly weaker signals than the frequency
1 A simple resonant circuit has two poles: s = ±j ω . A filter with four resonant circuits therefore
0
has eight poles, but is still called a 4-pole filter in practice since bandpass filters with a low-
pass/bandpass transformation are calculated on the basis of an equivalent lowpass filter with half
the number of poles.
25.2 Receivers 1249
range below the receive frequency; in this way it is easier to suppress the image frequency.
The inverted mode must be compensated for in the modulator or by an inverted mode in
the transmitter.
Pre-Amplifiers
A low-noise amplifier (LNA) is used in front of the RF filter to keep the noise figure of
the receiver low (see Fig. 25.8b). Without a pre-amplifier, the noise figure is according to
(4.201):
FRFF =DRFF
FM1 − 1 GA,RFF =1/DRFF
Fr
= FRFF + = DRFF FM1
GA,RFF
Here, FRFF is the noise figure and GA,RFF is the available power gain of the RF filter and
FM1 is the noise figure at the input of mixer M1. The latter is calculated with (4.201) from
the noise figure of the mixer and the noise figures of the subsequent components. An overall
impedance matching is assumed so that the noise figure of the filter corresponds to the
power attenuation DRFF in the pass band region, and the available power gain corresponds
to the reciprocal value of the power attenuation. With the typical values DRFF ≈ 1.6 (2 dB)
and FM1 ≈ 10 (10 dB), the noise figure becomes unacceptably high: Fr
≈ 16 (12 dB).
Using a pre-amplifier with noise figure FLNA and available power gain GA,LNA the noise
figure is:
Fr
− 1 DRFF FM1 − 1
Fr = FLNA + = FLNA +
GA,LNA GA,LNA
With a sufficiently high gain this value is much smaller than the noise figure without a
pre-amplifier and in the limiting case of a very high gain it approaches the noise figure of
the pre-amplifier.
In practice, the gain of the pre-amplifier cannot be increased without limits since at this
point it is still the entire receive signal of the antenna that is amplified. This means that both
the signal to be received and, under good receiving conditions, the signals of neighboring
channels can reach relatively high levels which may overdrive a pre-amplifier with too high
a gain. In addition, a high gain in the RF range is achievable with great effort only. Therefore,
the gain is selected at a level which is high enough to reduce the noise figure of the receiver
to an acceptable level. Typical values are FLNA ≈ 2 (3 dB) and GA,LNA ≈ 10 . . . 100
(10 . . . 20 dB). In the above example, these values lead to Fr ≈ 2.15 . . . 3.5 (3.3 . . . 5.4 dB)
compared to Fr
≈ 16 (12 dB) without pre-amplification.
IF Filters
Due to the mixer, the entire pass band region of the RF filter is shifted to the intermediate
frequency range (see Fig. 25.11). Here, the channel with the desired receive frequency is
filtered out by the IF filter. For this reason the IF filter is also known as the channel filter. It
must have very steep edges since the transition region between the pass band and the cutoff
band must not be wider than the region between adjacent channels. Particularly well suited
are surface acoustic wave (SAW ) filters which, despite extremely steep edges, have almost
no group delay distortions. In contrast, the group delay distortions of LC or dielectric filters
increase with rising edge steepness. Filters with ceramic resonators (ceramic filters) are
used in applications that are relatively insensitive to group delay distortions, such is the
1250 25 Transmitters and Receivers
f IF f IF
rAnt (t) R Ant
B B
~
~
f RF,im = f LO – f IF f LO f RF = f LO + f IF f
RF filter
f IF
RF filter
R RFF
B
rRFF (t)
~
~
f LO M1 f LO f RF = f LO + f IF f
R M1
B
rM1 (t)
~
~
f IF f
IF filter IF filter
R IF
B
rIF (t)
~
~
f IF f
Fig. 25.11. Signal spectra in a superhet receiver with one intermediate frequency
case in AM broadcasting for example. In digital modulation modes, on the other hand,
group delay distortions have to be kept as low as possible and thus the use of SAW filters
is usually mandatory.
f RF f IF1 f IF2
Pre- IF IF
amplifier RF filter filter 1 filter 2 VGA
M1 M2
r IF (t)
r Ant (t)
f LO1 f LO2
Gain switchover
Demo-
dulator r (t)
must be within the cutoff band of the filter. To prevent a backwards transmission of the
second local oscillator frequency
fLO 2 = fIF1 − fIF2
this frequency must also be within the cutoff band; consequently, the quality of the filter is:
fIF1
QIF1 ∼
fIF2
After conversion to the second intermediate frequency with the mixer M2, the desired
channel is filtered out by means of IF filter 2 which acts as the channel filter.
It is possible to operate one or both mixers in inverted mode by regarding the frequencies
fLO 1 −fIF1 or fLO 2 −fIF2 below the local oscillator frequencies as the receive frequencies.
In this case, the RF filter suppresses the image frequency fRF,im = fLO 1 + fIF1 while
IF filter 1 suppresses the image frequency fIF1,im = fLO 2 + fIF2 . If only one of the
mixers is operated in inverted mode, then the frequency sequence is inverted, due to
fIF1 = fLO 1 − fRF or fIF2 = fLO 2 − fIF1 . This must be taken into account in the
demodulator or must be compensated via an inverted mode in the transmitter. If both
mixers are operated in inverted mode, the overall receiver operates in noninverted mode.
The advantage of the superhet receiver with two intermediate frequencies is that the
quality for filtering out the desired channel can be distributed to two IF filters
fIF1 fIF1 fIF2
QIF ∼ = ∼ QIF1 QIF2
B fIF2 B
which is in contrast to the superhet receiver with one intermediate frequency where the task
must be performed by one IF filter. This is required whenever the receive frequency fRF
is very high, meaning that a high (first) intermediate frequency fIF1 is required in order to
limit the quality of the RF filter or if the bandwidth B of the receive signal is very low.
~
~
f RF,im = f LO1 – f IF1 f LO1 f RF = f LO1 + f IF1 f
RF filter
f IF1
RF filter
R RFF
B
rRFF (t)
~
~
R M2
B
rM2 (t)
~
~
f IF2 f
IF filter 2
IF filter 2
R IF2
B
rIF2 (t)
~
~
f IF2 f
Fig. 25.13. Signal spectra in a superhet receiver with two intermediate frequencies
the frequency of the first local oscillator is varied by adapting the divider factors of the
corresponding PLL.
25.2 Receivers 1253
VGA
vi (t) vo (t)
a Simplified diagram
VGA
Peak value
A (VR ) measurement
v^o
VR
ʃ v^ setpoint
25.2.3
Gain Control
A variable gain amplifier (VGA) and an amplitude detector are used for gain control as
shown by a simplified diagram in Fig. 25.14a. The VGA generates the voltage
with the variable gain A(VR ) and the control voltage VR . A peak value rectifier is usually
used to determine the amplitude. By comparing the rectifier output with the setpoint value
an integrator generates the control voltage VR from the difference. Figure 25.14b shows
the equivalent circuit for the gain control.
Control Characteristic
In steady state (operating point A) we have v̂o = v̂setpoint and VR = VR,A with:
v̂setpoint
|A(VR,A )| =
v̂i
For examination of the dynamic response we linearize (25.1) at the operating point:
d|A|
d v̂o = v̂i
dVR + |A(VR )|
d v̂i
dVR A A
d|A|
VGA
Vi (s) kF Vo (s)
VR (s) –1
kR
sTI Fig. 25.15. Linear model of the
gain control
HR
dB 3 dB
kF
20dB/Dek.
~
~
1 kR v̂i,A d|A|
f-3dB = = = (25.3)
2π TR 2πTI 2πTI dVR
A
Figure 25.16 shows the frequency response. Changes to the input amplitude with a fre-
quency, that is below the cutoff frequency, are better suppressed with decreasing fre-
quency; while changes with frequencies above the cutoff frequency are amplified with
kF = |A(VR,A )|. The cutoff frequency must be less than the lower cutoff frequency of the
amplitude modulation contained in the desired signal to prevent the desired signal from
being invalidated.
According to (25.3) the cutoff frequency is proportional to the input amplitude v̂i and
to the derivative of the gain characteristic |A(VR )|. In order to prevent the cutoff frequency
from being dependent on the operating point, the condition
d|A| v̂setpoint d|A|
kR = v̂i = = const.
dVR |A(VR )| dVR
must be met; it follows:
kR VR
d|A| kR
= |A(VR )| ⇒ |A(VR )| = A0 e v̂setpoint (25.4)
dVR v̂setpoint
25.2 Receivers 1255
Therefore, the VGA must have an exponential gain characteristic. In practice, the gain is
quoted in decibel, i.e. logarithmically, thus producing a linear relationship:
kR VR
A(VR ) [dB] = A0 [dB] + · 8.68 dB
v̂setpoint
Taking the subsequent amplifier with gain AV into account, the small-signal output volt-
age is:
AV iC1 (t)R7 AV gm1 R7 vi (t)
vo (t) = − AV iC3 (t)R7 = − = −
−
VR 1 + gm1 R1 −
VR
1 + e VT 1+e VT
Vb = 3 V Vb Vb Vb Vb Vb
R3 R7 R6 Amplifier
32 k Ω 8 kΩ 32 k Ω
I R7 1.4 V
1.3 V Vo
I C2 I C3 I C5 I C6 A V = 60 dB
VR
1.3 V T2 T3 T5 T6
1 1 1 1
0.75 V 0.75 V
I C1,A = 200 µ A I C4,A = 200 µ A
C1
0.7 V T1 T4 0.7 V
2 2
100 pF
Vi
R2 R1 R4 R5
10 k Ω 200 Ω 200Ω 10 k Ω
2 Current I
C1 corresponds to the quiescent current 2I0 of the differential amplifier.
1256 25 Transmitters and Receivers
A 180mV 0.33dB/mV
90
dB
80
70
60
50
60dB
40
30
20
10
0
– 300 – 200 – 100 0 100 V R
Fig. 25.18. Characteristic of the VGA of
mV Fig. 25.17 (f = 3 MHz)
A
90 VR = 0 V
dB
80
– 50mV
70
60 – 100mV
50
– 150mV
40
30 – 200mV
20
10 – 250mV
0
– 10 – 300mV
– 20 Fig. 25.19. Frequency
~
The control range is VR < − 2VT . Here, the constant value of one is negligible with respect
to the exponential function, and the desired exponential gain characteristic is:
V V
AV gm1 R7 VR AV gm1 R7 VR
vo (t) ≈ − e T vi (t) ⇒ A(VR ) ≈ − e T (25.5)
1 + gm1 R1 1 + gm1 R1
Figure 25.18 shows the characteristic of the VGA in Fig. 25.17 for a signal frequency of
3 MHz. The control range covers 60 dB with a slope of 0.33 dB/mV. It is limited upward by
the deviation from the exponential shape and downward by the reverse attenuation of the
VGA cell. The latter depends on the parasitic capacitances and becomes worse with a fre-
quency increase. Figure 25.19 shows the frequency response for different control voltages.
Above 10 MHz the gain drops at a rate of 20 dB/decade; thus, the control range narrows
accordingly. In this region, the minimum gain increases to 25 dB due to the declining
reverse attenuation of the VGA cell.
The change in the current distribution also changes the DC voltage at the output of
the VGA cell making the galvanic coupling with the subsequent amplifier difficult. The
change can be compensated by connecting a second VGA cell with the same quiescent
current (T4 . . . T6 , R4 . . . R6 ) in parallel and inversely controlling the differential amplifier.
Then we have
IR7,A = IC3,A + IC6,A = IC1,A = IC4,A
25.2 Receivers 1257
v̂setpoint
kR = (25.6)
VT
Here, v̂setpoint is the desired amplitude at the VGA output (see Fig. 25.14b). The time
constant TI of the integrator can be calculated from v̂setpoint and the cutoff frequency
f−3dB :
kR v̂setpoint
TI = = (25.7)
2πf-3dB 2πf-3dB VT
Level Detection
In addition to the amplitude-controlled useful signal, many systems require a measure
for the received level of the useful signal. Typical examples include the VHF broadcasting
system with automatic stereo/mono switchover controlled by the received level, and mobile
communication in which several base stations receive a signal transmitted from a mobile
unit and then the base station with the highest received level takes over the communication.
Level detection can be based on the control voltage of the gain control. If the control-
lable amplifier has an exponential characteristic, the control voltage VR is a logarithmic
measure for the received level. In steady state, (25.4) provides:
kR VR
v̂setpoint v̂setpoint
v̂setpoint = |A(VR )| v̂i = A0 v̂i e v̂setpoint ⇒ VR = ln
kR A0 v̂i
1258 25 Transmitters and Receivers
f LO
RF filter IF filter VGA
rIF (t)
rAnt (t)
RSSI
n VGA
Micro- A
controller D
The microcontroller can evaluate the received level by averaging relatively quickly
the RSSI signal, while at the same time taking into account the current amplifier setting.
The microcontroller can thus programme all controllable amplifiers in one step with high
accuracy, thus significantly reducing the transient time. Following this pre-setting, the
duration of averaging is increased so that only amplitude variations with frequencies below
the lower limit of the desired signal amplitude modulation are adjusted. In practice, the
gain is set by a central microcontroller that controls the overall system. Therefore, it
is particularly easy to adapt the performance to the given mode of operation (normal
reception, channel switching, search mode, etc.).
25.2.4
Dynamic Range of a Receiver
The dynamic range of a receiver corresponds to the difference between the maximum and
minimum received level. The maximum received level is determined by the maximum
permissible intermodulation distortions and depends on the intercept point of the receiver.
The minimum received level follows from the minimum signal-to-noise ratio at the input
of the demodulator and depends on the noise figure of the receiver. In turn, the intercept
point and noise figure of the receiver are dependent on the intercept points, the noise figures
and the gain factors of the individual components. Therefore, the main task in designing
a receiver is the selection of components with suitable characteristics. On one hand, the
performance of the signal processing chain is limited by its weakest member, while on
the other hand, components with unnecessarily high characteristics are either expensive
or have a high power consumption. Thus, the selection of components must be balanced
between the two extremes in order to achieve an optimum result.
In the example below, the dynamic range of the receiver shown in Fig. 25.21 is cal-
culated. It is assumed that the receiver picks up channels in the range of 434 MHz with
a bandwidth of B = 200 kHz and a channel spacing of C = 250 kHz. We use a receiver
with one intermediate frequency fIF = 70 MHz. Two identical RF amplifiers with gain
A = 12 dB are used in the RF stage where RF amplifier 1 corresponds to the pre-amplifier
in Fig. 25.8a. The RF filter for suppressing the image frequency
fRF,im = fRF − 2fIF = 434 MHz − 2 · 70 MHz = 294 MHz
is arranged between the two RF amplifiers and is designed as a two-circuit bandpass filter
with an attenuation of 6 dB (A = − 6 dB). A programmable attenuator performs the gain
switching to adapt the received level. The attenuator performance can be switched between
1 dB and 25 dB (A1 = − 1 dB, A2 = − 25 dB). It should be noted in this respect that
the noise figures of a passive reactive filter and an attenuator correspond to the respective
attenuation.A diode mixer with a conversion loss of 7 dB (A = −7 dB) and a noise figure of
7 dB is used as the mixer. Two identical IF amplifiers with gain A = 25 dB, and the IF filter
arranged between them, follow in the IF stage. The IF filter is a surface acoustic wave (SAW)
filter with a center frequency of 70 MHz and a bandwidth of 200 kHz. The attenuation is
24 dB (A = −24 dB). This is followed by a variable gain IF amplifier that provides the
RF RF IF IF variable
1260
v i (t) vo ( t)
A = 12 dB A = – 6 dB A1 = –1 dB A = 12 dB A = –7dB A = 25 dB A = – 24 dB A = 25 dB A 1 = 67 dB
F = 3 dB F = 6 dB F1 = 1 dB F = 3 dB F = 7 dB F = 4 dB F = 24 dB F = 4 dB F1 = 20 dB
IP3 = 8 dBm A 2 = – 25 dB IP3 = 8 dBm IP3 = 8 dBm IP3 = 18 dBm A 2 = 13 dB
IP 3 = –7dBm
Signal level min – 103 dBm – 91 dBm – 97 dBm – 98 dBm – 86 dBm – 93 dBm – 68 dBm – 92 dBm – 67 dBm 0 dBm
[dBm] max – 25 dBm – 13 dBm – 19 dBm – 44 dBm – 32 dBm – 39 dBm – 14 dBm – 38 dBm – 13 dBm 0 dBm
25 Transmitters and Receivers
Signal level min 1.6 µV 6.3 µV 3.2 µV 2.8 µV 11.2 µV 5 µV 89 µV 5.6 µV 100 µV 224 mV
[V] max 12.6 mV 50 mV 25 mV 1.4 mV 5.6 mV 2.5 mV 45 mV 2.8 mV 50 mV 224 mV
∑ FZ,r = 2.025
Calculation of vo,IP3 0.56 V 0.1 V 0.56 V 0.07 V 1.78 V
Fr ≈ 3 (4.8 dB)
intercept point Π |A| 4 2 0.11 0.45 0.2 3.5
IP3 (i)
vo,IP3 0.14 V 0.91 V 1.24 V 0.35 V 0.5 V ⇒ v i,IP3 = 0.124 V ( – 5.1 dBm)
25.2 Receivers 1261
subsequent demodulator with a constant output level of 0 dBm (veff = 224 mV) 3 . It is
based on the VGA of Fig. 25.17 and has a high noise figure of 20 dB which is typical for
VGA cells.
2
veff veff [V]
P = = 1 mW ⇒ veff = 223.6 mV ⇒ veff [dBm] = 20 log
50 0.2236 V
1262 25 Transmitters and Receivers
IM3 which is characterized by the intercept point IP3. These relationships are described in
Sect. 4.2.3 on page 426 by way of the amplitude of sinusoidal signals. In telecommunication
engineering, the levels are usually given in dBm or the corresponding rms values, but this
does not influence the intermodulation ratio IM3. From (4.184) it follows:
v̂i,IP3 2 vi,IP3 2
IM3 ≈ = (25.12)
v̂i vi
√ √
Here, vi,IP3 and vi are the rms values and v̂i,IP3 = 2vi,IP3 and v̂i = 2vi are the
amplitudes of the intercept point IP3 and the received signal, both of which are related
to the input of the receiver. In practice, the intermodulation ratio is quoted in decibel and
the rms values of the intercept point and the received signals are quoted in dBm. It thus
follows:
% &
IM3 [dB] ≈ 2 vi,IP3 [dBm] − vi [dBm] (25.13)
The intercept point is determined by means of a two-tone signal; therefore, the inter-
modulation ratios according to (25.12) and (25.13) are also only valid for a two-tone signal.
The receiver receives a very complex signal that is composed of the desired receive signal
and the signals of adjacent channels. For this reason, the intermodulation ratio cannot be
given; whereas, in practice, the two-tone intermodulation ratio is used as a substitute. For
this purpose, the permissible nonlinearity is determined for the case of two neighboring
channels with the same level, then the relevant two-tone intermodulation ratio and the in-
tercept point are calculated from this value. We will not go into details on this and assume
that the required two-tone intermodulation ratio is known.
The intercept point vi,IP3 of the receiver is calculated from the intercept points of the
components, but only the components up to the last IF filter are taken into account since,
behind this filter, the signals of all adjacent channels are suppressed. Figure 25.21 shows
the output intercept points of the components in dBm. The lower portion of the table
contains the resulting effective values vo,IP3 , which are converted to the input by means of
the related gain factors from the receiver input to the output of each component ( |A|):
(i) vo,IP3
vo,IP3 =
|A|
In Sect. 4.2.3 we demonstrated that the inverse square of the intercept points of 3rd order
of a series connection must be added (see page 439):
1 1
2
= (i) 2
vi,IP3 vo,IP3
For the receiver in Fig. 25.21, this leads to vi,IP3 = 0.124 V (− 5.1 dBm). QPSK usually
requires an intermodulation ratio of IM3 ≈ 10000 (40 dB); the maximum received level
is obtained with (25.13):
IM3 [dB]
Pi,max [dBm] = vi [dBm] = vi,IP3 [dBm] −
2
40 dB
= − 5.1 dBm − ≈ − 25 dBm
2
This corresponds to an rms value of 12.6 mV.
The component intercept points converted to the input indicate the contribution which
the components make to the intercept point of the receiver; a higher value is better than
1264 25 Transmitters and Receivers
a lower value. In Fig. 25.21, the contribution of the first RF amplifier dominates and is
enhanced by squaring the values for the inverted quadratic addition. The dominance of the
intercept point of the first RF amplifier is typical of receivers and an improvement at this
point is difficult to achieve and only possible at the cost of the noise figure or the current
consumption.
Dynamic Range
The maximum dynamic range of the receiver can be determined from the minimum and
maximum received levels:
Dmax [dB] = Pi,max [dBm] − P i,min [dBm] (25.14)
For the receiver in Fig.25.21 it follows:
Dmax = − 25 dBm − (− 103 dBm) = 78 dB
The signal levels for borderline cases are given in Fig. 25.21 in both dBm and Volt. One
should note that the signal levels are related to the portion of the received useful signal.
The overall levels may be significantly higher if adjacent channels with higher levels exist.
Only after the last IF filter are the levels of the useful signal and the overall system equal;
in this case, the signals of all adjacent channels are suppressed.
The available dynamic range depends on the signal levels in the adjacent channels and
may be much lower than the maximum dynamic range. We examine the case of the receiver
in Fig. 25.21 receiving an adjacent channel with a maximum level of Pi,max = − 25 dBm.
In this case, there are intermodulation distortions, some of which add to the received
channel and cause a noise-like interference with a level that is clearly higher than the
thermal noise level. Therefore, the level of the useful signal must be above the minimum
received level P i,min = − 103 dBm by the same factor to guarantee the required signal-
to-noise ratio. This reduction in sensitivity is particularly undesirable in radio receivers
and is the reason why weak stations which are located close to powerful stations can
not be received. The same problem occurs in base stations of mobile communication
systems that must be capable of receiving very different signal levels from several mobile
units. The mobile units themselves are less critical since they normally use the highest
received level in their communication with the base station. Blocking of one mobile unit
by other mobile units operating in the immediate vicinity is prevented by using a different
frequency range for communication from the mobile units to the base stations (uplink)
than from the base stations to the mobile units (downlink) (see Fig. 24.21). The separation
of uplink and downlink ranges is achieved by a duplexer consisting of two bandpass filters.
Figure 25.22 shows an example of a mobile unit for GSM900. The two ranges are separated
by a frequency gap which is needed as the transition region for the bandpass filters of the
duplexer. One disadvantage of this is the increase in the noise figure caused by the duplexer.
The noise figure is increased by the power attenuation DD of the duplexer:
(4.201) Fr − 1 FD =1/GA,D =DD
Fr
= FD + = DD + DD (Fr − 1) = DD Fr
GA,D
Here, Fr is the noise figure of the receiver without a duplexer. Consequently:
Fr
[dB] = DD [dB] + Fr [dB]
A typical value for duplexers is DD ≈ 3 . . . 4 dB. This means that the maximum dynamic
range is reduced by the factor DD when a duplexer is used. On the other hand, the available
25.2 Receivers 1265
to antenna
Duplexer
Uplink Downlink
890...915 935...960
MHz MHz
Fig. 25.22. Separation of the uplink and downlink regions by means of the duplexer in case of a
mobile unit for GSM900
dynamic range increases significantly when the unit is operated in the vicinity of other
mobile units since their comparably strong transmission signals can no longer reach the
receiver.
The available dynamic range depends on the stop-band attenuation of the RF and IF
filters. If, for example, the last IF filter has a stop-band attenuation of 50 dB, but the level of
the adjacent channel is 50 dB higher, then the levels of the desired and the adjacent channels
at the output of the filter are the same, and no reception is possible. The position of the
image frequencies and the levels that occur at these frequencies, which are determined by
the selected IF frequencies, also have an effect on the available dynamic range. Therefore,
besides the above-mentioned considerations, a multitude of additional considerations must
be taken into account when designing a receiver.
25.2.5
Receivers for Digital Modulation
Receivers for digital modulation methods have basically the same design as receivers for
analog modulation but differ in terms of the demodulator. While analog demodulators
process the IF signal directly, digital demodulators perform an additional frequency con-
version by means of an I/Q mixer in order to provide the quadrature components i(t) and
q(t).
The principle construction of a demodulator for digital modulation methods is illus-
trated in Fig. 24.69; Fig. 25.23a shows the same version with an additional gain control.
The input signal is given by the IF signal rIF (t) of a superhet receiver with one or two
intermediate frequencies (see Fig. 25.8b or Fig. 25.12) and corresponds to the carrier signal
sC (t) of Fig. 24.69. The quadrature components i(t) and q(t) are derived with the help of
an I/Q mixer and two lowpass filters and fed to the demodulator.
Compared to a receiver for analog modulation, the lowpass filters behind the I/Q mixer
act like an additional filter. Therefore, in a receiver for digital modulation, the desired
channel is normally not filtered out by the last IF filter but by the lowpass filters behind the
I/Q mixer; this is the reason they are called channel filters in Fig. 25.23a. With respect to the
filtering function, a receiver for digital modulation with one intermediate frequency already
has the same characteristics as a receiver for analog modulation with two intermediate
frequencies. Figure 25.24 shows the relevant signal spectra for the i branch which are the
same as for the q branch.
However, channel filtering behind the I/Q mixer has two disadvantages:
1266 25 Transmitters and Receivers
Lowpass filters
I/Q mixer (channel filters) Gain control
MI
i (t)
o
0 Demodulator
with
r IF ( t) f IF analog r ( n)
o inputs
90 q(t)
MQ
Anti-aliasing Digital
I/Q mixer filter channel filters
MI
A i ( n)
Demodulator
o D
0 with digital
inputs
rIF ( t) f IF and r( n)
o digital
90 q( n)
A gain control
D
MQ
i M ( n) i (n )
Demodulator
with digital
A rAD ( n) Digital inputs
rIF ( t ) I/Q and r( n)
D mixer digital
qM ( n) q( n) gain control
Fig. 25.23. Receiver for digital modulation methods (without RF and IF components refer to
Fig. 25.8b and Fig. 25.12)
– The gain control can be performed only after the lowpass filters since the IF signal
may still contain adjacent channels with considerably higher signal levels. Gain control
requires two variable gain amplifiers that amplify the mean of the absolute value
'
|rCB (t)| = i 2 (t) + q 2 (t)
of the complex baseband signal rCB (t) = i(t) + j q(t) to a setpoint value. An analog
realization of this gain control is rather complex.
– The lowpass filters for channel filtering must have very steep edges as the frequency gap
between the useful and the adjacent channels is very narrow. At the same time, the group
25.2 Receivers 1267
BIF
rIF (t) R IF
B
f IF f IF + BIF /2 f
f IF MI
R MI
r MI ( t ) BIF /2 f IF 2 f IF f
R MI
Lowpass filter
B/ 2 as channel filter
Lowpass BIF /2 f
filter
I
B/ 2
i (t)
Fig. 25.24. Signal spectra for a digital receiver with analog channel filters according to Fig. 25.23a
(shown for branch i only, branch q is identical)
delay in the useful channel must be as constant as possible since digital modulation is
very sensitive to group delay distortions. It is difficult to meet these demands with analog
lowpass filters.
Due to these disadvantages, a demodulator with analog inputs usually is used in com-
bination with channel filtering and gain control in the IF range. In this case, the lowpass
filters in Fig. 25.23a are only required to suppress the portions of the double IF frequency
and the gain control for i and q is unnecessary.
BIF
rIF (t) R IF
B
f IF f IF + BIF /2 f
f IF MI
R MI
2 f IF – ( BIF +B ) / 2
r AA (t) BIF /2 f
R AA
B/ 2
A BIF /2 f
D
R AD
B/ 2 BIF /2
r AD (n)
Digital
fAD – BZF /2 fAD /2 BIF /2 f AD f
channel
filter
I
B/ 2
i ( n) Main region Aliasing region
( f < fAD /2) ( f > fAD /2)
fAD /2 f
Fig. 25.25. Signal spectra of a digital receiver with digital channel filters according to Fig. 25.23b
(for branch i only, branch q is identical)
output signal of the mixers which is caused by asymmetries and crosstalk. Usually the
attenuation of the IF signal is high enough to prevent any interference. The level of the
local oscillator signal is substantially higher and must be reduced. This can be done in two
ways:
– Stop filters with a resonant frequency that is matched to the IF frequency are added to
the anti-aliasing filters (see Fig. 25.26).
25.2 Receivers 1269
V MI V MI
50Ω 50Ω
Fig. 25.26. Anti-aliasing filters with stop filters for the IF frequency to attenuate the local oscillator
signal
– The sampling frequency of the A/D converter is selected such that the difference between
the IF frequency and the harmonics of the sampling frequency is larger than half the
bandwidth of the desired signal (= B/2). After sampling, the IF frequency is then within
the stop-band of the digital channel filters.
A combination of both methods is also possible.
After anti-aliasing filtering, the signal has an upper cutoff frequency corresponding
to half the bandwidth of the IF filter (= BIF /2) (see Fig. 25.25). Therefore, a sampling
frequency fAD > BIF would be required for nonaliasing A/D conversion. Since the subse-
quent digital channel filter suppresses all portions above a frequency of half the bandwidth
of the desired signal (= B/2), aliasing may be permitted in this region. Consequently the
sampling frequency is:
BIF + B
fAD > (25.15)
2
Figure 25.25 is the borderline case of the minimum sampling frequency. The aliasing
components shown as a broken line go right up to the desired channel.
The IF signal and the signals behind the mixers still contain several adjacent channels;
therefore, the overall level of these signals may be significantly higher than the level of the
desired channel. In order to prevent overdriving of the A/D converters, it is necessary to
use a gain control for the IF signal in addition to the gain control for the desired channel
that is integrated in the demodulator. For this purpose, the gain control in the IF range,
which exists in the superhet receivers shown in Figs. 25.8b and 25.12, is used.
Dynamic range: The available dynamic range of the receiver depends primarily on the
resolution of the A/D converter. We demonstrate this for a desired channel of power PC and
an adjacent channel of power PAC . The corresponding spectrum at the output of an A/D
converter is plotted in Fig. 25.27. The powers of the channels correspond to the areas below
the respective curves5 . Pn,Q is the power of the quantization noise of the A/D converter and
is evenly distributed in the frequency interval from zero up to half the sampling frequency.
We assume that the power in the neighboring channel is significantly higher than the power
5 The power of a signal x(t) with the Fourier transform (two-sided spectrum) X(f ) is:
+∞
Px = |X(f )|2 df
−∞
This is called the Parseval equation. We use unilateral absolute spectra; this eliminates the negative
frequencies, and the lower limit of the integral becomes zero.
1270 25 Transmitters and Receivers
2
R AD PAC
PC
Pn,C Pn,Q
B/ 2 f AD /2 f
Fig. 25.27. Spectrum at the output of the A/D converter for a desired channel with the power PC
and an adjacent channel with the power PAC . Pn,Q is the power of the quantization noise, Pn,C is
the portion in the desired channel.
in the desired channel. Thus, the overall power is approximately equal to the power in the
adjacent channel:
PAC PC ,Pn,Q
P = PC + PAC + Pn,Q ≈ PAC
When fully modulated, an ideal A/D converter with a resolution of N bits achieves the
signal-to-noise ratio:
3 · 22N
SNR = ⇒ SNR [dB] = N · 6 dB + 4.8 dB − CF [dB] (25.16)
C2
The crest factor of the signal is
Peak value vmax
CF = = (25.17)
rms value veff
The crest factor ranges from CF = 1 (0 dB) for a square-wave signal to CF ≈ 4 (12 dB)
for a noise-like signal6 . This means that the achievable signal-to-noise ratio depends on
the signal in the adjacent channel. The power of the quantization noise can be calculated
from the overall power P and the signal-to-noise ratio:
P P P CF 2
SNR = ⇒ Pn,Q = =
Pn,Q SNR 3 · 22N
The portion is within the desired channel (see Fig. 25.27):
B P CF 2 B
Pn,C = Pn,Q =
fAD 3 · 22N fAD
In order to ensure correct demodulation of the desired signal, the signal-to-noise ratio SNR C
in the desired channel must be higher than the minimum signal-to-noise ratio SNR i,min of
the modulation method used:
PC
SNR C = > SNR i,min
Pn,C
6 For a sinusoidal signal with CF =
√
2 (3 dB) (25.16) produces the relationship SNR = N · 6 dB +
1.8 dB (see (18.9) on page 950).
25.2 Receivers 1271
SNR i,min P CF 2 B
PC > (25.18)
3 · 22N fAD
and the permitted ratio of the powers in the adjacent channel and the desired channel
(available dynamic range) is:
PAC ≈P
PAC P 3 · 22N fAD
≈ < 2 B
(25.19)
PC PC SNR i,min CF
The parameters SNR i,min , CF and B are defined by the modulation method used. Therefore,
the available dynamic range is essentially determined by the resolution N of the A/D
converter and the sampling frequency fAD . While the sampling rate is often increased in
audio applications to achieve a better signal-to-noise ratio (oversampling), this is usually
not possible in receivers due to their very high minimum sampling rate; here, it is the
resolution that must be increased if the available dynamic range is not large enough.
Owing to a number of interferences, the signal-to-noise ratio of a real A/D converter
is lower than that of an ideal A/D converter according to (25.16). In practice, one must use
the effective resolution Neff < N , which is quoted in data sheets, in place of the resolution
N . Instead of the effective resolution, many data sheets specify the signal-to-noise ratio
for a sinusoidal signal as a function of the signal and the sampling frequencies; in this
case, the effective resolution is
SNR [dB] − 1.8 dB
Neff = (25.20)
6 dB
Example: We consider a receiver for a QPSK system with a data rate rD = 200 kbit/s, a
roll-off factor r = 1 and a bandwidth of B = 200 kHz. The bandwidth of the last IF filter
is assumed to be BIF = 1 MHz. According to (25.15) the sampling frequency is
BIF + B
fAD > = 600 kHz
2
We choose fAD = 800 kHz. At a bit error rate of 10−6 QPSK requires a minimum signal-
to-noise ratio SNR i,min = 20 (13 dB); with r = 1 the crest factor is CF ≈ 1.25 (2 dB).
Furthermore, we assume an available dynamic range of PAC /PC = 106 (60 dB). Solving
(25.19) for N we obtain:
1 PAC SNR i,min CF 2 B 1 1
N > ld = ld 10 · 10.4 ·
6
≈ 10.7
2 PC 3 fAD 2 4
Thus, an A/D converter with an effective resolution of at least 10.7 bits at fAD = 800 kHz is
required. If operated with a sinusoidal signal, the signal-to-noise ratio according to (25.20)
is SNR = 10.7 · 6 dB + 1.8 dB = 66 dB. In practice, this means that a 12 bit converter is
required.
This example is typical of receivers with digital channel filters. A/D converters with
a comparably high resolution are required despite the fact that the signal-to-noise ratio
SNR i,min required in the desired channel is very low. This is necessary due to the high
signal levels in adjacent channels.
1272 25 Transmitters and Receivers
R IF BIF R AD BIF
f g = f IF + BIF /2
f IF ,D = f AD – f IF
R IF BIF R AD BIF
f AD /2 f IF fAD 3 f AD /2 2 f AD f f IF ,D fAD /2 f
f IF ,D = f IF – f AD
R IF BIF R AD BIF
f AD /2 f AD f IF 3 f AD /2 2 f AD f f IF ,D fAD /2 f
f IF ,D = 2 fAD – f IF
R IF BIF R AD BIF
f AD /2 f AD 3 f AD /2 f IF 2 f AD f f IF ,D fAD /2 f
sampling frequency:
BIF
fg = fIF + < fAD
2
For subsampling in the mth aliasing region, the IF signal must be fully within this frequency
range7 . Thus, at the lower limit this requires
BIF fAD
fIF − > m
2 2
and at the upper limit:
BIF fAD
fIF + < (m + 1)
2 2
In summary, the general condition for the sampling frequency fAD is:
2fIF + BIF 2fIF − BIF fIF 1
< fAD < with m ≤ − (25.21)
m+1 m BIF 2
For m = 0 this also applies to the main region but in this case the upper limit is excluded.
By inserting the maximum possible integer for m into (25.21), the minimum sampling
frequency fAD,min is obtained. This minimum value depends on the ratio fIF /BIF and lies
within the range:
BIF
2BIF < fAD,min < 2BIF 1 +
2fIF
The digital IF frequency fIF,D at the output of the A/D converter is:
⎧
⎪ fAD
⎪
⎨ fIF − m m even
2
fIF,D = (25.22)
⎪
⎪ fAD
⎩ (m + 1) − fIF m odd
2
It follows that with even m values, the IF signal is converted in noninverted mode and with
odd m values in inverted mode (see Fig. 25.28). The inverted mode must either be taken
into account in the demodulator or compensated by an inverted mode in the transmitter or
in the mixers of the previous superhet receiver.
The I/Q mixer generates the signals from the digital output signal rAD (n) of the A/D
converter:
fIF,D
iM (n) = rAD (n) cos 2πn
fAD
fIF,D
qM (n) = − rAD (n) sin 2πn
fAD
The digital quadrature components i(n) and q(n) are obtained after channel filtering. The
digital I/Q mixer becomes particularly simple if the digital IF frequency is equal to a
7 This condition applies only to cases where the entire IF signal is to be processed digitally. If limited
to the desired channel, aliasing can be allowed as long as the desired channel is not affected. This
will be detailed further below.
1274 25 Transmitters and Receivers
0
0 i M ( n)
i ( n)
1
A r AD (n) 0
r IF (t ) 1,0,1,0,...
D 1 1
0 qM ( n)
q ( n)
0
0,1,1,0,... 1
Fig. 25.29. Digital receiver with IF sampling where fIF,D = fAD /4. The switches operate in
synchronization with the A/D converter.
4fIF fIF 1
fAD = with m ≤ − (25.24)
2m + 1 BIF 2
BIF
r IF (t) R IF
B
A f AD /2 f AD f IF 3 f AD /2 f
D
R AD
B
r AD ( n)
Digital
f IF ,D = fAD /4 fAD /2 f AD f IF 3 f AD / 2 f
I/Q
mixer
IM
Main region
Aliasing regions
( f > fAD /2)
i M ( n) f AD /2 f
IM
B/ 2
Digital
f AD /2 f
channel
filter
I
B/ 2
i ( n)
f AD /2 f
Fig. 25.31. Signal spectra in a digital receiver with IF sampling for fIF,D = fAD /4 and
fAD = 4fIF /5 (m = 2)
Figure 25.31 shows the signal spectra of a digital receiver with IF sampling for fIF,D =
fAD /4 and fAD = 4fIF /5 (m = 2). We can see that no aliasing occurs when the condition
of (25.24) is met; this means that the entire IF signal is digitized without any loss. Thus, it is
possible to receive adjacent channels by using bandpass filters, instead of lowpass filters, as
channel filters and by once again converting the frequency of the output signal. This enables
1276 25 Transmitters and Receivers
R AD BIF,max = fAD – B
the reception of all channels that are completely within the pass band of the IF filter without
changing the local oscillator frequency. Switching the channel filters is particularly easy in
practice as the channel filtering is generally done by a digital signal processor (DSP); only
the coefficients for the filter must be exchanged. This method is of particular importance
in narrow-band systems since it enables an entire group of channels to be received with
the same local oscillator frequency. In extreme cases the entire frequency band of the
application is within the IF bandwidth; here, one can use a fixed local oscillator frequency
and perform the channel selection solely by switching the channel filters. If, however, only
the desired channel is to be processed, as in Fig. 25.31, aliasing can be permitted as long
as the desired channel is not affected; thus the condition for m in (25.24) can then be
widened. We illustrate this by enlarging the IF bandwidth in Fig. 25.31 to the point just
before aliasing in the desired channel occurs (see Fig. 25.32). It thus follows:
BIF,max = fAD − B ⇒ fAD > BIF + B (25.25)
Inserting (25.24) into (25.25) and solving for m yields the condition:
2fIF 1
m < − (25.26)
BIF + B 2
A comparison of (25.25) and (25.15) shows that with IF sampling the minimum sam-
pling frequency is twice as high than for sampling of the quadature components after analog
I/Q mixing. The reason for this is that the IF signal contains both quadrature components:
rIF (t) = i(t) cos(2πfIF t) − q(t) sin(2πfIF t)
This shows that it is possible to perform IF sampling with one A/D converter and a sampling
rate according to (25.25) or to perform sampling of the quadrature components with two
A/D converters and half the sampling rate.
the modulation method becomes more complex, the disturbing effects of the unavoidable
asymmetries in the analog I/Q mixer increase which in turn increases the bit error rate.
Thorough tuning of the I/Q mixer, in regards to the amplitude and phase of the two signal
paths, is essential if complex modulation methods are used. This adjustment must have
a high degree of stability in terms of temperature sensitivity and durability in order to
permanently satisfy the demands.
The digital I/Q mixer in the receiver with IF sampling works accurately. Thus, this
receiver yields the best results. If the condition fIF,D = fAD /4 is met, the mixer only
consists of three multiplexers and one inverter.
Gain control
I/Q mixer for A/D converter
MI
Pre- RF
amplifier filter o
0
rRF (t)
rAnt (t) fRF
o
90
MQ
Gain switching
A i (n)
D Demodulator
with digital
inputs
and r (n)
digital
A q (n) gain control
D
r Ant (t ) R Ant
B
~
~
f RF f
RF filter
BRF
R RF
RF filter B
r RF (t )
~
~
f RF MI f RF f
R MI
r MI (t ) BRF / 2 f
R MI
B/ 2 Anti-aliasing filter
Anti-
aliasing BAAF / 2 f
filter R AA
B/ 2
r AA (t)
A
BAAF / 2 f
D
R AD
B/ 2 BAAF /2
r AD (n )
Digital
f AD – BAAF /2 fAD /2 BAAF /2 f AD f
channel
filter I
B/ 2 Main
region Aliasing regions
i ( n) ( f < fAD /2) ( f > fAD /2)
fAD /2 f
Fig. 25.34. Signal spectra in a direct conversion receiver (for branch i only, branch q is identical)
There are no image frequencies in the direct conversion receiver. Therefore, the RF
filter is only required to limit the received band with the aim of limiting the received
power. In the superhet receiver, the bandwidth of the RF filter must be at least as large as
the frequency range to be received; it may be even larger as long as the additional received
power does not restrict the dynamic range of the subsequent components too much.
25.2 Receivers 1279
In addition to the portions with differential frequencies in the range of 0 ≤ f ≤ BRF /2,
the output signals of the I/Q mixer also contain portions of the sum frequencies in the range
of 2fRF . Furthermore, there are portions at the frequency fRF generated by crosstalk in
the mixers that are suppressed by the anti-aliasing filter.
The minimum sampling frequency of the A/D converter depends on the bandwidth B
of the desired channel and the bandwidth BAAF of the anti-aliasing filter or BRF of the RF
filter depending on which of the two bandwidths is lower:
⎧
⎪ B + BAAF
⎪
⎨ for BAAF < BRF
2
fAD > (25.27)
⎪
⎪ B + BRF
⎩ for BAAF ≥ BRF
2
In both cases, the desired signal remains free of any aliasing contents. The situation of
BAAF < BRF is shown in Fig. 25.34. However, the sampling frequency can also be selected
such that all channels are digitized in the pass band region of the RF filter without aliasing,
and the channel selection can be achieved by switching the digital channel filters where,
in this case, the following must apply: fAD > BRF . The anti-aliasing filter is then used
exclusively for suppressing the signal portions in the frequency range fRF and 2fRF .
The primary advantage of a direct conversion receiver is its reduced number of fil-
ters. It is particularly suitable for monolithic integration as only the RF filter is required
as an external component and RC filters are used as the anti-aliasing filters. At the same
time, only one local oscillator with an RC quadrature network (0◦ /90◦ ) is required, which
may also be integrated with the exception of a resonant circuit that determines the fre-
quency, and a variable capacitance diode for frequency tuning. The elimination of the IF
components substantially reduces the current consumption of the receiver. In particular,
power-consuming drivers required for the SAW-IF filters in the superhet receiver and the
subsequent amplifiers for compensating the relatively high attenuation of these filters are
eliminated.
Besides the advantages mentioned, the direct conversion amplifier also has three prob-
lems with negative effects that must be reduced to a noncritical level by additional circuitry:
– The local oscillator frequency corresponds to the received frequency which causes the
risk of the relatively strong local oscillator signal reaching the antenna via the RF filter
and the pre-amplifier and then being transmitted (see Fig. 25.35). In order to avoid this,
the pre-amplifier must have a particularly low reverse transmission. As an alternative,
a 3-gate circulator can be introduced between the pre-amplifier and the RF filter. This
eliminates the local oscillator signal at the third gate so that it no longer reaches the
output of the pre-amplifier (see Fig. 25.36).
– If the local oscillator signal reaches the RF path and is reflected, the result is a self-
mixing effect. This results in a DC component at the outputs of the I/Q mixer which
R MI F( f )
B/ 2 B B B B
50
40
Noise figure F
30
20
10
0
C 2C 3C 4C BRF / 2 f
Fig. 25.37. Plot of the spectral noise figure F (f ) of variable gain amplifiers over the channel
sequence in direct conversion amplifiers
superimposes the DC component of the desired signal. Since the removal of this dis-
turbing DC component is not possible, the total DC component must be removed in the
demodulator by a digital highpass filter with a very low cutoff frequency. This must be
done in such a way that the useful signal is affected as little as possible.
– The variable gain amplifiers operate as LF amplifiers in the frequency of 1/f noise, which
increases the noise figure significantly above that of an IF amplifier. The influence from
the noise figure of the receiver can, of course, be reduced by making the gain of the
RF pre-amplifier as high as possible, but this is limited by the fact that a high gain in
the RF range is only possible with several amplifier stages and comparably high power
consumption; at the same time this reduces the overload margin. Figure 25.37 shows
the plot of the spectral noise figure F (f ) of the variable gain amplifier over the channel
sequence. One possible way of improving the noise figure is to reduce the spectral noise
figure by using the mth adjacent channel at f = mC as the desired channel instead of
the channel at f = 0 8 .
8 According to Fig. 25.37 the bandwidth of the adjacent channels is twice as high as the bandwidth
of the channel at f = 0. However, these channels contain two RF channels, i.e. fRF + C and
fRF − C, which are separated by combining the quadrature components in the following digital
processing; only half of the noise power becomes effective so that factor 2 in the bandwidth is
compensated.
25.2 Receivers 1281
Maintaining the required amplitude and phase position accuracy of the I/Q mixer poses
an additional problem. The demands on I/Q mixers in direct conversion receivers and in
superhet receivers are identical, but meeting them is much more difficult in the I/Q mixer
with RF input than in the I/Q mixer with IF input due to the higher frequency.
Today, the problems outlined for direct conversion receivers are dealt with well. There-
fore, it may be assumed that direct conversion receivers will replace superhet receivers. In
this respect, considerations arise as to whether the receiver with IF sampling, according to
Fig. 25.23c, shall be also used as a direct conversion receiver by placing only an additional
pre-amplifier and an RF filter in front of the A/D converter. This is known as RF sampling.
Chapter 26:
Passive Components
26.1
High-Frequency Equivalent Circuits
When dimensioning and simulating high-frequency and intermediate-frequency circuits,
the high-frequency response of passive components must be taken into consideration. For
this purpose, the high-frequency equivalent circuits shown in Fig. 26.1 are used to model
resistors, inductors, and capacitors.
It is common practice to name the reactive components inductor and capacitor and
their ideal values inductance and capacitance, respectively. Resistive components are
called resistors and their value is called resistance.
Supplementary elements in the equivalent circuits are called parasitic elements or
parasitics. Their values depend on the construction of the given component. One of the
most important values is the parasitic inductance of the component body and the connecting
leads. It is roughly proportional to the length and amounts to approximately 1 nH/mm,
i.e. a standard resistor with an overall length of 15 mm (5 mm each for the body and the
two leads) has an inductance of LR ≈ 15 nH. Wound film capacitors have even higher
values since the wound film layers act as an inductance. For inductors, this portion can be
neglected if their main inductance is sufficiently high. Similar considerations apply to the
parasitic capacitance.
The values of parasitics can be minimized by producing miniaturized components
without connecting leads, which is the case for surface-mounted components (surface
mounted devices or SMD). Modern RF and IF circuits use SMD components only and
our explanations are restricted to this type. The frequency range, in which the equivalent
circuits are valid, depends on the size of the SMD component and increases with decreasing
size. For components of the size 1206 (3 mm × 1.5 mm) the equivalent circuits are valid
up to 1 GHz (with restrictions up to 2 GHz). We will state the impedances and reflection
factors up to 5 GHz to characterize the response of the equivalent circuits in this range.
The response of real components in this range is not only dependent on the characteristics
R L C
R LR RL(f ) L RC LC C
CR CL
of the component but also on the type of mounting. Therefore, the demands in terms of
mounting and soldering precision increase with frequency.
26.1.1
Resistor
Figure 26.1 a shows the equivalent circuit for an SMD resistor. It corresponds to the equiv-
alent circuit of a parallel resonant circuit with imperfect inductance. The impedance is:
1 R + sLR
ZR (s) = (R + sLR ) || = (26.1)
sCR 1 + sCR R + s 2 LR CR
Consequently:
# # $ $
R + j ω LR − CR R 2 − ω3 L2R CR
ZR (j ω) = # $2 (26.2)
1 − ω2 LR CR + ω2 CR2 R 2
The dominating response of the resistance depends on the sign of the term (LR − CR R 2 )
in the imaginary part of ZR (j ω):
√
R < LR /CR ⇒ inductive response
√
R > LR /CR ⇒ capacitive response
√
For R = LR /CR , the imaginary part is maximally flat and the impedance remains real
for as long as possible. For very high frequencies, there is always a capacitive response
because the capacitance CR dominates. But in this region, the equivalent circuit is no
longer valid.
Figure 26.2 shows the magnitude and phase of the impedance of SMD resistors of the
size 1206 with LR = 3 nH and CR = 0.2 pF. A maximally flat imaginary
√ part in which the
phase remains zero for as long as possible is obtained for R = LR /CR ≈ 120 . With
lower values, the resistors respond inductively (positive phase) and with higher values,
capacitively (negative phase). For R = 190 the magnitude remains flat up to high
frequencies.
In addition to the impedance, the reflection factor
ZR (j ω) − ZW
rR (j ω) = (26.3)
ZR (j ω) + ZW
is also of interest. ZW is the characteristic impedance of the connecting lines. Figure 26.3
shows a plot of the reflection factor of the resistors from Fig. 26.2. The maximally flat
phase of the impedance for the 120 resistor also results in a maximally flat phase of
the reflection factor; the curve of the reflection factor thus begins tangential to the real
part axis. In contrast, the curve of the 190 resistor with maximally flat magnitude runs
perpendicular to the real part axis.
Furthermore, it is noticeable that a wideband 50 termination is not possible with a
50 resistance. This requires a capacitance C ≈ 1 pF in parallel to make the imaginary
part maximal flat:
LR 3 nH
LR = (CR + C) R 2 ⇒ C = 2 − CR = − 0.2 pF ≈ 1 pF
R (50 )2
√
In this way, it is possible to compensate all resistances with R < LR /CR .
26.1 High-Frequency Equivalent Circuits 1285
ZR
1 kΩ
Ω 1k
500
190Ω
200
120Ω
100
50Ω
50
20
10Ω
10
~
~
Im { rR }
1
2 GHz 3 GHz
4 GHz
1 GHz
2 GHz 3 GHz
4 GHz 5 GHz
1 GHz 5 GHz
120 Ω 190 Ω 1k Ω 1
10 Ω 50 Ω 1 GHz 1 GHz Re { rR }
5 GHz 1 GHz
2 GHz
5 GHz 3 GHz
4 GHz
5 GHz
26.1.2
Inductor
The equivalent circuit of an inductor, as shown in Fig. 26.1 b, is essentially similar to the
equivalent circuit of a resistor, except that the magnitudes of the values are different. The
parasitic resistance RL is caused by the skin resistance (skin effect) of the winding and is
proportional to the square root of the frequency [26.1]:
RL (f ) = kRL f (26.4)
In SMD inductors√ with an inductance of up to 10 mH the loss resistance coefficient kRL
with the unit / Hz is approximately proportional to the inductance:
kRL ≈ kL L (26.5)
√ √
Typical values are kL ≈ 1200 /( Hz · H) for the size 1206 and kL ≈ 600 /( Hz · H)
for the size 1812 [26.2]. The following approximation applies to SMD inductors of size
1812 with an inductance above 10 mH [26.2]:
0.7
√ L
kRL ≈ 20 / Hz ·
H
In inductors, the parallel resonance is of high quality as the plot of the magnitude of
the impedance in the upper part of Fig. 26.4 indicates. At a frequency of
1 1
ωr = ⇒ fr = (26.6)
LCL 2π LCL
the quality is:
√
1 L 2π 4 L3
Qr = = (26.7)
RL (fr ) CL kRL CL
The quality of SMD inductors of size 1206 and 1812 is Qr ≈ 100 . . . 300. In regards to
the resonant frequency, one must distinguish between the phase resonant frequency
1
fr,ph = fr 1 − 2
Qr
and the magnitude resonant frequency [26.1]
1
fr,max ≈ fr 1 −
2Q4r
At the phase resonant frequency the impedance of the inductor is real. At the magnitude
resonant frequency the impedance reaches its maximum value:
ZL,max ≈ Q2r RL (fr )
Due to the high quality Qr , the frequencies fr , fr,ph and fr,max differ only very slightly.
In practice, the frequency fr is usually called the resonant frequency or self-resonating
frequency (SRF).
The inductor quality factor (QF) QL is more important than the quality Qr :
Im {ZL (j 2πf )} f <fr /4
2πf L 2π L
QL (f ) = ≈ = f (26.8)
Re {ZL (j 2πf )} RL (f ) kRL
26.1 High-Frequency Equivalent Circuits 1287
ZL
Ω 1M
1µ H
330nH 100nH
100k 33nH 10nH
10k
1k
100
10
1
~
~
10nH
150
33nH
100
100nH
330nH
50
1µ H
0
~
This value is a measure of the losses (high QL → low losses) and is defined for the
frequency range of inductive response only (f < fr,ph ). In the frequency range up to
fr /4, it is almost proportional to the root of the frequency and reaches its maximum at
approximately fr /2; above this maximum, it declines rapidly and becomes zero at the phase
resonant frequency. This is illustrated in the lower part of Fig. 26.4. The loss resistance
coefficient of SMD inductors with an inductance lower than 10 mH is kRL ≈ kL L. From
(26.8) it follows:
2π f/Hz
QL (f ) ≈ f ≈
kL 100 . . . 200
Factor 100 applies to the size 1812 and factor 200 to the size 1206.
1288 26 Passive Components
Due to the high quality factor QL and the high quality Qr , the impedance is almost
purely imaginary with the exception of a small region around the resonant frequency;
consequently, the magnitude of the reflection factor is approximately one:
Im{ZL (j ω)}
ZL (j ω) − ZW j π−2 arctan
rL (j ω) = ≈ e ZW
ZL (j ω) + ZW
The values for ω = 0 are Im {ZL (j 0)} = 0 and rL (j 0) ≈ − 1, i.e. the locus diagram
of the reflection factor for f = 0 commences in the short-circuit point of the r plane.
Figure 26.6 a shows the typical curve of the reflection factor taking an SMD coil with
L = 100 nH as an example.
26.1.3
Capacitor
The equivalent circuit of a capacitor is shown in Fig. 26.1 c which results in a lossy series
resonant circuit with the impedance:
1 1 + sCRC + s 2 LC C
ZC (s) = RC + sLC + = (26.9)
sC sC
The self-resonating frequency (SRF) is
1 1
ωr = ⇒ fr = (26.10)
LC C 2π LC C
with the quality
1 LC
Qr = (26.11)
RC C
The phase and magnitude resonant frequencies are equal to the resonant frequency fr ; a
differentiation, as in the case of an inductor, is not required. Figure 26.5 shows the plot of
ZC
10k
1k
100
10
Fig. 26.5. Magnitude of impedance for SMD capacitors of size 1206 with RC = 0.2 and
LC = 3 nH
26.2 Filters 1289
Im{ rL } Im{ rC }
50 MHz 100 MHz 3 GHz 4 GHz
40 MHz 200 MHz 2 GHz 5 GHz
30 MHz
300 MHz
400 MHz
1 GHz 1 GHz 0 Hz
0 Hz
Re{ rL } Re{ rC }
3 GHz
Fig. 26.6. Typical curve of the reflection factor for SMD conductors and SMD capacitors of size
1206
the magnitude of the impedance of SMD capacitors of size 1206 with RC = 0.2 and
LC = 3 nH.
The capacitor quality factor (QF) QC is more important than the quality Qr :
26.2
Filters
Besides amplifiers and mixers, filters are amongst the most important components of
telecommunication systems. With the exception of filters in low-frequency bands, the
filters used are passive because active filters are only suitable for IF and RF frequencies
in exceptional cases. The standard LC filters are being increasingly replaced by dielectric
filters or surface acoustic wave (SAW) filters. This is particularly the case in applications
with high unit numbers in which customer-specific or application-specific filters are used.
1290 26 Passive Components
The essential advantage of dielectric and SAW filters is that they are designed as a single
component which the vendor supplies with specified tolerances, so that they can be used in
very demanding applications without any adjustment. In contrast, LC filters are made up of
several components and can only be used, without adjustments, in noncritical applications.
SAW filters provide another very important advantage, namely an almost constant group
delay that can be achieved independent of the magnitude response.
26.2.1
LC-Filters
LC filters are often designed with the aid of filter catalogues. The first step is to select a
suitable filter characteristic (Butterworth, Thompsen, Chebyshev, etc.) in order to meet the
demands in terms of magnitude response, group delay and steepness. Then, the desired
order of the filter is determined. The filter structures and the normalized component values
for filters are listed in filter catalogues as, for example, in [26.3].
In most cases lowpass filters have the branching structure shown in Fig. 26.7 a and
are designed directly. For bandpass filters, however, an equivalent lowpass filter with the
desired properties is designed and then converted into the corresponding bandpass filter by
means of a lowpass/bandpass transformation. This procedure changes the lowpass structure
of Fig. 26.7 a into the bandpass structure of Fig. 26.7 b [26.3]. However, it is not always
successful as the lowpass/bandpass transformation is based on a nonlinear mapping of the
frequency axis which changes the curve of the group delay.
Rg= ZW
Vg RL= ZW
a Lowpass filter
Rg= ZW
Vg RL= ZW
b Bandpass filter
Rg= ZW C12
Vg L1 C1 C2 L2 Vo RL= Z W
Here, the off-resonance factor v takes the place of the radian frequency ω and the argument
jv is therefore used instead of j ω. The change-over from ω or f to v, according to (26.15),
corresponds to a bandpass/lowpass transformation. AB (jv) is the transfer function of the
equivalent lowpass filter.
When calculating the group delay one must start with the radian frequency ω and
the nonlinear relationship between ω and the off-resonance factor v, thus the following
applies:
d Im {AB (j ω)}
τGr (ω) = − arctan
dω Re {AB (j w)}
d Im {AB (j v)} dv
= − arctan
dv Re {AB (j v)} dω
1292 26 Passive Components
A
dB 0 10MHz
6 dB
3 dB
–10
– 20
0.6 k
– 30
1
1.5
– 40
– 50
– 60
433.4MHz
~
~
Fig. 26.9. Frequency response of a two-circuit bandpass filter with fC = 433.4 MHz and
B = 10 MHz for several coupling factors k.
AB,max 1
AB ±j 2
= √ = √
2 2 2
26.2 Filters 1293
A 10MHz ( k = 1 /1.5)
dB – 6 k = 1.5
–7 k=1
–8 k = 0.6
–9
–10
–11
–12 433.4MHz
~
~
60 k=1
50
k = 0.6
40
30
~
~
Fig. 26.10. Frequency response and group delay of a two-circuit bandpass filter with
fC = 433.4 MHz and B = 10 MHz in the passband for several coupling factors k
– Overcritically coupled (k > 1): The frequency response displays two maxima with
# $
1
AB,max =
AB ±j k 2 − 1
=
2
which are located at either side of a local minimum at the resonant frequency:
k 1
AB,0 = |AB (j 0)| = < for k > 1
1+k 2 2
The equivalent lowpass filter has a Chebyshev characteristic with the ripple:
AB,max 1 + k2
w = = > 1 for k > 1
AB,0 2k
Figure 26.11 shows the ripple plotted over the coupling√factor. The −3 dB cutoff fre-
quencies are reached at an off-resonance factor v = ± 2 k. Here, however, they are
1294 26 Passive Components
w
dB
4
1
~
0
1 1.5 2 2.5 3 k
related to the squared mean value of the maximum and the local minimum at the center
frequency:
- # $
1 -
# √ $
A2B,max + A2B,0
2 1 1
AB ±j 2 k
= √ = 1+ 2
2 4 w
The corresponding attenuation is therefore higher than 9 dB. In practice, overcritically
coupled filters are used whenever a high steepness at the transition to the stop-band is
required. However, this is achieved at the cost of the group delay which for k > 1 shows
a distinct ripple.
– Undercritically coupled (k < 1): As in the case of critical coupling, the frequency
response curve has a maximum value at the center frequency with the corresponding
attenuation higher than 6 dB:
k 1
AB,max = < for k < 1
1+k 2 2
On both sides of the maximum, the frequency response declines faster than in critically
coupled filters. The equivalent lowpass filter has a Bessel characteristic at k ≈ 0.6.
Undercritical coupling is used in cases that demand a constant group delay over the
entire passband. This is rarely the case as the filter is almost exclusively used as an RF
filter in transmitters and receivers. The bandwidth of the filter is substantially higher
than the bandwidth of one channel, and group delay variations within a channel are
sufficiently low even in critically or overcritically coupled filters.
The center frequency fC and the −3 dB bandwidth B are required to define the filter
and then the resonant frequency is calculated:
B2
fr = fC2 − (26.20)
4
The difference between the two frequencies results from the nonlinear relationship be-
tween the frequency f and the off-resonance factor v (26.15). This makes the filter curves
symmetrical with respect to the off-resonance factor v, but asymmetrical with respect to
the frequency f . For the two −3 dB cutoff frequencies (L/U: lower/upper cutoff frequency)
B B
fL = fC − , fU = fC +
2 2
26.2 Filters 1295
C12
nC1 nC2
Rg= ZW n –1 n –1
L1 L2
λ Resonator
∼ 4 2...3 mm Terminals bores
a Two poles b Three poles c Four poles
26.2.2
Dielectric Filters
In the frequency range of 800 MHz to 5 GHz, filters with coupled resonators of the length
λ/4 are used where one end of the resonator remains open, and the other is short-circuited.
To prevent the dimensions from becoming too large, a dielectric with a minimum of losses
and maximum relative dielectric coefficient is employed to reduce the wave length from
the free-space wave length λ0 = c0 /f to:
v c0
λ = = √ (26.23)
f r f
Such filters are called dielectric filters. In the frequency range up to 1 GHz, barium-
titanate with r ≈ 90 is used; therefore the length of the resonator is approximately 8 mm
(λ ≈ 32 mm at f = 1 GHz). For higher frequencies, dielectrics with a lower relative
dielectric coefficients are employed.
A dielectric bandpass filter with n resonators is called an n-pole filter. The number of
poles relates to the equivalent lowpass filter because the transfer function of the bandpass
filter has 2n poles (two per resonator). Typical construction of standard dielectric filters is
shown in Fig. 26.13 [26.4].
Figure 26.14 illustrates a cross section of a two-pole filter. It consists of two resonator
bodies made of barium-titanate with an axial resonator bore and a radial bore for coupling.
The resonator bodies are metalized with the exception of the no-load side, the bores for
coupling and a small gap for capacitive coupling of the resonators. It should be noted that
electromagnetic fields expand through the resonator bodies; no field exists in the resonator
bores. The length of the physical resonators is always slightly smaller than λ/4 as the fields
at the open end expand into the ambient space (stray field). Thus, the electrical length of
the resonator is longer than the physical length.
26.2 Filters 1297
Open-ended
Gap for
Resonator coupling the
Metalised bore resonators
Ground Ground
∼
4
Input Output
Short-circuited
Fig. 26.14. Cross section of a two-pole dielectric bandpass filter. The electromagnetic fields spread
in the hatched resonator bodies; no fields exist in the bores.
The equivalent circuit of a two-pole dielectric filter corresponds to the circuit diagram
of the two-circuit bandpass filter in Figs. 26.8 and 26.12. In the case of three-pole or multi-
pole filters, additional parallel resonant circuits with the same type of capacitive coupling
are added. Although, this correspondence only exists for the pass band and the adjacent
A
dB 0
–10
– 20
– 30
– 40
– 50
– 60 f C =1.92 GHz
~
~
Fig. 26.15. Frequency response of a three-pole dielectric bandpass filter with a center frequency
fC = 1.92 GHz [26.4]
1298 26 Passive Components
portions of the stop-band since, with all odd harmonics, the resonators also have a parallel
resonance causing additional regions with lower attenuation above the desired pass band.
Dielectric filters, predominantly two-pole or three-pole, are employed as RF filters
in transmitters and receivers. The small size of the filters is particularly important for
mobile units. Figure 26.15 shows the frequency response of a three-pole filter used in the
PCS mobile communication system with the center frequency fC = 1.92 GHz and the
dimensions 6.5 mm × 4.3 mm × 2 mm [26.4].
26.2.3
SAW Filters
A SAW (surface acoustic wave) filter is a transversal filter (FIR filter) which uses the transit
time of an acoustic surface wave across a piezo-electric crystal for a controlled delay. The
excitation of the surface acoustic wave by an electrical input signal and its conversion back
to an electrical output signal is achieved with piezo-electric transducers which are known as
interdigital transducers due to their comb-like interlocked electrodes. Figure 26.16 shows
the construction of a SAW filter with a weighted and nonweighted transducer separated
by the transit distance.
An acoustic surface wave (Rayleigh wave) is an elastic wave travelling across a solid
surface with a speed of v ≈ 3000 . . . 4000 m/s. On smooth surfaces, the speed, or propa-
gation, does not depend on the frequency of the wave, i.e. there is no dispersion, thus the
shape of the wave remains the same and the group delay is constant. The piezo-electric
crystal is mostly made of lithium-niobate (LiNbO3 ) with v = 3990 m/s. The resulting
wave lengths range from λ = 400 mm at f = 10 MHz to λ = 4 mm at f = 1 GHz.
The electrodes of the transducers are λ/4 wide and spaced at λ/2 making the electrode
dimensions for frequencies above 1 GHz smaller than 1 mm. Thus, the maximum operating
frequency of a SAW filter is determined by the minimum size allowed in the manufacturing
process. Today, SAW filters with center frequencies of up to 400 MHz are available and,
when frequencies are above 200 MHz, the lithium-niobate is replaced by a quartz crystal.
The transfer function of a SAW filter corresponds to that of a transversal or FIR filter
(see Sect. 19.3 on page 994). The filter coefficients result from the electrode length of
both transducers and as the two transducers become consecutively active, it is necessary to
form the convolution product for the electrode lengths of the two transducers. In practice,
one of the transducers is often nonweighted, i.e. all electrodes are of the same length, and
Weighted
transducer
Non-weighted
transducer
Piezo-electric
crystal
Metal
base plate Fig. 26.16. Construction of a SAW filter
26.2 Filters 1299
L1 L L2
L
Transit time tL =
v
t1 t t2 t
tL tL + t1 + t2 t
b Pulse response of the filter (convolution product of the pulse responses of the
transducers + transit time
A
dB
~
~
f
c Frequency response of the filter (magnitude of the Laplace
transformation of the pulse response)
Fig. 26.17. Relationship between geometry and absolute frequency response of a SAW filter
the other is weighted (see Fig. 26.16). Figure 26.17 explains the relationship between the
geometry and the frequency response.
The pass band attenuation of a SAW filter is relatively high. The transducer used as
the transmitter emits one wave in the direction of the receiving transducer and one wave
in the opposite direction. This results in a loss of half of the power, which corresponds to
an attenuation of 3 dB. For symmetry reasons, the attenuation of the receiving transducer
is also 3 dB so that the theoretical lower limit for the attenuation of a SAW filter is 6 dB.
In practice, the attenuation must be much higher to prevent too strong a degradation of
the transfer function caused by the waves reflected at the transducers and the ends of the
crystal. Particularly disturbing is the triple-transit echo, which is reflected once at each
transducer and therefore passes the transit distance three times. The triple-transit echo
undergoes an attenuation (in decibel) that is three times higher than that of the desired
1300 26 Passive Components
A A
– 21 –7
dB dB
– 22 –8
– 23 –9
– 24 – 10
– 25 – 11
~ ~
~
Fig. 26.18. Frequency response of a standard SAW filter and a low-loss SAW filter with a center
frequency fC = 70 MHz and a −3 dB bandwidth B = 1 MHz [26.5]
signal and must have an attenuation that is at least 40 dB higher. Thus, in standard filters
the attenuation of the desired signal is at least 20 dB. If the attenuation of the triple-transit
echo is insufficient, it will cause a ripple in the frequency response and in the group
delay. This is the case in low-loss filters in which the ripple is accepted in favor of the
lower attenuation. Figure 26.18 shows the frequency response of a standard SAW filter
compared to that of a low-loss SAW filter [26.5].
Figure 26.18 shows the frequency response of a SAW filter for the situation of
double-ended impedance matching to the characteristic impedance ZW = 50 . Without
impedance matching, neither the specified frequency response nor the specified attenua-
tion can be reached. The circuits for impedance matching are listed on the data sheet. The
impedance of the two transducers can be described by means of the electro-mechanical
equivalent circuit of a piezo-electric transducer as shown in Fig. 26.19 a. Rm , Lm and
Cm are substitute elements for describing the mechanical properties; Cstat is the static
capacitance of the transducer electrodes. At the center frequency, the impedance of the
electro-mechanical portion becomes real and then only the electro-mechanical resistance
Rm and the static capacitance Cstat are effective (see Fig. 26.19 b). The dimensions are such
that the impedance of the converter is ohmic capacitive not only at the center frequency but
throughout the entire pass band range and beyond. Generally the resistance Rm is higher
than 50 so that the impedance matching circuit consisting of the capacitance Cstat and
Static Electro-mechanic
capacitance of equivalent circuit
the transducer of the transducer
Lm Cm
Cstat Rm Cstat Rm
a For the range around the centre frequency b At the centre frequency
26.3
Circuits for Impedance Transformation
Circuits for impedance transformation are used for impedance matching and coupling.
In the matching process an input or output impedance of a component is matched to the
characteristic impedance of a line to prevent reflections and to maximize the transfered
power. In some cases a controlled mismatch is performed. In the coupling (or interfacing)
process, a load is connected to a resonant circuit with the purpose of transforming the
impedance of the load such that the quality of the resonant circuit reaches a predetermined
value.
26.3.1
Impedance Matching
The following section contains a description of simple reactive networks for loss-free
matching of an impedance to the characteristic impedance ZW of a line. This type of
matching is known as narrow-band impedance matching and is accurate at one frequency
only. This is sufficient for practical applications as long as the impedance matching band-
width is larger than the bandwidth of the useful signal. The criterion is the reflection factor
r, which must drop to zero (match) at the center frequency and must not exceed a certain
1302 26 Passive Components
value at the pass band borders; usually the demand is |r| < 0.1. The compliance check is
done by circuit simulation or by measurement in a sample circuit.
The bandwidth of the matching circuit declines with an increasing transformation
factor. Therefore, matching of impedances with |Z| ZW and |Z| ZW is possible
for a very narrow band only. If the bandwidth of the simple network is not sufficient,
more sophisticated networks must be used for wideband impedance matching. Often these
networks are not loss-free since both the reflection factor and the broadband frequency
response must be optimized. For a more detailed description refer to the literature [26.1].
1 1 G − jB
Z = = = 2
Y G + jB G + B2
G B
⇒ R = , X = − (26.24)
G2 + B 2 G2 + B 2
j X1 (Z + j X2 ) !
j X1 || (Z + j X2 ) = = ZW
Z + j (X1 + X2 )
By inserting Z = R + j X, then separating into real and imaginary parts and solving the
equation for X1 and X2 , the following conditions are derived:
ZW R
X1 = ± , X2 = ∓ R (ZW − R) − X (26.25)
R (ZW − R)
Here, the condition R < ZW must apply to keep the term under the square root positive;
thus, this network only allows a step-up transformation R → ZW > R. There are two
solutions according to the ± signs; the positive sign must be assigned to one reactance
j X2 j X1
ZW j X1 Z=R + jX ZW jX 2 Z = R + jX
Fig. 26.21. Impedance matching networks with two elements. Dimensions are defined with
(26.25) for R < ZW and with (26.27) forR > ZW .
26.3 Circuits for Impedance Transformation 1303
L2 C2
ZW C1 R < ZW ZW L1 R < ZW
Fig. 26.22. Step-up transformation of resistances. Dimensions are defined with (26.25) and
(26.26).
and the negative sign to the other. A positive reactance is produced by an inductance and
a negative reactance by a capacitance:
X1/2
X1/2 > 0 ⇒ L1/2 =
2πfC
(26.26)
1
X1/2 < 0 ⇒ C1/2 = −
2πfC X1/2
For resistances (Z = R, X = 0), the signs of X1 and X2 in (26.25) are different, thus
resulting in the versions with an inductance and a capacitance shown in Fig. 26.22. The
version in Fig. 26.22 a has the characteristic of a lowpass filter and that in Fig. 26.22 b the
characteristic of a highpass filter. For general impedances (X = 0), the sign of X2 is also
influenced by the reactance X which allows versions with two inductances (X1 , X2 > 0)
or two capacitances (X1 , X2 < 0). X2 = 0 eliminates the series element and impedance
matching is achieved with a parallel inductance (X1 > 0) or a parallel capacitance (X1 <
0).
For the network in Fig. 26.21 b we obtain the condition:
j Z (X1 + X2 ) − X1 X2 !
j X1 + (Z || j X2 ) = = ZW
Z + j X2
By inserting Z = R + j X, separating into real and imaginary parts and solving for X1
and X2 , we obtain the conditions:
R 2 + X2
X1 = ± ZW −1
ZW R
# $
∓ R 2 + X2 (26.27)
X2 =
R 2 + X2
R −1 ±X
ZW R
For resistances (Z = R, X = 0) the following applies:
ZW R
X1 = ± ZW (R − ZW ) , X2 = ∓ (26.28)
ZW (R − ZW )
Here, the condition R > ZW must apply to make sure that the term under the square
root remains positive. Therefore, this network only enables a step-down transformation
1304 26 Passive Components
L1 C1
ZW C2 R > ZW ZW L2 R > ZW
Fig. 26.23. Step-up transformation of resistances. Dimensions are defined with (26.28) and (26.26)
Example: Let us consider the impedance matching at the input side of the 70 MHz low-
loss SAW filter in Fig. 26.20 on page 1301. The equivalent circuit consists of a resistance
Rm = 143 and a parallel capacitance Cstat = 63 pF, which, at the center frequency
fC = 70 MHz, results in the admittance
1 ω=2π·70 MHz
Y = G + jB = + j ωCstat = (7 + j 27.7) mS
Rm
where G = 7 mS and B = 27.7 mS. Conversion with (26.24) yields the impedance Z
with R = 8.58 and X = − 33.9 . Matching to ZW = 50 must be achieved with the
network in Fig. 26.21 a due to R < ZW . From (26.25) it follows that X1 = ± 22.8 and
X2 = (∓ 18.9 + 33.9) . We select the lowpass characteristic with X1 = − 22.8 and
X2 = 52.8 in order to increase the attenuation at frequencies above the pass band; thus,
with (26.26) we obtain:
1 X2
C1 = ≈ 100 pF , L2 = ≈ 120 nH
2π · 70 MHz · 22.8 2π · 70 MHz
26.3 Circuits for Impedance Transformation 1305
120 nH 34 nH
Fig. 26.24. Impedance matching to ZW = 50 at the input side of a 70 MHz low-loss SAW filter
For the version with the characteristic of a highpass filter, this results in two inductances:
X1 = 22.8 → L1 ≈ 52 nH and X2 = 15 → L2 ≈ 34 nH. Owing to the series
inductance L2 , this causes a lowpass characteristic, meaning that the overall characteristic
is that of a bandpass filter. Figure 26.24 shows both versions.
Collins Filter
In practical applications, the π network in Fig. 26.25 with two parallel capacitances and
one series inductance is often used instead of the simple impedance matching network with
two elements. This is known as the Collins filter and has the characteristic of a lowpass
filter. The additional degree of freedom resulting from the third element can be used to
optimize the bandwidth or to shift the values of the elements into a region better suited for
the implementation.
Here we shall concentrate first on the matching of resistances. The following condition
is obtained for the center frequency ωC = 2πfC :
1 !
= ZW
1
j ω C C1 +
1
j ωC L +
1
j ω C C2 +
R
After multiplying out the terms, separating the real and imaginary parts and using the
transformation ratio
R
t = (26.29)
ZW
ZW C1 C2 R
B
fC t=1.5
0.5
2
0.2
4
0.1
10
0.05 20
100
0.02
0.01
~
~
Fig. 26.26. Relative bandwidth B/fC (|r| < 0.1 of a Collins filter for several transformation
ratios t
The Collins filter can also be used for matching general impedances Z. This is done
by starting with
Z = R || j X
(24.22) 2 2 !
ZW 1 ZW 1
Z1 = = = ZW
Z R + jX
In the matched state it should correspond with the characteristic impedance ZW of the
connecting line.
Figure 26.27 a shows the matching of a resistance (Z = R, X = 0). Here, the line of
the λ/4 transformer must have the characteristic impedance:
Z W 1 = ZW R
The transformation range is very limited since, in practice, the characteristic impedance
of a strip line can vary by a factor of 4 only (see Fig. 26.10 on page 1293); thus, for
ZW /2 < ZW 1 < 2ZW , the transformation range is ZW /4 < R < 4ZW .
1308 26 Passive Components
ZW1 = ZW R
l1 = λ/4
ZW R
a Transformation of a resistance
ZW1 = ZW R
jX 2
l1 = λ /4
ZW Z=R + j X
2
ZW1
ZW2 Z1
Z
l2
ZW1 = ZW R1 ZW2 = ZW
l 1= λ/4 l2 Z – ZW
ZW Z ,rZ =
R1, r1 = +
– rZ Z + ZW
Fig. 26.27. Examples of impedance matching with strip lines with the use of a λ/4 transformer
For a general impedance Z the structure in Fig. 26.27 b can be used. First, a λ/4
transformation to
√
ZW 1 = ZW R
2
ZW 1
Z=R+j X ZW R 1
Z1 = = =
Z R + jX 1 X
+j
ZW ZW R
is performed. The reactive portion is then compensated with a shunt reactance X2 . From
the condition Z1 || j X2 = ZW , it follows:
ZW R
X2 =
X
In the capacitive case (X < 0 → X2 < 0), the shunt reactance is implemented by a short
open-ended line and in the inductive case (X > 0 → X2 > 0) by a short short-circuited
line. The required length is derived from (26.23) in the capacitive case
λ ZW 1 λ ZW 1 X
l2 = arctan − = arctan − for X < 0
2π X2 2π ZW R
26.3 Circuits for Impedance Transformation 1309
It is real for
4π l2 λ # ϕz $
ϕz − = nπ ⇒ l2 = −n n integer
λ 4 π
The following values are selected to keep the line as short as possible:
ϕz > 0 ⇒ n=0 ⇒ r1 = |rZ |
ϕz < 0 ⇒ n = −1 ⇒ r1 = − |rZ |
The structures in Fig. 26.27 are designed so that the first step in the matching process
is done with a series line. This causes a spatial distance between the impedance to be
matched and the other components, thereby facilitating the arrangement of the strip lines
on the substrate. On the matched side, the arrangement of other components causes no
problem because a connecting line with the characteristic impedance ZW can be used for
the spatial separation.
Impedance matching with a λ/4 transformer is possible only for a very limited trans-
formation ratio and is thus not optimum with respect to the required line length. Better
results are achieved with the structures shown in Fig. 26.28. Let us first look at matching
with the series line according to Fig. 26.28 a. Here, we use (26.21) in order to obtain the
input impedance Z1 of a line with characteristic impedance ZW 1 and length l1 if terminated
with an impedance Z2 = Z = R + j X; then we set Z1 = ZW :
2πl1
Z + j ZW 1 tan !
λ
Z1 = = ZW
Z 2πl1
1+j tan
ZW 1 λ
1310 26 Passive Components
ZW1
l1
ZW Z = R + jX
ZW1
jX 2
l1
ZW Z = R + jX
ZW2
l2
Open or short-circuited
ZW1
jX 2
l1
ZW Z= R + jX
ZW2
l2
Open or short-circuited
Fig. 26.28. Examples of
impedance matching with strip
c With a series line and input compensation lines
j r plane
Matching Matching
–1 is possible is possible 1
must be met to ensure that the term under the square root in (26.36) is positive. This
condition is quite easily represented in the r plane where matching with a simple series
line is possible for all impedances with a reflection factor located in the r plane within the
two circular regions shown in Fig. 26.29.
Matching of impedances for which condition (26.37) is not met, requires the structures
in Figs. 26.28 b and 26.28 c. In Fig. 26.28 b, the structure’s reactance X is compensated by
the parallel reactance X2 such that condition (26.37) is complied with; thus, impedance
matching with a series line is possible. In the structure of Fig. 26.28 c, a parallel reactance
X1 at the input of the series line is permitted:
Z1 = ZW || j X1
26.3.2
Coupling
For power decoupling from a parallel resonant circuit, a load resistance must be coupled
to the circuit. For the quality of a parallel resonant circuit loaded with a load resistance RL
but otherwise loss-free the following equation applies:
-
C
Qr = RL
L
C1 CP
f >> f P, C
R L R L C R RP L C
C2 RL RP
L1 LP
f << f P, L
R C R C L R RP C L
L2 RL RP
L1
R C R C L RP R RP C L
L2 RL
The resonant circuit is loaded with the transformed resistance RP , which is arranged in
parallel to the resonant impedance R. The transformation is independent of the frequency.
1314 26 Passive Components
26.4
Power Splitters and Hybrids
When the output power of a matched amplifier is to be divided into two load resistances,
a power splitter (power divider) is required; the splitter allows loss-free matching to the
characteristic impedance ZW . The principle of splitting the power of a matched RF ampli-
fier as compared to an LF amplifier is shown in Fig. 26.31. Generally, LF amplifiers have a
very low output resistance ro . It is therefore possible to connect several load resistances to
the output as long as the permissible output current is not exceeded. The power delivered
by the amplifier depends on the load resistances. However, a matched RF amplifier must
always be operated with a load resistance RL = ZW to achieve the maximum output power
and to prevent reflections which could destroy the amplifier. Consequently, the delivered
power is constant and, where several load resistances are used, it must be divided by a
power splitter.
The following section describes power splitters with three and four terminals. The latter
are known as hybrids and may also be used as power combiners.
A typical application of power splitters and combiners are RF power amplifiers that
consist of two parallel stages (see Fig. 26.32). The input power is distributed to the two
stages by a power splitter, and the outputs of the two stages are added by a power combiner.
I 2I
ro ∼ 0 V RL ro ∼ 0 V RL RL
ZW
ro= Z W P RL= ZW
ZW
ZW
ZW P/ 2 RL = ZW
Power
splitter
ZW
ro = ZW P ZW
ZW P/ 2 RL = ZW
Pi / 2 Po / 2
Pi Po
Power Power
splitter combiner
Pi / 2 Po / 2
Fig. 26.32. Power splitter and power combiner in an RF amplifier with two stages in parallel
26.4.1
Power Splitter
Lossy Resistive Power Splitters
The lossy resistive power splitters shown in Fig. 26.33 are employed for wide-band power
splitting. They are fully matched but deliver only half of the input power to the outputs
while the other half is dissipated at the internal resistances. Since a quarter of the input
power is available at each output, such splitters are also knows as 6 dB power splitters.
An identification of the three terminals is not needed because of the symmetry.
P/ 4
ZW
P
ZW
ZW ZW ZW
Vg ZW
ZW
P/ 4 ZW
a Delta configuration
P/ 4
ZW / 3
P
ZW ZW / 3
ZW ZW
Vg ZW
ZW / 3
P/ 4 ZW
ZW 2 P/ 2
P a2
ZW l = l/4
ZW
2 ZW ZW
b2
a1 ZW 2
Vg ZW
b1
l = l /4 a3
P/ 2 ZW
b3
Wilkinson Splitter
Fully matched and loss-free properties characterize the Wilkinson splitter shown in
Fig. 26.34. It consists of two λ/4 lines and one resistance, thus having a narrow band-
width. The input must be identified because the splitter is asymmetrical, and a loss-free
operation is obtained only with the configuration shown in Fig. 26.34. Since half of the
input power is available at each output, this splitter is also known as 3 dB power splitter.
The response of the Wilkinson splitter is best described by means of the S parameters
[26.1]:
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤⎡ ⎤
b1 S11 S12 S13 a1 0 1 1 a1
− j
⎣ b2 ⎦ = ⎣ S21 S22 S23 ⎦ ⎣ a2 ⎦ = √ ⎣ 1 0 0 ⎦ ⎣ a2 ⎦ (26.46)
b3 S31 S32 S33 a3 2 1 0 0 a3
We can see that it is fully matched from the fact that the reflection factors at the three
terminals are zero: S11 = S22 = S33 = 0. If a wave a1 with the power
P1 = |a1 |2
occurs at terminal 1, then waves with the following powers are obtained at terminals 2
and 3:
|a1 |2 P1
P2 = |b2 |2 = |S21 |2 |a1 |2 = =
2 2
|a1 |2 P1
P3 = |b3 |2 = |S31 |2 |a1 |2 = =
2 2
It should be noted that in this case b1 = a2 = a3 = 0 due to the fully matched state. On
the other hand, if a wave a2 with the power P2 = |a2 |2 occurs at terminal 2, we obtain
P1 = |S12 |2 |a2 |2 = |a2 |2 /2 = P2 /2 and P3 = |S32 |2 |a2 |2 = 0, i.e., half the power is
available at terminal 1 while the other half is dissipated at the internal resistance. The same
situation arises if a wave occurs at terminal 3.
26.4.2
Hybrids
It can be demonstrated that a loss-free symmetrical power splitter that is fully matched to
the characteristic impedance can only be constructed with four terminals. In a configuration
with three terminals, the demands on the S parameters lead to inconsistencies [26.1]. Power
splitters with four terminals are known as hybrids or ring couplers. The power fed to one
26.4 Power Splitters and Hybrids 1317
terminal is divided and distributed to two of the other three terminals while the fourth
terminal carries no signal.
S Parameters of a Hybrid
The properties of a hybrid can be easily described by means of the S parameters. One must
distinguish between the 180◦ Hybrid with
⎡ ⎤ ⎡ ⎤⎡ ⎤
b1 0 0 1 1 a1
⎢ b2 ⎥ − j ⎢ 0 0 1 −1 ⎥ ⎢ a2 ⎥
⎢ ⎥ = √ ⎢ ⎥⎢ ⎥ (26.47)
⎣ b3 ⎦ 2 ⎣ 1 1 0 0 ⎦ ⎣ a3 ⎦
b4 1 −1 0 0 a4
and the 90◦ Hybrid with:
⎡ ⎤ ⎡ ⎤⎡ ⎤
b1 0 0 −j 1 a1
⎢ b2 ⎥ − j ⎢ 0 0 1 −j ⎥ ⎢ a2 ⎥
⎢ ⎥ = √ ⎢ ⎥⎢ ⎥ (26.48)
⎣ b3 ⎦ 2 ⎣ −j 1 0 0 ⎦ ⎣ a3 ⎦
b4 1 −j 0 0 a4
Both hybrids are fully matched: S11 = S22 = S33 = S44 = 0. Figure 26.35 shows the
symbols of the two hybrid versions.
Let us first look at the 180◦ hybrid. The power of a wave a1 occurring at terminal 1 is
distributed to terminals 3 and 4. With a2 = 0, it follows from (26.47):
− j a1 |a1 |2 P1
b3 = S31 a1 = √ ⇒ P3 = |b3 |2 = =
2 2 2
− j a1 |a1 |2 P1
b4 = S41 a1 = √ ⇒ P4 = |b3 |2 = =
2 2 2
The output waves b3 and b4 are in phase. The power of an incident wave a2 at terminal 2 is
also distributed to terminals 3 and 4 but the output waves b3 and b4 are in phase opposition.
With a1 = 0, it follows from (26.47):
− j a2 |a2 |2 P2
b3 = S32 a2 = √ ⇒ P3 = |b3 |2 = =
2 2 2
j a2 |a2 |2 P2
b4 = S42 a2 = √ ⇒ P4 = |b3 |2 = =
2 2 2
◦
The 180 phase shift between terminals 2 and 4 is indicated in the diagram shown in
Fig. 26.35 a. The 90◦ hybrid provides the following outputs if a wave occurs at terminal 1:
− a1 |a1 |2 P1
b3 = S31 a1 = √ ⇒ P3 = |b3 |2 = =
2 2 2
1 O
3 1 O
3
0 0
0O 90
O
0O 90O
180O 0O
2 4 2 4
− j a1 |a1 |2 P1
b4 = S41 a1 = √ ⇒ P4 = |b3 |2 = =
2 2 2
and if a wave occurs at terminal 2:
− j a2 |a2 |2 P2
b3 = S32 a2 = √ ⇒ P3 = |b3 |2 = =
2 2 2
− a2 |a2 |2 P2
b4 = S42 a2 = √ ⇒ P4 = |b3 |2 = =
2 2 2
In both cases there is a 90◦ phase shift between the two outputs. This is indicated in the
diagram shown in Fig. 26.35 b.
Here, the bandwidth amounts to only approximately 2% of the center frequency. For the
90◦ hybrid with inductor coupling, according to Fig. 26.36 c, the following condition must
apply:
ZW 1
L = , C = (26.51)
2πfC 2πfC ZW
Again, the bandwidth amounts to only approximately 2% of the center frequency.
2C 2C C2 C2
L L
1 3 1 3 1 3
L
L L C1 C1 C C
C C L L
4 2 2 4 2 4
C L C C2 C2
O O O
a 180 hybrid b 90 hybrid c 90 hybrid with transformer
1
1 3
/4 /8
ZW
/4 1 3
/4 /4 /4
4 ZW 2 2 4 ZW 2 2 C C
ZW
/4 2 4
3 /4 /4
/8
3
27.1
Integrated High-Frequency Amplifiers
In principle, integrated high-frequency amplifiers use the same circuitry as low-frequency
or operational amplifiers. A typical amplifier consists of a differential amplifier used as a
voltage amplifier and common-collector circuits used as current amplifiers or impedance
converters (see Fig. 27.1a). The differential amplifier is often designed as a cascode dif-
ferential amplifier to reduce its reverse transmission and its input capacitance (no Miller
effect). Such circuits are described in Chap. 4, Sect. 4.1. Since the transit frequency of
high-frequency transistors (fT ≈ 50 . . . 100 GHz) is approximately 100 times higher than
that of low-frequency transistors (fT ≈ 500 MHz . . . 1 GHz), the bandwidth of the ampli-
fier increases by approximately the same factor. This, however, presumes that the parasitics
of the bond wires and the connections within the integrated circuit can be reduced enough
so that the bandwidth is primarily determined by the transit frequency of the transistors
and is not limited by the connections. This is a key problem in both the design and use of
high-frequency semiconductor processes.
1 The construction of an HBT corresponds to that of a conventional bipolar transistor. Here, however,
different material compositions are used for the base and emitter regions in order to enhance the
current gain at high frequencies. The construction of a MESFET is shown in Fig. 3.26b on
page 196.
1322 27 High-Frequency Amplifiers
vi v1 = A V vi vo
vg RL ≈ 0.5...5 kΩ
i1
v1 = A V vi
io = Al i1
Rg i1
vo RL
vg vi
io = Al i1
2 I0 IE IE
Power amplifier
Rg= ZW ZW ZW
vi vo
vg RL = ZW
ri = ZW ro = ZW
Vb Vb
IB,A IC,A
Rg = ZW ZW ZW
vi vo
vg RL = ZW
27.1.1
Impedance Matching
Generally, the connecting leads within integrated circuits are so short that they can be con-
sidered as ideal connections even in the GHz range;2 therefore, it is not necessary to carry
2 These are electrically short lines (see Sect. 26.2). In this context the term ideal does not refer to
the losses; these are relatively high in integrated circuits due to the comparably thin metal coating
and the losses in the substrate.
27.1 Integrated High-Frequency Amplifiers 1323
out matching to the characteristic impedance within the circuit. In contrast, the external
signal-carrying terminals must be matched to the characteristic impedance of the external
lines to prevent any reflections. In the ideal case, the circuit is dimensioned such that input
and output impedances, including the parasitic effects of bond wires, connecting limbs
and the case, correspond to the characteristic impedance. Otherwise, external components
or strip lines must be used for impedance matching (see Sect. 26.3).
Figure 27.1a shows typical values of low-frequency input and output resistances of
the voltage and the current amplifier in an integrated high-frequency amplifier where it is
assumed that equivalent amplifiers are employed as signal source and load.
Rg = ZW ZW
vg
ZW R = 2 ZW
2 I0
vg ZW Zi >> 2 Z W
Rg = ZW
a With terminating resistance
Rg = ZW ZW
vg
I0 VB I0
ZW
vg ZW
Rg = ZW
b With common-base circuits ( I 0 ≈ 520 µA for ZW = 50 Ω)
this matches both inputs to ZW = 50 . This method is simple, easy to accomplish with
a resistor in the integrated circuit and acts across a wide band. A disadvantage is the poor
power coupling due to the dissipation of the resistor and the large increase in the noise
figure (see Sect. 27.1.2). Instead of placing a resistance R = 2ZW between the two inputs,
each of the two inputs can be connected to ground via a resistance R = ZW . However, this
means that a galvanic coupling to signal sources with a DC voltage is no longer possible
as the inputs are connected to ground with low resistance. The version with a resistance
R = 2ZW is thus preferred.
As an alternative, common-base circuits can be used for the input stages (see Fig. 27.2b);
then, the input impedance corresponds approximately to the transconductance resistance
1/gm = VT /I0 of the transistors. With a bias current I0 ≈ 520 mA, this resistance is
1/gm ≈ ZW = 50 . In this case, the power coupling is optimal. A disadvantage is the
comparably high noise figure (see Sect. 27.1.2).
Both methods are suitable for frequencies in the MHz range only. In the GHz range, the
influence of the bond wires, the connecting limbs and the casing have a noticeable effect.
The situation can be improved by using loss-free matching networks made up of reactive
components or strip lines that must be fitted externally. This will provide an optimum
power coupling with a very low noise figure. In practice, impedance matching focuses less
on optimum power transmission than it does on optimum noise figure, or a compromise
between both optima. This is described in more detail in Sect. 27.1.2.
27.1.2
Noise Figure
In Sect. 2.3.4 we showed that the noise figure of a bipolar transistor with a given collector
current IC,A is minimum if the effective source resistance between the base and the emitter
terminal reaches its optimum value:
RB →0
β VT VT VT β
Rgopt = RB + 2 + 2RB ≈ (27.1)
IC,A IC,A IC,A
27.1 Integrated High-Frequency Amplifiers 1325
Here, RB is the base spreading resistance and β the current gain of the transistor. For the
collector currents IC,A ≈ 0.1 . . . 1 mA, which are typical of integrated high-frequency
circuits, the source resistance for β ≈ 100 is in the region Rgopt ≈ 260 . . . 2600 . With
larger collector currents, Rgopt can be further reduced, e.g. to 50 at IC,A = 23 mA and
RB = 10 , but the noise figure reaches only a local minimum as shown in Fig. 2.52 on
page 90. This is caused by the base spreading resistance. Very large transistors with very
small base spreading resistances are used in low-frequency applications which enables the
global minimum of the noise figure to be nearly reached even with small source resistances.
However, in this case the transit frequency of the transistors drops rapidly; thus, in high-
frequency applications, this method can be used in exceptional cases only.
In impedance matching at the input side by means of a terminating resistance as
shown in Fig. 27.2a, the effective source resistance has the value Rg,eff = Rg ||R/2 =
ZW /2 = 25 for each of the two transistors in the differential amplifier due to the
parallel connection of the external resistances Rg = ZW and the internal terminat-
ing resistance R = 2ZW . It is thus clearly lower than the optimum source resistance
Rgopt ≈ 260 . . . 2600 . Furthermore, the noise of the terminating resistance causes the
noise figure to become relatively high. With impedance matching at the input side by means
of a common-base circuit as shown in Fig. 27.2b, the effective source resistance has the
value Rg,eff = Rg = ZW = 50 ; here, too, the noise figure is comparably high.
For impedance matching with reactive components or strip lines, the internal resistance
Rg of the signal source can be matched to the input resistance ri of the transistor by means of
a loss-free and noise-free matching network. If we disregard the base spreading resistance
RB , then ri = rBE . For the effective source resistance Rg,eff between the base and emitter
terminals this means that Rg,eff = rBE . For rBE = βVT /IC,A and Rgopt the following
relationship is obtained from (27.1) with RB = 0:
Rg,eff = rBE = Rgopt β (27.2)
Thus, with impedance matching, √ the effective source resistance is higher than the optimum
source resistance by a factor of β ≈ 10. This might make the noise figure lower than
that in the configurations with a terminating resistance or a common-base circuit, but it is
still clearly higher than the optimum noise figure.
The optimum noise figure is only obtained when noise matching is performed instead
of power matching. This means that the internal √resistance Rg = ZW of the signal source
is not matched to ri = rBE but to Rgopt = rBE / β. √ Conversely, the input resistance of the
(noise) matched amplifier is no longer ZW but ZW β. This leads to the input reflection
factor
(24.34) ZW β − ZW β − 1 β≈100
r = = ≈ 0.82
ZW β + Z W β +1
and a standing wave ratio (SWR):
1 + |r|
(24.42) β≈100
s = = β ≈ 10
1 − |r|
In most applications this is not acceptable. Therefore, a compromise between power and
noise matching is used in most practical cases where a low noise figure is of importance.
Power matching is generally
√ used if the noise figure is of no importance.
Above f = fT / β ≈ fT /10 the optimum source resistance decreases, as can be seen
from the equation for Rgopt,RF in Sect. 2.3.4. This does not mean that the matching methods
1326 27 High-Frequency Amplifiers
in Fig. 27.2 can achieve a lower noise figure in this range. Factor Rg,eff /Rgopt does go down
but the minimum noise figure increases as the equation for Fopt,RF in Sect. 2.3.4 shows.
We will not examine this range more closely as the noise model for bipolar transistors
with a transit frequency above 10 GHz as used in Sect. 2.3.4 will only allow qualitative
statements in this case. The range f > fT /10 is then entirely in the GHz range and some
secondary effects, such as the correlation between the noise sources of the transistor, which
were disregarded in Sect. 2.3.4, become significant, and the optimum source impedance
is no longer real.
Example: With the help of circuit simulation we have determined the noise figure of the
different circuit versions for an integrated amplifier with the transistor parameters in Fig. 4.5
on page 274. Owing to the symmetry, we can restrict the calculations to one of the two
input transistors; Fig. 27.3 shows the corresponding circuits. We use a transistor of size 10
and a bias current of IC,A = 1 mA. In the common-base circuit according to Fig. 27.3c, we
reduce the bias current to 520 mA in order to achieve impedance matching to ZW = 50 .
Vb Vb
Rg = Rg,opt = 575Ω
or
Rg = ZW = 50Ω Rg = ZW = 50Ω
vg vg R = 50Ω
1mA 1mA
– Vb – Vb
Vb Vb
Matching
network
Rg = ZW = 50Ω
Rg = ZW = 50Ω
vg
vg 520 µA 1 mA
– Vb – Vb
27.2
High-Frequency Amplifiers with Discrete Transistors
Figure 27.1b shows the principle design of high-frequency amplifiers made up of discrete
transistors. It is clear that the circuit design differs fundamentally from that of the inte-
grated amplifier shown in Fig. 27.1a. The actual amplifier consists of a bipolar transistor
in common-emitter configuration and circuitry for setting the operating point, which is
presented in Fig. 27.1b, by the two current sources IB,A and IC,A . The practical function-
ality will be further described below. Instead of a bipolar transistor, a field effect transistor
can also be used. Coupling capacitances are used in front of and behind the transistor to
prevent the operating point from being influenced by the additional circuitry. The networks
for impedance matching to the characteristic impedance of the signal lines include π el-
ements (Collins filters) with a series inductance and two shunt capacitances as shown in
Fig. 27.1b.
27.2.1
Generalized Discrete Transistor
The term discrete transistor should not be misunderstood in a limited sense because the
components used in practice often contain several transistors and additional resistances
and capacitances in order to simplify the process of setting the operating point. We call
these components generalized discrete transistors.3
Figure 27.4a shows the graphic symbol and the most important versions of a general-
ized discrete transistor without additional components for setting the operating point. A
Darlington circuit is often used to enhance the current gain at high frequencies.
Figure 27.4b presents some typical designs with additions for setting the operating
point. The version at the left can be used equally well for the Darlington circuits in
3 This can be related to the CC operational amplifier which may also be regarded as a generalized
discrete transistor (see Sect. 5.5 and Figs. 5.82 to 5.87).
1328 27 High-Frequency Amplifiers
Vb Vb
Vb
b Circuit configurations with additional elements for setting the operating point
Fig. 27.4a. The resistances provide a voltage feedback which, at sufficiently high-resistive
dimensions, becomes virtually ineffective at high frequencies if the impedance of the
collector-base capacitance falls below the value of the feedback resistor. The external
element is an inductance which represents an open circuit at the operating frequency and
consequently causes a separation of the signal path and the DC path. The version shown in
the center of Fig. 27.4b has an additional emitter resistance for current feedback; therefore,
it is particularly suitable for wideband amplifiers or amplifiers with a high demand in
terms of linearity.
The version shown at the right of Fig. 27.4b consists of a common-emitter circuit with
voltage feedback followed by a common-collector circuit. Strictly speaking, this does not
belong to the group of discrete transistors since, like the integrated amplifier in Fig. 27.1b, it
comprises a voltage amplifier (common-emitter circuit) and a current amplifier (common-
collector circuit). Nevertheless, we have included it since it usually comes in a casing that is
typical of discrete transistors. The voltage feedback is often operated with two resistances
and one capacitance. Only the resistance, which is directly connected between the base and
the collector, influences the operating point and is used for setting the collector voltage
at the operating point. The capacitance is given dimensions such that it functions as a
short circuit at the operating frequency, thus allowing the parallel arrangement of the two
resistances to become effective.
27.2 High-Frequency Amplifiers with Discrete Transistors 1329
The versions shown in Fig. 27.4 are regarded as low-integrated circuits and are termed
monolithic microwave integrated circuits (MMIC). They are made of silicon (Si-MMIC),
silicon-germanium (SiGe-MMIC) or gallium-arsenide (GaAs-MMIC) and are suitable for
frequencies of up to 20 GHz.
27.2.2
Setting the Operating Point (Biasing)
Generally, the operating point is set in the same way as for low-frequency transistors.
However, with high-frequency transistors, one attempts to make the resistances required
in order to set the operating point ineffective at the operating frequency otherwise they
will have an adverse effect on the gain and noise figure. For this reason, the resistances
are combined with one or more inductances which can be considered short-circuited with
regard to setting the operating point, and nearly open-circuited at the operating frequency.
A description of how the operating point is set in a bipolar transistor is given below.
The circuits described may equally well be used for field effect transistors.
DC Current Feedback
If we apply the above-mentioned principle to the operating point adjustment with DC
current feedback as shown in Fig. 2.75a on page 117, we obtain the circuit design shown
in Fig. 27.5a in which high-frequency decoupling is achieved for the base and the collector
of the transistor by means of inductances LB and LC respectively. The collector resistance
can be omitted in this case. Thus, there is no DC voltage drop in the collector circuit so
that this method is particularly suitable for low supply voltages. In extreme situations, one
may remove R1 and R2 and connect the free contact of LB directly to the supply voltage;
the transistor then operates with VBE,A = VCE,A . Due to the decoupled base, the noise of
resistors R1 and R2 have only very little influence on the noise figure of the amplifier at the
operating frequency; thus, this is a particularly low-noise design for setting the operating
point. This is especially the case if an additional capacitance CB is introduced which, at
Vb
Vb Vb Vb Vb
LC
R1 LC R1 LC
LB
RC
Co Ci Co R1
Ci Ci Co
R2 CB RE CE R2 RE CE R2
the operating frequency, acts almost as a short circuit. Where a slight increase in the noise
figure is not critical, it may not be necessary to decouple the base and thus the circuit
shown in Fig. 27.5b may be used.
With an increase in frequency decoupling becomes more and more difficult since
the characteristics of the inductors used to achieve the required inductance become less
favorable. In order to make the magnitude of the impedance as high as possible, an inductor
with a resonant frequency that is as close as possible to the operating frequency is used.
As a result, the resonant impedance is approximately reached which, however, decreases
with an increasing resonant frequency as shown in Fig. 28.4 on page 1366. For this reason,
in the GHz range, the inductances are replaced by strip lines of the length λ/4. These lines
are short-circuited for small signals at the end opposite the transistor by capacitance CB
or by connecting them to the supply voltage. The end closest to the transistor then acts as
an open circuit.
Particularly problematic is the capacitance CE which, at the operating frequency, must
perform as a short circuit. Here, too, a capacitance with a resonant frequency as close as
possible to the operating frequency is used, whereby doing so results in impedances with
a magnitude close to that for the series resistance of the capacitance (typically 0.2 ).
However, with increasing resonant frequency, the resonance quality of the capacitances
increases (see Fig. 28.5 on page 1366), thus making the adjustment more and more difficult.
As an alternative, an open-circuited strip line of length λ/4 could be used that acts as a
short circuit at the transistor end but, due to the unavoidable radiation at the open-ended
side (antenna effect), this method is not practical. A short-circuited strip line must also
be rejected as it provides a short circuit for the DC current and thus short-circuits the
resistance RE . Owing to these problems, the DC current feedback is used only in the MHz
range while in the GHz range the emitter terminal of the transistor must be connected
directly to ground.
DC Voltage Feedback
Figure 27.5c shows the method of setting the operating point by means of DC voltage
feedback. This is used in many monolithic microwave integrated circuits (see Fig. 27.4b).
A collector resistance RC is essential in order to render the feedback effective and to ensure
a stable operating point. The collector is decoupled by the inductance LC so that, at the
operating frequency, the output is not loaded by the collector resistance. The base can be
decoupled by adding series inductances to the resistances R1 and R2 ; however, this method
is not used in practice. A disadvantage is an increase in the noise figure due to the noise
contributions from R1 and R2 , but these can be kept low using high-resistive dimensioning.
Vb Vb Vb
D1 VD1 VRC ∼ VD1 RC
D1
RC
D2
I E2,A CC D2
T2 LC CC
LB Co T2 LC
RB
I B1,A I C1,A
Ci
T1 T1
IR2
R1
R1 R2 CB VBE1,A R2 CB
on/off
27.2.3
Impedance Matching for a Single-Stage Amplifier
Calculation of the matching networks for an amplifier with a generalized single transistor
is complex because the impedances at the input and output port depend on the circuitry
connected to the other port, respectively; this is due to the internal reactive feedback which
also leads to a non-zero reverse transmission. The calculation is usually based on the
S parameters of the transistor including the circuitry for setting the operating point.
1 + rL rL =r2∗ 1 + r2∗
ZL = ZW = ZW = Z2∗
1 − rL 1 − r2∗
The conditions for power matching are thus met.
Rg= ZW
Matching Matching
network network
vg RL = ZW
r =0 rg r1 r2 rL r=0
Z = ZW Zg Z1 Z2 ZL Z = ZW
a1 a2 = b2 rL a1 = b1 rg a2
ZL Zg
b1 b2 b1 b2
r1 rL rg r2
With a load with reflection factor rL connected to the output, the input reflection factor r1 is
determined by inserting the condition a2 = b2 rL from Fig. 27.8a and solving the equation
for r1 = b1 /a1 . Similarly, the output reflection factor r2 with a source with reflection
factor rg connected to the input is calculated by inserting the condition a1 = b1 rg from
Fig. 27.8b and solving the equation for r2 = b2 /a2 . This leads to:
S12 S21 rL
r1 = S11 + (27.4)
1 − S22 rL
S12 S21 rg
r2 = S22 + (27.5)
1 − S11 rg
Without reverse transmission (S12 = 0), there is no interdependence and the reflection
factors are r1 = S11 und r2 = S22 .
In (27.6) and (27.7) the negative sign applies to B1 > 0 or B2 > 0 and the positive sign to
B1 < 0 or B2 < 0.
1334 27 High-Frequency Amplifiers
Power Gain
For impedance matching on both sides with reactive, i.e., loss-free, matching networks,
the maximum available power gain (MAG) [27.1]
S21
# $
MAG =
k − k2 − 1 (27.12)
S12
can be determined from (27.8) with the stability factor k > 1. This and other power gains
are described in Sect. 27.4 in more detail.
Example: The task is to design a high-frequency amplifier with transistor type BFR93
matched at both sides for an operating frequency (center frequency) fC = 1.88 GHz. The
supply voltage is to be 3.3 V. We use automatic control of the operating point according
to Fig. 27.6a with a bias current of IC = 5 mA. For this bias current we obtain a minimum
noise figure as stated in the data sheet.4
Figure 27.9 shows the dimensioned components of a circuit for setting the operating
point. The following aspects were taken into consideration:
– Since the input impedance of the transistor is very low (Re {S11 } < 0 → Re {Zi } <
50 ), the inductive decoupling of the base is omitted; therefore, the inductance LB of
Fig. 27.6a is replaced by a resistor RB = 1 k.
– An inductor with LC = 33 nH and a parallel resonant frequency of approximately
1.9 GHz (C ≈ 0, 2 pF) is used for the inductive decoupling of the collector.
– A resistor RLC = 100 is placed in series with LC so that at frequencies below the
operating frequency it causes losses which increase the k factor in the frequency range
100 MHz . . . 1.8 GHz (see Fig. 27.10). This reduces the tendency to oscillate in this
frequency range.
– For capacitive blocking at the operating frequency, the capacitors CB1 and CC1 , whose
series resonant frequency is approximately 1.9 GHz, are used (C = 4.7 pF, size 0604:
L ≈ 1.5 nH).
4 The data sheet also specifies that the maximum transit frequency is reached with I = 20 mA
C
so that IC = 5 mA is not optimum. However, one should be careful, since the transit frequency
is measured with the output short-circuited, allowing only limited conclusions to be drawn as
regards to the power gain that can be achieved with impedance matching on both sides. In another
design, conducted in parallel to this one, for IC = 20 mA a power gain was achieved that was a
mere 0.2 dB greater, a value which does not warrant the higher bias current, especially since the
noise figure increases significantly.
1336 27 High-Frequency Amplifiers
Vb = 3.3 V Vb = 3.3V
D1
RC
1N4148
100 Ω
D2 2.8 V
1N4148
2.1 V T2
RLC CC1 CC2
BC557B
100 Ω 4.7 pF 100 pF
0.73 V
LC
R1 CB2 CB1 R2 RB 33 nH
2.3 V
2.7 kΩ 10 nF 4.7 pF 6.8 kΩ 1 kΩ
I C1,A = 5 mA
0.66 V T1
BFR93 Fig. 27.9. Circuit for setting
the operating point of
transistor BFR93
k
1.2
1 with RLC
0.8
0.6
without RLC
0.4
~
~
5 In this simulation, the high-frequency equivalent circuits of resistors and capacitors were taken
into consideration. Nevertheless, the results of the simulation cannot be used for a real circuit
design since the simulation model for transistor BFR93 provided by the manufacturer is not
accurate enough for this frequency range. In practice, the S parameters of the transistor, including
the network for setting the operating point, must be measured with a network analyser. In this
example we use the S parameters from the simulation so that it can repeated with PSpice.
27.2 High-Frequency Amplifiers with Discrete Transistors 1337
With (27.8) it follows k = 1.05 > 1, i.e., impedance matching on both sides is possi-
ble. The power gain to be expected is obtained with (27.12): MAG = 5.57 ≈ 7.5 dB.
Equations (27.6) and (27.7) lead to:
rg,m = − 0.6475 − j 0.402 , rL,m = 0.3791 + j 0.6
Then, using (27.10) and (27.11) we can calculate the input and output impedances of the
transistor with operating point setting in the matched condition:
Z1,m = (7.3 + j 14) , Z2,m = (33 − j 80)
For both impedances, the real part is smaller than ZW = 50 so that matching requires a
step-up transformation according to Fig. 26.21a on page 1302.
For matching at the input side we obtain from (26.25) with R = 7.3 and X = 14 :
X1 = ± 20.7 , X2 = ∓ 17.7 − 14
We select the highpass filter characteristic (X1 > 0, X2 < 0) according to Fig. 26.22b on
page 1303, because then the series capacitance C2 can simultaneously serve as a coupling
capacitor. From
X1 = 20.7 , X2 = − 31.7
it follows with (26.26):
L1,i = 1.75 nH , C2,i = 2.65 pF
The additional index i refers to the input side matching.
For matching at the output side we obtain from (26.25) with R = 33 and X =
− 80 :
X1 = ± 70 , X2 = ∓ 24 + 80
We now select the lowpass filter characteristic (X1 < 0, X2 > 0) according to Fig. 26.22a
on page 1303 so that the overall characteristic is that of a bandpass filter. From
X1 = − 70 , X2 = 104
it follows with (26.26):
C1,o = 1.2 pF , L2,o = 8.8 nH
The additional index o refers to matching at the output side. An additional coupling capac-
itor is required at the output. We use a 4.7 pF capacitor with a series resonant frequency of
1.9 GHz which, at the operating frequency fC = 1.88 GHz, it acts as a short-circuit and
thus has no influence on the matching effect.
Figure 27.11 shows the amplifier with the two matching networks. The elements of
the matching networks are ideal; at this stage the design is not ready for practical use. It is
necessary to check at which points inductors and capacitors can be connected and where
strip lines may be advantageous or are mandatory for functionality of the elements. This is
not discussed any further; please refer to the notes on impedance matching in multi-stage
amplifiers in the next section.
Finally we present the results achieved. The upper part of Fig. 27.12 shows the
magnitudes of the S parameters in the matched amplifier at the operating frequency
fC = 1.88 GHz. One can see that matching covers a relatively narrow frequency band.
If the requirements |S11 | < 0.1 and |S22 | < 0.1 hold for the reflection factors, then the
bandwidth is approximately 53 MHz. Matching at the input covers a narrower band than
1338 27 High-Frequency Amplifiers
Vb = 3.3 V Vb = 3.3V
D1
RC
1N4148
100 Ω
D2 2.8 V
1N4148
2.1 V T2
RLC CC1 CC2
BC557B
100 Ω 4.7 pF 100 pF
0.73 V
LC Output matching
R1 CB2 CB1 R2 RB 33 nH
2.7 kΩ 10 nF 4.7 pF 6.8 kΩ 1 kΩ Cc L2,o = 8.8 nH
2.3 V
I C1,A = 5 mA 4.7 pF
Input matching
at the output since the transformation factor for the real part of the impedance is higher:
7.3 → 50 at the input compared to 33 → 50 at the output. In the center of
Fig. 27.12 the magnitudes of the S parameters are plotted over a wider range. This shows
that the output is also nearly matched (|S22 | ≈ 0.1) in the range around 600 MHz. The
position of this range depends on the capacitance of the coupling capacitor at the output,
which can be used for adjustment. This can be useful if the amplifier is followed by a
mixer for conversion to a low intermediate frequency. A suitable choice of the coupling
capacitor can also provide a sufficient matching for the intermediate frequency. This indi-
cates that high-frequency circuit engineering often takes advantage of secondary effects.
The bottom diagram of Fig. 27.12 shows the gain in decibel. At the operating frequency it
reaches its maximum, which we have calculated with (27.12): MAG ≈ 7.5 dB. The gain
is comparatively low as the transistor type BFR93 has a transit frequency of only 5 GHz
and is operated at its performance limit in our example. Modern circuits for the frequency
range around 2 GHz use transistors with transit frequencies of about 25 GHz, resulting in
gains of 20 . . . 25 dB.
27.2.4
Impedance Matching in Multi-stage Amplifiers
Matching in multi-stage amplifiers is done in the same way as in single-stage amplifiers.
Each stage is matched at both sides and then arranged in series, where the matching
networks between the stages can often be simplified by combining the elements. In most
cases, however, this is not the optimum procedure. In practice, it is used only if, for
construction purposes, the stages are so far apart that the connections between the stages can
no longer be considered as electrically short lines as is especially the case in the GHz range.
In all other cases the output of each stage is matched directly to the input of the next
stage. The calculation of this type of impedance matching is complicated since an amplifier
27.2 High-Frequency Amplifiers with Discrete Transistors 1339
0.6 1.5
S11 S11
S22 S22
0.4 1
S12
S12
0.2 0.5
0.1
~
0 0
1.6 G 1.7G 1.8G 1.9G 2G 2.1G 2.2G f
Hz
S11 S21
S12 S11 S21 S22
1 2.5
S22 S11
0.8 2
S22
0.6 1.5
0.4 1
S21
0.2 0.5
S21 S12
~
0 0
50M 100M 200M 500M 1G 2G 5G f
Hz
S21
dB
10
7.5
0
– 10
– 20
– 30
– 40
~
~
with n stages, including n+1 matching networks (input side, output side and n−1 networks
between the stages), are interdependent due to the reverse transmission of the transistors.
The procedure is divided into two steps:
– In the first step, structures must be selected that, in principle, allow impedance matching
on the basis of the S parameters of the individual transistors. This must include all wiring
that is required for construction, i.e., the PC board layout of the amplifier must be roughly
outlined.
1340 27 High-Frequency Amplifiers
– In the second step, the values of the elements in the individual structures must be de-
termined by means of a simulation program. For this purpose, iterative optimization
methods (optimizers) are used to find the ideal dimensions with regard to the crite-
ria specified by the user. Often these criteria include maximizing |S21 | observing the
secondary conditions |S11 | < 0.1 and |S22 | < 0.1 in the specified frequency range.
If the reverse transmission of the transistors is not very high, the first run may already
provide a satisfactory result. Otherwise the structures must be varied before further runs
are carried out. These may become necessary solely because the established element values
cannot be achieved or arranged on the predetermined layout of the PC board.
In practice, this procedure is also used for single-stage amplifiers. The ideal matching
networks can, of course, be calculated directly by the procedure described in the previous
section, but practical operation on the basis of the properties of the real components and
the PC board layout require additional computer-aided optimization.
27.2.5
Neutralization
The main obstacle in impedance matching is the reverse transmission of the transistors
which reduces the stability factor k and prevents matching on both sides if k < 1. For
a transistor without reverse transmission S12 = 0 and k → ∞ holds, so both sides can
be matched provided the real parts of the input and output impedances are positive, i.e.,
|S11 | < 1 and |S22 | < 1. A transistor without reverse transmission operates unilaterally
which means that signal transmission takes place only in the forward direction.
Vb Vb LB RB CC LC
LE
Simplification
LB LC
Ci ri vi g'm1 vi ro Co
Vb Vb Vb
L2 L3
C1 T1 T2 C3
Simplified
equivalent
circuit
Fig. 27.13. Impedance matching of a two-stage amplifier with Collins filters utilizing the parasitic
elements of the transistors
Vb Vb
R1 –vC L1a
vC L1b
Cn ∼ CC
Co
CC
Ci
R2 RE CE
Vb
CC1
T1
L3a
L4
Vb
L3b
L2a R1 Cn1 Cn2
RE1 CE1
L1
L2b CB R2
CC2
T2
RE2 CE2
Cn1
Cn2
CC1 CC2
T1 T2
S21
− 1
2 S12
U =
. (27.13)
S21
− Re S21
S12
S12
Here, the S parameters of the transistor without neutralization and the stability factor k
from (27.8) on page 1334 are to be inserted. However, the S parameters of the neutralized
transistor can also be used, making S12,n = 0 and leading to:6
|S21,n |2
U = # $# $
1 − |S11,n |2 1 − |S22,n |2
27.2.6
Special Circuits for Improved Impedance Matching
If the methods described so far fail to provide acceptable matching, circulators or 90◦
hybrids can be used to improve matching. This is the case, for example, if noise matching
is carried out at the input of an amplifier in order to minimize the noise figure and at the
same time the lowest possible reflection factor is required.
6 This relationship is obtained by calculating the transfer gain G according to (27.30) on page 1357
T
with matching on both sides and without reverse transmission, thus giving S12 = 0, rg = S11 ∗
∗
and rL = S22 .
1344 27 High-Frequency Amplifiers
a1 a2 a4 a5
S11 = 0 S22 = 0
b1 b2 b4 b5
a3 = 0 b3 b6 a6 = 0
S11,A ≠ 0 S22,A ≠ 0
ZW ZW
existing input mismatch in the case of noise matching. The next section will describe noise
matching in more detail. Power amplifiers sometimes use a circulator at the output; the
circulator then performs two tasks:
– It reduces the reflection factor S22 at the amplifier output to zero.
– It prevents the wave reflected by the load from reaching the output of the amplifier;
instead, the wave is absorbed in the terminating resistance ZW .
The second task is of particular importance as the power amplifier can be destructed by
the reflected wave.
Amplifier 1
ZW
a1 a3
S11 = 0 O O
0 0
b1 90
O b3 90
O
S 11,A ≠0
a2 90O a4 90O
0O 0O
b2 b4
ZW
Amplifier 2
The hybrid at the output acts as a power combiner and adds the output powers of the two
amplifiers. Therefore, this version of matching is often used in power amplifiers despite
its comparatively elaborate circuit construction.
27.2.7
Noise
In relation to the noise figure of integrated high-frequency amplifiers, we explained in
Sect. 27.1 that bipolar transistors with (power) matching do not achieve the minimum
noise figure since the matching network transforms the source resistance Rg√to the input
resistance rBE of the transistor, while the optimum source resistance is rBE / β. In order
to minimize the noise figure, power matching can be replaced by noise matching although
this causes unacceptably high input reflection factors in most cases. The same applies to
field effect transistors where here, too, power and noise matching differ substantially.
rg − rg,opt
2
F = Fopt + 4 rn # $
2 (27.15)
1 − |rg |2
1 + rg,opt
1 − S11 rg (1 − S22 rL ) − S12 S21 rg rL
2
This results in circles with constant power gain. Normally the calculation is performed
with the aid of suitable simulation or mathematical programs.
Figure 27.19 shows the noise figure and the power gain of a GaAs MESFET CFY10
at f = 9 GHz. Power matching is achieved for rg = rg,m = − 0.68 + j 0.5 and noise
27.2 High-Frequency Amplifiers with Discrete Transistors 1347
Im{ rg }
Fopt + 1 dB Fopt
1 MAG= 12 dB
Fopt + 2 dB
Fopt = 1.6 dB
Fopt + 3 dB
MAG
MAG − 1 dB
MAG − 2 dB
MAG − 3 dB
1
Re{ rg }
Fig. 27.19. Noise figure and power gain of a GaAs MESFET CFY10 at f = 9 GHz
(ID,A = 15 mA, VDS,A = 4 V)
matching is achieved for rg = rg,opt = − 0.24 + j 0.33. The circles of constant noise
figures show that with power matching the noise figure is 3 dB higher than with noise
matching. Likewise one can see from the circles of constant power gain that with noise
matching the power gain is 3.1 dB lower than MAG . Now it is possible to draw a connecting
line between rg,m and rg,opt and to select a reflection factor rg for which the user-specific
requirements are met.
Where matching at both sides is not possible, one can often carry out noise matching at
the input and power matching at the output. For this purpose, the circles of constant noise
figures are first drawn in the r plane. Then the power gain is calculated for all rg values
for which stable operation is possible. The procedure is:
– Starting with a given reflection factor rg , the reflection factor at the output is calculated:
(27.5) S12 S21 rg
r2 = S22 +
1 − S11 rg
If |r2 | ≥ 1, stable operation with power matching at the output is not possible.
– If |r2 | < 1, power matching at the output is assumed: rL = r2∗ .
– The corresponding reflection factor at the input is calculated:
(27.4) S12 S21 rL S12 S21 r2∗
r1 = S11 + = S11 +
1 − S22 rL 1 − S22 r2∗
If |r1 | ≥ 1, stable operation with power matching at the output is not possible.
1348 27 High-Frequency Amplifiers
Im{ rg }
MSG = 22 dB 1
Fopt = 1.3dB
Fopt + 3 dB
Fopt + 2 dB
Fopt + 1 dB
Fopt
Unstable region
1
Re{ rg }
MSG
MSG – 1 dB
MSG – 2 dB MSG – 4 dB
MSG – 3 dB
Fig. 27.20. Noise figure and power gain of a bipolar transistor BFP405 at f = 2, 4 GHz
(IC,A = 5 mA, VCE,A = 4 V)
1 − S11 rg (1 − S22 rL ) − S12 S21 rg rL
2
This results in circles of constant power gain that are limited by a stability border which
is also circular. The stability border is the point at which the maximum stable power gain
MSG is obtained; this will be described in more detail in Sect. 27.4.
Figure 27.20 shows the noise figure and the power gain of a bipolar transistor BFP405
at f = 2.4 GHz. The stability factor is below one so that power matching at both sides is
not possible. Noise matching is obtained for rg = rg,opt = 0.32 + j 0.25. The circles of
constant power gain are limited by a stability border at which the maximum stable power
gain MSG is reached. The circles of constant power gain show that, with noise matching,
the power gain is 3.5 dB below MSG . Likewise the circles of constant noise figure indicate
that for operation with the power gain MSG , the noise figure is 1.8 dB above the minimum
noise figure. A suitable reflection factor rg can now be selected.
If the optimum reflection factor rg,opt with power matching at the output side is located
inside the instable region, one must do without power matching and shift the stability border
by a suitable choice of rL = r2∗ until rg,opt is within the stable region.
27.3 Broadband Amplifiers 1349
In practice, optimizing the parameters rg and rL in terms of noise, power gain and other
criteria is done by means of simulation or mathematical programs with which nonlinear
optimization processes can be carried out.
27.3
Broadband Amplifiers
Amplifiers with a constant gain over an extended frequency range are known as broadband
amplifiers. High-frequency amplifiers are called broadband amplifiers if their bandwidth
B is wider than the center frequency fC thus producing a lower cutoff frequency fL =
fC − B/2 < fC /2 and an upper cutoff frequency fU = fC + B/2 > 3fC /2 as well as a
ratio fU /fL > 3. Sometimes fU /fL > 2 is used as a criterion. The term broadband is given
to these amplifiers only because their bandwidth is clearly higher than the bandwidth of
reactively matched amplifiers that are typical of high-frequency applications and in most
cases have a ratio of fU /fL < 1.1. Furthermore, the wideband characteristic of high-
frequency amplifiers is also related to impedance matching. Therefore, it is not the −3dB
bandwidth that is used as the bandwidth, but the bandwidth within which the magnitude
of the input and output reflection factors remain below a given limit. While reactively
matched amplifiers usually require reflection factors of |r| < 0.1, broadband amplifiers
accept reflection factors of |r| < 0.2. The less stringent demand reflects the fact that
wideband matching in the MHz or GHz range is much more complicated than the narrow-
band reactive matching.
27.3.1
Principle of a Broadband Amplifier
The functional principle of a broadband amplifier is based on the fact that a voltage-
controlled current source with resistive feedback can be matched at both sides to the
characteristic impedance ZW . To implement the voltage-controlled current source, one of
the generalized discrete transistors from Fig. 27.4 on page 1328 is used.7 Figure 27.21
shows the principle of a broadband amplifier.
Let us first calculate the gain using the small-signal equivalent circuit shown in
Fig. 27.22a. The nodal equation at the output is:
vi − vo vo
= gm vi +
R RL
This leads to the gain:
vo RL (1 − gm R)
A = = (27.16)
vi R + RL
The input current is
vi − vo vi (1 − A)
ii = =
R R
which leads to the input resistance:
7 The version in the right portion of Fig. 27.4b cannot be used as it has no high-resistance output.
1350 27 High-Frequency Amplifiers
R R
Voltage-controlled
current source
vi
ZW ZW ZW g m vi ZW
ZW ZW ZW ZW
R R
ii io
vi vi vo vi vi vo
g m vi RL Rg g m vi
ri ro
Fig. 27.22. Equivalent circuits for calculating the gain as well as the input and output resistances
of a broadband amplifier
vi R + RL
ri = = (27.17)
ii 1 + g m RL
According to Fig. 27.22b, the output current is:
vo vo Rg vo
io = + g m vi = + gm
R + Rg R + Rg R + Rg
This leads to the output resistance:
vo R + Rg
ro = = (27.18)
io 1 + g m Rg
We set RL = Rg = ZW and calculate the reflection factors at the input and output:
ri − ZW
R − gm ZW 2
S11 = = (27.19)
ri − ZW
RL =ZW R + 2ZW + gm ZW 2
ro − ZW
R − gm ZW 2
S22 =
= = S11 (27.20)
ro − ZW Rg =ZW R + 2ZW + gm ZW 2
The reflection factors S11 and S22 are identical and become zero for:
R = gm ZW
2 (27.21)
This means that both sides are matched. The forward transmission factor is:
R
S21 = A
= − + 1 = − g m ZW + 1 (27.22)
RL =ZW , R=gm ZW 2
ZW
This is identical to the gain in a circuit which is matched at both ends. It can be influenced
only by means of the transconductance gm as the feedback resistance is linked to the
transconductance. A high transconductance results in a high gain.
27.3 Broadband Amplifiers 1351
27.3.2
Design of a Broadband Amplifier
Figure 27.23 shows the practical design of a broadband amplifier on the basis of an in-
tegrated Darlington transistor with resistances for the operating point adjustment. Resis-
tances R3 und R4 have values in the k range and are therefore negligible. This is especially
the case for the internal feedback resistance R3 which is higher by at least a factor of 10
than the resistance R required for impedance matching. The effective feedback resistance
is thus:
RR3
Reff = R || R3 ≈ R
Resistance RC serves to adjust the bias current. In terms of the small-signal parameters,
it is parallel to the amplifier output and acts like an additional load resistance. This means
that the amplifier no longer exactly fulfils the symmetry condition S11 = S22 of an ideal
broadband amplifier, in other words, the matching condition S11 = S22 = 0 can only be
approximately satisfied. RC must therefore be made as high as possible. In the region of the
upper cutoff frequency, the gain and matching can be improved by the inductances LR and
LC . The inductance LR also contains the parasitic inductances of the resistance R and the
coupling capacitance Cc . Therefore, Cc can be a capacitor with a relatively high capacitance
and inductance, i.e., with a low resonant frequency, without producing any negative effect.
Capacitances Ci and Co serve as coupling capacitances at the input and the output. These
are critical as most capacitors only achieve an impedance of |X| ZW = 50 in a
relatively narrow range around the resonant frequency (see Fig. 28.5 on page 1366). Thus,
the matching bandwidth is usually limited by the coupling capacitors.
Vb
RC
LC
Cc R LR
Co
I C1,A I C2,A
R3
Ci
T1
T2
R4 R1 R2
With the help of (27.22) we can use the desired gain to derive the necessary transcon-
ductance gm of the voltage-controlled current source which corresponds approximately to
the transconductance of transistor T2 when taking the current feedback via R2 into account:
gm2 IC2,A
gm ≈ with gm2 =
1 + gm2 R2 VT
The selection of the bias current IC2,A determines the maximum output power of the
amplifier. In practice, modulation with an rms value of up to Ieff ≈ IC2,A /2 is useful;
the distortion factor then remains below 10%. Consequently, the output power and the
quiescent current are:
2
IC2,A ZW 4 Po,max
Po,max = Ieff ZW ≈
2
⇒ IC2,A > (27.23)
4 ZW
However, the bias current must be high enough to achieve the necessary transconductance:
IC2,A ≥ gm VT . In this case, the resistance of the current feedback is:
IC,A >gm VT 1 VT
R2 = − (27.24)
gm IC2,A
The parasitic inductance of resistor R2 must be as low as possible in order to avoid un-
desirable reactive feedback and is of particular importance with values below 20 . If the
expected bandwidth is not achieved in a broadband amplifier with current feedback, the
reason is often because the parasitic inductance in the emitter circuit of T2 is too high.
The current feedback via R2 also influences the bandwidth by causing it to increase with
increasing feedback. This is the reason why amplifiers with a particularly wide bandwidth
make use of current feedback even if this is not required on the basis of the output power;
typical examples are broadband amplifiers for instrumentation.
Example: In the following, a broadband amplifier is designed according to Fig. 27.23 for a
50 system by using two transistors of the type BFR93 in Darlington configuration (see
Fig. 27.24). A gain of A = 16 dB and a maximum output power of Po,max = 0.3 mW =
− 5 dBm are required. For the supply voltage we assume Vb = 5 V. The gain is:
A [dB] 16 dB
|A| = |S21 | = 10 20 dB = 10 dB
20 = 6.3
With (27.22) we obtain the necessary transconductance:
! ZW =50 7.3
S21 = − gm ZW + 1 = 6.3 ⇒ gm = = 146 mS
50
For the quiescent current of T2 it follows IC2,A > gm VT = 3.8 mA. With (27.23) we obtain
from the maximum output power IC2,A > 4.9 mA. We select IC2,A = 5 mA. The resistor
R2 is calculated with (27.24) to be R2 = 1.6 . The resulting small current feedback is
not implemented for the moment as we must expect a loss in gain due to secondary effects.
For the bias current of transistor T1 we select IC1,A = 2 mA since, with smaller
currents, the transit frequency drops rapidly. As the base-emitter voltage of T2 is ap-
proximately 0.66 V and the base current IB2,A ≈ 50 mA (current gain approximately
100) is negligible compared to IC1,A = 2 mA, the value for the resistor R1 is obtained:
27.3 Broadband Amplifiers 1353
Vb = 5 V
RC
270 Ω
LC
Cc LR 270 nH
R
10 nF 440 Ω 47 nH Co
3V
I C1,A I C2,A
R3
2 mA 5 mA
Ci 5.6 kΩ
T1
1.3 V BFR93
0.66 V T2
BFR93
R4 R1
4.7 kΩ 330 Ω
Fig. 27.24. Example of a broadband amplifier
R1 ≈ 0, 66 V/2 mA = 330 . Concerning the voltage divider for operating point adjust-
ment we select R3 = 5.6 k and R4 = 4.7 k resulting in a voltage of 3 V at the collectors
of the transistors (see Fig. 27.24). To ensure that the desired bias current for T2 is achieved
(IC2,A = 5 mA), a collector resistor RC = 270 must be used for the supply voltage
Vb = 5 V.
After all resistors for operating point setting have been dimensioned, we can calculate
the transconductance gm . For this purpose we use the equation for the transconductance
of a Darlington transistor with resistance R from Sect. 2.3.4 and insert R = R1 :
1 + gm2 (rBE 2 || R1 )
gm ≈ gm1
1 + gm1 (rBE 2 || R1 )
For gm1 = IC1,A /VT = 77 mS, gm2 = IC2,A /VT = 192 mS and R1 = 330 , the
transconductance is gm ≈ 185 mS. From (27.21) the feedback resistance is thus R =
gm ZW2 = 463 .
Further dimensioning is done with the aid of circuit simulation. We have used the high-
frequency equivalent circuits for all resistances and inductances as well as the capacitor
Cc , only for the coupling capacitances Ci and Co have we assumed ideal capacitances.
First, the reflection factors S11 and S22 are optimized at low frequencies by finely tuning
the resistance R; the result is R ≈ 440 . Then, the gain and the impedance matching
at high frequencies is optimized by adding inductors LR and LC . For LR = 47 nH and
LC = 270 nH, the plots of the magnitudes of the S parameters are obtained as shown
in Fig. 27.25. The typical demand on broadband amplifiers of |S22 | < 0.2 is complied
with up to about 1 GHz. In this range |S11 | < 0.1, i.e., the input matching is extremely
good for a broadband amplifier. The desired gain |S21 | = 6.3 = 16 dB is reached up to
approximately 300 MHz. The −3dB cutoff frequency is at 700 MHz.
The current feedback calculated for transistor T2 with R2 ≈ 1.6 can be neglected
because the amplifier achieves the desired gain. Deviations from the calculated values
have two sources. First, the transconductance gm = 185 mS of the Darlington transistor is
lower than the transconductance gm2 = 192 mS of transistor T2 , and second, the transistor
BFR93 has a parasitic emitter resistance of approximately 1 .
1354 27 High-Frequency Amplifiers
S11 S 21
S22 0.35 7
S21
0.3 6
S22
0.25 5
0.2 4
S11
0.15 3
0.1 2
S22
0.05 1
S11
~
0 0
20M 50M 100M 200M 500M 1G 2G 5G f
Hz
S21
dB
16.1
15
13.1
10
0
700M
~
~
In practice, the very good overall performance of this amplifier can only be used in
a comparatively small frequency band as the coupling capacitances Ci and Co cannot
be given a wide-band low-resistance characteristic. If necessary, several capacitors with
staggered resonant frequencies must be used.
27.4
Power Gain
Usually the power gain is specified for high-frequency amplifiers. There are different
definitions of gain which relate to different parameters. Some of the related equations on
the basis of S or Y parameters are very complicated. We shall begin by explaining the
definitions of gain for an ideal amplifier and then extend these to cover a more general
situation. The complex equations on the basis of S and Y parameters are intended for
computer-aided evaluations only as manual calculation is very involved.
Figure 27.26 shows the ideal amplifier with the open-circuit gain factor A, the input
resistance ri and the output resistance ro ; there is no reverse transmission. The amplifier
27.4 Power Gain 1355
Ideal amplifier
Rg ro
vg vi vo
ri vi Avi RL
Fig. 27.26. Ideal amplifier with
signal source and load
Presentation with
S parameters or
Y parameters
Zg = 1/Yg
vg vi vo
ZL = 1/YL
Fig. 27.27. General amplifier
rg rL with signal source and load
is operated with a signal source of the internal resistance Rg and a load RL . For further
calculations we require the overall gain
vo ri RL
AB = = A
vg R g + r i ro + R L
and the gain under load:
vo RL
AL = = A
vi ro + R L
For the general situation, we look at an amplifier that is characterized by S and Y
parameters. It is operated with a source of the impedance Zg = 1/Yg and a load ZL = 1/YL
(see Fig. 27.27). For presentation with the help of the S parameters we also need the
reflection factors of the source and the load
Zg − ZW ZL − ZW
rg = , rL =
Zg + Z W ZL + Z W
and the determinant of the S matrix:
S = S11 S22 − S12 S21
It should be noted that the parameters rg and rL are reflection factors while ri and ro
are the resistances of the ideal amplifier from Fig. 27.26.
27.4.1
Direct Power Gain
Direct power gain refers to the power gain in the conventional sense:
PL Effective power absorbed by the load
G = =
Pi Effective power absorbed at the amplifier input
1356 27 High-Frequency Amplifiers
|Y21 |2 Re {YL }
= . (27.26)
Y12 Y21
Re Y11 − |Y22 + YL |2
Y22 YL
The direct power gain is independent of the signal source impedance and therefore contains
no indication regarding the impedance matching on the input side. Comparison of, say, two
amplifiers that use the same signal source, the same load, and output the same effective
power to the load reveals that the amplifier with the lower effective input power has a
higher direct power gain. In relation to high-frequency amplifiers, this property is not
useful; therefore, the direct power gain is rarely used in high-frequency engineering.
27.4.2
Insertion Gain
Insertion gain is the ratio of the effective powers absorbed by the load with or without
amplification:
PL Effective power absorbed by the load with amplifier
GI = =
PL,wa Effective power absorbed by the load without amplifier
Consequently, PL,wa is the effective power which the signal source can deliver directly to
the load. For the ideal amplifier from Fig. 27.26, we obtain:
vo2 vg2 RL
PL = , PL,wa = % &2
RL Rg + R L
Consequently:
2
vo Rg + R L 2 Rg + R L 2
GI = = AB 2
vg RL RL
2 2
ri Rg + R L
= A2 (27.27)
Rg + r i ro + R L
The corresponding calculation for the amplifier in Fig. 27.27 leads to:
2
|S21 |2
1 − rg rL
GI =
% &
1 − S11 rg (1 − S22 rL ) − S12 S21 rg rL
2
A B
2
|Y21 |2 Re Yg Re {YL }
Yg + YL
=
% &
(27.28)
Y11 + Yg (Y22 + YL ) − Y12 Y21
2
Yg YL
The insertion gain depends on the impedance of the signal source and the load and therefore
takes the input and output impedance matching into account. However, the maximum gain
is generally not reached with matching at both sides. This can be exemplified with the ideal
amplifier. With two-sided matching, Rg = ri and RL = ro where insertion into (27.27)
leads to:
2 2
1 2 Rg + R L
GI,match = A
2 2RL
This shows that, despite impedance matching at both sides, the insertion gain depends on
the ratio Rg /RL . A constant insertion gain is achieved only in the special case of equal
resistances at the input and output, i.e., Rg = ri = ro = RL . Owing to this characteristic,
the insertion gain is hardly used.
27.4.3
Transfer Gain
Transfer gain specifies the ratio of the effective power absorbed by the load to the available
(effective) power at the signal source:9
PL Effective power absorbed by the load
GT = =
PA,g Available power at the signal source
For the ideal amplifier from Fig. 27.26 we obtain:
vo2 vg2
PL = , PA,g =
RL 4Rg
This leads to:
2 2
vo 4Rg 2 4Rg ri 4Rg RL
GT = = AB = A2 (27.29)
vg RL RL Rg + r i (ro + RL )2
The corresponding calculation for the amplifier in Fig. 27.27 leads to:
# $# $
|S21 |2 1 − |rg |2 1 − |rL |2
GT =
% &
1 − S11 rg (1 − S22 rL ) − S12 S21 rg rL
2
A B
4 |Y21 |2 Re Yg Re {YL }
=
% &
(27.30)
Y11 + Yg (Y22 + YL ) − Y12 Y21
2
9 The available power is an effective power by definition and thus does not have to be explicitly
specified as being effective.
1358 27 High-Frequency Amplifiers
The transfer gain depends on the impedance of the signal source and the load and becomes
maximum with impedance matching at both sides. This can be demonstrated with (27.29):
∂GT ∂GT
= 0 , = 0 ⇒ Rg = ri , RL = ro
∂Rg ∂RL
Thus, the transfer gain meets the reasonable demands expected of a gain definition.
27.4.4
Available Power Gain
Available power gain specifies the ratio of the available powers of the amplifier to the
signal source:
PA,amp Available power of the amplifier
GA = =
PA,g Available power of the signal source
For the ideal amplifier from Fig. 27.26 we obtain:
(Avi )2 vg2
PA,amp = , PA,g =
4ro 4Rg
This leads to:
2
Avi 2 Rg ri Rg
GA = = A2 (27.31)
vg ro Rg + r i ro
The corresponding calculation for the amplifier in Fig. 27.27 leads to:
# $
|S21 |2 1 − |rg |2
GA = # $ A % &B
∗
1 − |S22 |2 + |rg |2 |S11 |2 − |S |2 − 2Re rg S11 − S S22
A B
|Y21 |2 Re Yg
= A%% & &% &B (27.32)
Re Y11 + Yg Y22 − Y12 Y21 Y11 + Yg
The available power gain is independent of the load and includes no indication with regard
to impedance matching at the output side. It is required for noise calculations since these
are based on the available power. The available power gain has already been described in
Sect. 4.2.4 in connection with the calculation of the noise figure of amplifiers connected
in series (see (4.200) and (4.201) on page 452).
27.4.5
Comparison of Gain Definitions
Specific properties of the various gain definitions have already been described in the
relevant sections; for this reason, we shall restrict ourselves to a brief comparison here.
Direct power gain G is of no relevance in high-frequency amplifiers since the optimum
use of the available power of the signal source is required and because impedance matching
at the input side necessary for this purpose has no bearing on the direct power gain. In
fact, it reaches its maximum if the amplifier absorbs as little power from the signal source
as possible, i.e., with the poorest possible impedance matching. The direct power gain is
relevant for low-frequency amplifiers since, in these cases, the aim is to achieve the highest
27.4 Power Gain 1359
possible voltage gain, which means a minimum load on the signal source. In high-frequency
amplifiers such mismatches are undesirable because of the resulting reflections.
The insertion gain GI is of no real significance for matched amplifiers. This will be
explained for the ideal amplifier in Fig. 27.26. With matching at both sides and different
resistances at the input and the output a mismatch occurs in the direct connection of the
signal source and the load that, in practice, would be corrected by a matching network. For
this reason, the two operating modes which are compared in the definition of the insertion
gain are not practical but theoretical alternatives only. With impedance matching at both
sides and equal resistances at the input and output, the matched condition (Rg = RL )
exists even with a direct connection between the signal source and the load, but in this case
the available power of the signal source is delivered to the load and the insertion gain GI
corresponds to the transfer gain GT .
Due to its properties, the transfer gain GT is the preferred definition of gain in high-
frequency engineering and is simply referred to as gain. However, it is not to be confused
with voltage gain or power gain. Only in the case of impedance matching at both sides and
the same resistances at the input and output are the voltage gain, current gain and transfer
gain identical in their decibel values.
The available power gain GA is required for noise calculations, as mentioned above,
but beyond this it is of no importance.
27.4.6
Gain with Impedance Matching at Both Sides
With identical resistances at the input and output, matching at both sides means that, for the
ideal amplifier in Fig. 27.26, Rg = ri = ro = RL = ZW . In this case, all gain definitions
are identical:
A2
G = GI = GT = GA = = 4 A2B (27.33)
4
This is also true for a general amplifier which can be demonstrated by comparing the
equations on the basis of the S and Y parameters, taking into account the given matching
conditions. Due to the length of the required calculations, proof thereof is not given here.
Using the S parameters for an amplifier matched at both sides with Rg = RL = ZW
leads to:
S11 = S22 = rg = rL = 0 ⇒ G = GI = GT = GA = |S21 |2
This is a simple relationship because the measuring condition RL = ZW for determining
S21 is equal to the operating condition.
When using the Y parameters, the two-sided match to 1/Yg = 1/YL = ZW is reached
when certain conditions are met:10
Y11 = Y22 , (Y11 Y22 − Y12 Y21 ) ZW
2
= 1 (27.34)
Then:
|Y21 |2 ZW
2
G = GI = GT = GA = (27.35)
|1 + Y11 ZW |2
10 These conditions are determined by calculating the Y parameters according to Fig. 24.40 on page
1186 from the S parameters while taking into account the condition S11 = S22 = 0.
1360 27 High-Frequency Amplifiers
For an amplifier without reverse transmission Y12 = 0; from the above conditions it
follows Y11 = Y22 = 1/ZW , i.e., the input resistance ri = 1/Y11 and the output resistance
ro = 1/Y22 must be equal to the characteristic resistance ZW . This case corresponds to the
ideal amplifier in Fig. 27.26 from which the matching conditions ri = ZW and ro = ZW
can be directly derived if Rg = RL = ZW .
27.4.7
Maximum Power Gain with Transistors
Sect. 27.2 showed that a generalized discrete transistor can be matched at both sides if the
stability factor is
1 + |S11 S22 − S12 S21 |2 − |S11 |2 − |S22 |2
k = > 1 (27.36)
2 |S12 S21 |
and the secondary conditions
|S12 S21 | < 1 − |S11 |2 , |S12 S21 | < 1 − |S22 |2 (27.37)
are met; here S11 , . . . , S22 are the S parameters of the transistor. The conditions for the Y
parameters are
2 Re {Y11 } Re {Y22 } − Re {Y12 Y21 }
k = > 1 (27.38)
|Y12 Y21 |
and:
Re {Y11 } ≥ 0 , Re {Y22 } ≥ 0 (27.39)
S21
# $
Y21
# $
MAG = |S21,match | =
2
k − k 2 − 1 =
k − k2 − 1 (27.40)
S
12
Y
12
At high frequencies, MAG is inversely proportional to the square of the frequency:
MAG ∼ 1/f 2 , which corresponds to a declining rate of 20 dB/decade. This is caused
by the frequency dependence of the S and Y parameters.
2
|S 21,match | = MAG
Fig. 27.28. Maximum available power gain (MAG) of an amplifier matched at both sides
27.4 Power Gain 1361
S21
Y21
MSG =
=
(27.41)
S12
Y12
S21
− 1
2
S12 |Y21 − Y12 |2
U =
. = (27.42)
S21
4 (Re {Y11 } Re {Y22 } − Re {Y12 Y21 })
k
− Re S21
S12
S12
This assumes that the transistor is neutralized by suitable circuitry, i.e., it has no reverse
transmission; it then operates unilaterally. Circuits for neutralization are described in
Sect. 27.2. At high frequencies, the unilateral power gain is approximately inversely
proportional to the square of the frequency: U ∼ 1/f 2 , which corresponds to a decline
rate of 20 dB/decade.
Limit Frequencies
The maximum available power gain (MAG) assumes the value 1 or 0 dB at the transit
frequency fT of the transistor. The unilateral power gain (U ) is higher than one even
above the transit frequency since the reverse transmission is eliminated. The frequency at
which U assumes the value 1 or 0 dB is called the maximum oscillation frequency fmax .
This represents the maximum frequency at which the transistor can be operated as an
oscillator.
Example: Figure 27.29 shows the maximum power gains for transistor BFR93 at VCE,A =
5 V and IC,A = 30 mA. The maximum available power gain (MAG) is only defined for
f > 500 MHz as only here does the stability factor k rise above one. It declines at a rate
of 20 dB/decade and assumes the value 1 or 0 dB at the transit frequency fT = 5 GHz.
For f < 500 MHz the maximum stable power gain (MSG) is obtained which, at lower
frequencies, declines at a rate of 10 dB/decade. At high frequencies the unilateral power
gain U is higher than MAG by approximately 7.5 dB and assumes the value 1 or 0 dB at
fmax = 12 GHz.
In transistors with transit frequencies above 20 GHz, the collector-base capacitance
CC or the gate-drain capacitance CGD are usually reduced to such an extent that the
transistor can be regarded as having no reverse transmission even without neutralization.
In this case, the maximum oscillation frequency fmax is only slightly higher than the transit
frequency fT .
1362 27 High-Frequency Amplifiers
MAG
dB 50
MSG
dB 40
U
dB 30
MSG U
20
– 10dB/Dek. – 20dB/Dek.
10
k=1 MAG
fT f max
0
~
~
Fig. 27.29. Maximum power gains for transistor BFR93 at VCE,A = 5 V and IC,A = 30 mA
Chapter 28:
Mixer
Mixers are required for frequency conversion in transmitters and receivers and together with
amplifiers and filters are among the essential components of radio transmission systems.
The following sections explain the functional principle of a mixer and then describe the
circuits used in practical applications.
28.1
Functional Principle of an Ideal Mixer
An ideal mixer corresponds to a multiplier (see Fig. 28.1). The signal to be converted along
with the local oscillator signal required for the conversion are fed into the inputs; in the
ideal case the latter signal is sinusoidal. The output provides the converted signal as well
as additional components generated in the conversion process. The unwanted components
must be suppressed by filters in further processing. For this reason, one or two filters are
required in addition to the mixer for frequency conversion. Usually the input for the signal
to be converted is called the input and the input for the local oscillator signal is called the
local oscillator input.
The process of converting the input signal to a higher frequency is called up-conversion
and the corresponding mixer is known as up-conversion mixer. Correspondingly, the terms
down conversion and down-conversion mixer are used when the input signal is converted
to a lower frequency. Figure 28.2 shows the characteristic frequencies in up-conversion
and down-conversion mixers.
Ideal mixer
= multiplier
Signal to be converted Converted signal
= input signal = output signal
– The intermediate frequency (IF) fIF is the lower of the two carrier frequencies, i.e. the
carrier frequency of the input signal of the up-conversion mixer or the carrier frequency
of the output signal in the down-conversion mixer. In up conversion of the signal from
the base band or down conversion of the signal into the base band, the intermediate
frequency is fIF = 0, which is the case, for example, in I/Q mixers.
– The radio frequency (RF) fRF is the higher of the two carrier frequencies, i.e. the carrier
frequency of the output signal in the up-conversion mixer or the carrier frequency of the
input signal in the down-conversion mixer.
– The local oscillator frequency (LO) is the frequency of the required local oscillator
signal and corresponds to the frequency shift achieved by the conversion.
The corresponding signals are called IF, RF and LO signals.
A distinction must be made between the frequencies related to the individual mixers
and the frequencies in a specific transmitter or receiver. In a transmitter, each IF frequency
of the transmitter occurs at the IF frequency of one of the mixers. Consequently, each IF
frequency and the transmit frequency of a transmitter is generated by a mixer and therefore
occurs as the RF frequency of the respective mixer. The same applies to the receiver. In
the following we look at the frequencies of a single mixer while the meaning of these
frequencies in a specific transmitter or receiver remains open.
28.1.1
Up-Conversion Mixer
An IF signal1
sIF (t) = a(t) cos [ωIF t + ϕ(t)]
is fed to the input of an up-conversion mixer and multiplied by the local oscillator signal
sLO (t) = 2 cos ωLO t
(see Fig. 28.3). The amplitude of the local oscillator signal is set to 2 so that the following
equations contain no factors of 1/2; this has no influence on the basic relations. The output
signal is:
sRF (t) = sIF (t) · sLO (t) = a(t) cos [ωIF t + ϕ(t)] · 2 cos ωLO t
= a(t) cos [(ωLO + ωIF ) t + ϕ(t)] + a(t) cos [(ωLO − ωIF ) t − ϕ(t)]
3 45 6 3 45 6
Upper band (f > fLO ) Lower band (f < fLO )
in noninverted mode in inverted mode
The component of frequency fLO + fIF is called the upper band and is of the same
frequency sequence as the IF signal; this is known as noninverted mode. The compo-
nent at frequency fLO − fIF is called the lower band and is of inverted frequency se-
quence compared to the IF signal; this is known as inverted mode. Each of the two bands
can serve as the output signal. The undesired band must be suppressed by means of a
filter.
1 Here, the amplitude modulation a(t) and the angular modulation ϕ(t) are used for the nota-
tion since this is more compact than the expression with the quadrature components: sIF (t) =
i(t) cos ωIF t − q(t) sin ωIF t.
28.1 Functional Principle of an Ideal Mixer 1365
sIF (t) = a(t) cos [wIF t + (t)] sRF (t) = a(t) cos [(wLO + wIF ) t + (t)]
+ a(t) cos [(wLO – wIF ) t – (t)]
2 cos wLO t
S IF SRF f IF f IF SRF
28.1.2
Down-Conversion Mixer
An RF signal
sRF (t) = a(t) cos [ωRF t + ϕ(t)]
is fed to the input of the down-conversion mixer and multiplied by the local oscillator
signal
sLO (t) = 2 cos ωLO t
(see Fig. 28.4). The output signal is:
sM (t) = sIF (t) · sLO (t) = a(t) cos [ωRF t + ϕ(t)] · 2 cos ωLO t
⎧
⎪ a(t) cos [(ωRF − ωLO ) t + ϕ(t)] Noninverted mode (fRF > fLO )
⎪
⎪
⎪
⎪ + a(t) cos [(ωRF + ωLO ) t + ϕ(t)]
⎨
=
⎪
⎪ a(t) cos [(ωLO − ωRF ) t − ϕ(t)] Inverted mode (fRF < fLO )
⎪
⎪
⎪
⎩ + a(t) cos [(ωLO + ωRF ) t + ϕ(t)]
In addition to the desired components at the differential frequency, the output signal con-
tains an extra component at the summation frequency which must be suppressed by a filter.
The IF signal of interest is therefore:
,
a(t) cos [(ωRF − ωLO ) t + ϕ(t)] Noninverted mode (fRF > fLO )
sIF (t) =
a(t) cos [(ωLO − ωRF ) t − ϕ(t)] Inverted mode (fRF < fLO )
If the RF frequency is higher than the LO frequency, then the result is an IF signal in
noninverted mode of the same frequency sequence (see Fig. 28.4a). Otherwise we obtain
an IF signal in inverted mode with an inverted frequency sequence (see Fig. 28.4b).
With down-conversion mixers it often occurs that the signal supplied to the RF input,
in addition to the desired RF signal of the frequency fRF = fLO ± fIF , also contains an
image signal with the image frequency fRF,im = fLO ∓ fIF , which is likewise converted
to the IF frequency. In this case, the mixer operates in the noninverted mode and in
the inverted mode. Figure 28.5 shows a down-conversion mixer with the RF frequency
fRF = fLO + fIF in noninverted mode and the image frequency fRF,im = fLO − fIF in
1366 28 Mixer
sIF (t)
sRF (t) = a(t) cos [!RF t + (t)] sM (t) = a(t) cos [(!RF – !LO ) t + (t)]
+ a(t) cos [(!RF + !LO ) t + (t)]
2 cos !LO t
S IF f IF S RF
sIF (t)
sRF (t) = a(t) cos [!RF t + (t)] sM (t) = a(t) cos [(!LO – !RF ) t + (t)]
+ a(t) cos [(!LO + !RF ) t + (t)]
2 cos !LO t
S IF S RF f IF
S IF
SRF,im f IF f IF S RF
f IF = f RF – f LO f RF,im f LO f RF f RF,im + f LO f RF + f LO f
= f LO – f RF,im
Fig. 28.5. Image frequency fRF,im in a down-conversion mixer in noninverted mode. The
frequency sequence of the image signal |S RF,im | is reversed due to the inverted mode
28.2 Functional Principles of Practical Mixers 1367
inverted mode where the frequency sequence of the image signal is reversed due to the
inverted mode. In order to ensure that the mixer only converts the desired RF signal, the
image signal must be suppressed by an image frequency filter arranged in front of the
mixer; this will be explained in more detail in Sect. 27.2 on receivers. The existence of the
image signal is a consequence of the functional symmetry of the up- and down-conversion
mixers. The up-conversion mixer converts one IF signal into two RF signals, one of which
must be selected behind the mixer. Similarly the down-conversion mixer converts two RF
signals into one IF signal, meaning that one of the RF signals must be selected in front of
the mixer.
28.2
Functional Principles of Practical Mixers
Multipliers are rarely used in practice. Practical multipliers feature a high linearity for both
inputs which is not required for frequency conversion as explained below. Thus, practical
multipliers used as mixers are even undesirable because they have a high noise figure due
to the complex circuitry that is required to obtain the high linearity. In fact, in most cases
the noise figure that occurs when a multiplier is used as a mixer is unacceptably high.
For a practical mixer it is sufficient for the signals of an ideal up- or down-conversion
mixer to be contained in the voltages or currents of the mixer. The voltages and currents
may contain other signals as long as these are separated from the useful signals with respect
to the frequency so that they can be suppressed by filters at the output. Here, a distinction
must be made between additive and multiplicative mixing. The two types of mixing will
be described in the sections below taking an up-conversion mixer as an example.
28.2.1
Additive Mixing
In additive mixing, the IF and the LO signals are added together, provided with a suitable
DC component V0 and fed to a component with a nonlinear characteristic. The nonlinearity
results in a multitude of mixing frequencies including the desired RF frequency, which
is separated by means of a bandpass filter. Figure 28.6. shows the principle of additive
mixing.
Description by Equations
In practice, the current-voltage characteristic I (V ) of a diode or transistor is usually used
for the nonlinear characteristic, i.e. the input signal is a voltage and the output signal is a
Bandpass
Non-linear filter for
Summation
component RF signal
vIF (t) = v^IF cos wIF t
vLO (t) = v^LO cos wLO t vRF (t) = v^RF cos wRF t
V0
dI
1 d 2 I
I (V ) = I (V0 ) + (V − V0 ) +
(V − V0 )2
dV
V =V0 2 dV 2
V =V0
1 d 3 I
1 d 4 I
+
(V − V0 ) +
3
(V − V0 )4 + · · ·
6 dV 3
24 dV 4
V =V0 V =V 0
+ ···
The quadratic term
2 a2 v̂IF v̂LO cos ωIF t cos ωLO t
= a2 v̂IF v̂LO [cos (ωLO + ωIF ) t + cos (ωLO − ωIF ) t]
contains the desired expression. The current i(t) is fed to a bandpass filter that separates
the signal portion at fRF = fLO + fIF (noninverted mode) or at fRF = fLO − fIF (inverted
mode) in order to convert it into an output voltage
vRF (t) = v̂RF cos ωRF t = RBP a2 v̂IF v̂LO cos ωRF t
Here, RBP is the transmission impedance of the bandpass filter in the passband.2
The amplitude v̂RF of the output voltage is proportional to the coefficient a2 of the
nonlinear characteristic, where a2 should be as high as possible so that the oscillator
amplitude v̂LO required for a certain output amplitude is kept low.
Nonlinearity
If we evaluate other terms of the current i(t), we see that all coefficients ai with an even
index i contribute a component at frequency fRF . For example, the term
% &4
a4 v 4 (t) = a4 v̂IF cos ωIF t + v̂LO cos ωLO t
contains the components
3 3
a4 v̂IF v̂LO [cos (ωLO + ωIF ) t + cos (ωLO − ωIF ) t]
2
2 Please note the units: [R ] = , [a ] = A/ V2 and [v̂ ] = [v̂ ] = V; consequently
BP 2 IF LO
[RBP a2 v̂IF v̂LO ] = V.
28.2 Functional Principles of Practical Mixers 1369
and:
3 3
a4 v̂IF v̂LO [cos (ωLO + ωIF ) t + cos (ωLO − ωIF ) t]
2
The amplitude of the first component is proportional to v̂IF and adds to the desired output
signal. In contrast, the amplitude of the second component is proportional to v̂IF 3 and
a4 v̂ v̂LO
a2 v̂IF v̂LO
⇒ v̂IF
2 a2
3
2 IF
3 a
By evaluating additional terms of the current i(t), more conditions for v̂IF are found that
depend on the coefficients a6 , a8 , . . . . Normally, additive mixing is nonlinear and can
be regarded as quasi linear only for low IF amplitudes. Conversely, it is strictly linear
only if all coefficients with an even index i ≥ 4 are zero as is the case with a quadratic
characteristic of i = a2 v 2 .
The frequencies generated in additive mixing can be presented systematically in the
form of a frequency pyramid (see Fig. 28.7). The coefficients ai produce frequency groups
(m, n) with nonnegative integer values for m and n and m + n = i; the coefficients a2 thus
produce the groups (2,0), (1,1) and (0,2). The frequencies belonging to one group (m, n)
are determined by calculating the sum
±f ± · · · ± fLO ±fIF ± · · · ± fIF
3 LO 45 6 3 45 6
m addends n addends
for all possible sign configurations that evaluate to nonnegative values. For group (1, 1)
the addends are
fLO + fIF , fLO − fIF , − fLO + fIF , − fLO − fIF
and the frequencies under the assumption fLO > fIF are:
fLO + fIF , fLO − fIF
In groups with higher values for m and n, the number of frequencies increases. All fre-
quencies of the group (m, n) are also contained in the groups (m + 2, n) and (m, n + 2).
The recursive application shows that all frequencies of one coefficient ai are also produced
by the coefficients a(i+2) , a(i+4) , a(i+6) , . . . . Therefore, besides the coefficient a2 , the co-
efficients a4 , a6 , a8 , . . . are also of importance in additive mixers. The amplitudes of a
RF = fLO ± fIF
group (m, n) are proportional to v̂LO m v̂ n . The desired output signal with f
IF
lies in the group (1,1) and is thus proportional to v̂LO v̂IF . Other components of the same
frequency occur, for example, in the groups (3,1) and (1,3). The component in group (3,1)
3 v̂ and is thus linear with respect to v̂ . The component in group
is proportional to v̂LO IF IF
(1,3), on the other hand, is proportional to v̂LO v̂IF 3 and is thus nonlinear with respect to v̂ .
IF
The nonlinearity of additive mixing not only results in a nonlinear relation between
the IF amplitude v̂IF and the RF amplitude v̂RF , but causes additional intermodulation
distortions in modulated IF signals. For this we replace the constant IF amplitude v̂IF by
an amplitude-modulated signal without carrier of the modulation frequency fm ; then:
vIF (t) = v̂IF cos ωm t cos ωIF t
v̂IF
= [cos (ωIF + ωm ) t + cos (ωIF − ωm ) t]
2
1370 28 Mixer
(1,0) (0,1)
a1 f LO f IF
...
...
A calculation that is not shown here in detail reveals that all coefficients ai with an even in-
dex i provide components at the desired output frequencies fLO ±fIF ±fm . The coefficients
ai with i = 4, 6, 8, . . . supply additional components at the frequencies fLO ± fIF ± 3fm
and those with i = 6, 8, . . . supply components at fLO ± fIF ± 5fm , etc. These unwanted
components, known as intermodulation products, are proportional to higher powers of the
IF amplitude and must therefore also be reduced to an acceptable level by limiting the IF
amplitude. For this case, too, a frequency pyramid can be built by utilizing the frequencies
fIF + fm and fIF − fm instead of the frequency fIF in group (0,1) and then calculating
the other groups in the usual fashion. The number of frequencies in the groups increases
compared to Fig. 28.7 because the following summations must be calculated:
±f ± · · · ± fLO ±(fIF ± fm ) ± · · · ± (fIF ± fm )
3 LO 45 6 3 45 6
m addends n addends fIF ± fm
28.2 Functional Principles of Practical Mixers 1371
(1,0) (0,1)
f LO f IF + f m
a1
f IF – f m
..
..
..
Figure 28.8 shows a portion of the frequency pyramid for intermodulation products with
the intermodulation products of third order (fLO ± fIF ± 3fm ) in group (1,3) which are
3 .
proportional to v̂LO v̂IF
Basically, the relationships are the same as in a nonlinear amplifier. Therefore, the
nonlinear characteristic parameters (compression and intercept points) can be used for
mixers in the same way. However, the relationships between the characteristic parameters
and the coefficients of the nonlinear characteristic are different because in an amplifier the
coefficients ai with an uneven index i are important unlike those with an even index i as
is the case in a mixer. A nonlinear mixer can be regarded as a nonlinear amplifier with an
additional frequency shift. For the qualitative aspects please refer to Sect. 4.2.3.
Practical Circuits
Figure 28.9 shows some typical circuits for additive mixing. In the circuit with a diode in
Fig. 28.9a, the IF signal, together with the voltage V0 for operating point setting, is supplied
1372 28 Mixer
v LO
V0 + vIF f RF v RF
f RF
Vb
Vb
f RF
f RF
v RF
v LO
v RF V0 + vIF
V0 + vIF v LO
b With bipolar transistor summation by means c With bipolar transistor and summation
of a transformer by means of a separate supply to the
base and emitter
Fig. 28.9. Typical circuits for additive mixing. The parallel resonant circuits are tuned to the RF
frequency fRF
directly whereas the LO signal is added via a transformer. A parallel resonant circuit is used
as the bandpass filter for the RF signal. The voltage at the diode corresponds to the sum
V0 + vIF + vLO since, at the IF and LO frequencies, the parallel resonant circuit acts as a
short circuit.At the output, all current components, with the exception of the RF component,
are short-circuited by the parallel resonant circuit. The RF component generates the RF
output voltage vRF at the resistance of the resonant circuit. The disadvantage of this circuit
is that the RF current must flow through the IF and LO signal sources in order to close the
RF current loop. This can be prevented by connecting a series-resonant circuit between
the transformer and the diode to ground in order to short-circuit the RF current at this
point (see Fig. 28.9a). For optimum operation, complete decoupling of the IF, LO and RF
terminals is necessary. To do so, the IF and the LO circuit must also be provided with a
parallel resonant circuit which acts as an open circuit for the respective frequency, while
for all other frequencies it acts almost as a short circuit. Section 28.3 explains this in more
detail. Since the diode is a passive component, the RF power available at the output is
always lower than the supplied IF power, i.e. there is a conversion loss.
A conversion gain can be achieved by using a bipolar transistor (see Fig. 28.9b).
Here, the RF current does not flow through the IF or LO signal source. The RF portion is
decoupled from the other components relatively well. This decoupling can be enhanced
by introducing a cascode transistor. Since the RF current of the transistor can be picked
28.2 Functional Principles of Practical Mixers 1373
up at a separate connection, namely the collector, the IF and the LO signals can be added
by supplying one signal to the base and the other to the emitter (see Fig. 28.6c) which
eliminates the need for a transformer. However, the RF current then flows through the LO
signal source.
The exponential characteristic
⎛ ⎞
V
I (V ) = IS ⎝e nVT − 1⎠
1 d i I
1 I0 1 I0
ai =
= ⇒ a2 = , ...
i ! dV i
i ! (nVT ) i 2 (nVT )2
V =V0
The coefficients are proportional to the bias current such that if I0 = 100 mA and n = 1,
a2 ≈ 74 mA/V2 . The bias current I0 and the amplitudes of the IF and the LO signals must
be selected such that the peak current
⎛ ⎞
V0 +v̂LO +v̂IF V0 +v̂LO +v̂IF v̂LO +v̂IF
Imax = IS ⎝e nVT − 1⎠ ≈ IS e nVT = I0 e nVT
is not too high considering that v̂LO + v̂IF = 100 mV and n = 1 already cause a current
of Imax ≈ 47 I0 . In practice, the levels are chosen so that the maximum IF signal ampli-
tude is clearly lower than the local oscillator amplitude; then, the peak current is almost
independent of the IF signal.
Instead of a bipolar transistor it is also possible to use a field effect transistor. This is even
more advantageous since the field effect transistor only produces very low intermodulation
distortions due to its approximately parabola-shaped transfer characteristic (a2 = 0 and
ai ≈ 0 for i > 2). Figure 28.10a shows a commonly used circuite with a junction FET
which allows the bias voltage V0 to be omitted and the operating point to be set by resistance
RS . From the transfer characteristic
K
ID = (VGS − Vth )2
2
it follows:
1 d 2 ID K
a2 = 2
=
2 dVGS 2
This means that coefficient a2 does not depend on the bias current but rather on the size
of the FET represented by the transconductance coefficient K. Even in very large-sized
FETs, it is clearly lower than in a bipolar transistor with typical operating point values.
Typical values lie in the range of a2 ≈ 1 . . . 10 mA/V2 .
Another circuit with an approximately parabola-shaped characteristic is the current
squaring circuit with bipolar transistors shown in Fig. 28.10b where IC4 ∼ I12 . For transis-
1374 28 Mixer
Vb Vb Vb
I0 / 2 + iIF I0 / 2 + i LO
fRF f RF
I1 = I 0 + i IF + i LO
Fig. 28.10. Additive mixers with approximately parabola-shaped characteristic. The parallel
resonant circuits are tuned to the RF frequency fRF .
tors of equal size (identical saturation reverse current IS ) and if base currents are neglected,
the following applies:
I1 I0 IC4
VBE 1 = VBE 2 = VT ln , VBE 3 = VT ln , VBE 4 = VT ln
IS IS IS
From the loop equation
VBE 1 + VBE 2 = VBE 3 + VBE 4
we obtain:
I12 I12
VBE 4 = VBE 1 + VBE 2 − VBE 3 = VT ln ⇒ IC4 =
I0 I S I0
The input current I1 is composed of the bias current I0 , the IF current iIF and the LO current
iLO . It originates from two current sources with the bias current I0 /2 and the small-signal
currents iIF and iLO .
A commonly used additive mixer is the one shown in Fig. 28.11a, which is provided
with two MOSFETs in cascode configuration and functions in most cases by means of
a dual-gate MOSFET (DGFET) (see Fig. 28.11b). We assume that the MOSFETs are of
equal size, i.e. both MOSFETs have the same transconductance coefficient K. The lower
MOSFET is operated in the ohmic region; thus, its drain current ID1 not only depends on
the gate-source voltage VGS1 , but also to a high degree on the drain source voltage VDS1 .
For V1 = VGS1 the drain current is:
(3.2)
VDS1
ID1 = KVDS1 V1 − Vth −
2
The channel length modulation can be neglected. The upper MOSFET is operated in the
pinch-off region. With respect to the lower MOSFET the upper MOSFET operates in
common drain configuration (source follower) and thus sets the voltage VDS1 . From
(3.3) K
ID2 = (VGS2 − Vth )2
2
28.2 Functional Principles of Practical Mixers 1375
Vb Vb
fRF fRF
I D2
vRF vRF
T2
V2 = V2,A + vLO
VGS2 V2,A + vLO
I D1
T1 VDS1
Insertion of this equation into the equation for ID1 results in, among other things, the term
KV1 V2 , which contains the desired product of the IF and LO signals due to V1 = V1,A +vIF
and V2 = V2,A + vLO . Since VDS1 also depends on ID1 , the resulting equation cannot be
solved for ID1 ; therefore, the drain current and the RF signal filtered out by the parallel
resonant circuit can only be determined numerically. The LO signal and the operating point
voltage V2,A are chosen such that the ohmic range of the lower MOSFET is used to its full
extent which results in the maximum IF signal. In literature, this mixer is often listed under
multiplicative mixers as the IF and LO signals are supplied to separate connections and
are not added explicitly. We regard it as an additive mixer since, because of its dependence
on ID1 , the drain source voltage VDS1 not only depends on V2 but also on V1 , where
VDS1 = VDS1 (V1 , V2 ). Thus, the drain current contains components that are nonlinear
with respect to V1 .
In modern transmitters and receivers, additive mixers are seldom used. There are es-
sentially two reasons for this:
– A practical mixer must be designed such that the inputs and outputs are decoupled as
well as possible and can be matched to the characteristic impedance; also the gain should
be as high as possible and the noise figure as low as possible. When using an additive
mixer this is feasible with major limitations only.
– Owing to the general nonlinearity, the intermodulation distortions are comparatively
high which limits the dynamic range.
1376 28 Mixer
28.2.2
Multiplicative Mixers
In multiplicative mixers the IF signal is multiplied by the LO signal. In contrast to an
ideal mixer, the multiplicative mixer uses a general periodic LO signal of the fundamental
frequency fLO instead of a sinusoidal signal. The desired RF frequency is obtained from
the output frequencies by a bandpass filter.
Multiplicative mixing becomes very simple if square-wave LO signals are used be-
cause multiplication can then be achieved with switches. Figure 28.12 shows the principle
Bandpass filter
with RF signal
Multiplier
sM (t)
sIF (t) sRF (t)
sLO (t)
Unipolar Bipolar
square-wave square-wave
signal signal
t
0 –1
t
vIF
vIF vM –1 vM
– vIF
fLO f LO
vM
vIF vM vIF
f LO
f LO
Fig. 28.12. Principle of multiplicative mixing (top) and special cases with square-wave LO signals
28.2 Functional Principles of Practical Mixers 1377
of multiplicative mixing and the special situations for unipolar and bipolar square-wave
signals. In the case of the unipolar square-wave signal, the IF signal is only multiplied
by 0 and 1 which can be implemented by an On/Off switch that operates as a series or
short-circuiting switch. In the case of the bipolar square-wave signal, the IF signal is mul-
tiplied by +1 and −1. For this purpose one can generate −vIF by means of an inverting
amplifier and use a change-over switch to alternate between vIF and −vIF . A two-pole
switch can be used as an alternative. For this purpose electronic switches that are driven
by a square-wave signal of the frequency fLO are used.
Description by Equations
The signal sM (t) at the output of the multiplier in Fig. 28.12 is obtained by a Fourier series
expansion of the LO signal:
sM (t) = sIF (t) · sLO (t)
= sIF (t) · [ c0 + c1 cos (ωLO t + ϕ1 ) + c2 cos (2ωLO t + ϕ2 ) + · · · ]
! ∞
"
0
= sIF (t) · c0 + cn cos (nωLO t + ϕn )
n=1
The IF signal is multiplied by the fundamental wave (c1 ) and the harmonics (c2 , . . .)
of the LO signal. Furthermore, there is a direct transmission corresponding to the DC
component (c0 ).
The following Fourier series are obtained from the square-wave signals in Fig. 28.12:
⎧
⎪
⎪ 1 2 2
⎨ + cos ωLO t − cos 3ωLO t + · · · unipolar
2 π 3π
sLO (t) =
⎪
⎪ 4 4
⎩ cos ωLO t − cos 3ωLO t + · · · bipolar
π 3π
⎧ ∞
⎪
⎪ 1 2 0 (− 1)n
⎪
⎪ + cos (2n + 1) ωLO t unipolar
⎨ 2 π 2n + 1
= ∞
n=0 (28.1)
⎪
⎪
⎪ 4 0 (− 1)n
⎪
⎩ π cos (2n + 1) ωLO t bipolar
2n + 1
n=0
This only produces odd multiples of the LO frequency. Furthermore, the bipolar square-
wave signal has no DC component. With the modulated IF signal
sIF (t) = a(t) cos [ωIF t + ϕ(t)]
the output of the multiplier provides in the case of a unipolar square-wave signal:
a(t)
sM (t) = cos [ωIF t + ϕ(t)]
2
a(t)
+ {cos [(ωLO + ωIF ) t + ϕ(t)] + cos [(ωLO − ωIF ) t − ϕ(t)]}
π
a(t)
− {cos [(3ωLO + ωIF ) t + ϕ(t)] + cos [(3ωLO − ωIF ) t − ϕ(t)]}
3π
+ ···
1378 28 Mixer
S IF
f IF f
f IF f IF fIF fIF f IF f IF
SM
Filter
f IF f LO 2 f LO 3 f LO 4 f LO 5 f LO f
f LO – f IF f LO + f IF 3 f LO – f IF 3 f LO + f IF 5 f LO – f IF 5 f LO + f IF
f IF f IF f IF f IF f IF f IF
SM
Filter
f LO 2 f LO 3 f LO 4 f LO 5 f LO f
f LO – f IF f LO + f IF 3 f LO – f IF 3 f LO + f IF 5 f LO – f IF 5 f LO + f IF
In the case of a bipolar square-wave signal there is no IF frequency component and the
amplitudes of all other components are doubled. Figure 28.13 shows the corresponding
spectra. At the LO frequency and all odd multiples of the LO frequency there is an upper
band in noninverted mode and a lower band in inverted mode. The amplitudes decrease
with increasing frequency according to the Fourier coefficients of the LO signal. The upper
and lower bands of the LO frequency are used as the RF signal, i.e. fRF = fLO ± fIF
while all other components are suppressed by a filter. Generally, components of higher
frequencies can also be used for the RF output signal.
If the square-wave signals are asymmetric (duty cycle = 50%), the Fourier series also
contains components at even multiples of the LO frequency. The output of the mixer then
supplies upper and lower bands even at these frequencies, e.g. at 2fLO ± fIF .
signal sLO that is used in the multiplication. This does not affect the basic function of the
28.2 Functional Principles of Practical Mixers 1379
s’LO (t)
Switching
t
characteristic
of the switch 1/ fLO
sLO
sLO (t)
t
mixer as the distortion only causes a change in the Fourier coefficient of the LO signal,
which can be accepted as long as the fundamental wave of the effective LO signal remains
sufficiently high.
In practice, the supplied LO signal sLO is also not usually a square-wave signal as the
generation of high frequency square pulses is difficult and causes significant interferences;
instead, the virtually sinusoidal signal of a high frequency oscillator is used. The effective
LO signal then depends on the switching characteristic of the electronic switches that are
controlled by sinusoidal signals.
With an increasing LO frequency the effect of the nonideal switching characteristic
becomes more noticeable. With frequencies above 10 GHz even the fastest switching diodes
no longer function as accurate switches but rather operate only within the transition region;
thus, the multiplicative mixer changes into an additive mixer.
Nonlinearity
The description by equations shows that multiplicative mixing is linear with regard to the
relationship between IF and RF amplitudes; consequently, there are no intermodulation
products. In practice, this is not the case as the required electronic switches are not exactly
linear in operation and have modulation limits. This is the reason why multiplicative
mixers are also characterized by nonlinear parameters (compression and intercept points).
The relationships are basically the same as in additive mixers; however, the nonlinearity
is usually much lower since it is caused solely by secondary effects while in additive
mixers nonlinearity is a precondition for the mixer function. Therefore, the compression
and intercept points achieved with multiplicative mixers are higher.
1380 28 Mixer
Practical Circuits
By principle, every additive mixer can also be used as a multiplicative mixer by using a
square-wave LO signal and selecting the operating point and the LO signal amplitude such
that the diode or transistor of the additive mixer switches between the nonconductive and
the conductive state when driven by the LO frequency. At the same time, the amplitude
of the input signal is selected so small that the small-signal modulation prevails; then the
input signal is amplified by the respective small-signal gain, i.e. alternating between zero (=
small-signal gain in the nonconductive state) and a constant value (= small-signal gain in the
conductive state). The result is a multiplicative mixer with On/Off switch and, depending
on the circuit, additional gain or attenuation. Figure 28.15 illustrates this using the mixer
with diode from Fig. 28.9a as an example. Thus the diode acts as an electronic On/Off
switch with a forward resistance corresponding to the small-signal resistance rD (v̂LO ) in
the conductive state.
The multiplicative mixers used in practice are described in the following sections.
vLO 1/f LO
^
vLO Diode is conductive
t
–v^LO Diode is nonconductive
vLO
vIF vRF
fRF
rD ( v^LO )
vIF vRF
f RF
f LO
28.3
Mixers with Diodes
Mixers with diodes are widely used and predominantly found in circuits made up of discrete
components. They operate almost exclusively as multiplicative mixers, i.e. the diodes are
used as switches. Since the diode is a passive component these mixers always have a typical
conversion loss of 5 . . . 8 dB. For this reason they are also known as passive mixers.
Due to the high frequencies, diodes with excellent switching performance are required.
These are usually special Schottky diodes (mixer diodes) with very low junction capaci-
tance; the diffusion capacitance of Schottky diodes is negligibly low. In order to minimize
the junction capacitance, the area of the metal-semiconductor contact must be minimized
and doping must be reduced in comparison to standard diodes which consequently increases
the spreading resistance. Mixer diodes are thus characterized by a very low capacitance
and a relatively high spreading resistance.
The following section describes the mixer with a single diode in more detail as all
mixers with diodes can be reduced to this construction in terms of their transmission
characteristic.
28.3.1
Unbalanced Diode Mixer
Figure 28.16a shows the circuit of a mixer with a single diode called an unbalanced (diode)
mixer. The LO voltage VLO is a large-signal voltage that is used to periodically switch over
the operating point of the diode between the forward and reverse state. The IF voltage vIF
is a small-signal voltage and is transferred to the RF output according to the small-signal
response of the diode. The LO and the IF voltages are added by a 1:1 transformer.
Frequency separation at the three connections is achieved by means of three narrow-
band parallel resonant circuits. At the resonant frequency they act like an open circuit
and are thus ineffective, while at all other frequencies they nearly cause a short circuit.
Consequently, these connections only carry voltages and currents at the relevant resonant
frequency. The capacitance of the diode is considered an integral part of the parallel
resonant circuit which eliminates the need to account for it separately.3
The method for calculating the properties is basically the same as for all small-signal
circuits. First, the operating point is determined and the circuit linearized, it is then possible
to calculate the small-signal behavior. Unlike amplifiers, mixers do not have a constant
operating point, but one that changes periodically in accordance with the LO voltage
resulting in a time-variable operating point. The calculation of this time-variable operating
point is based on the LO circuit.
LO Circuit
Figure 28.16b shows the LO circuit of the unbalanced diode mixer which assumes that
the parallel resonant circuits at the IF and the RF terminals act as short circuits for the
LO frequency and multiples thereof. The 1:1 transformer supplies the LO voltage to the
3 The combined effect of capacitance and spreading resistance of the diode cause losses that are
proportional to the frequency and are neglected in our simplified approach. A detailed calculation
can be found in [28.1].
1382 28 Mixer
Rg,LO I LO
Vg,LO VLO
f LO
1:1
Rg,IF i IF ID i RF
Rg,LO I LO I D,LO
Vg,LO VLO
f LO
VLO
I D,LO
VLO VD = VLO
b LO circuit
Rg,IF i IF gD ( t ) iD i RF
diode. It is sinusoidal with the LO frequency fLO since the LO parallel resonant circuit
suppresses all harmonics at multiples of fLO :
VD (t) = VLO (t) = v̂LO cos ωLO t
From the voltage the current ID,LO (t) of the time-variable operating point
ID,LO (t) = ID (VLO (t))
is determined by means of the diode characteristic and has the maximum value:
ID,max = ID (v̂LO )
It cannot be calculated using the simple exponential diode characteristic according to
(1.1) as the mixer diodes are operated in a region in which the spreading resistance has a
noticeable influence. Figure 28.17 shows the typical curves of VLO (t) und ID,LO (t). The
amplitude v̂LO must be higher than the forward voltage VF of the diode in order to cause
an accountable current flow.
28.3 Mixers with Diodes 1383
VLO 1/f LO
v^LO
VF
I D,LO
I D,max
t
gD
gD,max
Fig. 28.17. Unbalanced diode mixer: voltage VLO (t) at the LO circuit, current ID,LO (t) of the
diode and the resulting curve of the small-signal conductivity gD (t). VF is the forward voltage of
the diode.
Here, ID,0 is the DC component and îD,1 is the amplitude of the fundamental wave at
frequency fLO . The series contains only cosine components since in Fig. 28.17 the current
is an even function of the time (ID,LO (−t) = ID,LO (t)). In this case, the coefficients of
the Fourier series are:
1/fLO
ID,0 = fLO ID,LO (t) dt
0
1/fLO
îD,n = 2fLO ID,LO (t) cos nωLO t dt
0
In practice, the coefficients can be determined with the help of a circuit simulation by
performing a time domain simulation for the LO circuit, presenting the current ID,LO (t)
spectrally4 and reading the amplitudes of the components.
Figure 28.18a shows the DC component ID,0 , the fundamental component îD,1 and
the maximum current ID,max of a Schottky diode of type BAS40 dependent upon the LO
amplitude v̂LO . Above v̂LO = 0.3 V, the components are no longer exponential due to the
spreading resistance.
4 In PSpice this is done using the FFT function of the Probe program.
1384 28 Mixer
I v^LO /( nVT ) R
A ∼e Ω
10k
10m 5k
I D,max
2k
1m 1k
^
i D,1 RLO
I D,0 500
100µ 200
ZW,M
100
10µ 50
~ ~
~
~
0.3 0.4 0.5 0.6 0.7 v^LO 0.3 0.4 0.5 0.6 0.7 v^
LO
V V
a Current of the diode: DC component b Resistances for power matching: R LO
^ at the LO terminal and Z W,M at the
I D,0 , fundamental component i D,1 and
maximum current I D,max IF and RF terminals
PLO g D, i
W S
3m 100m
g D,1
1m 30m
300 µ
10m
100 µ g D,0
3m
30 µ
1m
10 µ
~ ~
~
0.3 0.4 0.5 0.6 0.7 v^LO 0.3 0.4 0.5 0.6 0.7 v^
LO
V V
c Power PLO at the LO terminal d Small-signal conductance: DC component
gD,0 and fundamental component g D,1
Fig. 28.18. Characteristic parameters of an unbalanced diode mixer with a Schottky diode of type
BAS40
The DC component and harmonics of current ID,LO (t) are short-circuited by the LO
parallel resonant circuit where the resonant circuit is noneffective only for the fundamental
wave. Consequently, the current ILO (t) at the LO terminal corresponds to the fundamental
wave of the current ID,LO (t):
Since both VLO (t) and ILO (t) are sinusoidal, at a constant LO amplitude the LO circuit
28.3 Mixers with Diodes 1385
dID
dID
gD,max = gD (v̂LO ) =
dVD
VD =v̂LO
The small-signal conductivity is used because, in the reverse region, the small-signal
resistance rD (t) = 1/gD (t) tends toward infinity and thus cannot be adequately shown in
the figures.
Figure 28.17 shows the plot of the small-signal conductivity. With small currents it is
proportional to ID,LO (t) since here the conductivity according to (1.3) is
1 ID,LO (t)
gD (t) = ≈ (28.6)
rD (t) nVT
With large currents the spreading resistance has an influence. The conductivity no longer
increases proportionally to the current; therefore, the peaks in the conductivity curve are
less pronounced than in the current curve.
The small-signal conductivity is also expanded in a Fourier series:
∞
0
gD (t) = gD,0 + gD,n cos nωLO t (28.7)
n=1
1386 28 Mixer
As for the current ID,LO (t), the coefficients can be calculated using the integral equations
for the Fourier series expansion. In practice, this is not necessary as the required coefficients
can be determined by means of circuit simulation which will be further described below.
Figure 28.18d shows the DC component gD,0 and the fundamental component gD,1 for a
Schottky diode of type BAS40 dependent upon the LO amplitude.
Small-Signal Response
In the following, the mixer is operated in noninverting mode with fRF = fLO + fIF and
the small-signal current iD (t) of the diode is calculated. From Fig. 28.16c it follows:
iD (t) = gD (t) vD (t) = gD (t) (vIF (t) − vRF (t)) (28.8)
The voltages vIF (t) and vRF (t) only contain components at the IF or RF frequencies since
the parallel resonant circuits short-circuit all other frequencies:
vIF (t) = v̂IF cos ωIF t , vRF (t) = v̂RF cos ωRF t (28.9)
Inserting (28.7) and (28.9) into (28.8) results in:
∞
0 % &
iD (t) = gD,0 + gD,n cos nωLO t v̂IF cos ωIF t − v̂RF cos ωRF t
n=1
% &% &
= gD,0 + gD,1 cos ωLO t + · · · v̂IF cos ωIF t − v̂RF cos ωRF t
= gD,0 v̂IF cos ωIF t − gD,0 v̂RF cos ωRF t
+ gD,1 v̂IF cos ωLO t cos ωIF t
− gD,1 v̂RF cos ωLO t cos ωRF t
+ ···
= gD,0 v̂IF cos ωIF t − gD,0 v̂RF cos ωRF t
gD,1 v̂IF
+ [ cos ( ωLO + ωIF ) t + cos (ωLO − ωIF ) t ]
2 3 45 6
ωRF
gD,1 v̂RF
− [ cos (ωRF + ωLO ) t + cos ( ωRF − ωLO ) t ]
2 3 45 6
ωIF
+ ···
It can be seen that the fundamental component gD,1 of the small-signal conductivity gD (t)
produces the desired frequency conversion from fIF to fRF by yielding, at the frequency
fLO +fIF = fRF , a component that is proportional to the IF amplitude v̂IF . Similarly, there
is a conversion from fRF to fIF , i.e. at frequency fRF −fLO = fIF a component is produced
that is proportional to the RF amplitude v̂RF . As a result of the harmonic components of
the small-signal conductivity, additional components are generated at higher frequencies
which are not relevant for further calculations.
The small-signal current iD (t) of the diode flows through the IF and RF circuits.
The parallel resonant circuits short-circuit all components at f = fIF in the IF circuit
28.3 Mixers with Diodes 1387
and all components at f = fRF in the RF circuit; thus, only the components at the
respective resonant frequencies are available at the terminals. The small-signal currents
iIF (t) and iRF (t) become available by extracting the components at fIF and fRF from the
current iD (t):
gD,1 v̂RF
iIF (t) = iD (t)
= gD,0 v̂IF − cos ωIF t
f =fIF 2
gD,1 v̂IF
iRF (t) = − iD (t)
= gD,0 v̂RF − cos ωRF t
f =fRF 2
The following relationships for the voltage and current phasors are then derived:
gD,1 v RF
i IF = gD,0 v IF − (28.10)
2
gD,1 v IF
i RF = gD,0 v RF − (28.11)
2
These equations correspond to the four-pole equations in Y notation. The small-signal
response of the mixer can thus be described by a Y matrix:
⎡ gD,1 ⎤
gD,0 −
i IF ⎢ 2 ⎥ v IF
= ⎣ gD,1 ⎦ v (28.12)
i RF − gD,0 RF
2
All parameters of interest, e.g. the small-signal gain and the input and output resistances
at the terminals, can be calculated by means of this Y matrix. The frequency conversion
of the mixer is no longer explicitly apparent.
A method for determining the coefficients gD,0 and gD,1 by means of a circuit simula-
tion follows directly from theY notation of the mixer. For this method, the mixer is operated
according to Fig. 28.19 with an LO voltage source with the intended amplitude v̂LO and
an IF voltage source with the small-signal amplitude v̂IF v̂LO . In the circuit simulation,
these two voltage sources can be connected directly in series and thus the transformer is no
longer required. The parallel resonant circuits at the LO and IF terminals are also omitted
since the voltage sources only contain components of the respective frequencies, while all
other frequencies are short-circuited, thus they adopt the same function as the resonant
^ ^
ID (t) = ... + iIF cos wIF t – iRF cos wRF t...
VLO (t) = v^LO cos wLO t
^ ^
vIF (t) = v^IF cos wIF t i IF 2 i RF
gD,0 = g D ,1 =
with v^IF << v^LO v^IF v^IF
Fig. 28.19. Circuit simulation to determine the coefficients gD,0 and gD,1 of unbalanced mixer.
Amplitudes îIF and îRF are obtained from the spectral display of current ID (t).
1388 28 Mixer
circuits. The RF output is short-circuited which renders the RF parallel resonant circuit
unnecessary. In this mode of operation it follows from (28.12) with v RF = 0 that:
gD,1
i IF = gD,0 v IF , i RF = − v IF
2
By inserting the small-signal amplitudes we obtain the defining equations for the coeffi-
cients:
îIF 2îRF
gD,0 = , gD,1 = −
v̂IF v̂IF
The small-signal amplitude v̂IF is given by the IF voltage source. The current ID (t) of
the diode is now determined by means of a time domain simulation. The small-signal
amplitudes îIF (component at fIF ) and îRF (component at fRF ) can be determined from
the spectral display of ID (t).
Conversion Gain
In most cases, mixers are used in matched systems; but, in this case, the internal resistance
Rg,IF of the IF voltage source and the RF load resistance RL,RF correspond to the char-
acteristic impedance ZW of the system: Rg,IF = RL,RF = ZW . The related power gain is
called the conversion gain GM and corresponds to the gain GT of an amplifier. With the
Y parameters
gD,1
Y11 = Y22 = gD,0 , Y12 = Y21 = −
2
of the mixer and the source and load conductivities
A B 1 1
Re Yg = , Re {YL } =
Rg,IF RL,RF
Equation (27.30) leads to:
2
gD,1 Rg,IF RL,RF
GM = 2
% &% & 1 2
1 + gD,0 Rg,IF 1 + gD,0 RL,RF − gD,1 Rg,IF RL,RF
4
28.3 Mixers with Diodes 1389
–4 –5
–6 – 10 G M (50)
Ideal – 15
–8 switch
– 8,9 – 20
– 10 – 25
4 /π ∼ 1.27
~ ~
~
~
1 1.2 1.4 1.6 1.8 2 gD,1 0.3 0.4 0.5 0.6 0.7 v^ LO
gD,0 V
Fig. 28.20. Maximum available power gain Fig. 28.21. Conversion gain GM (50)
(MAG ) dependent upon the ratio of the (Rg,IF = RL,RF = 50 ) and MAG of an
coefficients gD,1 and gD,0 of the small-signal unbalanced diode mixer with a Schottky diode
conductivity gD (t) of type BAS40
With small LO amplitudes, the conversion gain GM (50) is very small due to the strong
mismatch (ZW,M 50 ), but increases rapidly with an increasing LO amplitude and
reaches a wide maximum in the region v̂LO = 0.5 . . . 0.6 V. Above this maximum GM (50)
and MAG are almost identical since ZW,M tends toward50 in this region.
In most cases, mixers with diodes are operated without any specific matching circuits.
The LO amplitude is selected such that the maximum conversion gain GM (50) is almost
reached and thus a good compromise between conversion gain and required LO power
is achieved. The reasonable operating range for the BAS40 diode is shown in Fig. 28.21;
here, the conversion gain reaches GM (50) ≈ − 5 dB. This does not warrant any matching
efforts as the MAG is no more than approximately 1 dB higher.
The interaction of the spreading resistance and the capacitance of the diode produce ad-
ditional frequency-proportional losses which were not taken into account here. Depending
on the diode and frequency, this can reduce the conversion gain to GM (50) ≈ − 5 . . .− 8 dB.
These losses are detailed in [28.1].
or MAG ≈ − 8.9 dB. Therefore, with a square-wave LO voltage, the MAG is clearly
lower than with a sinusoidal LO voltage (MAG ≈ − 4 . . . − 5 dB). At first, this seems
surprising, but the reason is that MAG is only influenced by the ratio gD,1 /gD,0 . In an ideal
switch, this ratio amounts to 4/π ≈ 1.27 and is thus lower than in typical diode mixers
with a sinusoidal LO voltage (gD,1 /gD,0 ≈ 1.7 . . . 1.8) (see Fig. 28.20). For practical
applications, this result is advantageous as the LO voltage is produced by a high frequency
oscillator with almost sinusoidal output voltage.
28.3.2
Single Balanced Diode Mixers
Figure 28.22a shows the circuit of a mixer with two diodes that is known as a single
balanced diode mixer. The LO voltage VLO regularly switches the operating point of the
diodes alternately between the forward and the reverse region. The IF small-signal voltage
vIF is added to the voltage of diode D1 and subtracted from the voltage of diode D2 by
the 1:1:1 transformer TR1. This means that, with respect to small signals, the diodes are
driven in balanced mode, i.e. both diodes carry the same amount of small-signal current
iD , but of opposing directions. On the RF side, the small-signal current iD is decoupled
by the 1:1:1 transformer TR2 and fed to the RF filter.
In the single balanced diode mixer the IF and the RF circuits are decoupled from the LO
circuit. The small-signal current iD of the diodes that flows through the IF and the RF circuit
does not contain any portions of the LO frequency fLO or multiples thereof. On the other
hand, no currents of the IF or RF frequency flow through the LO circuit. Figure 28.23 shows
this on the basis of the operating modes of transformer TR1. Figure 28.23a shows that,
with a symmetrical load, the centre tap of the secondary side is current-free; therefore,
the IF voltage vIF generates no current in the LO circuit. Figure 28.23b shows that a
symmetrical modulation on the secondary side has no effect on the primary side since the
magnetic fluxes of the two secondary coils eliminate each other because their currents flow
in opposite directions. Thus, the LO voltage VLO generates no current in the IF circuit.
This means that both the IF and LO circuits are decoupled. The RF and LO circuits are
also decoupled in the same way. Consequently, the filters in the IF and RF circuits are only
required to suppress the frequency of the other circuit and must no longer suppress the LO
frequency. This reduces the requirements for the RF filter as it no longer faces the task of
separating the closely adjacent frequencies fRF and fLO . Similarly, the demands on the LO
filter are reduced as it only has to suppress the harmonics at multiples of fLO . If harmonics
1392 28 Mixer
D1 ID,LO + i D
TR1 TR2
i IF 1:1:1 vIF 1:1:1 i RF
2 iD 2 iD
vIF vRF
f IF f RF RL,RF
vIF
D2
I D,LO – i D
I LO
VLO
f LO
D1 I D,LO
2 I D,LO
D2 I D,LO
I LO
VLO
f LO
b LO circuit
Fig. 28.23. Voltages and currents of transformer TR1 in the case of symmetrical loads on the
secondary side
28.3 Mixers with Diodes 1393
at the LO terminal are acceptable, the LO filter can be omitted. This, however, changes the
waveforms of the LO voltage and LO current and thus the small-signal characteristics.
LO Circuit
Figure 28.22b shows the LO circuit of the single balanced diode mixer. The same cur-
rent ID,LO (t) as in the single-phase mixer flows through both diodes with the same LO
amplitude v̂LO . Since both diodes are connected in parallel, the current ILO (t) at the LO
terminal is twice as high as in an unbalanced diode mixer:
f =fLO
gD1 (t) iD
TR1 TR2
1:1:1 1:1:1
2 iD 2 iD
vIF 2 vIF 2 vRF vRF
f IF f RF RL,RF
iD gD2 (t)
vIF vRF
f IF f RF RL,RF
This results in the small-signal equivalent circuit shown in Fig. 28.24 that corresponds to a
small-signal equivalent circuit of an unbalanced diode mixer. The small-signal conductivity
gD (t) is twice as high as in an unbalanced diode mixer with the same LO amplitude:
This also makes the coefficients of the Fourier series of gD (t) twice as high:
The Y matrix of the single balanced diode mixer can now be determined according to
(28.12) and all other parameters can be calculated by using (28.13)–(28.16) for the unbal-
anced diode mixer.
The MAG has the same magnitude as that of the unbalanced diode mixer with the same
LO amplitude since here only the ratio of gD,1 and gD,0 are effective according to (28.16).
However, the related characteristic impedance is lower by a factor of 2 (see 28.15):
1
ZW,M = ZW,M (unbal) (28.22)
2
The conversion gain GM (50) in a 50 system is similar to that of an unbalanced diode
mixer. Figure 28.25 shows a comparison of different mixers with Schottky diodes of type
BAS40. In the single balanced diode mixer, the maximum value of GM (50) is slightly
higher than in the unbalanced diode mixer and is achieved at a lower LO amplitude.
–4
–5
–6
–7
~
~
Fig. 28.25. Conversion gain GM (50) of an unbalanced, a single balanced and a double balanced
diode mixer with Schottky diodes of type BAS40 in a 50 system
28.3 Mixers with Diodes 1395
28.3.3
Double Balanced Diode Mixer
Figure 28.26 shows the circuit of a mixer with four diodes known as a double balanced
diode mixer or ring modulator. It consists of the anti-parallel connection of two single
balanced diode mixers (D1 /D2 and D3 /D4 ) connected crosswise (see Fig. 28.26a). A
modification of the circuit diagram results in a diode ring as shown in Fig. 28.26b which
led to the name ring modulator. For reasons of clarity in the following explanation, we
shall use two single balanced diode mixers in our illustrations.
Owing to the anti-parallel connection of the two single balanced diode mixers, the
double balanced diode mixer uses both half waves of the LO voltage where diodes D1
and D2 are conductive in the positive phase and diodes D3 and D4 are conductive in
the negative phase. Figure 28.27 shows the two phases of the LO circuit. The crosswise
connection causes the polarity of the small-signal current 2iD on the RF side to change with
each half wave of the LO voltage. As a result, the double balanced diode mixer generally
operates as a multiplicative mixer with a bipolar square-wave signal.
Section 28.2.2 showed that the RF signal of a multiplicative mixer with bipolar square-
wave signal has no component at the IF frequency (see Fig. 28.13c on page 1378); corre-
spondingly, the IF signal has no part at the RF frequency. This means that the IF and RF
circuits in the double balanced diode mixer are decoupled. Since the LO circuit is already
decoupled from the IF and RF circuits due to the characteristics of the two single balanced
diode mixers, all three circuits are decoupled. Nevertheless, the number of filters required
cannot be reduced as both the IF and RF signals have components at multiples of the LO
frequency.
LO Circuit
Figure 28.27 shows the LO circuit of the double balance diode mixer for the two half
waves of the LO voltage. Currents ID1,LO (t), . . . , ID4,LO (t) of the diodes, as well as the
total current ILOD (t), are plotted in Fig. 28.28. The same current flows through each of the
diodes as in the unbalanced diode mixer with the same LO amplitude. The total current
ILOD (t) has no DC component due to the symmetry and contains portions at odd multiples
of the LO frequency:
ILOD (t) = îLOD,1 cos ωLO t + îLOD,3 cos 3ωLO t + îLOD,5 cos 5ωLO t + · · ·
The harmonics of ILOD (t) are short-circuited by the LO filter. The portion of the funda-
1396 28 Mixer
D1
TR1 TR2
i IF 1:1:1 D3 1:1:1 i RF
vIF D4 vRF
fIF f RF RL,RF
D2
I LO
VLO
f LO
a Circuit diagram showing two single balanced diode mixers in anti-parallel connection
TR1 TR2
i IF 1:1:1 D3 D1 1:1:1 i RF
vIF vRF
f IF f RF RL,RF
D2 D4
I LO
VLO
f LO
With the same LO amplitude, the amplitude îLOD,1 in the double balanced diode mixer is
higher than in an unbalanced diode mixer by a factor of four. This is true because a current
flows through a set of two diodes connected in parallel during both half waves of the LO
voltage, thus reducing the resistance RLO by a factor of four:
v̂LO 1
RLO = = RLO (unbal)
îLOD,1 4
D1 I
D1,LO + i D
TR1 TR2
1:1:1 D3 1:1:1
2 iD 2 iD
D4
D2
I LO
VLO > VF
f LO
a Positive LO voltage
I D3,LO – i D D1 I D4,LO + i D
TR1 TR2
1:1:1 D3 1:1:1
2 iD – 2 iD
D4
D2
V LO < – VF
f LO
b Negative LO voltage
Fig. 28.27. LO circuit of a double balanced diode mixer. VF represents the forward voltage of the
diodes
by inserting two change-over switches to represent the polarity reversal. The small-signal
conductivity gD (t) consists of the small-signal conductivities of the two single balanced
diode mixers that are phase-shifted by half the LO period. The plot of gD (t) can be seen
in Fig. 28.28.
Calculating the small-signal behavior is more complex than for an unbalanced or a
single balanced diode mixer due to the polarity reversal. First we determine the small-
signal currents iD (t) and iD,S (t) on the IF and RF side. From Fig. 28.29 it follows:
,
gD (t) [ vIF (t) − vRF (t) ] VLO ≥ 0
iD (t) =
gD (t) [ vIF (t) + vRF (t) ] VLO < 0
1398 28 Mixer
V LO 1/f LO
v^ LO
VF
t
– VF
I D1,LO
I D2,LO
I D,max
I D3,LO t
I D4,LO
I D,max
t
I LOD
2 I D,max
– 2 I D,max
gD 1/ 2 f LO
gD,max
t
gD,S 1/f LO
gD,max
t
– gD,max
Fig. 28.28. Ring mixer with voltage VLO (t) of the LO circuit, currents of diodes D1 . . . D4 , total
current ILOD of the diodes, small-signal parameters gD (t) and gD,S (t). VF represents the forward
voltage of the diodes
iD (t) = gD (t) [ vIF (t) − vRF (t) ] VLO ≥ 0
iD,S (t) =
2 i RF
vIF f IF f RF vRF RL,RF
1
2
1: VLO <
–0
2: VLO < 0
iD,S (t) : fLO + fIF , fLO − fIF , 3fLO ± fIF , 5fLO ± fIF , . . .
3 45 6 3 45 6
fRF fRF,im
The small-signal current iD (t) on the IF side contains no component at the RF frequency
and the small-signal current iD,S (t) on the RF side contains no component at the IF
frequency, i.e. the IF and RF circuits are decoupled as already mentioned.
From the small-signal currents iD (t) and iD,S (t), the respective portions of the IF and
RF currents can be determined by extraction:
f =fIF f =fRF
All other components are short-circuited by the filters. The calculation, which will not be
explained here in detail, is similar to that for the unbalanced diode mixer and also leads to
a Y matrix:
⎡ gD,1 ⎤
gD,0 −
i IF ⎢ 2 ⎥ v IF
= ⎣ gD,1 ⎦ v
i RF − gD,0 RF
2
Coefficient gD,0 corresponds to the DC component in gD (t) and coefficient gD,1 corre-
sponds to the fundamental component in gD,S (t). Using (28.24), this leads to the relation-
ship with the coefficients of an unbalanced diode mixer of the same LO amplitude:
gD,0 = 4 gD,0 (unbal) , gD,1 = 4 gD,1 (unbal) (28.25)
This allows the Y matrix of the double balanced diode mixer to be determined and all other
parameters to be calculated using (28.13)–(28.16) for the unbalanced diode mixer.
The MAG has the same magnitude as in the unbalanced diode mixer or the single
balanced diode mixer of the same LO amplitude since only the ratio of gD,1 and gD,0
becomes effective according to (28.16). However, the related characteristic impedance is
four times lower (see (28.15)):
1
ZW,M = ZW,M (unbal) (28.26)
4
The conversion gain GM (50) in a 50 system has a similar characteristic as that of an
unbalanced or a single balanced diode mixer. Figure 28.25 on page 1394 shows a compar-
ison of mixers with Schottky diodes of type BAS40. In a double balanced diode mixer, the
maximum value for GM (50) is always somewhat higher than in an unbalanced or a single
balanced diode mixer and is reached at a lower LO amplitude.
Broadband Operation
Double balanced diode mixers are often operated in the mode shown in Fig. 28.30. The
IF and RF filters are then separated from the mixer by amplifiers. The unwanted portions
of the small-signal currents iD (t) at the input and iD,S (t) at the output of the mixer are no
longer short-circuited by the filters but have an influence in accordance with the impedances
ZIF (s) and ZRF (s) of the amplifiers. This is especially the case with the image frequency
component at the frequency fRF,im , which has the same size as the component at the
28.3 Mixers with Diodes 1401
Double balanced
IF IF diode mixer RF RF
filter amplifier without filter amplifier filter
fLO
RF frequency fRF . Like all other interfering components, these are amplified by the RF
amplifier shown in Fig. 28.30 and are suppressed only by the subsequent RF filter.
The small-signal response of the mixer depends on the values of the impedances ZIF (s)
and ZRF (s) at all frequencies involved and can only be determined using numeric meth-
ods or circuit simulation. This is called the broadband operation of the mixer. There are
qualitatively similar relationships as in narrow-band operation as discussed above, but the
quantitative changes for the broadband operation are:
– The conversion gain GM (50) and the MAG are approximately 1 . . . 2 dB lower.
– The characteristic impedances ZW,M (IF) and ZW,M (RF) for power matching at the IF
and/or RF terminal are higher than in narrow-band operation and are no longer equal.
Typical values are:
ZW,M (IF) ≈ (2 . . . 3) ZW,M , ZW,M (RF) ≈ (1.2 . . . 1.5) ZW,M
Here, ZW,M is the characteristic impedance in narrow-band operation (see (28.15)).
This means that the characteristic impedance ZW,M (IF) at the IF terminal is twice as
large as the characteristic impedance ZW,M (RF) at the RF terminal.
No separate matching circuits are used in broadband operation. The conversion gain
GM (50) is optimized by a suitable choice of the LO amplitude and the transformation
ratios of the transformers.
28.3.4
Diode Mixers in Practical Use
In practice, double balanced diode mixers are used predominantly. They are available as
discrete components and, in addition to the four diodes, also contain two transformers.
Figure 28.31 shows a common design with a total of four terminals: LO, IF, RF and
common ground. In some models the ground terminals are not tied to a common ground;
the double balanced diode mixer then has six terminals.
Since the terminals are decoupled and the circuit is symmetrical, the connections of the
double balanced diode mixer can be interchanged. Of course, this alters the voltages and
currents of the diodes, but the small-signal equivalent circuit and the operating parameters
(RLO , ZW,M , MAG , etc.) remain the same if the transformers are symmetrical. In double
balanced diode mixers used in practical applications, the LO and IF terminals are often
interchanged (see Fig. 28.31). Then transformer TR1 is no longer supplied with the IF
frequency but with the significantly higher LO frequency, thus reducing the transformer
1402 28 Mixer
TR1 TR2
D3 D1
IF RF
better: LO
D2 D4
Exchange
terminals
Fig. 28.31. Double balanced
LO diode mixer as a discrete
better: IF component
TR1 i IF TR2
D3 D1
RF
i IF
V LO > 0
D2 D4
i IF
IF
a Positive LO voltage
TR1 TR2
D3 D1
RF
– i IF
V LO < 0
D2 D4
i IF
i IF
size. In this case, a very low IF frequency can be used. Figure 28.32 shows the related
current distribution. In the LO circuit, current flows alternately through diodes D1 /D4
(VLO > 0) and D2 /D3 (VLO < 0) which causes the IF current to flow alternately through
both secondary windings of transformer TR2.
Discrete double balanced diode mixers are always designed for a certain LO power.
For this power, the conversion gain GM (50) in a 50 system reaches its maximum and
the terminals are optimally matched to 50 . This is achieved by using suitable diodes and
matching the transformation ratios of the transformers. The double balanced diode mixer
is then no longer symmetrical, i.e. the terminals must be used as designated by the manu-
facturer. Instead of the conversion gain, the data sheet specifies the conversion loss in dB:
28.3 Mixers with Diodes 1403
PLO
PLO [dBm] = 10 log
1 mW
1 3
RF
0O
0O
IF f RF
0O
O
2 180 4
f IF LO
28.4
Mixers with Transistors
Multiplicative mixers with transistors are used almost exclusively in integrated circuits. In
these mixers the input signal is converted to a current by a voltage-to-current converter and
connected to the output by means of one or two differential amplifiers operated as change-
over switches. In the following section, the two commonly used circuits are described. The
first is the single balanced mixer and the second is the double balanced mixer which is also
named Gilbert mixer after its inventor B. Gilbert. Both circuits can be made up of bipolar
transistors or MOSFETs. The following explanations are based on bipolar transistors.
28.4.1
Single Balanced Mixer
Figure 28.34 shows the circuit diagram of a single balanced mixer operated as an up-
conversion mixer. It consists of a common-emitter circuit with current feedback (T3 ,RE )
that operates as a voltage-to-current converter (V/I converter) and a differential amplifier
(T1 ,T2 ) that acts as a switch and supplies the output current alternately to the RF output
and the supply voltage. The small-signal IF voltage vIF , together with a DC voltage V0 ,
is supplied to the input to determine the operating point. The differential amplifier is
switched by the LO voltage VLO , which, under ideal conditions, is rectangular. From the
mixed products in current IC2 the small-signal RF current iRF is separated by an RF filter
and supplied to the RF load resistance RL,RF via a coupling capacitance Cc .
The functional principle of the single balanced mixer is shown in Fig. 28.35. One
can see that in terms of the transfer characteristic, the change-over switch functions as
an On/Off switch only. In this form, the single balanced mixer works as a multiplicative
mixer with a unipolar square-wave signal as shown in a comparison with Fig. 28.12 on
page 1376.
Vb Vb Vb
f RF
i RF
I C1 I C2 Cc vRF
RL,RF
T1 T2
VLO
I C3
T3
V0 + vZF
RE
Fig. 28.34. Single balanced mixer
with transistors
28.4 Mixers with Transistors 1405
iC2 i RF
vIF i C3 = gm vIF vRF
f RF RL,RF
f LO
Fig. 28.35. Functional principle (= small-signal equivalent circuit) of a single balanced mixer with
transistors
I I C3
T2 conductive T2 conductive
T1 non- Switching T1 non-
conductive region I C1 conductive
I C2
– 5 VT 5 VT
Fig. 28.36. Current
– 200 –150 –100 – 50 0 50 100 150 200 VLO
characteristics of the
mV differential amplifier
1406 28 Mixer
Switching 0.5
characteristic
– 5 VT 5 VT
Fig. 28.37. Single balanced mixer as a multiplicative mixer taking into consideration the switching
characteristic of the differential amplifier
sLO (t)
This shows that the single-balanced mixer operates as a multiplicative mixer where the
IF signal sIF (t) is multiplied by the LO signal sLO (t). In addition, a DC component occurs
that corresponds to the bias current IC3,A , which is also multiplied by sLO (t). The LO
signal sLO (t) results from the LO voltage VLO (t) when taking the switching characteristic
of the differential amplifier into consideration. Figure 28.37 illustrates this.
Rectangular LO Voltage
First we look at the operation with a bipolar rectangular LO voltage of the amplitude v̂LO .
The LO signal sLO (t) is also rectangular with the following values:
VLO =±v̂LO
1 VLO tanh(−x)=− tanh x 1 v̂LO
sLO = 1 − tanh = 1 ∓ tanh
2 2VT 2 2VT
Figure 28.38a shows the plot of VLO (t) and sLO (t) for different amplitudes. For v̂LO >
5VT = 130 mV the signal sLO (t) is approximately a unipolar square-wave signal with the
values 0 and 1. In this case, the mixer can be considered an ideal switch.
For further calculations, the signal sLO (t) is expanded into a Fourier series:
Besides the DC component c0 , the series contains only cosine components of the LO
frequency fLO and uneven multiples thereof since sLO (t) is an even function of time
according to Fig. 28.38 (sLO (t) = sLO (−t)) with a pulse duty ratio of 50 %. Signal sLO (t)
28.4 Mixers with Transistors 1407
VLO VLO
200 200
mV mV
100 100
50 50
– 50 v^LO t
– 50
t
–100 –100
– 200 – 200
sLO
´ ´
sLO
1 1
1 v^LO
tanh
2 2 VT
0.5 0.5
0 0
t t
Fig. 28.38. LO voltage VLO (t) and LO signal sLO (t) for the amplitudes
v̂LO = 50 mV / 100 mV / 200 mV
can be viewed as the sum of a DC component c0 = 1/2 and a bipolar square-wave signal
of the amplitude:
1 v̂LO
tanh
2 2VT
ci Rectangular
Sinusoidal 2/ ∼ 0.64
0.7
c1
0.6
0.5
0.4
0.3 2 / (3 ) ∼ 0.21
0.2 c3
0.1
5 VT
0 Fig. 28.39. Coefficients c1 and |c3 | of the Fourier
0 50 100 150 200 v^LO
Using the series expansion for a bipolar square-wave signal in (28.1) leads to the coeffi-
cients:
1 2 v̂LO 2 v̂LO
c0 = , c1 = tanh , c3 = − tanh , ··· (28.33)
2 π 2VT 3π 2VT
In Fig. 28.39 the coefficients c1 and |c3 | are plotted over the amplitude v̂LO , where, for
v̂LO → ∞, they assume the state of coefficients of a unipolar square-wave signal. Essen-
tially this is the case at v̂LO > 5VT .
Sinusoidal LO Voltage
Generating a rectangular LO voltage becomes more and more difficult with an increasing
LO frequency. For this reason, at high frequencies the nearly sinusoidal output voltage of a
high frequency oscillator is used. Figure 28.38b shows the curves of VLO (t) and sLO (t) for
this case. Here, too, with an increasing amplitude the LO signal sLO (t) enters the state of a
unipolar square-wave signal. With the exception of the DC component c0 the coefficients
of the Fourier series are smaller than for a rectangular LO voltage of the same amplitude.
Fig. 28.39 compares the coefficients c1 und |c3 |.
Small-Signal Response
Now the small-signal dynamic response can be calculated by inserting a sinusoidal IF
voltage
vIF (t) = v̂IF cos ωIF t
and the Fourier series for sLO (t) from (28.32) into (28.31):
* +
c1
iRF (t) = IC2 (t)
= gm v̂IF cos ωRF t
f =fRF =fLO +fIF 2
and the RF voltage:
c1
vRF (t) = − RL,RF iRF (t) = − gm RL,RF v̂IF cos ωRF t
2
28.4 Mixers with Transistors 1409
I C2 c1 I C3,A
c0 I C3,A
c1 gmv^ IF c3 I C3,A
c0 gmv^IF c1 gm v^IF
2 RF filter
2 c3 gmv^IF c3 gmv^IF
2 2
f IF f LO – f IF f LO f LO + f IF 2 f LO 3 f LO – f IF 3 f LO 3 f LO + f IF f
Fig. 28.40. Spectrum of current IC2 (t) in the case of a sinusoidal IF voltage
This assumes that the output resistance of transistor T2 can be neglected. The voltage
phasors are:
c1
v RF = − gm RL,RF v IF (28.34)
2
Bandwidth
We have calculated the mixer voltage gain AM for the static condition only, i.e. without
accounting for the capacitances of the transistors. Therefore, strictly speaking, it applies
to low frequencies only. However, the bandwidth of the single balanced mixer is usually
very wide for three reasons:
– The common-emitter circuit with current feedback, together with the transistors of the
differential amplifier, forms a cascode circuit and therefore reaches a cutoff frequency
1410 28 Mixer
that is between the transconductance cutoff frequency fY 21e and the transit frequency
fT of transistor T3 .
– With respect to the small-signal current, transistor T2 operates in common-base config-
uration with the α cutoff frequency fα ≈ fT .
– The output capacitance of transistor T2 can be regarded as an integral part of the RF
filter capacitance and thus has no interfering effect.
Therefore, even for higher frequencies, the mixer voltage gain can be evaluated by means
of the static voltage gain.
Impedance Matching
For high frequencies, the single balanced mixer must be matched at all sides to the charac-
teristic impedance ZW of the connecting lines in order to avoid undesirable reflections and
impedance transformations. This is possible with the same methods used in amplifiers:
– Circuits for impedance transformation described in Sect. 28.3.1.
– Methods for matching integrated amplifiers described in Sect. 27.1.1 (see Fig. 27.2 on
page 1323).
Figure 28.41 shows a typical example:
– The common-emitter circuit at the input is replaced by a common-base circuit with bias
current I0 and input impedance:
1 VT
= (28.36)
gm3 I0
I0 ≈ 520 mA provides an impedance matching to ZW = 50 . If a current feedback is
required to improve the linearity, one may select a higher bias current and introduce an
Vb Vb Vb Vb
C2
vRF L RV
C1 ´
vRF
RL,RF = ZW
T1 T2
VLO RLO
= 2 ZW
I C3,A ∼ I 0
T3
RE
V0
vIF
I0
r i = ZW
Vb Vb Vb Vb Vb Vb Vb
C2
v RF L RV v RF L C RV vRF RP
C1 ´
vRF RL,RF ro
=ZW
T2 T2
5 The output resistance of the transistor is defined here as the reciprocal of the real part of the output
admittance: ro = 1/Re {Yo }. For low frequencies, ro = rCE = VA /IC,A (see (2.13)); for high
frequencies the output resistance is clearly lower.
1412 28 Mixer
Conversion Gain
Now we can determine the conversion gain GM of the single balanced mixer. It corresponds
to the gain GT of the amplifier and is calculated with (27.29):
2
ri 4Rg RL
GM = GT = A2
Rg + r i (ro + RL )2
Here, ri = input resistance, A = open-circuit gain, ro = output resistance of the single
balanced mixer. Additionally, Rg is the internal resistance of the signal source and RL is
the load resistance. We assume that the mixer is matched at both ends with ri = Rg = ZW
and ro = RL ; then:
A2 ZW
GM = (28.40)
4 ro
Since the transformation is loss-free we can determine the open-circuit gain and the output
resistance by means of the transformed diagram in Fig. 28.42b where RP and RL,RF absorb
the same power. As RV represents the loss resistances by definition from Fig. 28.42b it
follows:
ro = RV (28.41)
In an open circuit, i.e. without RP , RV acts as the load resistance. The open-circuit voltage
gain A can thus be calculated with (28.35) by replacing RL,RF by RV :
1
A = − c1 gm RV (28.42)
2
Here, gm is the transconductance of the voltage-to-current converter:
,
gm3 Without current feedback
gm = (28.43)
gm3 / (1 + gm3 RE ) With current feedback
Inserting (28.41) and (28.42) into (28.40) leads to the conversion gain of a single balanced
mixer with impedance matching at both ends:
1 2 2
GM = c g Z W RV (28.44)
16 1 m
This is proportional to the loss resistance RV . At low frequencies, RV is very high and
it may be necessary to reduce it by an additional parallel resistance in order to prevent
the voltage of the parallel resonant circuit from becoming too high. RV decreases with
an increasing frequency. A benefit is that RV only has a linear effect on the conversion
gain (= power gain). Therefore, with a decrease√ in loss resistance in a matched circuit, the
voltage gain only decreases in proportion to RV and not in proportion to RV as is the
case for the open-circuit voltage gain in (28.42).
At the output we have RV = RP . The required capacitance ratio of the capacitive
voltage divider results from (28.39):
C2 2 C2 RV
RV = ZW 1 + ⇒ = −1 (28.45)
C1 C1 ZW
28.4 Mixers with Transistors 1413
Example: Let us look at a matched single balanced mixer with common-base circuit as
shown in Fig. 28.41. Here, according to (28.38) the relation gm = 1/ZW applies. For
a fully modulated differential amplifier (c1 = 2/π ) and ZW = 50 , substitution into
(28.44) leads to:
gm =1/ZW 1
2 RV 1 2 2 RV RV
GM = c1 = =
16 ZW 16 π 50 1974
This means that for a conversion gain of GM = 4 (6 dB), a loss resistance of RV ≈ 7.9 k
is required. The capacitance ratio of the capacitive voltage divider is calculated with (28.45)
which yields C2 /C1 ≈ 11.6. This represents the limit of what can be achieved in practical
applications. The reason for this is the fundamental relation of gm = 1/ZW , which applies
to matched single balanced mixers with common-base circuit. The transconductance gm ,
which has a quadratic effect on the conversion gain, is thus limited to a comparably low
value.
Better results are achievable with a single balanced mixer with common-emitter circuit
as shown in Fig. 28.43. Here, any value can be selected for the transconductance gm and
the relatively high input resistance can be matched by a terminating resistance R1 ≈ ZW
independent of gm . The current feedback is omitted and I0 = 2 mA is selected so that
gm = gm3 = I0 /VT ≈ 77 mS. For β3 = 100, the input resistance of transistor T3
is rBE 3 = β3 /gm3 ≈ 1.3 k; for R1 = 52 we have ri = (R1 || rBE ) = 50 . By
substituting into (28.44) and using c1 = 2/π we obtain:
1 2 2 RV
GM = (77 mS)2 · 50 · RV =
16 π 133
We assume that the loss resistance RV is caused by the output resistance of transistor T2 .
Since the bias current is higher than in the single balanced mixer with common-base circuit,
we can assume a reduced loss resistance of RV = 7.9 k · (520 mA/2 mA) ≈ 2050 .
Vb Vb Vb Vb
C2
L RV
C1 ´
vRF
RL,RF = ZW
T1 T2
VLO RLO
= 2 ZW
I C3,A ≈ I 0
T3
vIF
R1 ≈ ZW
I0
ri = ZW
– Vb
Fig. 28.43. Single balanced mixer with common-emitter circuit and impedance matching
1414 28 Mixer
Thus the conversion gain achieved is GM ≈ 15 (12 dB). From (28.45) for the capacitive
voltage divider it follows C2 /C1 ≈ 5.4.
In this example the conversion gain of the single balanced mixer with common-emitter
circuit is higher than that of the single balanced mixer with common-base circuit by a
factor of 4 (6 dB). A disadvantage is the increase in the noise figure due to the terminating
resistance R1 . Therefore, this design is not used for front-end down-conversion mixers in
receivers.
Practical Design
Figure 28.44 shows the design of a single balanced mixer for practical applications with
all components required for operating point setting and matching to ZW = 50 . The
voltages V0 and V1 for setting the operating point are generated with resistors R1 , R2 and
R3 ; C3 and C6 serve as blocking capacitors. Resistors R4 and R5 supply the voltage V1
to the inputs of the differential amplifier and at the same time serve as LO terminating
resistors R4 = R5 = 50 . The series connection of R4 and R5 corresponds to resistance
RLO = 2ZW in Figs. 28.41 and 28.43. The LO voltage is provided via the coupling
capacitors C4 and C5 . The bias current IC3,A ≈ 520 mA required for impedance matching
to 50 is set with resistor R6 . No current feedback is used. The IF voltage is supplied
via the coupling capacitor C7 . The circuitry at the output with capacitive voltage divider
C1 , C2 and the resonant impedance RV is adopted from Fig. 28.41. However, in this case
capacitor C2 is not connected to the supply voltage Vb (small-signal ground) but to ground.
With very high voltage divider factors this causes the RF output current which flows almost
entirely through C2 not to flow onto the supply voltage line.
The symmetric LO voltage can be generated by an oscillator with differential output.
It often occurs that only an asymmetric LO voltage is available, then the supply methods
shown in Fig. 28.45 can be used. In Fig. 28.45a the LO voltage is supplied asymmetrically to
one of the two LO inputs while the other input is short-circuited (C5 to ground) with respect
Vb Vb Vb Vb
2
C2
R1 vRF L RV RV = RL,RF 1+
C1
C3 V1 R4 = R5 = 50 Ω C1
R4 R5 v ´RF
C4 C2 RL ,RF = 50 Ω
T1 T2
VLO R2
C5
I C3,A ≈ 520 m A
C6 V0
T3
C7
R3
vIF
R6
ri = 50 Ω
Vb Vb
R1 R1
C3 V 1 R4 = R5 = 50Ω C3 V1 R4 = R5 = 25Ω
R4 R5 R4 R5
C4 C4
1:1
V1 + VLO I1 V1 + VLO /2
VLO VLO
C5 V1 C5 I2 = I1 V1 – VLO /2
to small signals. But the asymmetry is disadvantageous to the distortion characteristic of the
mixer; therefore, the symmetric supply by means of a balanced-to-unbalanced transformer
(balun) is often used in practice. The balun requires I1 = I2 so that there is pure differential
modulation at the LO inputs.
In integrated circuits, balancing of an asymmetric LO voltage is achieved by means of
a differential amplifier with an asymmetric input and a symmetric output. At the same time,
this differential amplifier serves as an amplifier for the LO signal and is directly coupled to
the differential amplifier of the single balanced mixer. Figure 28.46 shows a typical design.
The inductance L and the capacitances C1 , . . . , C4 are generally not integrated, rather they
are connected externally. Matching at the LO side is done with resistance RB ≈ ZW . The
differential amplifier T4 , T5 is driven into saturation and generates an almost rectangular
LO voltage VLO of the amplitude I1 RC > 5VT from the sinusoidal voltage VLO . The
Vb Vb Vb Vb
R1 RV L
C1
V1
RC RC ´
vRF
C2
T1 T2
VLO
C4 RB
T4 T5 T3
´
VLO C3
RB
ZW vIF
I1 I0
RB ∼ ZW
ZW
– Vb – Vb
sLO (t)
TR2
I1 i RF
Vb
vRF
f RF RL,RF
V1 R V´
I C1 I C2
TR1 T1 T2
VLO
T3
V0 + vIF
RE
In this case, the LO signal sLO (t) has no DC component and twice the amplitude as that of a
single balanced mixer without output transformer. For the coefficients of the Fourier series
of sLO (t) this means that coefficient c0 becomes zero, while all other coefficients increase
by a factor of 2. Consequently, the mixer voltage gain AM rises by a factor of 2 as well
since, according to (28.35), it is proportional to the coefficient c1 . According to (28.44),
the conversion gain GM in a matched circuit is proportional to the square of coefficient c1
and must thus increase by a factor of 4. In practice this is usually not the case since the
output resistance of transistor T1 becomes effective, too, and causes a reduction in the loss
resistance RV . In extreme cases the loss resistance is caused solely by the transistors, and
then the conversion gain only rises by a factor of 2.
Transformer TR2 is also used for output matching. To this end, the transformation ratio
n is selected such that the loss resistance RV
= RV /n2 , as related to the secondary side,
is equal to the load resistance RL,RF .
28.4.2
Double Balanced Mixer (Gilbert Mixer)
Figure 28.48 shows the circuit of a double balanced mixer, which is also named Gilbert
mixer after its inventor B. Gilbert. It is the mixer of choice in integrated circuits since
it can be operated without any filters connected directly to the mixer. Suppression of
unwanted components in the output voltages is then achieved in subsequent components.
The following description is based on an up-conversion mixer.
A comparison of the double balanced mixer in Fig. 28.48 with the single balanced
mixer in Fig. 28.34 on page 1404 shows that the double balanced mixer consists of two
single balanced mixers whose outputs are connected: T1 , T2 , T5 and T3 , T4 and T6 . The
common-emitter circuits with current feedback (T5 and T6 ) that operate as voltage-to-
current converters (V/I converters) are combined to a differential amplifier with current
feedback and are inversely modulated by the IF voltage vIF . This makes the connection
1418 28 Mixer
Vb Vb Vb Vb
RC RC
I1 I2
Vo1 Vo2
I C1 I C2 I C3 I C4
T1 T2 T3 T4
VLO
I C5 I C6
T5 T6
RE RE
vIF
2 I0
point between the two feedback resistances RE a virtual ground (small-signal ground).
The bias currents are set by means of a current source 2I0 : IC5,A = IC6,A = I0 . Ide-
ally, the LO voltage VLO is rectangular and is supplied to the differential amplifiers
(T1 , T2 and T3 , T4 ) which operate as switches with opposite polarity. This portion of
the circuit is called the Gilbert cell. Two collector resistances RC are used instead of
the RF filter; therefore, the output voltages are not filtered at this point, and in addition
to the desired RF components, they still contain all other components that are generated
in the conversion process. Common-collector circuits at the output are usually used as
impedance converters. Then the RF filters follow; in most cases dielectric or SAW filters
are used.
The doubled balanced mixer in Fig. 28.48 corresponds to a differential amplifier with
current feedback and collector resistances in which the polarity between the IF inputs and
the outputs can be reversed. Similar to a differential amplifier, the double balanced mixer
allows asymmetric operation by supplying a constant potential to one of the two IF inputs
or using only one output or a combination of both. Also, the LO input can be operated
asymmetrically, although this has a negative effect on the distortion characteristic. For this
reason, asymmetrical IF or LO voltages are usually converted into symmetric voltages by
a balun or an asymmetric differential amplifier before they reach the mixer. This method
is shown in Figs. 28.45b and 28.46 for the example of a single balanced mixer with
asymmetric LO voltages. In the case of an asymmetric output, the collector resistance
usually remains at the unused output.
Figure 28.49 shows the functional principle of the double balanced mixer. It can be
seen that both single balanced mixers are inversely modulated with half the IF voltage each.
28.4 Mixers with Transistors 1419
– vIF /2 RC vo2
i C4
iC6 = – gm vIF /2
i2
i C3
V/I converter Switch
( T6 , RE) ( T3 , T4)
Fig. 28.49. Functional principle (= small-signal equivalent circuit) of a double balanced mixer
with transistors
The double balanced mixer operates as a multiplicative mixer with a bipolar square-wave
signal as a comparison with Fig. 28.12 on page 1376 shows.
Switching – 5 VT 5 VT
character-
istics – 200 0 100 200
–100 VLO
mV
– 0.5
VLO ( t )
Fig. 28.50. Double balanced mixer as a multiplicative mixer taking into account the switching
characteristics
It can be seen that in the double balanced mixer only the small-signal components are
switched over while the bias current I0 remains constant. This is an essential advantage as
compared to the single balanced mixer in which the bias current is switched over as well
(see (28.31) on page 1406). Due to this fact, we can restrict the following considerations
to the small-signal currents:
1 VLO (t)
i1 (t) = gm vIF (t) tanh , i2 (t) = − i1 (t) (28.52)
3 45 6 2 2VT
sIF (t) 3
45 6
sLO (t)
We can see that the double balanced mixer operates as a multiplicative mixer: the IF signal
sIF (t) is multiplied by the LO signal sLO (t). The LO signal sLO (t) is derived from the
voltage VLO (t) when taking the switching characteristics into consideration. Figure 28.50
illustrates the characteristic behavior.
The relationship between a rectangular or sinusoidal LO voltage VLO (t) and the LO
signal sLO (t) has already been explained for the single balanced mixer. In the double
balanced mixer sLO (t) contains no DC component since the plot of the switching char-
acteristics is symmetrical to the original.6 Thus, the coefficient c0 of the Fourier series
be a part of sLO (t) but is treated separately. The coefficients are then higher by a factor of 2 but
this is compensated for in the course of the calculation by the separate factor 1/2. In this context
the definition of the transconductance gm must be thoroughly examined in order to determine
whether a symmetric or asymmetric output signal is used.
28.4 Mixers with Transistors 1421
2 v̂LO 2 v̂LO
c1 = tanh , c3 = − tanh , ···
π 2VT 3π 2VT
Figure 28.39 on page 1407 presents the coefficients c1 and |c3 | for a rectangular and a
sinusoidal LO voltage.
Small-Signal Response
We can now calculate the small-signal output voltages
vo1 (t) = − RC i1 (t) , vo2 (t) = − RC i2 (t) = − vo1 (t)
for a sinusoidal IF voltage:
vIF (t) = v̂IF cos ωIF t
Inserting the small-signal currents from (28.52) and the Fourier series expansion from
(28.53) leads to
vo1 (t) = − gm RC v̂IF cos ωIF t [ c1 cos ωLO t + c3 cos 3ωLO t + · · · ]
c1
= − gm RC v̂IF [ cos(ωLO + ωIF )t + cos(ωLO − ωIF )t ]
2
c3
− gm RC v̂IF [ cos(3ωLO + ωIF )t + cos(3ωLO − ωIF )t ]
2
− ···
with the RF component:
c1
vRF (t) = vo1 (t)
= − gm RC v̂IF cos ωRF t (28.54)
f =fRF =fLO +fIF 2
Figure 28.51 shows the related spectrum which corresponds to the spectrum of a multi-
plicative mixer with bipolar square-wave signal in Fig. 28.13c on page 1378. It contains
no undesirable components at the LO frequency fLO and at multiples thereof as they are
caused in the single balanced mixer by switching the bias current. This can be seen in a
comparison with Fig. 28.40 on page 1409.
The maximum value of the output voltage vo1 (t) is
1
vo1,max = max|vo1 (t)| = gm RC v̂IF
2
and with an ideal switch (c1 = 2/π ), it exceeds the amplitude of the RF component in
(28.54) by a factor of 1/c1 = π/2 ≈ 1.57 (4 dB) only. This allows the entire output
v o1 c1gm RCv^IF
2 c3 gm RCv^IF
RF filter
2
f IF f LO – f IF f LO f LO + f IF 2 f LO 3 f LO – f IF 3 f LO 3 f LO + f IF f
Fig. 28.51. Spectrum of the output voltage vo1 (t) for a sinusoidal IF voltage
1422 28 Mixer
voltage to be further processed without major limitations to the dynamic range and the RF
component to be filtered out at a later stage. The demands on the RF filter are lower than
in the single balanced mixer since there is no component at the LO frequency as can be
seen by comparing Fig. 28.51 with Fig. 28.40 on page 1409.
v RF c1 (28.48) c1 gm5 RC
AM = = − g m RC = − (28.56)
v IF 2 2 1 + gm5 RE
Only the RF component of the output voltage vo1 (t) is taken into account when determining
the mixer voltage gain, which thus corresponds to the differential gain AD of a differential
amplifier. In most cases, however, the differential output voltage vo (t) = vo1 (t) − vo2 (t)
is used. The mixer voltage gain is then twice as high:
In the following, we refer to AM as the single-ended mixer voltage gain and to AM,diff as
the differential mixer voltage gain.
The single-ended mixer voltage gain AM of the double balanced mixer corresponds to
the mixer voltage gain of the single balanced mixer in (28.35) on page 1409 if we assume
that RC = RL,RF , i.e. equal load resistances for the RF component. Typical values are in
the range of |AM | ≈ 2 . . . 10 (6 . . . 20 dB).
Bandwidth
With regard to bandwidth, basically the same considerations apply to the double balanced
mixer as to the single balanced mixer. However, in a double balanced mixer with col-
lector resistances and subsequent impedance converters, it is not possible to compensate
for the output capacitances of transistors T1 , . . . , T4 because, together with the collector
resistances, they form lowpass filters and thus limit the bandwidth of the output. This is
particularly apparent in up-conversion mixers with output signals of a high frequency fRF .
This negative effect is much less in down-conversion mixers whose output frequency fIF
is significantly lower. To solve this, one can increase the transconductance gm and reduce
the resistances RC accordingly; however, this is at the cost of the current consumption.
As an alternative, inductances can be used to compensate the capacitances; Fig. 28.52
shows two ways of doing this. In both cases we obtain parallel resonant circuits at the
outputs which are tuned to the RF frequency in up-conversion mixers so that their function
corresponds to that of the RF filter in a single balanced mixer. This method is particularly
interesting in integrated circuits if the required inductances are so small that they can be
integrated or implemented by means of bond wires; otherwise external inductances must
be used.
28.4 Mixers with Transistors 1423
Vb Vb Vb Vb Vb Vb
RC L1 L2 RC RC RC
LQ
T1 T2 T3 T4 T1 T2 T3 T4
Integrated Circuit
Input Double Output
Filter amplifier balanced mixer amplifier Filter
ZW LO ZW
amplifier
f LO
Output amplifier
Double balanced mixer
RC
T3 T4
T6
RE
RE
T1 T2
T5
RC
Input amplifier
LO amplifier (limiter circuit)
Fig. 28.54. Example of a double balanced mixer with amplifiers in an integrated circuit
Impedance Matching
In order to allow universal use of integrated double balanced mixers, they are used with-
out input and output amplifiers. In this case, the input and output of the mixer must be
matched to the characteristic impedance. To do this, the same methods are used as in sin-
gle balanced mixers. Figure 28.55 shows some examples of impedance matching at the
28.4 Mixers with Transistors 1425
T5 T6 T5 T6 T5 T6
R1 RE RE RE RE RE RE
1:1
I0 I0 I0 I0
2 I0
– Vb – Vb – Vb – Vb
Fig. 28.55. Examples of impedance matching at the input of a double balanced mixer
T5 T6 T5 T6
r i < ZW r i < ZW
C1 C2 C1 C2
I0 I0 I0 I0
L=L 1+ L2
L1 – Vb – Vb L2 – Vb – Vb
ZW ZW
2 ZW
Fig. 28.56. Impedance matching at the input of a double balanced mixer by means of matching
networks
input side. As in the single balanced mixer, common-base circuits are often used instead of
common-emitter circuits. Where an asymmetric input is required, a balance-to-unbalance
transformer (balun) can be added. The matching networks described in Sect. 28.3.1 can
also be used as an alternative. In the case of a symmetric input, either two asymmetric
or one symmetric matching network may be applied. Fig. 28.56 illustrates the use of the
matching network from Fig. 28.22b on page 1392 for a step-up transformer from ri < ZW
to ZW .
The matching networks from Sect. 28.3.1 are also used at the output. Figure 28.57
shows the use of the matching network from Fig. 28.23b on page 1392 for a step-down
transformation. In down- or up-conversion mixers with low RF frequencies, the output
impedance of transistors T1 , . . . , T4 is very high at the output frequency. In this case,
the collector resistances are required to limit the voltage amplitudes at the collectors;
at the same time they allow a feasible transformation ratio RC /ZW . This is shown in
Fig. 28.57a in connection with a symmetric matching network. In up-conversion mixers
with high RF frequencies, the output impedance of the transistors is often so low that the
1426 28 Mixer
Vb Vb Vb Vb
RC RC
C1 C2 L1 C1 C2 L2
L
T1 T2 T3 T4 T1 T2 T3 T4
Fig. 28.57. Impedance matching at the output of a double balanced mixer by means of matching
networks
TR3
Vb
vRF
f RF
V1
TR2 T1 T2 T3 T4
VLO
L2
TR1 T5 T6
RE RE
vIF
L1
V0 2 I0
collector resistances can be omitted. Matching is then achieved as in Fig. 28.57b with two
asymmetric matching networks since the inductances of the impedance matching networks
can be used to supply the operating voltage at the same time.
Besides baluns, 1:1:n and n:n:1 transformers are also used to convert asymmetric signal
sources and loads to symmetric inputs and outputs of double balanced mixers. Impedance
matching can then be achieved fully or in part by a suitable choice of the transformer ratio.
Figure 28.58 shows an example with three transformers. Since the input admittance of
the transformers is ohmic-capacitive, the primary side of transformers TR1 and TR2 also
carry capacitive admittances; this necessitates an additional compensation of the capacitive
component to match the characteristic impedance. In the most simple case, this can be done
by tuning the inductances L1 and L2 to resonance. The output admittance on the secondary
side of transformer TR3 also has a capacitive component which, however, can be regarded
as an integral portion of the RF filter.
Conversion Gain
To calculate the conversion gain of a circuit matched at both sides, we combine the collector
resistances RC and the output resistances of transistors T1 , . . . , T4 into two loss resistances
RV . The load resistances RL1 = RL2 = ZW are transformed by the matching networks
into two resistances RP which are connected in parallel to the loss resistances. In the
matched state we have RV = RP . Figure 28.59 shows the transformation at one of the
two outputs. At both outputs this produces the same conditions as at the output of a single
balanced mixer (see Fig. 28.42 on page 1411). According to (28.40) the conversion gain is:
A2 ZW ro =RV A 2 ZW
GM = = (28.58)
ro RV
Here, ZW is the input resistance at one input and RV is the transformed load resistance at
one output. Therefore, it is necessary to use either the open-circuit gain from one input to
one output or the differential open-circuit gain for the open-circuit gain A in the equation.
The differential open-circuit gain is determined from the differential mixer voltage gain
AM,diff found by replacing RC by RV :
(28.57)
A = AM,diff
= − c 1 gm RV
RC =RV
By insertion into (28.58), the conversion gain of the double balanced mixer with impedance
Output
impedance of Matching
T 1 and T 3 network
C1 f = fmatching
RV =
rT CT RC L1 RL1 = ZW RP = R V
RC || r T
ro
Fig. 28.59. Small-signal equivalent circuit for transformation of the load resistance at one of the
two outputs
1428 28 Mixer
1 2 2
GM = c g Z W RV (28.59)
4 1 m
A comparison with the conversion gain of a single balanced mixer in (28.44) shows that,
with the same loss resistances RV , the conversion gain of the double balanced mixer is four
times higher. The same loss resistances, however, can be assumed only at low frequencies
because only then are the output resistances of the transistors negligible and the loss
resistances correspond to the collector resistances. In this case, the double balanced mixer
achieves double the output voltage and four times the output power due to its differential
output, but at high frequencies the output resistances of the transistors have a dominating
effect. Since there are two transistors connected in parallel at each output of the double
balanced mixer, the loss resistances then have half the value of that for the single balanced
mixer in Fig. 28.42. In this case, the conversion gain of the double balanced mixer is only
twice as high as that of the single balanced mixer.
q ( t) q( t )
qM (t)
Lowpass
f IF = 0 – sin Ct f LO = f C f LO = f C – sin Ct f IF = 0
filter
RC RC
VLO,i VLO,q
vi vq
Fig. 28.61. I/Q up-conversion mixer with two double balanced mixers and current summation
Chapter 29:
Appendix
29.1
PSpice – Brief User’s Guide
29.1.1
General
PSpice from OrCAD (previously MicroSim) is a circuit simulator from the Spice family
(Simulation Program with Integrated Circuit Emphasis) for the simulation of analog, dig-
ital, and mixed analog/digital circuits. In 1970 Spice was developed at Berkely University
and is available today in the version 3F5 for use without licence. On this basis, commercial
offshoots evolved which contain specific expansions and additional modules for entering
circuits graphically, presenting results and controlling processes. The most common ones
are PSpice and HSpice. While HSpice from Synopsys (previously Meta Software) was
designed for developing integrated circuits comprising several thousand transistors and is
used in many IC design systems as the simulator, PSpice is a particularly well-priced and
easy to operate simulation environment for developing small and medium-sized circuits
on PCs with a Microsoft Windows operating system.
The following short instructions apply to the demo version of PSpice 8 (PSpice Eval
8) for Microsoft Windows 98/ME/2000/XP and Vista.
29.1.2
Programs and Files
Spice
Every simulator in the Spice family uses netlists. A netlist is a description of a circuit
prepared by an editor which contains component lists and circuit topology data augmented
by simulation instructions and references to model libraries. Fig. 29.1.1 shows the programs
and files involved in the circuit simulation process.
– The netlist of the circuit to be simulated is prepared by an editor and stored in the circuit
file <name>.CIR (CIRcuit).
– The circuit file is read in by the simulator (PSpice or Spice 3F5), which then performs
the simulations according to the simulation instructions; this may include the use of
models from component libraries <xxx>.LIB (LIBrary).
– The simulation results and (error) messages are stored in the output file <name>.OUT
(OUTput) and can be displayed and printed with the aid of an editor.
PSpice
In addition to the simulator PSpice, the PSpice software package also contains a program
for the graphic input of circuit diagrams (Schematics) and a program for the graphic
representation of the simulation results (Probe). Figure 29.1.2 shows the process with the
programs and files involved:
1432 29 Appendix
Editor
(Notepad,
PFE or similar)
Circuit
<name>.CIR
Circuit simulator
Models (PSpice or Results and errors
<xxx>.LIB Spice 3F4) <name>.OUT
– The Schematics program allows you to enter the circuit diagram for the circuit to be
simulated and store it in the schematics file <name>.SCH (SCHematic); this process
uses schematic symbols from symbol libraries <xxx>.SLB (Schematic LiBrary).
– By starting the simulation (Analysis/Simulate) or creating the netlist (Analysis/Create
Netlist), the Schematics program generates the circuit file <name>.CIR; at the same time,
the netlist is stored in the file <name>.NET and incorporated by an Include instruction.
Another file <name>.ALS is generated which contains a list of alias names, but is of no
relevance for the user.
– The PSpice simulator is activated in the Schematics program by starting the simulation
(Analysis/Simulate); as an alternative PSpice can be started manually and the circuit
file can be selected with File/Open. The simulation uses models from the component
libraries <xxx>.LIB.
– The simulation results that may be presented graphically are stored in the data
file <name>.DAT ; nongraphic results and messages are stored in the output file
<name>.OUT and can be displayed by selecting Analysis/Examine Output in the
Schematics program or using an external editor.
– The simulation results can be presented graphically using the Probe program; this allows
you to directly display individual signals or to perform calculations with one or more
signals. The commands required to create a graph may be stored in the display file
<name>.PRB using the Options/Display Control function and are available for later
retrieval. If the simulation has been started via Analysis/Simulate in the Schematics
program, the Probe program will start automatically upon completion of the simulation;
29.1 PSpice – Brief User’s Guide 1433
Add Load,
Part Save
Circuit entry
Circuit symbols Circuit diagram
(Schematics)
<xxx>.SLB <name>.SCH
Create Netlist
or Simulate
Open
Circuit
simulator Messages
Models
(PSpice) and errors
<xxx>.LIB
<name>.OUT
Simulation
results
<name>.DAT
Open
Display
Evaluation Control
and display Settings for
(Probe) graphics
<name>.PRB
in this case the data file <name>.DAT is loaded automatically. If it has been started
manually, the data file must be selected via File/Open.
With PSpice you can also work with netlists directly. To do so the circuit diagram
should not be entered graphically; instead, the circuit file <name>.CIR should be created
with the help of an editor. Unlike Spice, this offers the advantage of being able to represent
1434 29 Appendix
the simulation results graphically using Probe. This procedure is often used when creating
new models since the experienced user can eliminate any errors that may occur during
model testing much faster in the circuit file than via the graphic diagram entry.
29.1.3
A Simple Example
A small-signal amplifier with AC coupling is taken as an example to illustrate how a circuit
diagram is entered and circuit simulation is performed; Fig. 29.1.3 shows the corresponding
circuit diagram.
Vb
R3
39k
R1
75k Vout
C1
Vin Vbase T1
BC547B
2.2u
Ve
Rg R4
50 4.7k Cp
R2 4p
18k Ve1
Vb OFFSET=0V
Vg AMPLITUDE=0.2V
15V R5 C2
f=1kHz 5.6k 3.3u
Inserting the components: Use the Get New Part tool to open the dialog window Part
Browser Basic; use Advanced to open the Part Browser Advance dialog window shown in
Fig. 29.1.5. If the component name is known, it can be entered in the Part Name field; the
component appears in the preview window and can be accepted by clicking on the Place
or Place & Close button. If the name is not known, the component list must be searched
to find the desired component. The Libraries button opens a dialog window containing
the component list arranged according to libraries; here, however, a preview is not shown
until a component has been selected and confirmed with Ok.
After accepting the component with the Place or Place & Close button the part is
inserted into the diagram by clicking the left mouse button. Prior to insertion the symbol
of the component can be rotated using Ctrl-R and/or mirrored using Ctrl-F. The insert
mode continues until you have clicked the right mouse button or pressed the Esc key.
The names of some important passive and active components are listed in the table on
page 1437
Configuring the components: Most components have to be configured following
insertion. In the case of passive components such as resistors, capacitors and inductors this
means entering the value; in the case of voltage and current sources this involves entering
the parameters of the signal shape (amplitude, frequency, etc.) and for controlled sources
the control ratio must be entered. For integrated transistors the size of the transistor and the
name of the substrate node must be entered while for operational amplifiers parameters like
transit frequency and slew-rate are needed. Some components such as standard transistors
(e.g. BC547B) need not be configured since they are assigned a reference to a model in a
model library which contains all the data required.
The value of a passive component can be changed by double clicking on the indicated
value; this opens the Set Attribute Value option for entering the value (see Fig. 29.1.6).
Clicking on Edit Attributes or double clicking the component symbol opens the Part
dialog window shown in Fig. 29.1.7 which contains a list of all parameters. Parameters not
29.1 PSpice – Brief User’s Guide 1437
marked by an asterisk can be selected, modified in the Value field and saved by pressing
the Save Attr button. Pressing Change Display allows you to ascertain whether or not and
how the selected parameters are presented in the circuit diagram; usually only the value,
e.g. 1k, or the parameter name and the value, e.g. R = 1k, is displayed.
Numeric values can be entered as an exponential expression (e.g. 1.5E-3) or can be
provided with the following suffixes:
Suffix f p n u m k Mega G T
Name Femto Pico Nano Micro Milli Kilo Mega Giga Terra
Value 10−15 10−12 10−9 10−6 10−3 103 106 109 1012
There is no distinction between upper and lower case letters. A common error is the use
of M for Mega which is then interpreted by PSpice as Milli.
Adding the connecting wires: After all the components have been placed and con-
figured, the connecting wires must be entered using the Draw Wire tool; here, the cursor
takes the form of a pencil. First fix the starting point of the wire with a left mouse click.
The wire is then shown as a broken line and can be confirmed from point to point by
clicking the left mouse button (see Fig. 29.1.8). The most simple case is a straight line
between starting point and end point; in this case the course is determined automatically.
By fixing intermediate points the course can be influenced. Placing a point on a component
terminal or any other wire causes the program to assume that the wire entry is complete
and conclude the entry. As an alternative the entry can be interrupted at any given point by
clicking the right mouse button or pressing Esc.
Vb
off
on
v
Vb
Vg
v
Normally ground wires are not drawn. Each point connected to ground is marked with
a ground symbol GND. In the netlist the nodal name 0 is assigned to any point that is part
of the GND net.
A nodal point 0 must exist in every diagram; every circuit diagram must therefore
contain at least one ground symbol GND.
A name is assigned automatically to every node. These names appear in the netlist
and are required in the Probe program to select the signals to be displayed. Since the
names assigned automatically do not appear in the circuit diagram and are thus not known
without referencing the netlist, an appropriate name should be given to any nodal point
in the diagram that is of particular interest; this is done by double clicking on one of the
wires that is connected to the node and entering an appropriate name.
After all the components have been entered, connecting wires inserted and nodal names
assigned, the completed circuit diagram is presented as shown in Fig. 29.1.9; if you have
not already done so, store the diagram by clicking on the File/Save button.
1440 29 Appendix
– Transient: length of the time interval to be simulated and, if applicable, the increment
for numeric integration
The simulation instructions are finalized with the Setup Analysis tool. Here, a list of
analysis set-up options appears. This list is shown in Fig. 29.1.11. In addition to the AC
Sweep, DC Sweep and Transient analyses already described, there are other analysis types
and capabilities which will be explained later. The Bias Point Detail analysis calculates
the operating point on the basis of the DC sources and saves the result in the output file
<name>.OUT ; this analysis is activated by default. For our example the AC Sweep and
Transient analyses must be activated.
Selecting the field AC Sweep opens the AC Sweep dialog shown in Fig.29.1.12, in
which the frequency range is entered. In our example the frequency range 1 Hz to 10 MHz
with 10 points per decade is to be investigated.
Fig. 29.1.12. Setting the frequency range for Fig. 29.1.13. Setting the parameters
AC Sweep for Transient
1442 29 Appendix
Selecting the Transient field opens the Transient dialog shown in Fig. 29.1.13. Here, the
end point of the simulation is entered in the Final Time field and the maximum increment for
numeric integration in the Step Ceiling field. The start time for recording the results is to be
entered in the No-Print Delay field; normally this is 0 to allow every value calculated to be
displayed graphically. If only the steady-state condition of circuits with long transient times
is to be investigated, the estimated transient time can be entered in the No-Print Delay field
so that recording does not begin until after the transient time. The Print Step parameter still
exists for historical reasons and is not needed. Nevertheless, this parameter must not be set
to 0 and must be less than or equal to the Final Time. In addition, the output signal V(Vout)
undergoes a Fourier analysis at a fundamental frequency of 1 kHz, which corresponds to
the frequency of the source; in doing so, five harmonics are determined, which, together
with the resulting harmonic content, are saved in the output file <name>.OUT.
After the simulation instructions have been entered, the schematics file is complete and
can be stored by pressing the File/Save button.
Fig. 29.1.15. Selecting the analysis type in the Probe start-up window
Figure 29.1.17 shows the Add Traces dialog window displaying a list of signals on the
left and a selection of mathematical functions on the right. The following designations are
used:
Click on the desired signals or functions to select. These will then be shown in the
Trace Expression field, where they can be edited if necessary. The following options are
available for AC signals:
29.1 PSpice – Brief User’s Guide 1445
In our example Vdb(Vout) represents the magnitude of the output voltage (see Fig. 29.1.18).
Since the driving voltage source has an amplitude of 1 V (AC = 1) this value represents
the small-signal gain of the circuit. The scale factor of the x and y-axes can be altered via
the menu options Plot/X Axis Settings and Plot/Y Axis Settings.
Other signals that use the same scale factors can be added to the display without
further alteration. If signals with different scale factors, e.g. the phase Vp(Vout), are to be
represented in a meaningful manner, it is first necessary to create another y-axis using the
Plot/Add Y Axis menu option. The active Y-axis is marked with a » and can be selected with
a mouse click; after using Plot/Add Y Axis the new y-axis is automatically active. Adding
the phase Vp(Vout) displays the window shown in Fig. 29.1.19.
Finally we also wish to display the results of the large-signal transient analysis. To do
so, we must switch over using the Plot/Transient menu option; an empty window appears
which already has a time scale corresponding to the time interval simulated. If the voltages
V(Vin), V (V base), V (V e) and V(Vout) are added via the Add Traces option, the display
shown in Fig. 29.1.20 appears.
The settings for a particular display can be saved with the Tools/Display Control menu
option and can be retrieved at a later date. The settings for the various analyses are saved
separately so that only those settings are shown that belong to the analysis selected. The
settings used last can be called up via the Last Session option.
The Tools/Cursor/Display menu option allows you to display two markers simultane-
ously, which are moved with the left or right mouse button; at the same time the x and y
values of the marker position are shown in an additional window. For more information
refer to Cursor in the help index. The markers can also be turned on and off with the Toggle
Cursor tool:
After an extensive circuit diagram has been entered, the first step is usually to check
the operating point by performing a simulation using a Bias Point Detail Analysis, which
is activated by default, and to examine the results. This ensures that the circuit has been
entered correctly and is functional, before other analyses, which may be somewhat time-
consuming, are performed. When following this procedure the Probe display program
does not start automatically because the Bias Point Detail analysis does not produce any
graphic data.
NAME Q_T1
MODEL BC547B
IB 8.54E-07
IC 2.19E-04
VBE 6.24E-01
VBC -3.56E+00
VCE 4.18E+00
BETADC 2.57E+02
GM 8.45E-03
RPI 3.47E+04
RX 0.00E+00
RO 3.03E+05
CBE 4.02E-11
CBC 2.82E-12
CJS 0.00E+00
BETAAC 2.93E+02
CBX 0.00E+00
FT 3.13E+07
DC COMPONENT = 6.461795E+00
This file contains the parameters for the models used (here: BJT Model Parameters),
operating point information (Small Signal Bias Solution) with the small-signal param-
eters of the components (Operating Point Information) and the results of the Fourier
analysis (Fourier Analysis).
29.1.4
Further Examples
Characteristics of a Transistor
Figure 29.1.23 shows the circuit used in this example. DC Sweep is activated in the Setup
Analysis dialog window (see Fig. 29.1.24). Then the parameters shown in Fig. 29.1.25 are
entered:
– In the internal loop DC Sweep the collector-emitter voltage source VCE is varied in
steps of 50 mV through a range of 0…5 V.
– In the external loop DC Nested Sweep the base current source IB is varied in steps of
1 mA through a range of 1…10 mA.
After the parameters have been entered, the simulation is started using Simulate and
the collector current is displayed via Add Traces in the Probe program (see Fig. 29.1.26).
29.1 PSpice – Brief User’s Guide 1451
Using Parameters
Often it is desirable to perform the same analysis several times while varying one circuit
parameter, e.g. the value of a resistor. The example in Fig. 29.1.27 shows the circuit diagram
of an inverter with variable base resistance RB. This means that the RB value has to be
replaced by a parameter in curly brackets (here: R), which then has to be defined. This is
done by means of the Parameters component that has been inserted in the upper left corner
of the diagram in Fig. 29.1.27. Double clicking on the Parameter symbol opens the Param
dialog window shown in Fig. 29.1.28. You will then be prompted to enter the name of the
parameter and its default value; the default value is used in analyses without parameter
variations.
In the Setup Analysis dialog, DC Sweep must be activated in order to simulate the char-
acteristics; similarly, Parametric must be activated to vary the parameter (see Fig. 29.1.29).
The corresponding parameters are shown in Fig. 29.1.30. In DC Sweep a given parameter
can also be varied via the Nested Sweep option; however, this is not the most flexible
of methods as the Nested Sweep option is only available in DC Sweep analyses, while
Parametric allows variation in any analysis.
After simulation using Simulate the Probe program initially displays the window shown
in Fig. 29.1.31 for selecting the curves or parameter values to be indicated; in the default
setting all curves are selected. Entering V (V out) results in the characteristics shown in
Fig. 29.1.32. The individual characteristic curves are marked by different symbols, which
are shown underneath the graph in the order of the parameter values.
29.1.5
Integrating Other Libraries
A library comprises two sections (see Fig. 29.1.2):
– The symbol library „xxx“.SLB contains the circuit symbols of the components and
information on the representation of the components in the netlist.
– The model library „xxx“.LIB contains the component models; these are either elemen-
tary models with the parameters given as .MODEL instructions or macro models which
1456 29 Appendix
comprise several elementary models that are combined to a sub-circuit and are contained
in the model library in the form of .SUBCKT„name“„terminals“„circuit“.ENDS.
A symbol library is integrated in the Schematics program via the Options/Editor Con-
figuration option. This opens the Editor Configuration dialog window shown in the left
hand side of Fig. 29.1.33, which lists the existing symbol libraries and their corresponding
paths. Selecting the Library Settings option opens the dialog box shown in the right hand
side of Fig. 29.1.33. This option is used to add, edit or delete symbol libraries. It is possible
to enter the name and path (drive and directory) of the library in the Library Name field or
to search for the desired library using Browse. Clicking on Add* adds the symbol library
to the list. To close the dialog window click on Ok.
Similarly, a model library is added in the Schematics program via the option Analysis/
Library and Include Files. The name and path of the library are entered in the same way
and the library can then be added with Add Library* (see Fig. 29.1.34).
To add libraries, always use the asterisk commands Add* or Add Library* since this
adds the libraries permanently and makes them automatically available when the program
is re-activated. Since in the demo version of PSpice both the number of libraries and the
number of library elements are limited it is necessary to exchange libraries if other libraries
are required for further simulations and the total capacity has been used up.
29.1.6
Some Typical Errors
The typical errors are described on the basis of the circuit diagram shown in Fig. 29.1.35
which contains several errors. If an error occurs, the MicroSim Message Viewer appears
showing the relevant error message either before or after the simulation (see Fig. 29.1.36).
– Node < node name> is floating: The voltage of a node cannot be determined since
it is indeterminate; this is the case for node N2 in Fig. 29.1.35. This error message
appears at any time when only capacitors and/or current sources are connected to a
node; consequently, Kirchhoff’s node law is not fulfilled. Every node must have a DC
path to ground in order to clearly determine the nodal voltage. In the case of node N2 in
Fig. 29.1.35 the error can be rectified, for example, by adding a high-resistive resistor
between N2 and ground.
– Voltage and/or inductor loop involving „component“: A loop formed by voltage sources
and/or inductors which goes against Kirchhoff’s loop law exists; in the example shown in
Fig. 29.1.35 the voltage source Vb1 is short-circuited for DC voltages by the inductor L1.
29.2 ispLEVER – Brief User’s Guide 1459
29.2
ispLEVER – Brief User’s Guide
29.2.1
Outline
Programming PLDs as described in Chap. 10.4 requires the creation of a so called fusemap
containing a list of the desired connections. This may be done manually using a text
editor or, more comfortably, with the aid of a design environment like the one pro-
vided by ispLEVER. The starter software of ispLEVER can be downloaded by Lattice
(www.latticesemi.com) and can be licensed there at no cost.
This program supports the entry of the circuit by either using a programming language
or by generating a circuit diagram. Furthermore, in order to test the functionality of the
design simulation with graphic output is carried out. In addition to this, the propagation
delay time can be tested with the help of timing analysis and optimised for different design
goals.
Input types:
Abel HDL Schematic VHDL
*.abl *.sch *.sym *.vhd
Hardware description Schematic drawing Hardware description
language (text editor) (graphic editor) language (text editor)
Fitter
JEDEC-Generation
*.jed
Output devices:
Programming device
In System Programming
(ISP)
The design environment itself has come a long way. It was initially developed by Data
I/O under the name Synario. The software was then acquired by MINC, who were bought
out by Vantis, who proceeded to rename the product DesignDirect. After the merger with
Lattice it was named DesignExpert and finally became ispLEVER.
The flow chart in Fig. 29.2.1 shows the interconnections between and procedures
carried out in the various input modes as well as the analysis process and the generation of
output files. The project navigator is used to start all the activities and to setup the various
options.
The hardware description language Abel HDL (High Definition Language) is en-
tered via the text editor. In addition, circuit diagrams can be drawn with the help of the
Schematic’s input option. A description in the programming language VHDL is also pos-
sible, but this is not explained here. The interdependence of the source files is managed
and represented in the form of a hierarchy in the source window.
The integrated compiler converts the source files independent of the source into a uni-
form machine format and generates a process-related report. Several reports are displayed
during each phase of the programming process in the output window below. Such reports
include not only design errors but also design analysis to show additional information such
as timing analysis and utilisation of the chip resources.
The ispLEVER software package comprises various program areas. Their combined
features enable the generation of complex designs. The integrated development environ-
ment is the project navigator shown in Fig. 29.2.2 which is started via the program is-
pLEVER. The project navigator shows all the files related to a given project and is used
to start up all processes required for the project.
As is common with Windows, the menus and their various options and commands are
activated by a left mouse click or a designated key combination. For reasons of simplicity
the following description is limited to the mouse operations. In the source or the process
window (see Fig. 29.2.2 and Fig. 29.2.3, respectively) you can perform a specific action
or review a report by double clicking the left mouse button.
For example, a double click on the project title “Untitled” allows you to modify the
project title. With a double click on the device name you can select the device, while double
clicking a file name opens the file.
Both a text and a graphic editor are available for data input. The system provides
text reports on the success of the individual compilation steps and the analysis results
in the Automake Log window. The functional simulation results may also be presented
graphically. The results of the timing analysis appear in the form of a table. Finally a
fusemap is generated as a JEDEC file for device programming.
The window at the button of the Project Navigator is for revision control purpose,
which is not reasonable for small designs and therefore neglected in this tutorial.
ispLEVER is a very comprehensive software package, which means that this descrip-
tion can only cover the most important commands and features. Detailed information is
available in the Help function of each element of the software package.
29.2.2
Circuit Entry
Every new project begins with data input. In principle it is unimportant which device
is used at this stage of the design process. But the chip-family should suit because this
determines which libraries are available. ispLEVER supports almost every device from
Lattice, beginning with the simple PLDs to the complex CPLDs and FPGAs. For our
example, we have opted for the model ispM4A5-64/32 to illustrate the various design
steps, starting with an empty project and proceeding to circuit input and design analysis
1462 29 Appendix
and finishing with the creation of the JEDEC file. This device is a CPLD that consists of
four PLDs of the 33V16 type and an additional programmable interconnection matrix. The
programming logic is located on the chip; it is thus programmable within the circuit (ISP –
In System Programmable) as is the case with all newer devices. The only thing required
is a passive download cable and a download program as described in Sect. 29.2.5.
The following table shows the meaning of the file name extensions (also refer to
Fig. 29.2.1). To archive a project only the files shown in bold print must be copied. The
remaining files will be regenerated by the program as required.
abl Abel HDL file sch Schematic file
abv Abel test vectors sym Symbol for schematic
fit Fitter report syn ispLEVER project file
jed JEDEC file (Fusemap) wav simulation output
rpt Report
rst ce EN2
rst R 2CT=5 carry
ce ce
0 1 2 clk C1+
ce ce
[1] q0
5 ce
4 ce
3 [2] q1
[3] q2
Type in project
name.
Choose the
favored directory.
Go on with the
next dialog.
Choose correct
device family.
Select the desired
device.
Possibility to add
former modules.
Go on.
been entered and confirmed via the OK button, the text editor for entering the Abel code
will open automatically.
The module name is the identification code within each project, while the file name is
the name under which the data record is saved in the directory. Both entries are mandatory;
the title, on the other hand, is an optional description of the function. It makes sense to use
identical names for module and file in order to facilitate later retrieval. The title should be
as detailed as possible so that the function of the module can be readily identified later.
The following program sample shows the complete generation of an Abel HDL module.
Of particular importance are the key words shown in bold print, which must exist in every
Abel file. If you wish to format the text as shown in the example below spaces or tabs
should be used since these are ignored by the compiler.
Enter the example from Fig. 29.2.9 using the names from Fig. 29.2.8 and save the
project. If you want to avoid typing you can import the source files from the sample
directory (my files\ispLEVER examples that the TS-installer has generated) using the
29.2 ispLEVER – Brief User’s Guide 1465
END
Fig. 29.2.9. Example 1. Abel
module for cnt_3bit
Source/Import option in the Project Navigator. The files can be used to avoid typing and to
prevent possible errors. However, the project should not be opened in the existing directory
as this would prevent you from being able to follow the individual steps.
Inputs and outputs in the Abel module: The description of the behaviour of a circuit
(see Fig. 29.2.9) starts immediately underneath the module name and title with a list of
the input and output signals and their interrelations. Pin numbers may also be assigned at
this point, which will be explained further below.
Abel recognises output signals from the key word “istype”. It would be more correct to
speak here of a resulting signal since it is not necessary available at a pin (see Fig. 29.2.10).
The outputs q2 to q0 were declared “istype ‘reg’”, which means that each of these outputs
is made available via a register (to build a sequential circuit). The carry signal is in the form
of a combinatorial circuit which calls for a corresponding designation. The syntax here is
“istype ‘com’”. The “istype” command has many more variations. But only the “,‘reg’”
(registered) and “ ‘com’” (combinatorial) types are required for this brief user’s guide. A
list of some more variations is contained in the on-line help menu of the Text Editor. The
outputs are not necessarily available at a pin of the chip; several signals are only intended
for internal feedback. Therefore it would be more accurate to speak of resulting signals.
1466 29 Appendix
DECLARATIONS
As " istype 'reg' " marked signals
clk pin 11; (registered signals) are computed
Clock every clock cycle. Signals marked
Reset rst pin 2; as " istype 'com' " (combinatorial
signals) follow immediately every
Count Enable change of the input signals.
ce pin 3;
counter = [q2..q0];
Declaration of a 3 bit wide bus using
the counter bits q0 to q2. q2 is the
".." operator most significant bit. The bus may be
used for arithmatical operations.
The “..” operator is an abbreviation for sequences, meaning that not every element
of a sequence needs to be described in detail. In this example (Fig. 29.2.9) the benefit is
minimal since only the element q1 needs not to be entered.
Behaviour of the circuit in the Abel module: The following syntax of the functional
description is only one among many, but it is nevertheless a very powerful method for a
counter. Figure 29.2.11 shows this method of defining the counter. The second and third
program lines tell the compiler to permanently transmit the input signals “clk” and “rst”
to the bus “counter”. Normally a decimal value is assigned to the contents of the bus.
The next five lines contain the actual program for the desired function. The carry signal
is theAND operation of the first and third bit and therefore becomes active at 5dec = 101dual .
Afterwards the counter reading is checked to determine whether or not it has reached the
final count of 5. If it is true the counter is reset to 0 otherwise it is incremented.
The “.” operator allows the sub-elements of signals and busses to be accessed. The
“:=” operator assigns a signal that is synchronous to the clock (registered), while “=”
sets the signal directly (combinatorial) to the given value. This works only with the ap-
propriate variables, e.g. a combinatorial signal cannot be assigned to a registered signal
(combinatorial).
The “when then else” expression is used to distinguish between different situations.
Any logic expression can serve as a condition; ensure that parentheses are used intelligently
to avoid erroneous operations. The operators are processed in the sequence given by the
compiler; but parentheses make it easier to understand the intention of the programmer.
After closing the text editor you can compile the Abel module by double clicking on
“Compile Logic” in the process window of the project navigator. The compiler automati-
cally checks the syntax.
The results of the compilation can be checked in the report window by opening the
desired object in the “Processes for current source” window in the Project Navigator.
Equations:
! = Negation
q2 := (q2 & !ce & = AND
# !q2 & q0 & q1 & ce # = OR
# q2 & !q0 & !q1 & ce);
q2.AR = (rst);
...
Reverse-Polarity Equations:
!q2.AR = (!rst);
!q2.C = (!clk);
...
You can view the reports in the Automake Log window (see Fig. 29.2.2) that imforms
you about the success of compiling your design. If you follow this procedure for the
“Compiled Equations”, you will receive a report as shown in Fig. 29.2.12. Please note
that the Project Navigator automatically performs all preparatory steps. In our example
the Project Navigator would cause the compiler to begin compilation (Compile Logic) as
soon as the report is requested. Use “Generate Schematic Symbol” to create a symbol for
later use.
MODULE cnt_3bit
DECLARATIONS
clk pin 11;
rst pin 2;
ce pin 3;
Equivalent to first example
q2..q0 pin 14..16 istype 'reg';
carry pin 24 istype 'com';
"bus definition
counter = [q2..q0];
EQUATIONS
counter.clk = clk;
carry = q2 & q0;
MODULE oct27seg
DECLARATIONS
oct0..oct2 pin 2,3,4;
seg0..seg6 pin 14..20 istype 'com';
"bus definition
ziffer = [oct2..oct0]; Input signal(s) Output signal(s)
TRUTH_TABLE (ziffer -> [seg0, seg1, seg2, seg3, seg4, seg5, seg6])
0 -> [ 1, 1, 1, 1, 1, 1, 0];
Values for 1 -> [ 0, 1, 1, 0, 0, 0, 0];
input signal(s) 2 -> [ 1, 1, 0, 1, 1, 0, 1];
3 -> [ 1, 1, 1, 1, 0, 0, 1];
4 -> [ 0, 1, 1, 0, 0, 1, 1];
5 -> [ 1, 0, 1, 1, 0, 1, 1];
6 -> [ 1, 0, 1, 1, 1, 1, 1];
7 -> [ 1, 1, 1, 0, 0, 0, 0];
Library device
Only the most important steps for the design of a circuit diagram will be outlined
here. For further information refer to the Help menu. To perform an operation first select a
command from the tool box and then the object that is to be used. If you want to terminate
a command, click on the circuit diagram with the right mouse button. If an entire region
is marked, the command is applied to several objects simultaneously. To do so, press and
hold the left mouse button and draw a box around the desired objects. If you want to cancel
the last step click on Undo (either in the Edit menu or in the symbol bar).
The following steps must be executed to create the sample design:
– Insert the toggle flip-flops (library REGS.LIB, component G_TC) selected from the
Symbol Libraries (Add Symbol) into the diagram. The desired library can be selected
in the upper section of the popped up window. Then mark the component in the lower
section and place it three times in the desired positions in the circuit diagram.
– Add the required logic gates to the diagram as shown in Fig. 29.2.19. You find them in
the GATES.LIB; select the gates G_INV, G_2OR and G_3AND.
– Use Add Wire to connect components at the respective red dots.
– Assign “net names” to all inputs and outputs so they can be addressed in the superordinate
modules or for testing. After selecting the command (Add Net Name) enter the desired
name at the bottom of the Schematic Editor, confirm it by clicking on Enter and click
on the red dot at the end of the wire.
– Now use the I/O markers (Add I/O Marker) to define the signals as input or output
signals. In the case of several I/Os, this can also be done for all of them in one step by
marking the relevant region.
– Save the schematic diagram.
– Use File/Matching Symbol to create a schematic symbol.
In order to illustrate how a bus is generated the outputs q0 up to q2 are connected to one
bus (see Fig. 29.2.21).
1472 29 Appendix
Component libraries
In Schematic the options File/Matching Symbol and Add Symbol make it possible to reuse
an existing design. For example the 3 bit counter could be connected with the 7-segment
decoder. In this context we wish to draw your attention to the existing component libraries.
29.2 ispLEVER – Brief User’s Guide 1473
Library Description
vanprim.lib selguide.pdf
vanttl.lib vanttl.pdf
vanfunc.lib vanfunc.pdf
They contain not only simple gates but also complex components such as entire counters,
multiplexers and adders. The use of these modules saves a lot of time in the designing and
testing process; in this way the design process is performed on the same level as before
when using complex TTL devices.
Which of the libraries are available depends on the PLD used. Figure 29.2.22 shows a
brief list of the libraries that are most important for the MACH modules. Libraries for the
ispLSI1k...8k family are described in the files ispmacro.pdf and 58kmcr.pdf.
In our exercise a BCD counter and a 7-segment decoder from the TTL library are used.
Again we generate a counter with a 7-segment output. The corresponding circuit diagram
is shown in Fig. 29.2.23.
Enter the sample component with the names from Fig. 29.2.24 or use the Source/Import
option to get it from the directory my files\ispLEVER examples. Follow the steps described
below to create the circuit diagram:
Active Low
– Insert the TTL modules V74162 and V7449 (library VANTTL.LIB) selected from the
Symbol Libraries (Add Symbol) into the diagram.
– Add the required logic gates, inverters and VCC (library GATES.LIB, G_INV and VCC)
to the diagram according to Fig. 29.2.23.
– Draw the wires using Add Wire.
– Assign net names to all inputs and outputs. After selecting the Add Net Name option
enter the desired names, confirm using Enter and click on the wire; ensure that you
click onto the red dot at the end of the line. For the segment designations use the name
“seg0+”. This causes the program to increase the final digit by one after each designation
process.
– Define the designated I/O signals as inputs or outputs using the I/O marker (Add I/O
Marker). In the case of several I/Os, all of these can be defined in one step by marking
an entire region.
Hierarchy
With ispLEVER you can link or interleaf several modules that have been designed in Abel
or Schematic. This is similar to a C or Pascal program with various different functions and
procedures. ispLEVER shows the hierarchy using a tree structure.
The simplest method of linking hierarchy levels is to develop the upper-most module
as a schematic as shown in Fig. 29.2.25. The desired sub-modules are then integrated into
this. In the hierarchy tree these modules appear below the main module. To do so, it may
be necessary to import the Abel files used.
In this exercise we will use the counter and the octal-to-seven-segment decoder to
create the design shown in Fig. 29.2.26.
Top module
Enter the example with the name from Fig. 29.2.27 or use the Project Navigator:
Source/Import option to import it from the ispLEVER examples.
Import the previously created Oct27Seg decoder and the first Abel counter (from
cnt_3bit_abel_1). Use menu option Source/Import in the Project Navigator and select
the corresponding Abel file. Then open a new Schematic design using Source/New…/
Schematic. Add the two symbols from the “local” library to the design.
Finally draw the necessary wires, define them as inputs/outputs and mark them with
the corresponding I/O markers.
Use the File/Matching Symbol to generate a symbol for the entire design before quitting
the Schematic Editor. This is effectively the same as using Generate Schematic Symbol in
the Project Navigator.
29.2.3
Pin Assignment
If no pins are assigned, the Device Fitter will perform the pin assignment itself when
generating the JEDEC file, so that it is best for the internal wiring (also refer to Sect. 29.2.5
Optimization). Manual pin assignment is done either directly in the source file (e.g. Abel
or Schematic) as in the examples described or by entering the values in the Constraint
Editor.
If the pin assignment is to be taken from a source file, the import option must be chosen
in the dialog popping up during the fitting procedure. The fitter then adopts the assignments
from the top-level module, i.e. all pin assignments in the sub-modules are discarded.
The information of the specific chip is first fetched from the chip-library when it is
used for simulation and for converting the net lists into JEDEC format. Only in such cases
does it actually make sense to assign pins.
The device selection window is opened by double clicking on the current device name
in the source window. This opens the Device Selector window (Fig. 29.2.28). Don’t forget
to choose the correct package type.
1476 29 Appendix
Package type
Abel
InAbel pins are assigned simply by entering the desired pin number after the key word “pin”
(see Fig. 29.2.29). The Oct27Seg decoder is used as an example; here, the pin numbers are
shown in bold print. The numbers can be entered either individually or by using the “..”
operator.
By assigning the pin numbers directly in the Abel HDL module, which is also used for
the function, the designer is forced to adapt the pin numbers accordingly when changing
to another module or when importing the module into another project.
Schematic
Pin assignment can be done in any Schematic file. However, it is easier and more common
to combine the entire circuit in a single block and to assign the pins in a top-level schematic
(see Fig. 29.2.30).
Create a new project and import the files shown in Fig. 29.2.31. Generate a new
Schematic module named “pins” and perform the following steps:
29.2 ispLEVER – Brief User’s Guide 1477
Output pad
Input pad
Pin number
Constraint Editor
By the use of the Constraint Editor the pin numbers are assigned to the signals by hand. To
start this editor mark the device entry in the source window and double-click at Constraint
Editor in the process window.
Double click at the “carry” signal in the left window (see Fig. 29.2.33). The selected
signal appears as entry on the right. Then double click t the “pin” cell and enter the pin
number 25. In the GLB cell the logic block of the device is assigned automatically, which
indicates the geometrical placement of the signal on the chip.
The “carry” signal is now assigned to pin 25. In order to undo an assignment click at
the signal with the right mouse button and choose “clear selected”.If you have assigned
the pin numbers in the schematic or ABEL file these numbers are taken for initializing the
constraint editor. You are also able to change the pin numbers in this case by the help of
the constraint editor.
Depending on the device selected other location assignments are possible, e.g. the
allocation to a special PAL-block or the macro cell which is to calculate the signal.
Open the “pins” project and start the Constraint Editor. Click on the “Loc” button and
mark the “carry” signal. Then mark pin number 25. Confirm the assignment by clicking
on Add. The “carry” signal is now included in the list underneath. In order to undo an
assignment use the cancel command. Alternative you mark the “carry” signal, select pin
number 24 and confirm this with the “Update” button. If the Import Source Constraint
option is active, the signals are shown instantly in the list underneath.
The assignment can be changed in 3 steps:
– mark the signal in the window at the bottom
– Modify transfers the signal to the upper window
– here it can be connected to the desired pin
– confirm with update
X [t] Y [t]
x0 y0
. .
. Circuit .
. .
xn yn
29.2.4
Simulation
Trouble shooting is the prime purpose of design analysis. The behaviour of the newly
designed circuit is checked by way of simulation. On the other hand the timing/frequency
analysis is suitable for checking the functionality under given conditions, such as the
maximum frequency. ispLEVER features an integrated simulator which applies the input
signals described in a test vector file to the design and calculates the resulting signal forms
(see Fig. 29.2.34).
These input signals are the test vectors that are applied to the design one after the other.
Another option is to compare the simulation results with a given result vector.
Whether the results of such simulation are useful and the extent (in terms of percentage)
to which they test the design depends on the selection of suitable input signals. Therefore
it is important to define suitable start conditions. For example the registers should be reset
prior to initial use.
Flip-flops accept input signals only in the event of a rising clock pulse edge. For this
reason, signals that occur after a positive test pulse edge and disappear before the next
pulse have no effect. Furthermore, the input signals of flip-flops must not change during
the positive clock pulse edge since otherwise they do not comply with the set-up and hold
time, which would result in undefined conditions.
The results of the simulation can be viewed in the Waveform Viewer. All signal forms
can be examined here.
Test vectors
Test of combinatorial circuits: Open the Oct27Seg project and open the Abel file. Insert
the text after the truth table as shown in Fig. 29.2.35. Only one input and one output signal
must be given in the test vector. All the other output signals can be viewed in the Waveform
Viewer.
Now mark the Oct27Seg vectors in the Project Navigator and start the functional
simulation. In the Simulator Control Panel the simulation is activated via the Simulate/Run
button. The Waveform Viewer opens and shows the simulation result. A more detailed
description of the operation is given further below.
Testing sequential logic: In this example a test vector file that is separate from the Abel
file is to be created in order to test the counter.
Open the “cnt_3bit” project in the “cnt_3bit_abel_1” directory, create a new source
test vectors module and enter the program according to Fig. 29.2.36. The test vectors are
saved in a separate file, which provides the advantage that the test vector file can be used
1480 29 Appendix
DECLARATIONS
This line defines a macro named
clk pin; test_counter using the key word MACRO.
rst pin; Optionally a variable (e.g. i) may be used
ce pin; in the macro.
carry pin istype 'com';
in different projects. In other words, this test vector file can be used for all three “cnt_3bit”
projects.
Test vectors are also used for stimulating counters. In this case macros are used. This
allows longer simulations to be set up without every time increment having to be specified
explicitly.
First you define the “test_counter” macro via the key word “macro”. Please note that
the entire macro is in parentheses. The variable (i in this example) is used to specify the
number of repetitions. The macro itself has a structure similar to that of a common test
29.2 ispLEVER – Brief User’s Guide 1481
vector. In order to avoid having to specify every single step, the repeat command ensures
that the test vector is repeated i number of times. Several test vectors could also be used –
the important thing is to pay close attention to the use of parentheses. To activate the
macro, its designated name must be entered; the value in parenthesis specifies the number
of repetitions desired.
The simulator knows how to interpret the values “.c.” and “.x.”. “.c.” means that the
simulator has to generate a clock pulse, while “.x.” (don’t care) instructs the simulator to
ignore the values.
The test vector must again contain at least one input and one output element. Other
signals can be selected and viewed in the Waveform Viewer. The test vector file created
can now be used for the two other counter projects and can be imported into the library,
hierarchy and pins projects; these projects are then ready for testing. Observe: there must
exist only one test vector file in a project.
Waveform Viewer
Open the hierarchy project, import the test vector file of the counter and start the simulation.
This opens the Simulator Control Panel (see Fig. 29.2.37), which allows several settings
to be made before the actual simulation is started (Simulate/Run). The Waveform Viewer
(see Fig. 29.2.38) is the most suitable tool in the ispLEVER software package for signal
representation.
All signals available for viewing (see Fig. 29.2.39) can be displayed via the Show com-
mand in the Edit menu. Instances indicates the current level of signals; the corresponding
signals can be found under Nets. Double clicking on a net displays the given signal.
Alternatively you can mark one or more signals and view them via the Show option. A
marked signal shown in the plot-window can be deleted when it is marked by the edit/hide
command.
In order to show a summary of the corresponding output signals when viewing a counter,
you have to use the bus option. In our example we wish to view the internal counter output
only. This is done by double clicking on D underneath Instances. This displays additional
internal signals. Expand the window by pressing the Bus button and enter a bus name.
Then mark the Nets N_1, N_2 and N_3. The signals marked are added to the active bus
via Add Net(s). The sequence of the signals, i.e. the significance within the bus, can be
reversed using the Reverse button. With Save Bus the bus can be saved and subsequently
displayed via Show. The sate of the bus is shown in decimal notation. New Bus allows
you to create additional buses.
The Waveform Viewer allows you to zoom in on an image. Click on the Zoom In option
in the View menu. Mark the desired area in the display or simply click in the display to
increase the zoom.
Functional simulation performs solely a functional analysis, while timing simulation
takes the real chip and rooting into consideration. This requires that the signals are con-
nected to pins. However, calculations become far more complex, which take much com-
putation time in large-scale projects.
29.2 ispLEVER – Brief User’s Guide 1483
one of the seven points for several seconds, you will receive more information. The Run
option starts the analysis chosen.
The analysis result is presented in the window to the right. Detailed information on the
transit times in the chip can be obtained by double clicking on one of the elements (see
Fig. 29.2.41).
29.2.5
Optimization
There are different ways of improving a design. First you have to define the intended design
goal. Is it to consume as few resources as possible or is it to achieve the highest possible
clock frequency?
With the information obtained from the analysis you can alter the allocated modules
in order to improve the design.
A simple way of increasing the permissible frequency is to choose a faster chip. How-
ever, such a chip may not be readily available on the market or may draw a too much
current or may be too expensive.
Often the only remaining option is to change the existing design. This may involve
changing the pin assignment (Constraint Editor) or redefining the operating conditions of
the Fitter in the “Optimization Constraint Editor” in the process window of the device
entry.
Maximum frequency versus minimum space requirement: By activating the “Opti-
mization Constraint Editor” from the process window of the device entry you are able
to configure the fitting constraints. By double clicking each table entry respectively you
are able to select the possible values from the drop down menu. By changing line 6
(“node_collapsing_mode”) from “Speed” to “Area” the fitter is optimized for bringing big
designs in a small chip: The logic cells of the chips are used to full capacity. Use this option
if the fitting process failes with “need more pins”. Be aware that the area option optimized
packing the design at the cost of signal transition time.
In small designs, as in the examples described here, these effects are not noticeable as
the module capabilities are not fully utilised.
Without fixed pin assignment: If the desired maximum frequency cannot be achieved
by the methods described above, you still have the possibility of not defining the pin
assignment. This enables the fitter to spread the design more suitably on the chip. But this
means that the printed circuit board has to be adapted to the pin assignment of the PLD.
Constraint Editor: The Constraint Editor not only allows you to change the pin assign-
ment in some chip types but also to influence the assignment of the macrocells, blocks
and segments. So you can force corresponding functions in one PAL-block and reduce
transmission times on the chip.
29.2.6
Programming
Generally, the newer PLDs are programmable in the circuit (ISP – In System Pro-
grammable). This makes programming equipment unnecessary since the required program-
ming logic is located on the chip. Sometimes “ISP” is included in the model designation,
29.2 ispLEVER – Brief User’s Guide 1485
D1 1 TCK
3 13
D5 2
7
D2 3 TMS
4 32
GND 4 GND 1...
D0 5 TDI
2 10
6 VCC 44...
ACK 7 TDO
10 35
8
D4 9 TRST
6
D3 10 ispEN
5
18
25 2 4 6 8 10
1 3 5 7 9
DB-25P
Top view
Fig. 29.2.42. Passive JTAG download cable with connections according to Lattice
as is the case with the chip from the ispM4A family used in our example. Programming is
done via the standardised JTAG interface1 which is also used to test the circuits.
To program the component all you need is a download program to transfer the JEDEC
file to the chip via a download cable. The download program for Lattice-products is named
ispVM. ispLEVER software contains the download program ispVM which uses the parallel
port (printer interface). The necessary connections are shown in Fig. 26.2.42.
A download cable can be ordered from Lattice under the order number HW-DL-3C and
is used to connect the PC parallel interface to the standardised JTAG plug on the printed
board of the PLD. As an extra measure, the Lattice cable also features integrated drivers
(74VHC244) to guarantee the correct levels at the PLD even in unfavourable conditions.
In most cases a simple passive cable like the one shown in Fig. 29.2.42 is sufficient. For
1 IEEE 1149.1 Boundary Scan Test Interface from the Joint Test Action Group (JTAG)
1486 29 Appendix
Selected device
Selected operation
the JTAG connection the download cable has a 10-pin socket like those commonly used
on interface cables for PC mainboards. The matching 10-pin plug is located on the printed
board of the PLD to be programmed. The names and meaning of the signals in the JTAG
interface are listed in Fig. 29.2.43. As seen in our example, the TRST and the ENABLE
signals are not required in many PLDs.
To program a device start the program ispVM (Fig 29.2.44).
– At first install the download-cable, insert the PLD and switch the power on for the board.
– Press the “Scan” button in Fig. 29.2.44. Now all PLDs in the chain of the circuit are
scanned and displayed in the right order.
– Double click on the Device in the list to be changed. Insert in the dialog the information
on the Device by choosing Select (resulting in the menu in Fig. 29.2.46), the Data File
(JEDEC-file) and programming operation as seen in Fig. 29.2.45.
29.2 ispLEVER – Brief User’s Guide 1487
– With Options/Cable and I/O Port Setup you select the PC port and the type of the
download-cable (Fig. 29.2.47).
– Start programming with “Go” from the main window (Fig. 29.2.44)
29.2.7
Outlook
We hope that the information provided in this chapter will enable you to design your own
circuits using ispLEVER and program them in a PLD. For those of you requiring extra or
additional information we suggest using the help menue in ispLEVER.
If a design is not only to be simulated but also to be tested in a circuit, the evaluation
board from Lattice is particularly useful. The provided configuration differs for the mounted
devices CPLDs and FPGAs.
1488 29 Appendix
29.3
Passiv RC and LRC Networks
RC Networks are of fundamental importance to circuit design. As their effect is the same
in all circuits, their operation will be described in some detail.
29.3.1
The Lowpass Filter
A lowpass filter is a circuit which passes low-frequency signals unchanged and attentuates
at high frequencies, introducing a phase lag. Figure 29.3.1 shows the simplest type of RC
lowpass filter circuit.
Frequency-domain analysis
To calculate the frequency response of the circuit, we use the voltage divider formula,
written in complex notation as:
Vo 1/(sC) 1
A(s) = = = (29.3.1)
Vi R + 1/(sC) 1 + sRC
Factoring according to
A = |A|ej ϕ
we obtain the frequency response of the absolute value or magnitude and of the phase shift:
1
|A| = √ , ϕ = − arctan ωRC (29.3.2)
1 + ω2 R 2 C 2
The two curves are shown in Fig. 29.3.2.
To calculate the 3 dB cutoff frequency fc , we substitute
1 1
|A| = √ = '
2 1 + ωg2 R 2 C 2
Vi Vo
Fig. 29.3.1. Simple lowpass filter Fig. 29.3.2. Bode plot of a lowpass filter
29.3 Passiv RC and LRC Networks 1489
ωg 1
fc = = (29.3.3)
2π 2πRC
Time-Domain Analysis
In order to analyze the circuit in the time domain, we apply a step function of voltage to
the input, as shown in Fig. 29.3.3. To calculate the output voltage, we apply Kirchhoff’s
current law to the (unloaded) output and obtain in accordance with Fig. 29.3.1
Vi − V o
− IC = 0
R
With IC = C V̇o , we obtain the differential equation
Vr for t > 0 in Case a
RC V̇o + Vo = Vi = (29.3.4)
0 for t > 0 in Case b
It has the following solutions
Case
% a: & Case b:
(29.3.5)
Vo (t) = Vr 1 − e−t/RC Vo (t) = Vr e−t/RC
This curve is also plotted in Fig. 29.3.3. We can see that the steady-state values Vo = Vr
or Vo = 0 are only attained asymptotically. As a measure of the response time, a time
constant τ is therefore defined. This indicates how long it takes for the deviation from the
Vi Vi
Vr Vr
Vo Vo
Vr Vr
t = RC t = RC
steady-state value to equal 1/e times the step magnitude. From (29.3.5) the time constant is
τ = RC (29.3.6)
The response time for smaller deviations can also be derived from (29.3.5). Figure 29.3.4
lists a number of important parameters.
If a square-wave voltage of period T is applied as the input signal, the e-function is
truncated after time T /2 by the subsequent step. Which final value is obtained at the output
depends on the ratio between the time T /2 and the time constant τ . This characteristic is
clearly illustrated by the oscillogram in Fig. 29.3.5.
Lowpass filter as an integrating circuit: In the previous section we saw that the alter-
nating output voltage is small compared with the input voltage if a signal frequency f fc
is selected. The lowpass filter operates then as an integrating circuit. This property can be
inferred directly from differential equation (29.3.4). Assuming that |Vo | |Vi |, it follows:
RC V̇o = Vi ,
t
1
Vo = Ue (t˜)d t˜ + Vo (0)
RC
0
T
1
Vi = Vi (t) dt
T
0
where T is the period of the input voltage. If all the higher-order terms of the Fourier series
are combined, a voltage Vi
(t) is obtained whose characteristic corresponds to that of the
input voltage, but which is displaced from zero such that its arithmetic mean is zero. The
input voltage may therefore be expressed in the form
29.3 Passiv RC and LRC Networks 1491
Vi (t) = V i + Vi
(t)
For voltage U ein
(t), the condition f fc can be satisfied; it is integrated, whereas the
DC component is transferred linearly. The output voltage therefore becomes
t
1
Vo = Vi
(t˜) d t˜ + Vi (29.3.7)
RC
0
3 45 6 3456
residual ripple mean value
Rise time and cutoff frequency: Another parameter for characterizing lowpass filters
is the rise time tr . This denotes the time taken for the output voltage to rise from 10 to 90%
of the final value when a step is applied to the input. From the e-function in (29.3.5)
tr = t90% − t10% = τ (ln 0.9 − ln 0.1) = τ ln 9 ≈ 2.2τ
Consequently, with fc = 1/2πτ
1
tr ≈ (29.3.9)
3fc
fci
fc ≈ √ (29.3.11)
n
29.3.2
The Highpass Filter
A highpass filter is a circuit which passes high-frequency signals unchanged and attenuates
at low frequencies, introducing a phase lead. Figure 29.3.6 shows the simplest form of RC
1492 29 Appendix
highpass filter circuit. The frequency response of the gain and phase shift is again obtained
from the voltage divider formula:
Vo R 1
A(s) = = = (29.3.12)
Vi R + 1/(sC) 1 + 1/(sRC)
This yields
1 1
|A| = and ϕ = arctan (29.3.13)
1 + 1/ω2 R 2 C 2 ωRC
The two curves are shown in Fig. 29.3.7. For the cutoff frequency, we obtain as with the
lowpass filter:
1
fc = (29.3.14)
2π RC
At this frequency the phase shift is +45◦ .
As in the case of the lowpass filter, the amplitude-frequency response can be easily
plotted on a double-logarithmic scale using the asymptotes:
1) At high frequences f fc , |A| = 1 =7 0 dB.
2) At low frequencies f fc , from (29.3.13) |A| ≈ ωRC, i.e. the gain is proportional to
√ of the asymptote is therefore +20 dB/decade or +6 dB/octave.
the frequency. The slope
3) For f = fc , |A| = 1/ 2 = 7 − 3 dB, as with the lowpass filter.
To calculate the step response, we apply Kirchhoff’s current law to the (unloaded)
output:
d Vo
C· (Vi − Vo ) − = 0 (29.3.15)
dt R
With V̇i = 0, this yields the differential equation
RC V̇o + Vo = 0 (29.3.16)
the solution of which is
Vi Vo
.
Fig. 29.3.6. Simple highpass filter Fig. 29.3.7. Bode plot of a highpass filter
29.3 Passiv RC and LRC Networks 1493
Vi Vi
Vr Vr
Vo Vo
Vr
t = RC
t = RC
Vr
t
Vo (t) = Vo0 e− RC (29.3.17)
The time constant is therefore τ = RC, as in the case of the lowpass filter.
In order to determine the initioal value Vo0 = Vo (t = 0), we have to consider that at the
instant when the input voltage changes abruptly, the capacitor charge remains unchanged.
The capacitor therefore acts as a voltage source of value V = Q/C. The output voltage
accordingly shows the same stem V as the input voltage. If Vi goes from zero to Vr , the
output voltage likewise jumps from zero to Vr , (see Fig. 29.3.8 a) then decays exponentially
to zero again in accordance with (29.3.17)
If the input voltage now goes abruptly from Vr to zero, Vo jumps from zero to −Vr
(see Fig. 29.3.8 b). Note that the output voltage assumes negative values even though the
input voltage is always positive. This distinctive characteristic is frequently used in circuit
design.
29.3.3
Compensated Voltage Divider
It is frequently the case that a resistive voltage divider is capacitively loaded, making it
a lowpass filter. The lower the resistance selected for the voltage divider, the higher the
cutoff frequency of the filter. However, limits are imposed in that the input resistance of
the divider should not be reduced below a specified value.
Another possible way of raising the cutoff frequency is to use a highpass filter to
compensate for the effect of the lowpass filter. This is the purpose of capacitor Ck in
Fig. 29.3.10. It is dimensioned such that the resultant parallel-connected capacitive voltage
divider has the same division ratio as the resistive voltage divider. Consequently, the same
voltage division is produced at high and low frequencies. This means that
Ck R2
=
CL R1
For optimum adjustment of Ck , it is useful to test with the step function. For the correct
value Ck , the step response becomes ideal.
Vi
Vo
29.3.4
Passive RC Bandpass Filter
By connecting a highpass and a lowpass filter in series, we obtain a bandpass filter whose
output voltage is zero for high and low frequencies. One widely used combination is shown
in Fig. 29.3.11. We shall now calculate the output voltage at medium frequencies and the
phase shifts introduced. In complex notation, the formula for the unloaded voltage divider
yields:
1
1
+ sC
Vo R sRC
= =
Vi 1 1 1 + 3sRC + s 2 R 2 C 2
+R+
1 sC
+ sC
R
Simplifying with sn = sRC, we obtain for the gain
Vo sn
A(sn ) = = (29.3.20)
Vi 1 + 3sn + sn2
1 1 − ωn2
|A| = 2 , ϕ = arctan (29.3.21)
1 3ωn
− ωn +9
ωn
1
fr = (29.3.22)
2π RC
The quantity ωn initially introduced for simplification expresses the normalized frequency
ω f
ωn = =
ωr fr
The phase shift at resonance is zero and the gain Ar = 13 . The frequency response of |A|
and ϕ is shown in Fig. 29.3.12.
Vi
Vo
29.3.5
Wien–Robinson Bridge
If the bandpass filter shown in Fig. 29.3.11 is modified by inserting resistors R1 and 2R1 , as
shown in Fig. 29.3.13, a Wien–Robinson bridge is obtained. The resistive voltage divider
furnishes the voltage 13 Vi irrespective of frequency. At the resonant frequency, the output
voltage is therefore zero. Unlike the bandpass filter response, the frequency response of the
Wien–Robinson bridge is then minimum. The circuit is therefore useful for suppressing a
given frequency band. The output voltage can be calculated from (29.3.20):
Vo 1 sn
= −
Vi 3 1 + 3sn + sn2
Hence
1 1 + sn2
A(sn ) = · (29.3.23)
3 1 + 3sn + sn2
Vi
Vo
29.3.6
Parallel-T Filter
The parallel-T filter in Fig.29.3.15 has a very similar frequency response to that of the Wien–
Robinson bridge and is therefore also suitable for suppressing a given range of frequencies.
However, it differs from the Wien–Robinson bridge in that the output voltage can be
measured with respect to ground. For high and low frequencies Vo = Vi . High frequencies
are transferred without attenuation via the two capacitors C and low frequencies via the
two resistors R.
To calculate the frequency response we apply Kirchhoff‘s current law to points 1, 2
and 3 in Fig. 29.3.15 and obtain, for an unloaded output:
Vi − V1 V − V1
Node 1: + o − V1 · 2sC = 0
R R
2V
Node 2: (Vi − V2 )sC + (Vo − V2 )sC − 2 = 0
R
V1
Vi V2 Vo
V1 − Vo
Node 3: (V2 − Vo )sC + = 0.
R
Thus, be eliminating V1 and V2 and with the normalization sn = sRC, we obtain for
the gain:
1 + sn2
A(sn ) = (29.3.24)
1 + 4sn + sn2
|1 − ωn2 | 4ωn
|A| = , ϕ = arctan
(1 − ωn2 )2 + 16ωn2 ωn2 − 1
The two curves have been plotten in Fig. 29.3.16. It can be seen that the gain here is also
zero at the resonant frequency fr = 1/(2πRC). For low and high frequencies the gain is
|A| = 1; the Q-factor here is even lower with Q = 41 .
29.3.7
Resonant Circuit
Series resonant circuit Parallel resonant circuit
Impedance:
1 R + sL
Z = R + sL + Z =
sC 1 + sRC + s 2 LC
sL
Z≈
1 + sRC + s 2 LC
Resonant frequency:
1 1
fr = √ fr ≈ √
2π LC 2π LC
29.3 Passiv RC and LRC Networks 1499
Quality factor:
- -
fr 1 L fr 1 L
Q = = Q = ≈
B R C B R C
Fig. 29.3.17a. Series resonant circuit Fig. 29.3.17b. Parallel resonant circuit
29.4
Definitions and Nomenclature
We hope that the following list of definitions will help to avoid confusion and enable a
better understanding. Where possible, the definitions are based on IEC recommendations
Voltage. A voltage between two points x and y is denoted by Vxy . It is defined as being
positive if point x is positive with respect to point y, and negative, if point x is negative
with respect to point y. Therefore, Vxy = −Vyx . The statement
VBE = − 5 V or
−VBE = 5 V or
VEB = 5V
thus indicates that there is a voltage of 5 V between E and B where E is positive with
respect to B. In a circuit diagram, the double indices are often omitted and the notation
Vxy is replaced by a voltage arrow V pointing from node x to node y
Potential The Potential V is the voltage of a node with relation to a common reference
node 0 or ground:
Vx = Vx0
In electrical circuits, the referenc epotential is denoted by a ground symbol. Often Vx is
used when actually implying Vx . Specialists then speak, although not quite correctly, of
the voltage of a node, e.g. the anode voltage. For the voltage between two nodes, x and y,
Vxy = Vx − Vy
Current. The current is indicated by a current arrow on the connecting line. One defines
the current I as being positive if the current in its conventional sense, i.e. the transport
of positive charge, flows in the direction of the arrow. I is thus positive if the arrow of
the current flowing through a load points from the larger to the smaller potential. The
directions of the current and voltage arrows in a circuit diagram are not important as long
as the actual values of V and I are given the correct signs.
If current and voltage arrow of a circuit element have the same direction, Ohm’s law,
with the above definitions, is R = V /I . If they have opposite directions it changes to
R = −V /I . This fact is illustrated in Fig. 29.4.1.
Resistance. If the resistance is voltage- or current-dependent, a static resistance R = V /I
and an incremental resistance r = ∂V /∂I ≈ V /I can be defined. These formulas are
valid if the voltage and current arrows point in the same direction. If the directions are
opposed, a negative sign must be inserted, as in Fig. 29.4.1.
Voltage and current source. A real voltage source can be described by the equation
Vo = V0 − Rint Io (29.4.1)
V V
V V
Fig. 29.4.1. Ohm’s law
29.4 Definitions and Nomenclature 1501
Io
V0 Vo Vo
Fig. 29.4.2. Equivalent circuit of a real Fig. 29.4.3. Equivalent circuit of a real
voltage source current source
where V0 is the no-load voltage and Rint = −dVo /dIo the internal resistance. This is
represented by the equivalent circuit in Fig. 29.4.2. An ideal voltage source is characterized
by the property Rint = 0, i.e. the output voltage is independent of the current.
A different equavalent circuit for a real voltage source can be deduced by rewriting
(29.4.1):
V0 − Vo Vo
Io = = Isc − (29.4.2)
Rint Rint
where Isc = V0 /Rint is the short-circuit current. The appropriate circuit is shown in
Fig. 29.4.3. It is obvious that, the large Rint , the less the output current depends on the
output voltage. For Rint → ∞, one obtains an ideal current source.
According to Figs. 29.4.2 and 29.4.3, a real voltage source can be represented either
by an ideal voltage source or by an ideal current source. Which representation is chosen
depends on whether the internal resistance Rint is small or large in comparison with the
load resistance RL
Kirchhoff’s current law (KCL). For the calculation of the parameters of many electronic
circuits, we use Kirchhoff’s current law. It states that the sum of all currents flowing into
a node is zero. Currents fldue toward the node are counted as being positive, and currents
flowing from the node are negative. Figure 29.4.4 demonstrates this fact. It can be seen
that, for Node N
0
Ii = I1 + I2 − I3 = 0
i
V3
V1 V2
V1
V2
V3
V4
Kirchhoff’s voltage law (KVL). Kirchoff’s voltage law states that the sum of all voltages
around any loop in an electrical network is zero. The voltage is entered in the appropriate
equation with a positive sign if its arrow points in the direction in which one proceeds
around the loop; the voltage is entered with a negative sign if the voltage arrow points
against this direction. For the example in Fig. 29.4.5
0
Vi = V1 + V4 − V2 − V3 = 0
i
AC circuits (alternating-current circuits). If the circuit can be described by a DC (direct-
current) transfer characteristic Vo = f (Vi ), this relationship necessarily also holds for any
time-dependent voltage, i.e. Vo (t) = f [Vi (t)], as long as the changes in the input voltage
are quasi stationary, i.e. not too fast. For this reason, we use upper-case letters for DC as
well as for time-dependent quantities, e.g. V = V (t).
However, cases exist wehre a transfer characteristic is only valid for alternating voltages
without DC components and it is therefore sensible to have a special symbol to dinstinguish
such alternating voltages. We use the lower-case letter u to denote their instantaneous
values.
A particularly important special case is that of sinusoidally alternating voltages, i.e.
v(t) = V 7 cos(ωt + ϕu ) (29.4.3)
7
where V is the peak value (amplitude). Other values for the
√caracterization of the voltage
magnitude are the root-mean-square value Veff = V 7/ 2 or the peak-to-peak value
VSS = 2 V7.
The calculus for trigonometric functions is rather involved, but that for exponential
functions fairly simple. Euler’s theorem
ej α = cos α + j sin α (29.4.4)
enables a sine function to be expressed by the imaginary part of the complex exponential
function
sin α = Im{ej α }
Equation (29.4.3) can therefore also be written as
v=V 7 · Im{ej tωt+ϕu } = Im{U7ej ϕu ej ωt = Im{U
7ej ωt }
where V = V7ej ϕv is the complex amplitude. Its magnitude is given by
'
7 7
|V| = V · |e | = V cos2 ϕv + sin2 ϕv = V
j ϕv 7
29.4 Definitions and Nomenclature 1503
i.e. it is equal to the peak value of the sine wave. Time-dependent currents are treated in
an analogous way. The corresponding symbols are then
I, I (t), i, Iˆ, I
Arrows can also be assigned to alternating voltages and currents. Of course, the direction of
the arrow then no longer indicates polarity but denotes the mathematical sign with which
the values must be entered in the formulas: The rule illustratede in Fig. 29.4.2 for DC
voltages also applies in this case.
In analogy to the resistance in a DC circuit, a complex resistance is defined as the
impedance Z
V 7ej ϕu
V 7
V
Z = = = ej (ϕu −ϕi ) = |Z|ej ϕ
I I7ej ϕi I7
where ϕ is the phase angle between current and voltage. If the voltage is leading with respect
to the current, ϕ is positive. For a purely ohmic resistance, Z = R; for a capacitance
1 j
Z = =−
j ωC ωC
and for an inductance Z = j ωL. Ths laws for the DC circuit quantities can be applied to
complex quantities as well
One can also define a complex gain
Vo 7o ej ϕa
V 7o j (ϕ −ϕ )
V
A = = = e a e = |A|ej ϕ
Vi 7
Vi e j ϕe 7
Vi
where ϕ is the phase angle between input and output voltage. If the output voltage is
leading with respect to the input voltage, ϕ is positive; for a lagging output voltage, it is
negative.
Power. There are different definitions of power. Their relationship is summarized here:
The instantaneous power is defined as
p(t) = v(t) · i(t) . (29.4.5)
The effective power is the mean instantaneous power. It is found by averaging over one
period:
T
1
P = v(t) · i(t) dt . (29.4.6)
T
0
For a cosine-form time behavior, this gives us
T
1
p(t) = V̂ cos(ωt + ϕu ) · Iˆ cos(ωt + ϕi ) dt
T (29.4.7)
0
= 1
2 V̂ Iˆ cos(ϕu − ϕi ) = Veff Ieff cos(ϕv − ϕi ) .
The apparent power is the effective power for the case that ϕv − ϕi would occur.
1 ˆ
S = V̂ I = Veff Ieff . (29.4.8)
2
1504 29 Appendix
The reactive power is obtained from the equation which gives the effective power by
substituting the angle ϕi − 90◦ for the angle ϕi .
T
1
Q(t) = V̂ cos(ωt + ϕv ) · Iˆ cos(ωt + ϕi − 90◦ ) dt
T (29.4.9)
0
= 1
2 V̂ Iˆ sin(ϕv − ϕi ) = Veff Ieff sin(ϕu − ϕi ) .
This results in the relationship
P 2 + Q2 = S 2 . (29.4.10)
The power factor is a measure of how large the effective power component is:
P (29.4.10) P
PF = = = cos(ϕv − ϕi ) = cos ϕ . (29.4.11)
S P2 + Q2
The maximum power factor is given for Q = 0; the condition PF = 1 then applies, and
the effective power is equal to the apparent power.
The relationship between the different power parameters can be made clear very easily
with the use of phasors: from the definition of the apparent power
1 ∗
S = VI = Veff I∗eff (29.4.12)
2
√ √ √
it follows with I = 2 Ieff ej ϕi and I∗ = 2 Ieff e−j ϕi and V = 2 Veff ej ϕv that
S = Veff Ieff ej (ϕu −ϕi ) = Veff Ieff cos(ϕv − ϕi ) + j Veff Ieff sin(ϕv − ϕi ) = P + j Q .
Fig. 29.4.7. Table for conversion of Fig. 29.4.8. Conversions of absolute power
voltage ratios: A[dB] = 20 dB lg A P
P[dBm] = 10 dBm lg 1 mW = 20 dBm lg 0.224 V
V
The symbol ∼ represents a proportional relationship; the symbol ≈ stands for approxi-
mately equal to; the symbol = ˆ means corresponding to. The symbol || means parallel. We
use it to indicate that resistors are connected in parallel, i.e.
R1 R2
R1 ||R2 =
R1 + R 2
29.5
Types of the 7400 Digital Families
The importance of primitive logic circuits has decreased since most functions are realized
by PLDs and FPGAs today. Therefore the number of manufacturers and types has decreased
also. Some manufacturers that produce the 7400 family until today are Fairchild, National;
On-Semiconductor; Philips, ST-Microelectronics and naturally Texas Instruments.
29.6
Standard Series
E3 E6 E 12 E 24 E 48 E 96 E3 E6 E 12 E 24 E 48 E 96
±20% ±20% ±10% ±5% ±2% ±1% ±20% ±20% ±10% ±5% ±2% ±1%
1.0 1.0 1.0 1.0 1.00 1.00 3.3 3.3 3.3 3.32 3.32
1.02 3.40
1.05 1.05 3.48 3.48
1.07 3.57
1.1 1.10 1.10 3.6 3.65 3.65
1.13 3.74
1.15 1.15 3.83 3.83
1.18 3.9 3.9 3.92
1.2 1.2 1.21 1.21 4.02 4.02
1.24 4.12
1.27 1.27 4.22 4.22
1.3 1.30 4.3 4.32
1.33 1.33 4.42 4.42
1.37 4.53
1.40 1.40 4.64 4.64
1.43 4.7 4.7 4.7 4.7 4.75
1.47 1.47 4.87 4.87
1.5 1.5 1.5 1.50 4.99
1.54 1.54 5.1 5.11 5.11
1.58 5.23
1.6 1.62 1.62 5.36 5.36
1.65 5.49
1.69 1.69 5.6 5.6 5.62 5.62
1.74 5.76
1.78 1.78 5.90 5.90
1.8 1.8 1.82 6.04
1.87 1.87 6.2 6.19 6.19
1.91 6.34
1.96 1.96 6.49 6.49
2.0 2.00 6.65
2.05 2.05 6.8 6.8 6.8 6.81 6.81
2.10 6.98
2.15 2.15 7.15 7.15
2.2 2.2 2.2 2.2 2.21 7.32
2.26 2.26 7.5 7.50 7.50
2.32 7.68
2.37 2.37 7.87 7.87
2.4 2.43 8.06
2.49 2.49 8.2 8.2 8.25 8.25
2.55 8.45
2.61 2.61 8.66 8.66
2.67 8.87
2.7 2.7 2.74 2.74 9.1 9.09 9.09
2.80 9.31
2.87 2.87 9.53 9.53
2.94 9.76
3.0 3.01 3.01
3.09
3.16 3.16
3.24
29.7
Color code
none ± 20 %
silver ± 10 %
gold ± 5%
black 0 ± 20 %
brown 1 1 ± 1%
red 2 2 ± 2%
orange 3 3
yellow 4 4
green 5 5
blue 6 6
violet 7 7
grey 8 8
white 9 9
silver
gold ± 5 %
red 2 2 2 ± 2 % ± 50 ppm/K
orange 3 3 3 ± 15 ppm/K
yellow 4 4 4 ± 25 ppm/K
blue 6 6 6 ± 10 ppm/K
violet 7 7 7 ± 5 ppm/K
grey 8 8 8 ± 1 ppm/K
white 9 9 9
29.8
Manufacturers
Here are listed the most important manufacturers for semiconductors and integrated cir-
cuits. The first line gives the short form notation, the second line specifies the internet
address of the homepage, the third line contains the most important products. Some older
manufacturers that are in the meantime transferred to other firms are also mentioned to
enable the reader to find older components.
Actel
http://www.actel.com/
FPGAs
Agilent
http://www.semiconductor.agilent.com/
Optoelectronics, RF-semiconductors
Allegro
http://www.allegromicro.com/
Power-drivers, Hall-Sensors
Altera
http://www.altera.com/
PLDs, FPGAs
AMD, Advanced Micro Devices
http://www.amd.com/
CPUs for PCs, Flash-memory
Analog Devices
http://www.analog.com/
OPAmps, AD-DA-converters, Signalprocessors, sensors
Apex
http://www.apexmicrotech.com/
Power-OPAmps, PWM-amplifiers
Atmel
http://www.atmel.com/
Flash memories, SRAMs, microcontroller
Benchmarq ⇒ Texas Instruments
Burr Brown ⇒ Texas Instruments
Cherry Semiconductor ⇒ ON Semiconductor
Coilcraft
http://www.coilcraft.com
Inductors
Cypress
http://www.cypressmicro.com/
SRAMs, PLDs
29.8 Manufacturers 1519
Dallas ⇒ Maxim
Datel
http://www.datel.com/
AD-DA-converter
Elantec ⇒ Intersil
Exar
http://www.exar.com/
Line Transceivers, UARTs
Fairchild
http://www.fairchildsemi.com/
Power mosfets, EEPROMs, 7400-Logic
Freescale
http://www.freescale.com/
Microcontrollers, DSPs, sensors
Fuji
http://www.fujisemiconductor.com/
Power mosfets and IGBTs, power modules
Fujitsu
http://www.fujitsumicro.com/
Memories, microcontrollers
General Electric
http://www.gesensing.com/
Sensors
Harris ⇒ Intersil
Honeywell
http://content.honeywell.com
Sensors
Hitachi
http://www.halsp.hitachi.com/
Memories, microcontrollers
Honeywell
www.honeywell.com/sensing
Sensors
IDT
http://www.idt.com/
SRAMs, Fifos, dual port memories
Infineon
http://www.infineon.com/
Power mosfets, memories, optoelectronics
Inmos ⇒ ST Microelectronics
1520 29 Appendix
Intel
http://www.intel.com/
CPUs for PCs, flash memories
Intermetall ⇒ Visahy
International Rectifier
http://www.irf.com/
Power mosfets, IGBTs, diodes
Intersil
http://www.intersil.com/
AD-DA-converters, OPAmps, MOS-drivers
IXYS
http://www.ixys.com/
Power mosfets, IGBTs, Diodes
Lattice
http://www.latticesemi.com/
PLDs, FPGAs
LEM
http://www.lem.com
Current sensors
Linear Technology
http://www.linear.com/
OPAmps, voltage regulators, SC-filters, AD-DA-converters, sensors
Maxim
http://www.maxim-ic.com/
OPAmps, voltage regulators, SC-filters, AD-DA-converters
Memories, sensors
Microchip
http://www.microchip.com/
Microcontrollers
Micron
http://www.micron.com/
DRAMs, flash memories
Microsemi
http://www.microsemi.com/
Diodes, transistors, voltage regulators, RF semiconductors
Mini-Circuits
http://www.minicircuits.com/
RF-Amplifiers, -oscillators, -mixer, -transformers, -attenuators
MIPS
http://www.mips.com/
IPs for Signal processors
29.8 Manufacturers 1521
Mitsubishi
http://www.mitsubishichips.com/
SRAM, DRAM, flash-memories
Monolithic Memories (MMI) ⇒ AMD
Motorola ⇒ Freescale, On Semiconductor
Murata
http://www.murata.com/
Sensors, passive components
National
http://www.national.com/
OPAmps, voltage regulators, AD-DA-converters
NEC
http://www.necel.com/
Microcontrollers, AD-DA-converter, MOS-power transistors, optoelectronics
Novasensor ⇒ General Electric
Oki
http://www.okisemi.com/
SRAMs. DRAMs, microcontrollers
Omega
http://www.omega.com
Temperature sensors
ON Semiconductor
http://www.onsemi.com/
Analog ICs, 7400 logic, ECL logic, discrete semiconductors
Optek
http://www.optekinc.com
Optoelectronics
Osram
http://www.osram-os.com/
Optoelectronics, LEDs
Philips
http://www.semiconductors.philips.com/
Microcontrollers, 74oo logic, sensors
Power Integrations
http://www.powerint.com/
Off-line switchers
Pulse
http://www.pulseeng.com
Inductors, pulse transformers, RF transformers
1522 29 Appendix
QuickLogic
http://www.quicklogic.com/
FPGAs
Raytheon ⇒ Fairchild
RCA ⇒ Intersil
Renesas
http://www.renesas.com/
SRAMs, flash-memories
Rohm
http://www.rohm.com/
Liner ICs, CMOS logic, transistors, resistors
Samsung
http://www.samsungsemi.com/
SRAM, DRAM, Flash-memories, microprocessors
Semikron
http://www.semikron.com/
Power modules
SensorTechnics
http://www.sensortechnics.com/
Pressure sensors
SGS-Thomson ⇒ ST Microelectronics
Signetics ⇒ Philips, ON semiconductor
Siemens ⇒ Infineon
Siliconix ⇒ Vishay
Sharp
http://www.sharp-sme.com/
DRAM, SRAM, Flash-memories, microcontrollers, optoelectronics
Silicon Systems ⇒ Texas Instruments
Sipex
http://www.sipex.com/
Power management
Silicon General ⇒ Microsemi
Sony
http://www.sony.com/semi
AD-DA-converters, optoelectronics, microwave ICs
Spansion
http://www.spansion.com/
Flash memories
29.8 Manufacturers 1523
Sprague ⇒ Allegro
SPT – Signal Processing Technologies ⇒ Fairchild
ST Microelectronics
http://www.st.com/
Memories, microcontrollers, power semiconductors
Supertex
http://www.supertex.com/
High voltage MOS components
Telefunken ⇒ Vishay
Temic ⇒ Vishay
Texas Instruments
http://www.ti.com/
OPAmps, AD-DA-converters, signal processors, 7400 logic
Thomson ⇒ ST Microelectronics
Toshiba
http://www.toshiba.com/taec/
DRAMs, SRAMs, Flash-memories, microcontrollers, 7400 CMOS logic
Triquint
http://www.triquint.com/
GaAs-semiconductors
Unitrode ⇒ Texas Instruments
Valvo ⇒ Philips
Vantis ⇒ Lattice
Vishay
http://www.vishay.com/
Mosfets, voltage regulators, diodes, Z-diodes, bipolartransistors, optoelectronics
Waferscale ⇒ ST Microelectronics
Xicor ⇒ Intersil
Xilinx
http://www.xilinx.com/
CPLDs, FPGAs
Zilog
http://www.zilog.com/
Microprozessors, signal processors
Bibliography
Chapter 1:
[1.1] Sze, S.M.: Physics of Semiconductor Devices, 2nd Edition. New York: John Wiley & Sons,
1981.
[1.2] Hoffmann, K.: VLSI-Entwurf. München: R. Oldenbourg, 1990.
[1.3] Löcherer, K.-H.: Halbleiterbauelemente. Stuttgart: B.G. Teubner, 1992.
[1.4] MicroSim: PSpice A/D Reference Manual.
[1.5] Antognetti, P.; Massobrio, G.: Semiconductor Device Modeling with SPICE. New York:
McGraw-Hill, 1988.
[1.6] Zinke, O.; Brunswig, H.; Hartnagel, H.L.: Lehrbuch der Hochfrequenztechnik, vol. 2, 3rd
Edition. Berlin: Springer, 1987.
[1.7] Bauer, W.: Bauelemente und Grundschaltungen der Elektronik, 3rd Edition. Müchen: Carl
Hanser, 1989.
[1.8] Kesel, K.; Hammerschmitt, J.; Lange, E.: Signalverarbeitende Dioden. Halbleiter-
Elektronik vol. 8. Berlin: Springer, 1982.
[1.9] Mini-Circuits: Frequency Mixers, datasheets.
Chapter 2:
[2.1] Gray, P.R.; Meyer, R.G.: Analysis and Design of Analog Integrated Circuits, 2nd Edition.
New York: John Wiley & Sons, 1984.
[2.2] Sze, S.M.: Physics of Semiconductor Devices, 2nd Edition. New York: John Wiley & Sons,
1981.
[2.3] Rein, H.-M.; Ranfft, R.: Integrierte Bipolarschaltungen. Halbleiter-Elektronik vol. 13.
Berlin: Springer, 1980.
[2.4] Antognetti, P.; Massobrio, G.: Semiconductor Device Modeling with SPICE. New York:
McGraw-Hill, 1988.
[2.5] Getreu, I.: Modeling the Bipolar Transistor. Amsterdam: Elsevier, 1978.
[2.6] MicroSim: PSpice A/D Reference Manual.
[2.7] Hoffmann, K.: VLSI-Entwurf. München: R. Oldenbourg, 1990.
[2.8] Schrenk, H.: Bipolare Transistoren. Halbleiter-Elektronik vol. 6. Berlin: Springer, 1978.
[2.9] Müller, R.: Rauschen. Halbleiter-Elektronik vol. 15. Berlin: Springer, 1979.
[2.10] Motchenbacher, C.D.; Fitchen, F.C.: Low-Noise Electronic Design. New York: John Wiley
& Sons, 1973.
[2.11] Thorton, R.D.; Searle, C.L.; Pederson, D.O.;Adler, R.B.;Angelo, E.J.: Multistage Transistor
Circuits. Semiconductor Electronics Education Committee, vol. 5. New York: John Wiley
& Sons, 1965.
1526 Bibliography
Chapter 3:
[3.1] Sze, S.M.: Physics of Semiconductor Devices, 2nd Edition. New York: John Wiley & Sons,
1981.
[3.2] Hoffmann, K.: VLSI-Entwurf. München: R. Oldenbourg, 1990.
[3.3] Antognetti, P.; Massobrio, G.: Semiconductor Device Modeling with SPICE. New York:
McGraw-Hill, 1988.
[3.4] Spenke, E.: pn-Übergänge. Halbleiter-Elektronik vol. 5. Berlin: Springer, 1979.
[3.5] MicroSim: PSpice A/D Reference Manual.
[3.6] Müller, R.: Rauschen. Halbleiter-Elektronik vol. 15. Berlin: Springer, 1990.
Chapter 4:
[4.1] Gray, P.R.; Meyer, R.G.: Analysis and Design of Analog Integrated Circuits, 2nd Edition.
New York: John Wiley & Sons, 1984.
[4.2] Geiger, L.G.; Allen, P.E.; Strader, N.R.: VLSI – Design Techniques for Analog and Digital
Circuits. New York: McGraw-Hill, 1990.
[4.3] Antognetti, P.; Massobrio, G.: Semiconductor Device Modeling with SPICE. New York:
McGraw-Hill, 1988.
[4.4] Weiner, D.D.; Spina, J.F.: Sinusoidal Analysis and Modeling of Weakly Nonlinear Circuits.
New York: Van Nostrand, 1980.
[4.5] Maas, S.A.: Nonlinear Microwave Circuits. Norwood: Artech House, 1988
[4.6] Motchenbacher, C.D.; Fitchen, F.C.: Low-Noise Electronic Design. New York: John Wiley
& Sons, 1973.
[4.7] Müller, R.: Rauschen. Halbleiter-Elektronik vol. 15. Berlin: Springer, 1979.
[4.8] Haus, H.A.; Adler, R.B.: Circuit theory of noisy networks. New York: John Wiley & Sons,
1959.
[4.9] Vanisri, T.; Toumazou, C.: Integrated high frequency low-noise current-mode optical tran-
simpedance preamplifiers: theory and practice. IEEE Journal of solid state circuits, vol. 30,
no. 6, June 1995, p. 677.
Chapter 24:
[24.1] Zinke, O.; Brunswig, H.: Lehrbuch der Hochfrequenztechnik. vol. 1, 4th Edition. Berlin:
Springer, 1990.
[24.2] Ebeling, K.J.: Integrierte Optoelektronik. 2nd Edition. Berlin: Springer, 1992.
[24.3] Grau, G.; Freude, W.: Optische Nachrichtentechnik. 3rd Edition. Berlin: Springer, 1991.
[24.4] Weinert, A.: Kunststofflichtwellenleiter. Erlangen: Publicis MCD, 1998.
[24.5] Pehl, E.: Digitale und analoge Nachrichtenübertragung. Heidelberg: Hüthig, 1998.
[24.6] Huber, J.: Digitale Übertragung I & II. Script of a lecture. University Erlangen-Nürnberg,
Lehrstuhl für Nachrichtentechnik II, 1999.
[24.7] Lee, J.S.; Miller, L.E.: CDMA Systems Engineering Handbook. Boston: Artech House,
1998.
Bibliography 1527
Chapter 25:
[25.1] Pettai, R.: Noise in Receiving Systems. New York: John Wiley & Sons, 1984.
[25.2] Huber, J.: Digitale Übertragung I & II. Script of a lecture. University Erlangen-Nürnberg,
Lehrstuhl für Nachrichtentechnik II, 1999.
Chapter 26:
[26.1] Zinke, O.; Brunswig, H.: Lehrbuch der Hochfrequenztechnik. vol. 1, 4th Edition. Berlin:
Springer, 1990.
[26.2] Coilcraft: SMD Inductors 1206 CS and 1812 CS, datasheets.
[26.3] Saal, R.: Handbuch zum Filterentwurf. 2nd Edition. Heidelberg: Hüthig, 1988.
[26.4] Toko: Chip Dielectric Filters, datasheets.
[26.5] Sawtek: SAW Filters, datasheets.
[26.6] Kupferschmidt, K.H.: Die Dimensionierung des π-Filters zur Resonanztransformation. Fre-
quenz 24, 1970, pp. 215–218.
[26.7] Larson, L.E.: RF and Microwave Circuit Design for Wireless Communications. Boston:
Artech House, 1996.
Chapter 27:
[27.1] Zinke, O.; Brunswig, H.: Lehrbuch der Hochfrequenztechnik. vol. 1, 4th Edition. Berlin:
Springer, 1990.
[27.2] Hewlett Packard: S-Parameter Design. Application Note 154.
Chapter 28:
[28.1] Meinke, Gundlach: Taschenbuch der Hochfrequenztechnik. 5th Edition. Berlin: Springer,
1992.
[28.2] Mini-Circuits: Frequency Mixers, datasheets.
Index
– mirror with cascode 294 – amplifier 327, 349, 476, 557, 624, 748,
– on demand 508, 554 1405
– source 277, 471, 1501 – amplifier with a current mirror 356
– source bank 287 – amplifier with cascode current sources 353
current-controlled – amplifier with current feedback 335, 339,
– current source 778 359
– voltage source 768 – amplifier with simple current sources 351
current-time area 11 – amplifier with symmetric output 340
current-to-voltage converter 113, 149 – discriminator 1208
cutoff – equation 738
– current 47 – gain 329, 492, 525
– frequency 81, 122, 318, 512, 1488 – nonlinearity 984
– region 33 – output amplifiers 583
cycle stealing 696, 697 – pressure sensor 1078
– quadri-phase shift keying (DQPSK) 1218
– resistance 10, 22
D flip-flops 660
– signaling 634
damping 571
– switch 937
Darlington
– equation oscillator 857
– circuit 159, 387
differentiating 1493
– transistor 159, 166
differentiator 735, 998
data
diffusion
– error 705 – capacitance 20, 70
– lockout 663 – current 16
– rate 1216 – range 16
– retention 694 – voltage 17
DC digit at a time 966
– coupling 121, 138 digital 587
– sweep 1440 – analog converter 945
DC-coupled 483 – channel filter 1267
DC/DC converter 908 – demodulator 1213
De Morgans 612 – families 1508
debouncing 678 – filter 980, 987
decoder 643 – gain control 1258
DECT 1168 – modulation 1209, 1243
deglitchers 966 – modulator 1212
delay – sample and hold 968
– element 988, 989, 993 – signal processor (DSPs) 1028
– time 589 – transfer function 990
– transmitter 1243
– converter 980 – voltmeter 983
– modulator 980 diode 3, 275
demodulation 1056, 1199, 1206 – substrate 200
demodulator 1056, 1149, 1213 – bridge 934
demultiplexer 643, 644, 930 – mixer 1381, 1395
dependency notation 686 – switch 933
depletion 169 – transistor logic (DTL) 620
device fitter 717, 720 Dirac impulse 945
diamond transistor 486, 554 direct
diaphragm 1077 – conversion receiver 1277
dielectric – feedback 562
– constant 176 – modulation 1237
– filter 1290, 1296 direct-detection receiver 1246
– insulation 932 direct-detector 1246
die 13 directivity 1164
difference equation 990 discrete transistor 55
differential discriminator 1206
Index 1533
– SAW 1240, 1289, 1298 – response 156, 245, 318, 369, 550, 806, 993
– single feedback 811 – shift keying (FSK) 1193, 1210
finite impulse response 1222 – tuning 29
– filter (FIR) 994, 1013, 1298 frequency-sensitive phase detector 1120
first harmonic wave 428 full duplex operation 1227
first-in-first-out memories (FIFO) 700 full-adder 651
fixed-point 656 full-scale error 963
– binary number 640 full-wave rectifier 1042
flash fully compensated 517
– converter 969 function
– method 966 – generator 859, 961
flip-flop 591, 597, 659 – generator memories 689
floating – generator network 739, 750
– current source 777 fundamental wave 428
– gate 706 fuse map 716
– gate driver 927
– ground 1037 GaAs 1321
– load 769 gain 806, 1095, 1446
– point 658 – bandwidth 572
– point binary numbers 640 – bandwidth product (GBW) 124, 246, 321,
– point formats 641 371, 515
– point numbers 655 – control 1253, 1257
– voltage 898 – crossover frequency 1105
flourescent lamp 1128 – expansion 432
flow – margin 511
– diagram 684 gallium-arsenide 1321
– velocity 1061 galvanic coupling 121, 240, 266, 1252
flyback gate 169
– controller 927 – array 689, 719
– converter 908, 916 GBW = Gain BandWidth product 321, 371,
foldback current limiting 894 536
folded cascode 364 Gilbert
force 1061 – cell 756
form factor 1047 – mixer 1404, 1417
forward glass fibre 1168
– converter 908, 918 glitch 964
– gain 417 global adder 992, 1026
– recovery 9 gradient fibre 1171
– region 33 gray code 1216
– transconductance 417 ground plane 570
– voltage 4 grounded loads 771
– wave 1155, 1174 group delay 796, 806, 827, 828, 990, 996,
four-quadrant 1029, 1291
– multiplication 960 GSM 1168
– multiplier 755, 761 Gummel–Poon
– operation 878 – model 58, 70, 71
four-wire measurement 1091 – plot 37
Fourier Analysis 1450 gyrator 559, 781
FPGA 1461
frequency 1167 H matrix 44
– compensation 509, 512 half bridge driver 928
– division duplex operation (FDD) 1235 half duplex operation 1227
– division multiplex 1227 half-adder 649, 650
– domain 988, 1488 half-band filter 1005, 1007
– hopping 1229 half-bridge configuration 919
– modulation 1191, 1202 half-wave rectifier 886, 1043
– multiplier 1123 hall-effect 1060
Index 1535