Designing Dynamic Circuit Response, Volume 2 - Analog Circuit Design Series (PDFDrive)

Analog Circuit Design Series Volume 2
Designing Dynamic
Circuit Response
D. Feucht
Innovatia Laboratories
Raleigh, NC.
Published by SciTech Publishing, Inc.
911 Paverstone Drive, Suite B
Raleigh, NC 27615
(919) 847-2434, fax (919) 847-2568
scitechpublishing.com
Copyright © 2010 by Dennis Feucht. All rights reserved.
No part of this publication may be reproduced, stored in a retrieval system or transmitted in any
form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise,
except as permitted under Sections 107 or 108 of the 1976 United Stated Copyright Act, without
either the prior written permission of the Publisher, or authorization through payment of the
appropriate per-copy fee to the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA
01923, (978) 750-8400, fax (978) 646-8600, or on the web at copyright.com. Requests to the
Publisher for permission should be addressed to the Publisher, SciTech Publishing, Inc., 911
Paverstone Drive, Suite B, Raleigh, NC 27615, (919) 847-2434, fax (919) 847-2568, or
email editor@scitechpub.com.
The publisher and the author make no representations or warranties with respect to the accuracy
or completeness of the contents of this work and specifically disclaim all warranties, including
without limitation warranties of fitness for a particular purpose.
Editor: Dudley R. Kay

Production Manager: Robert Lawless
Typesetting: SNP Best-set Typesetter Ltd., Hong Kong
Cover Design: Aaron Lawhon
Printer: Docusource
This book is available at special quantity discounts to use as premiums and sales promotions, or for
use in corporate training programs. For more information and quotes, please contact the publisher.
Printed in the United States of America
10 9 8 7 6 5 4 3
ISBN: 9781891121838
Series ISBN: 9781891121876
Library of Congress Cataloging-in-Publication Data

Feucht, Dennis.
Designing dynamic circuit response / D. Feucht.
p. cm. -- (Analog circuit design series ; v. 2)
Includes bibliographical references and index.
ISBN 978-1-891121-83-8 (pbk. : alk. paper) – ISBN 978-1-891121-87-6 (series)
1. Frequency response (Dynamics) 2. Transients (Dynamics) 3. Electronic circuit
design. I. Title.
TK7867.F48 2010
621.3815′35--dc22
2009028289
Preface
Solid-state electronics has been a familiar technology for almost a half century,
yet some circuit ideas, like the transresistance method of finding amplifier gain
or identifying resonances above an amplifier’s bandwidth that cause spurious
oscillations, are so simple and intuitively appealing that it is a wonder they are
not better understood in the industry. I was blessed to have encountered them
in my earlier days at Tektronix but have not found them in engineering text-
books. My motivation in writing this book, which began in the late 1980s and
saw its first publication in the form of a single volume published by Academic
Press in 1990, has been to reduce the concepts of analog electronics as I know
them to their simplest, most obvious form, which can be easily remembered and
applied, even quantitatively, with minimal effort.
The behavior of most circuits is determined most easily by computer simula-
tion. What circuit simulators do not provide is knowledge of what to compute.
The creative aspect of circuit design and analysis must be performed by the
circuit designer, and this aspect of design is emphasized here. Two kinds of
reasoning seem to be most closely related to creative circuit intuition:
1. Geometric reasoning: A kind of visual or graphic reasoning that applies to
the topology (component interconnection) of circuit diagrams and to graphs
such as reactance plots.
2. Causal reasoning: The kind of reasoning that most appeals to our sense of
understanding of mechanisms and sequences of events. When we can trace
a chain of causes for circuit behavior, we feel we understand how the circuit
works.
These two kinds of reasoning combine when we try to understand a circuit by
causally thinking our way through the circuit diagram. These insights, obtained
viii Preface
by inspection, lie at the root of the quest. The sought result is the ability to write
down accurate circuit equations by inspection. Circuits can often be analyzed
multiple ways. The emphasis of this book is on development of an intuition
into how circuits work with a perspective that can be applied more generally to
circuits of the same class.
This second volume of the Analog Circuit Design series builds upon Designing
Amplifier Circuits by extending consideration to include reactances and their
time- and frequency-related behavioral consequences. Retaining a design-
oriented analysis, this volume begins with circuit fundamentals involving capaci-
tance and inductance and lays down the approach using s-domain analysis.
Though there is overlap with the contents of passive networks textbooks, addi-
tional concepts and perspectives fill in the picture required for circuit design.
Circuit characterization in both time and frequency domains provides a means
of assessing dynamic circuit performance. Dynamic circuit analysis is simplified
by use of the graphical methods of reactance plots. Methods of compensating
amplifiers, including feedback amplifiers, are kept as simple as possible using
reactance plots and s-domain transfer functions that mainly require algebraic
skill.
What started the entire series is my favorite topic and one that, after all these
years, is still hardly covered in engineering schools: the impedance gyrations
that occur in circuits above the bandwidth of the active devices. This compli-
cated topic reduces quickly to more simple, intuitive insights that can be applied
by inspection to get at the cause of why spurious oscillations occur in amplifiers,
why those 47 Ω resistors appear in the bases of fast amplifier stages, why emitter
followers or capacitively loaded amplifiers tend to oscillate, and how to emulate
inductance for bandwidth extension using transistors and resistances in inte-
grated circuits (ICs). With these concepts in circuit dynamics firmly in mind,
the reader is prepared for the next volume, Designing High-Performance Amplifiers,
third in this Analog Circuit Design series.
Much of what is in this book must be credited in part to others from whom
I picked up essential ideas about circuits at Tektronix, mainly in the 1970s. I am
particularly indebted to Bruce Hofer, a founder of Audio Precision Inc.; Carl
Battjes, who founded and taught the Tek Amplifier Frequency and Transient
Response (AFTR) course; Laudie Doubrava, who investigated power supply
topics; and Art Metz, for his clever contributions to a number of designs, some
Preface ix
extending from the seminal work on translinear circuits by Barrie Gilbert, also
at Tek at the same time. Then there is Jim Woo, who, like Battjes, is another
oscilloscope vertical amplifier designer; Ian Getreu and Bob Nordstrom, from
whom I learned transistors; and Mike Freiling, an artificial intelligence researcher
in Tektronix Laboratories whose work in knowledge representation of physical
systems influenced my broader understanding of electronics.
In addition, in no particular order, are Fred Beckett, Lee Jalovec, Wayne
Kelsoe, Cal Diller, Marv LaVoie, Keith Lofstrom, Peter Starič, Erik Margan, Tim
Sauerwein, George Ermini, Jim Geddes, Carl Hollingsworth, Chuck Barrows,
Dick Hung, Carl Matson, Don Hall, Phil Crosby, Keith Ericson, John Taggart,
John Zeigler, Mike Cranford, Allan Plunkett, Neldon Wagner, and Paul Magerl.
These and others I have failed to name have contributed personally to my
knowledge as an engineer and indirectly to this book. Most of all, I am indebted
to the creator of our universe, who made electronics possible. Any errors or
weaknesses in this book, however, are my own.
Contents
Chapter 1 Transient and Frequency Response . . . . . . . . . . . . . . . . . . . . . . . 1

Reactive Circuit Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
First-Order Time-Domain Transient Response . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Complex Poles and the Complex Frequency Domain . . . . . . . . . . . . . . . . . . . . 7
Second-Order Time-Domain Response: RLC Circuit . . . . . . . . . . . . . . . . . . . . 10
Forced Response and Transfer Functions in the s-Domain . . . . . . . . . . . . . . . 16
The Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Time-Domain Response to a Unit Step Function . . . . . . . . . . . . . . . . . . . . . . . 29
Circuit Characterization in the Time Domain. . . . . . . . . . . . . . . . . . . . . . . . . . 37
The s-Plane Frequency Response of Transfer Functions . . . . . . . . . . . . . . . . . 41
Graphical Representation of Frequency Response . . . . . . . . . . . . . . . . . . . . . . 43
Loci of Quadratic Poles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Optimization of Time-Domain and Frequency-Domain Response . . . . . . . . . 53
Reactance Chart Transfer Functions of Passive Circuits . . . . . . . . . . . . . . . . . . 61
Closure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Chapter 2 Dynamic Response Compensation . . . . . . . . . . . . . . . . . . . . . . . 75

Passive Compensation: Voltage Divider . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Op-Amp Transfer Functions from Reactance Charts . . . . . . . . . . . . . . . . . . . . 78
Feedback Circuit Response Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Feedback Circuit Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Compensation Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Compensator Design: Compensating with Zeros in H . . . . . . . . . . . . . . . . . . 105
Compensator Design: Reducing Static Loop Gain . . . . . . . . . . . . . . . . . . . . . 118
Compensator Design: Pole Separation and Parameter Variation . . . . . . . . . 120
vi Contents
Two-Pole Compensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

Output Load Isolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
Complex Pole Compensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
Compensation by the Direct (Truxal’s) Method . . . . . . . . . . . . . . . . . . . . . . . 162
Power Supply Bypassing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Chapter 3 High-Frequency Impedance Transformations . . . . . . . . . . . . . .167

Active Device Behavior above Bandwidth. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
BJT High-Frequency Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Impedance Transformations in the High-Frequency Region. . . . . . . . . . . . . 170
Reactance Chart Representation of b-Gyrated Circuits. . . . . . . . . . . . . . . . . . 177
Reactance Chart Stability Criteria for Resonances . . . . . . . . . . . . . . . . . . . . . 179
Emitter-Follower Reactance-Plot Stability Analysis. . . . . . . . . . . . . . . . . . . . . . 181
Emitter-Follower High-Frequency Equivalent Circuit . . . . . . . . . . . . . . . . . . . 183
Emitter-Follower High-Frequency Compensation . . . . . . . . . . . . . . . . . . . . . . 186
Emitter-Follower Resonance Analysis from the Base Circuit . . . . . . . . . . . . . 190
Emitter-Follower Compensation with a Base Series RC . . . . . . . . . . . . . . . . . 191
BJT Amplifier with Base Inductance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
The Effect of rb′ on Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
Field-Effect Transistor High-Frequency Analysis . . . . . . . . . . . . . . . . . . . . . . . 197
Output Impedance of a Feedback Amplifier . . . . . . . . . . . . . . . . . . . . . . . . . . 198
Closure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
1
Transient and Frequency Response
REACTIVE CIRCUIT ELEMENTS

Reactive circuit elements require complex-number analysis, and circuit behavior
can be expressed as either functions of time (time-domain analysis) or frequency
(frequency-domain analysis). These domains are united by expressing circuit
response in terms of a complex frequency.
What makes circuit behavior different with inductances and capacitances is
that they can store energy and release it later. Their behavior depends on the
rate of change of electrical quantities applied to them. The definitions of induc-
tance L and capacitance C are
dλ
L definition: L ≡
di
dq
C definition: C ≡
dv
where l is the magnetic flux linkage or circuit-referred flux, and q is the electric
charge. Given that
dλ
v=
dt
and
dq
i=
dt
2 Chapter 1
then
d λ di di
L v -i relation v = ⋅ =L⋅
di dt dt
dq dv dv
C v -i relation i = ⋅ =C ⋅
dv dt dt
L and C differ by an interchange of v and i; they are duals. Besides the defini-
tion and v-i relation for L and C, their expression based on geometry is also
important:
µ⋅A
L=
l
ε ⋅A
C=
l
where A is the coil loop or capacitor plate area, l is the coil length or capacitor
plate separation, m is the permeability of the material inside the coil, and e is
the permittivity (or dielectric constant) of the material between the plates. Not all
inductors or capacitors are constructed of solenoidal coils or parallel plates, but
the form of these equations is generally correct.
It is of interest to note that conductance can be expressed similarly:
σ ⋅A
G=
l
where s is conductivity of the conductive material. This equation is the recipro-

cal of the more common formula for resistance:
ρ ⋅l
R=
A
where r is resistivity = 1/s, and R = 1/G.

Transient and Frequency Response 3
The energy W stored by an inductor or capacitor can be derived by noting

that power P is the rate of energy flow,
dW
P=
dt
and that power is, by definition,
Watt’s Law: P = v ⋅ i
Combining and solving for W,
W = ∫ v ⋅ i ⋅ dt
Substituting results in
dλ
inductive energy, W = ∫ ⋅ i ⋅ dt = ∫ i ⋅ d λ
di
dq
capacitive energy, W = ∫ ⋅ v ⋅ dt = ∫ v ⋅ dq
dv
For linear elements, the definitions of L and C can be simplified to
linear L , λ = L ⋅ i
linear C , q = C ⋅ v
Substituting these expressions into W, the energy stored in a linear L or C is
1
W = ∫ i ⋅ d (L ⋅ i ) = L ⋅ ∫ i ⋅ di = ⋅ L ⋅i 2
2
1
W = ∫ v ⋅ d (C ⋅ v ) = C ⋅ ∫ v ⋅ dv = ⋅C ⋅ v 2
2
4 Chapter 1
The energy stored in an inductor is proportional to the square of the current

through it; for a capacitor, it is the square of the voltage across it. The propor-
tionality constants are half L and C.
The v-i relations for L and C can also be expressed as
1 t 1 t
v= ⋅ ∫c i ⋅ d τ , i = ⋅ ∫c v ⋅ d τ
C L
where t is a local (dummy) variable of integration, and t is time.
FIRST-ORDER TIME-DOMAIN TRANSIENT RESPONSE

Using the relations for L and C from the previous section, the time-domain
response (that is, the response as a function of time) can be determined. Time
is the independent variable of circuits such as that in (a), shown below.
R
vo
+
vi c Ii R L I0
(a) (b)
This is a simple RC integrator or low-pass filter. The response can be found using
Kirchhoff’s current law (KCL) at the output node:
vo − vi dv
+C ⋅ o = 0
R dt
or, arranging into the form of a standard differential equation,
dvo  1 
⋅v o = 
1 
+ ⋅v
dt  RC   RC  i
This is an ordinary, linear, constant-coefficient differential equation and can be

solved by using the substitution
v 0 = V ⋅ e st
where s is a variable. The input function driving the circuit, vi, appears only on
the right side of the differential equation. If a specific function is substituted
for vi(t), the output time response can be found for that input. The output
response also depends on the characteristic of the circuit itself, as represented
by the equation’s left side. By setting the input function to zero and solving for
vo, the natural response of the circuit is
dvo  1 
+ ⋅v o = 0
dt  RC 
Substituting for vo gives an algebraic equation:
V ⋅s +
1  st
⋅e = 0
 RC 
The exponential cannot be zero and V ≠ 0. Consequently, the middle factor

must be zero, or
1
s+ =0
RC
This algebraic equation in s is the characteristic equation of the differential equa-

tion and of the circuit. Solving for s gives
1
s=−
RC
and substituting for s, the natural response is
vo (t ) = V ⋅ e −(1 RC )⋅t
6 Chapter 1
If the capacitor has an initial charge of vo(t) = V0 and vi = 0, then
V0 = V ⋅ e 0 = V
and
vo = V 0 ⋅ e −(1 RC )⋅t
The response, shown below, is very common and is called exponential decay.
The output voltage decays, asymptotically approaching zero at infinite time.
This is a common natural response of circuits. Given an initial energy stored in
the reactive elements, it is eventually dissipated by any resistive elements. Con-
sequently, circuit response to a nonzero initial condition – the natural response
– is also called the transient response.
The time scale of the decay is measured as the time constant, the value of −1/s
from s above, and is
τ = R ⋅C
V0
t t
The initial slope of vo(t) is projected to the t-axis in the above figure and inter-
cepts the axis at t. At t = t, vo(t)/V0 = e−1 ≅ 36.7%. After 5 ⋅ t, vo is within 1% of
zero.
The RL circuit of Figure (b) above is the dual of (a), and its time constant is
t = L/R. For ii = 0 and an initial inductor current of I0, io(t) is about 37% of its
final value after t. Replacing V0 with I0 in the figure showing exponential decay
results in the response of io(t).
COMPLEX POLES AND THE COMPLEX-FREQUENCY DOMAIN

For circuits with more than one reactive element, the differential equations
describing them can also be solved by substitution of
v 0 = V ⋅ e st
The complication is in solving the characteristic equation. Its degree is equal to

the number of reactive circuit elements. After its roots are found, the time-
domain response is obtained by substitution. For repeated roots, linear combi-
nations of the expression for vo are required.
The roots of the characteristic equation are called poles. For single-pole
response, s is real. For two or more poles, s can be real, imaginary, or complex.
In general, s is a complex variable. Complex numbers can be expressed in rect-
angular or polar form:
rectangular form: s = σ + j ω
polar form: s = s ⋅ e jφ
where
j = −1
In rectangular form, s is the sum of real and imaginary numbers. In polar form,
s is expressed by a magnitude and phase angle. The polar and rectangular forms
are related through Euler’s formula,
e jφ = cos φ + j ⋅ sin φ
and the Pythagorean theorem

8 Chapter 1
s = σ 2 + ω2
The rectangular components are expressed in polar components as
σ = s ⋅ cos φ
ω = s ⋅ sin φ
Therefore,
s = σ + j ω = s ⋅ cos φ + j ⋅ s ⋅ sin φ
= s ⋅ (cos φ + j ⋅ sin φ ) = s ⋅ e jφ
Dividing w by s and solving for f results in
ω
φ = tan−1  
σ
With these expressions we arrive at the general form of a quadratic characteristic

equation for
s = −α ± j ωd
which is
s 2 + 2 ⋅ ζ ⋅ ωn ⋅ s + ωn2 = (s + α + j ωd ) ⋅ (s + α − j ωd )
= s 2 + 2 ⋅ α ⋅ s + (α 2 + ωd2 ) = 0
Equating terms,
α = ζ ⋅ ωn
ωn2 = α 2 + ωd2
The general solution of the quadratic equation is
s1,2 = −ζ ⋅ ωn ± j ωn ⋅ 1 − ζ 2
where
ωd = ω n ⋅ 1 − ζ 2
The quantity wn is called the natural frequency, a the damping factor, and z the
damping ratio.
Combining the equations for wn and a, z can be directly related to the pole
angle f as
ζ = cos φ
The roots, s1,2, can be expressed in polar form using the above expressions for
||s||, f and
tan θ = − tan (180° − θ )
to obtain
s1,2 = ωn ⋅ e ± jφ
This pole pair is shown plotted below.

Complex poles always occur in conjugate pairs, as shown. In the complex
plane of s (the s-plane or s-domain), poles are represented by × marks.
The units of s must be frequency, and because s is complex, it is a complex fre-
quency. The s-domain is called the complex-frequency domain.
10 Chapter 1
jw
s1 × jwd

s1
=
w
n
f
-a −f s
wn
=
2
s
s2 × −jwd
SECOND-ORDER TIME-DOMAIN RESPONSE: RLC CIRCUIT

The series RLC circuit shown below is a typical circuit with a second-order dif-
ferential equation and, consequently, a second-degree characteristic equation.
L C
vo
+
vi i R
−
Using KCL, its differential equation is
d 2i  R  di  1  1 di
+ ⋅ + ⋅i = ⋅ i
dt 2  L  dt  LC  L dt
The transient response is found by setting ii(t) = 0. Solving for the characteristic
equation using i = I ⋅ est,
s 2 +   ⋅s + 
R 1 
=0
L  LC 
Using the quadratic formula to solve for the poles in s, the roots are
R 1  R 2
s1,2 =− ±j −
2L LC  2L 
These poles are of the form
s1,2 = −α ± j ωd
where
R 1  R 2
α= , ωd = −
2L LC  2L 
Each pole contributes a solution to the differential equation when substituted

into I ⋅ e st. Because the response is the solution to a linear differential equation,
then by superposition, the independent pole solutions can be combined linearly
to form a complete solution.
The transient response of a second-order circuit depends on the value of its
elements. For the RLC circuit shown above, the poles are real when for wd,
2
 R  − 1 ≥0
 2L  LC
or
L
R ≥ 2⋅
C
When distinct poles are real, response is exponential, and the natural
frequency wn equals the real frequency a. Distinct real poles are located at
s = −a ± wd.
12 Chapter 1
A special case of real poles is for them to be equal, or repeated. Their solutions
cannot be combined by superposition to produce a response because they are
not independent. In this case, the general form of the pole solutions is, for n
poles,
t n +1
⋅ e −α⋅t
(n − 1)!
For the RLC circuit, n = 2, and the response is
i (t ) = I 1 ⋅ e −α⋅t + I 2 ⋅ t ⋅ e −α⋅t
For imaginary poles, wn = wd, a = 0, and the natural response is a sinusoid. This
is the case of oscillators and is a conditionally stable response.
The last case to consider is that of complex poles. The solution is
i (t ) = c1 ⋅ e (−α + jωd )⋅t + c 2 ⋅ e (−α − jωd )⋅t
This can be expressed as
i (t ) = e −α⋅t ⋅ (c1 ⋅ cos ωd ⋅ t + jc1 ⋅ sin ωd ⋅ t + c 2 ⋅ cos ωd ⋅ t − jc 2 ⋅ sin ωd ⋅ t )
= e −α⋅t ⋅ [(c1 + c 2 ) ⋅ cos ωd ⋅ t + j (c1 − c 2 ) ⋅ sin ωd ⋅ t ]
If c1 and c2 are complex conjugate constants, and
I 1 = c1 + c 2 , I 2 = j (c1 − c 2 )
then the response is
i (t ) = e −α⋅t ⋅ (I 1 ⋅ cos ωdt + I 2 ⋅ sin ωdt )
with real I1 and I2 determined by initial conditions. This can also be expressed
as a single sinusoid with a phase angle J:
i (t ) = I e −α⋅t ⋅ sin (ωd ⋅ t + ϑ )
where
I 
I = I 12 + I 22 , ϑ = tan−1  1 
 I2 
This response is a damped sinusoid (shown below).

For left half-plane (LHP) poles, the sinusoid decays with time due to the
decaying exponential factor. This factor is the envelope of the sinusoid and is
shown dotted. For right half-plane (RHP) poles, the response is unstable – an
exponentially growing sinusoid. The limited range of actual circuits causes non-
linear limiting of such a response and distorts the sine wave.
i(t)
The pole angle f and its related parameter z most explicitly express the kind
of response a circuit will have. The pole angle is
ω
φ = tan−1  d 
α
As z decreases, poles move toward each other and then split off the real axis,
increasing in pole angle, as shown below.
14 Chapter 1
Pole movement jw
as z decreases
Poles split off w

n
s axis here:
z=1
s
z = 0:
t t Undamped
z > 1: z > 1:
Overdamped Overdamped
0 < z < 1: Underdamped
As the poles leave the real axis, the time-domain response begins to show a
sinusoid, with noticeable “ringing.” The larger the pole angle, the more sinu-
soidal cycles occur before being damped out by the decaying exponential factor.
What results are the following categories of response:
z > 1 overdamped response: poles real and distinct
z = 1 critically damped response: two equal real poles
0 < z < 1 underdamped response: complex pole pair
z = 0 undamped response: poles imaginary
As z decreases below one, the poles move in a circular arc of radius wn
and increasing pole angle until, at z = 0, they are located on the jw axis at
±jwd = ±jwn. The damped frequency, wd, is less than the natural frequency wn when
z > 0 (that is, when damping exists due to R).
The real component of s, −a, is related to the exponential factor in the
response. It has units of 1/time (or frequency) and is half the reciprocal of the
time constant of the exponential factor. For large a, the exponential response
is fast. For poles on the jw axis, a = 0 and no exponential decay occurs.
The RLC circuit has
1 R
α= =
2τ 2L
1
ωn =
LC
and
α R
ζ= =
ωn 2 L C
The RLC circuit is critically damped at z = 1. Solving z for R, we obtain the pre-
vious equation for R:
L
R = 2⋅
C
Now, define
L
Zn =
C
This is the characteristic impedance of the LC elements. L and C also

determine the equation for wn. A combination of elements L and C can give
rise to an underdamped response or resonance. The behavior of resonant
circuits exhibiting sinusoidal response is oscillatory. Resonance occurs at
a frequency of wn and characteristic impedance of Zn. Circuits
with more than two reactive elements have more than one resonance. These
circuits can have more than one complex pole-pair.
A resonance is either a series resonance or parallel resonance. The previous
RLC circuit is series resonant because L and C are in series with R. For critical
damping of a series resonance, R must be equal to the sum of the (equal) reac-
tances of L and C at resonance, or 2 ⋅ Zn. A circuit with R, L, and C in parallel is
16 Chapter 1
parallel resonant, and for critical damping, once again, R must equal the com-
bined reactance of L and C or be half of Zn.
FORCED RESPONSE AND TRANSFER FUNCTIONS IN THE S-DOMAIN

The natural or transient response of a circuit is due to initial nonzero energy
storage in reactive elements. At t > 0, the circuit responds to this energy inde-
pendent of external input sources. With resistive elements in the circuit, this
initial energy is dissipated and eventually goes to zero. For a transient circuit
quantity (a voltage or current) xtr, then
lim xtr = 0
t →∞
When the circuit is driven by a source, it continues to respond indefinitely to

the source. This is the forced or steady-state response. It continues after the transient
response has decayed away and is
x ss = lim x
t →∞
For linear circuits, the total response is the superposition of the transient and
steady-state responses, or
x (t ) = xtr (t ) + x ss (t )
The transient response can be found, as in the previous section, by solving the
circuit differential equations for zero input. This (homogeneous) solution can
then be used to find the (particular or complementary) solution with a nonzero
input, resulting in the total response.
For linear circuits, input sinusoids always result in output sinusoids. A differen-
tial circuit equation describing an output quantity xo and input xi is, in general,
D1(xo ) = N 1(xi )
The transient response is found by setting N1 = 0 and xo = Xo(s) ⋅ est, where Xo is

a complex parameter of s and constant with t. Then,
D1(X o e st ) = 0
The exponential can be factored out of D, leaving the characteristic equation

D(s) = 0. This factorization is
D1(X o e st ) = D (s ) ⋅ X o e st = D (s ) ⋅ xo
If xi is also made a complex exponential, then a similar factorization on the

right side of N1 produces
N 1(xi ) xi =Xie st = N (s ) ⋅ xi
Substituting N1 and D1 into the general equation and solving for the output/
input ratio, or transfer function, results in
X o N (s )
=
X i D (s )
where
xi , xo = X (e st )
This is an extremely important result. Whenever the input is a complex expo-

nential function, the output will also be, as modified by the factor N(s)/D(s). A
sinusoidal input is a special case in which s = jw. This equation is a general result
because complex exponentials can be summed in a Fourier series to create
arbitrary functions. Therefore, this is the key to finding the transfer functions
of linear circuits in general, expressed in s.
N and D can be factored into the canonical form of a transfer function:
N (s ) (s + z1 ) (s + z2 )" (s + zm )
=K⋅ , m ≤n
D (s ) (s + p1 ) (s + p2 )" (s + pn )
The roots of D(s) are called poles, −pi, and the roots −zi of N(s) are called zeros
because N/D is zero at s = −zi. (The word pole fits the idea that since the poles
18 Chapter 1
make N/D infinite, a plot of N(s)/D(s) typically looks like a tent with poles
holding it up at the poles.) Poles and zeros are called critical frequencies. (Poles
are also mathematically known as singularities, for they result in no finite value.)
For actual (causal) circuits, the degree of N does not exceed that of D, or m ≤
n. K is a constant, but it is not the static (dc) transmittance. The transfer func-
tion equation can be expressed in normalized form by factoring out zi and pi. The
factors are normalized to unity at s = 0, and the constant K is the static
transmittance:
N (s )
=K⋅
(s z1 + 1) (s z2 + 1)" (s zm + 1) , m ≤ n
D (s ) (s p1 + 1) (s p2 + 1)" (s pn + 1)
To illustrate these general concepts by example, consider the RC differentiator

circuit shown below.
C
+ +
vi R vo
− −
Using KCL at the output node,
d (vo − vi ) vo
C⋅ + =0
dt R
This can be rearranged as
dvo  1  dv
+ ⋅v o = i
dt  RC  dt
The right side of the equation describes the effect of the input and is related
to the steady-state response. The left side characterizes the transient response.
Letting both vo and vi be complex exponentials and factoring
N (s ) s s
M (s ) = = = RC ⋅
D (s ) s + (1 RC ) sRC + 1
There is a zero at the origin (s = 0) and a pole at −1/RC. D(s) is the same as
that of the previous first-order RC circuit. Because the poles characterize the
transient response, it is identical for this RC circuit.
The numerator describes the transformation of the input. Because
d st
e = s ⋅ e st
dt
s functions as a differentiation operator for complex exponentials and, more

generally, for functions that can be described in terms of complex exponentials.
In M(s), N(s) = s can be interpreted as a differentiation of the input function
to the circuit – hence the name “RC differentiator”.
Applying this method of finding the transfer function in s to the RC integra-
tor of (a), described previously by
dvo  1 
⋅v o = 
1 
+ ⋅v i
dt  RC   RC 
R
vo
+
vi c Ii R L I0
(a) (b)
it follows that the transfer function of the RC integrator is
N (s ) 1 R ⋅C 1
M (s ) = = =
D (s ) s + (1 R ⋅C ) sR ⋅ C + 1
20 Chapter 1
The dual circuit in (b) will have a transfer function of a similar form, with pole
at L/R.
THE LAPLACE TRANSFORM

The RC integrator and differentiator are described by one differential equation
resulting from application of KCL in the time domain. More complicated cir-
cuits result in multiple differential equations. The total solution requires sub-
stitution of a particular function for the input. As in the previous section, for
complex exponential inputs, the response is also a complex exponential, and
the transfer function in the s-domain can be found.
The Laplace transform is an extension of the Fourier transform and enables
us to readily reduce differential circuit equations to algebra. Better yet, it allows
us to avoid differential equations altogether.
Fourier-series descriptions of arbitrary functions can be given in terms of
complex exponentials, making this a generally useful result. But our goal is to
determine the time-domain response of a circuit described by transfer function
M(s) for a given input. Specifically, we need a way of converting a function x(t)
to its equivalent in the s-domain, X(s). Then we can multiply Xi(s) by M(s) to
get the resulting output Xo(s):
X o (s ) = M (s ) ⋅ X i (s )
Then if we can transform Xo(s) to the time domain, we have the desired result,
xo(t).
The Fourier series is limited to periodic waveforms but in the limit becomes
the Fourier integral. This integral is a transform from the time domain to the
jw domain. By adding a real component s to jw, we have an extended transform
in s, the Laplace transform, defined as
∞
L { f (t )} ≡ ∫0 f (t ) ⋅ e − st ⋅ dt
Some transformed functions are given below:

f (t ) F (s ) = L { f (t )}
δ (t ) 1
1
u (t )
s
1
e −αt
s +α
ωd
sin (ωd ⋅ t )
s + ωd2
2
s
cos (ωd ⋅ t )
s + ωd2
2
ωd
e −αt ⋅ sin (ωd ⋅ t )
(s + α )2 + ωd2
n!
tn
s n +1
These are among the functions useful as inputs to circuits for characterizing
their responses. The unit impulse function d(t) is defined as
 0, t ≠ 0
δ (t ) ≡ 
∞, t = 0
Although d(0) is infinite, ∫d dt = 1. At t = 0+, the circuit responds as it would to

nonzero initial conditions with the transient response. In practice, it is not easy
to generate d, and a step input is used to characterize response instead. The
unit step function u(t) is defined as
0, t < 0
u (t ) = 
1, t > 0
Neither d(t) nor u(t) is periodic. Step functions are physically approximated as
a square wave (a periodic function) with a relatively long period, allowing tran-
sient effects to decay away while the step persists.
22 Chapter 1
The Laplace transform of operations, such as differentiation and integration,

can also be taken:
t-Domain s-Domain
K ⋅ f (t) K ⋅ F(s) scale invariance
∑fi(t) ∑Fi(s) superposition
df (t ) sF(s) − f (0+)
dt
d 2f ( t ) df (0 + )
s2 F ( s) − sf (0 + ) −
dt 2 dt
t F ( s)
∫ f (τ ) ⋅ dτ
c
s
e−a ⋅ t f(t) F(s + a)
n
t f(t) d n F ( s)
(−1)n ⋅
ds n
−t ⋅ s
u(t − t) ⋅ f (t − t) e ⋅ F(s) shifting theorem
t F(s) ⋅ G(s) convolution
∫ f (τ ) ⋅ g (t − τ ) ⋅ dτ
c
lim f (t ) lim s ⋅ F ( s) final-value theorem

t →∞ s→ 0
lim f (t ) = f (0 + ) lim s ⋅ F ( s) initial-value theorem

t →0 s→∞
The first two equations establish the linearity of the Laplace transform. The
transform of a derivative is consistent with the use of s as a differentiation opera-
tor in the s-domain (as described in the previous section). The initial condition,
f(0+), is part of the transformed derivative. The shifting theorem expresses the
effect of shifting f in time by a delay of t. The convolution integral offers an
alternative to the inverse Laplace transform for finding xo(t) from
X o (s ) = M (s ) ⋅ X i (s )
but is usually not as easy to use.

The Laplace transform can be applied to the v-i relations of R, L, and C to

find their s-domain impedances. They can then be used in basic circuit analysis.
KCL, Kirchhoff’s voltage law (KVL), and Ohm’s law (ΩL) can be applied directly
in the s-domain with no need for intermediate steps involving differential equa-
tions. For R, C, and L,
L {v } V (s )
L {v } = L {R ⋅ i } = R ⋅ L {i } ⇒ Z R (s ) = = =R
L {i } I (s )
L {v } = L { 1
C
∫ } 1 L {i }
i ⋅ dt = ⋅
C s
⇒ Z C (s ) =
L {v }
=
1
L {i } s ⋅ C
L {i } = L { 1
L
∫ } 1 L {v }
v dt = ⋅
L s
⇒ Z L (s ) =
L {v }
L {i }
= s ⋅L
Initial conditions for L and C in the s-domain can be accounted for by Laplace
transforming the time-domain expressions for L and C with initial conditions,
or
1 V (s ) iL (0 + )
iL (t ) = ⋅ ∫ v L (t ) ⋅ dt + iL (0 + ) 
L
→ I L (s ) = L +
L sL s
1 I (s ) vC (0 + )
vC (t ) = ⋅ ∫ iC (t ) ⋅ dt + vC (0 + ) 
L
→VC (s ) = C +
C sC s
The s-domain equivalent circuit for L with initial current iL(0+) is a current
source of iL(0+)/s in parallel with L. The s-domain equivalent circuit for C with
initial voltage vC(0+) is a voltage source of vC(0+)/s in series with 1/sC, shown in
figure (a) below. By transforming the derivative form of the L and C v-i relations,
we obtain an equivalent circuit that accounts for initial conditions, shown in
(b).
24
Chapter 1
sL
i L (0 + )
iL (0 + ) L L i L (0 + ) sL
s +
LiL (0 + ) = λ (0 + )
–
1
C
+ sC
1 +
vC (0 + ) C + + CvC (0 )
– sC +
vC (0 + ) vC (0 + ) = q (0 )
s
– –
(a) (b)
By working directly in s, we can find the transfer function of the RC integrator

and differentiator from inspection, by treating them as voltage dividers. For the
RC integrator,
Vo (s ) 1 sC 1
M (s ) = = =
Vi (s ) R + 1 sC sRC + 1
and for the RC differentiator,
Vo (s ) R sRC
M (s ) = = =
Vi (s ) R + 1 sC sRC + 1
Writing circuit equations directly in the s-domain avoids the need either to
transform or to solve differential equations.
Example: Series RC Circuit

R1 C1
+
+
R2
vi vo
− C2
−
The figure shows a passive circuit with series RC divider impedances. The trans-
fer function is found by directly writing out the voltage-divider formula using
s-domain impedances. Both impedances are of the form: R + 1/sC. The transfer
function is
Vo (s )  C 1  sR 2 ⋅ C 2 + 1 C ⋅C
=  ⋅ , C1 C 2 = 1 2
Vi (s )  C 1 + C 2  s ⋅ (R1 + R 2 ) ⋅ (C1 C 2 ) + 1 C1 + C 2
where || is a mathematical operator, not a topological circuit descriptor.

26 Chapter 1
Example: Wien-Bridge Filter
R1 C1
+
+
vi C2 R2 vo
−
−
The figure is that of a filter topology used as the feedback path of the
Wien-bridge oscillator. It is another voltage divider, for which the transfer func-
tion is
Vo (s ) s ⋅ R 2C 2
= 2
Vi (s ) s ⋅ R1R 2C1C 2 + s ⋅ [R1C1 + R 2(C1 + C 2 )] + 1
For a Wien-bridge filter, R1 = R2 = R and C1 = C2 = C. Then
Vo (s ) s ⋅ RC
= 2
Vi (s ) s ⋅ (RC ) + s ⋅ [3RC ] + 1
2
For this filter, wn = 1/RC and z = 1.5.
Example: Inverse of Wien-Bridge Filter

In the figure, the divider topology of the Wien-bridge filter (shown previously)
is inverted. The transfer function is
Vo (s ) (sR1C1 + 1) ⋅ (sR 2C 2 + 1)
= 2
Vi (s ) s ⋅ R1R 2C1C 2 + s ⋅ [R1 ⋅ (C1 + C 2 ) + R 2C 2 ] + 1
C1
+
+
R2
vi R1 vo
− C2
−
For the Wien-bridge conditions, this reduces to
(sRC + 1)2
s 2 ⋅ (RC )2 + s ⋅ [3RC ] + 1
The poles are in the same place as in the Wien-bridge example, but instead of
a zero at the origin repeated zeros appear at 1/RC.
Example: Shunt-Series RC Circuit
R2
Z R1 C1
C2
This figure shows another example of an RC circuit. It has a terminal impedance

of
s ⋅ R 2C 2 + 1
Z = R1 ⋅
s ⋅ [R1C1R 2C 2 ] + s ⋅ [R1C1 + (R1 + R 2 ) ⋅ C 2 ] + 1
2
This circuit is sometimes the external emitter network of common-emitter (CE)

amplifiers, in which R1 is the emitter resistor, and C1, R2C2 is frequency
compensation.
28 Chapter 1
Example: Crystal Equivalent Circuit
Cp Cs
The figure above shows the equivalent circuit of a quartz crystal. The terminal
impedance is
1 s 2 ⋅ LC s + s ⋅ RC s + 1
Z xtal = ⋅
C p + C s s ⋅ [s 2 ⋅ L ⋅ (C s C p ) + s ⋅ R ⋅ (C s C p ) + 1]
where (as usual) || is an algebraic operator not a topological descriptor; Cs and

Cp are in series in the quadratic pole.
Resonance occurs when the phase is zero. The phases of numerator and
denominator are
{ ω ⋅ RC s
} −1  1 − ω ⋅ L ⋅ (C s C p ) 
2
φN = tan−1 , φ = tan  
 ω ⋅ R ⋅ (C s C p ) 
D
1 − ω 2 ⋅ LC s
Then ∠Z(jw) = fN − fD. Setting this to zero and simplifying, we must solve for
resonant frequency wr in
ω r4 L2Cs (Cs C p ) − ω r2 [ R 2Cs (Cs C p ) + L ⋅ (Cs C p ) + LCs ] + 1 = 0
Solving for w2r with
ω s2 = 1 L ⋅ Cs , ω 2p = 1 L ⋅ (Cs C p )
then
1  2 R  1  2 R 
2 2
ω = ⋅ ωs + ω p −
2 2  ± ⋅ ωs + ω p +
2  − 4 ⋅ ω s2 ⋅ ω 2p
r
2    
L  2   
L  
The expression under the radical can be rewritten as

2 4
(ω s2 − ω 2p )2 + 2 ⋅  L  ⋅(ω s2 + ω 2p ) +  L 
R R
For R = 0,
ωr = ωs ,ω p
The two resonant frequencies are at the series resonance ws and parallel reso-
nance wp of the crystal.
TIME-DOMAIN RESPONSE TO A UNIT STEP FUNCTION

To find the time-domain response to a given input function, take the inverse
Laplace transform, L−1, of Xo(s), where
X o (s ) = M (s ) ⋅ X i (s )
or apply the convolution integral. After Xo(s) is in the form of a known trans-
form, it is inverse-Laplace-transformed to produce the time-domain response.
Because M(s) is a rational function, partial-fraction expansion is the usual
method of expressing Xo so that it can be inverse transformed.
The Laplace-transformed impulse function when multiplied by M(s) yields
the s-domain transient response. It is a difficult function to generate and observe;
the step function is the dominant alternative. It is approximated in practice by
a square wave with a period much longer than the duration of significant tran-
sient response (and thereby is effectively aperiodic). Various characteristics of
circuit response to the step are of interest, and all are time related. This
approach to circuit characterization is time-domain analysis.
Transfer functions represent the complex dynamic behavior of circuits but
are an abstraction of actual circuit behavior. The response of a circuit under
30 Chapter 1
controlled conditions produces features that characterize the circuit. We now

investigate the characterization of circuits by their time-domain response to a
unit step input, u(t). The time-domain response can be determined by multiply-
ing the transfer function by the Laplace transform of u(t), which is 1/s, and
inverse transforming the result.
The RC integrator response is calculated as
vo (t ) = L−1{ 1
⋅
sRC + 1 s
1
}
The s-domain expression is partial-fraction expanded to
A B 1 RC
+ = −
s sRC + 1 s sRC + 1
This inverse transforms (using the u(t) and e−a ⋅ t Laplace transforms) to
vo (t ) = u (t ) − e −t RC = 1 − e −t RC , t > 0
which is plotted against t/t as curve (b) below.

For the RC differentiator,
vo (t ) = L−1 { RCs 1
} 
⋅ = L−1 
sRC + 1 s
1  −t RC
=e
 s + (1 RC ) 
This response is graphed as curve (a), with time scaled in time constants.
The response of a circuit with complex poles is demonstrated by the following
RLC circuit.
sL
+ +
1
Vi(s) sC R Vo(s)
− −
Its transfer function can be written by treating it as a voltage divider. Then
Vo (s )
=
(1 sC ) R = 1 ⋅ 1
Vi (s ) (1 sC ) R + sL LC s + s (1 RC ) + (1 LC )
2
1
=
s LC + s (L R ) + 1
2
where
1 1
K = 1, ωn = , α=
LC 2RC
For Vi(s) = 1/s, the step response of the RLC circuit is
1 1 
v step (t ) = L−1  ⋅ 2 
 s s LC + s (L R ) + 1 
32 Chapter 1
The quadratic factor is of the form
N (s ) 1 N (s )
= 2⋅ 2 2
s + 2α ⋅ s + ωn ωn (s ωn ) + (2α ωn2 ) ⋅ s + 1
2 2
The denominator can be factored so that
N (s)
=
( s + α + jω d ) ⋅ ( s + α − jω d )
1 N(s)
⋅
ω n [( s ω n ) + (α ω n ) + j (ω d ω n )] ⋅ [( s ω n ) + (α ω n ) − j (ω d ω n )]
2
This can be expressed as a partial-fraction expansion:
N (s ) A* A
= +
(s − p ) ⋅ (s − p *) (s − p ) (s − p *)
where X* is the complex conjugate of X and
p = −α + j ωd
This form can be shown to be valid by letting
A = a + jb

N (s ) = cs + d
This is the most general form N can take, with its degree one less than the
denominator. (If the expanded expression is a transfer function of a circuit
with zero magnitude at infinite frequency, the fraction must be less than 1, or
m < n.) Then the partial-fraction expansion coefficients are
c 1  d − α ⋅c
A= ⋅ ⋅ e jϑ , ϑ = tan −1 
2 cos ϑ  c ⋅ ω d 
The time-domain response of the expanded expression is found by making use

of the polar form of A. Substituting A = ||A|| ⋅ e jJ and A* = ||A|| ⋅ e−jJ results in
 A* A 
L−1  +  = A* ⋅ e + A ⋅ e
pt p *t
s − p s − p* 
= A ⋅ e −α⋅t ⋅ (e − jϑ ⋅ e jωdt + e jϑ ⋅ e − jωdt )
= A ⋅ e −α⋅t ⋅ (e j (ωdt −ϑ ) + e − j (ωdt −ϑ ) )
= 2 ⋅ A ⋅ e −α⋅t ⋅ cos (ωdt − ϑ )
Therefore, the general transform involving complex pole pairs is
 A ⋅ e jϑ A ⋅ e − jϑ 
 = 2 ⋅ A ⋅ e ⋅ cos (ωdt − ϑ )
− α ⋅t
L−1  +
 s + α + j ωd s + α − j ωd 
Returning to the equation for vstep(t), its partial-fraction expansion is
1 1 1 A B* B 
⋅ 2 = ⋅ + +
s s LC + s (L R ) + 1 LC  s s − p s − p * 
where p = −a + jwd. Solving for the numerators of the above equation,
A=
1
ωn2
, B=−
1
2ωn2

⋅ 1 +

 α  1
j   = − 2 ⋅
 ωd  
1
2ωn cos γ
⋅ e j γ , γ = tan−1
α
ωd { }
Inverse transforming the equation, using the Laplace transform of u(t), then
for complex pole-pairs,
v step (t ) = 1 −
1
sin γ
⋅ e −αt ⋅ cos (ωdt − γ ) , γ = tan−1
α
ωd { }
Applying the trigonometric relation
34 Chapter 1
1
tan ϑ =
tan (90 D −ϑ )
the step time response becomes
v step (t ) = 1 −
1
sin φ { }ω
⋅ e −α⋅t ⋅ sin (ωdt + φ ) , φ = tan−1 d
α
The factor 1/sin f can be expressed as
α
2
1
= 1 +   = [1 − ζ 2 ]
−1 2
sin φ  ωd 
Expressed in circuit-element values for the RLC circuit,
Zn LC
ζ= =
2R 2R
An alternative approach to the inverse Laplace transformation of the qua-

dratic factor is to complete the square for the quadratic denominator:
s 2 + 2 ⋅ α ⋅ s + ωn2 = (s + α )2 − (α 2 − ωn2 ) = (s + α )2 + ωd2
Then, for N(s) = c ⋅ s + d, the quadratic factor becomes
N (s ) cs d
= +
s + 2 ⋅ α ⋅ s + ωn (s + α ) + ωd (s + α )2 + ωd2
2 2 2 2
(s + α )  αc  ωd  d  ωd
= c⋅ −  ⋅ + ⋅
( s + α ) + ω d  ω d  ( s + α ) + ω d  ω d  ( s + α )2 + ω d2
2 2 2 2
N (s ) (s + α )  d − αc  ωd
=c⋅ +  ⋅
s + 2 ⋅ α ⋅ s + ωn
2 2
(s + α ) + ωd  ωd  (s + α )2 + ωd2
2 2
Using the Laplace transforms for an exponentially decaying sine and cosine,
L−1 { N (s )
s + 2 ⋅ α ⋅ s + ωn
2 2 }  d − α ⋅ c  −α⋅t
= c ⋅ e −α⋅t ⋅ cos ωdt + 
 ωd 
⋅ e ⋅ siin ωdt
For the sum of a sine and cosine,
a ⋅ cos ϑ + b ⋅ sin ϑ = a 2 + b 2 ⋅ sin ϑ + tan−1   

a
  b  
This equation can be applied to express the previously inverse-transformed

expression as a single sinusoid. After some manipulation,
L−1 { cs + d
s + 2 ⋅α ⋅ s + ωn
2 2
= }
c
sin φ
⋅ e −α ⋅t ⋅ sin (ω d t − φ ) , φ = tan −1 
c ⋅ωd 
 α ⋅c − d 
If we apply this method to vstep(t), the partial-fraction expansion is
1 s + 2 ⋅α
− 2
s s + 2 ⋅ α ⋅ s + ωn2
From this expression, c = −1 and d = −2 ⋅ a. Substituting into the above inverse-

transformed expression for the quadratic term yields
L−1 { 1
− 2
s + 2 ⋅α
s s + 2 ⋅ α ⋅ s + ωn
2
= 1− }
1
sin φ
ω
⋅ e −α⋅t ⋅ sin (ωdt + φ ) , φ = tan−1 d
α { }
This is the same result as the first derivation.
For the case of repeated real poles (critical damping),
v step (t ) = 1 − (1 + α ⋅ t ) ⋅ e −α⋅t
and for distinct real poles, whenever z > 1,
p1,2 = −α B ωd = −α B ωn ⋅ ζ 2 − 1
36 Chapter 1
These are real roots. Their corresponding step response is
 p1 p2 
v step (t ) = 1 −  ⋅ e p2⋅t − ⋅ e p1⋅t 
 p1 − p2 p1 − p2 
Example: Magnetic Deflection Yoke Coil Circuit
L
Ii C Io
The diagram shows a simplified cathode ray tube (CRT) deflection circuit. The
deflection yoke consists of horizontal and vertical deflection coils that magneti-
cally deflect the CRT electron beam. A yoke coil has significant series resistance
and intrawinding capacitance, modeled as shown. If ii(t) is a ramp function
(producing a horizontal or vertical sweep needed for raster scanning of the CRT
screen by the electron beam), then it can be expressed as
ii (t ) =   ⋅t = m ⋅ t
I
T 
where I is the peak ramp current, and T is the ramp duration (or the period of
an ideal sawtooth function). The output current io(t) is the current that flows
through L, creating the deflection field. Our goal is to find a general expression
in s for Io(s) and also to find the time-domain response.
The current-divider formula is used here and yields
I o (s ) 1
= 2
I i (s ) s LC + sRC + 1
For Ii(s) = m/s2,
1 m
I o (s ) = ⋅ 2
s LC + sRC + 1 s
2
and
ωn2 = 1 LC , α = ζ ⋅ ωn = 1 (2L R )
Io can be written by completing the square of the quadratic pole and expressing
Io as a partial-fraction expansion. Because of multiple roots at s = 0, it is neces-
sary to take the derivative of the partial-fraction equation to find the coefficient
k for the k/s term. Then,
(2ζ wn ) ⋅ s + m ⋅ (4ζ 2 − 1) m 2ζ ⋅ m ωn
I o (s ) = + −
(s + α )2 + ωd2 s2 s
Using the expression to perform L−1 on the first term,
 2ζ  −α⋅t  2ζ 
io (t ) =   ⋅ e ⋅ sin (ωdt − φ ) + m ⋅  t − 
 ωn ⋅ sin φ   ωn 
io(t) is a ramp delayed by 2z/wn. Superimposed on this ramp is a decaying sinu-

soid, the first term. When the response of the horizontal deflector is too under-
damped, the resulting ringing causes the picture on the left side of the CRT
screen to show an alternating compaction and expansion until the sinusoid dies
out.
CIRCUIT CHARACTERIZATION IN THE TIME DOMAIN

Prior derivations have worked out both first- and second-order responses to a
unit step function. Higher-order responses are combinations of first- and second-
order responses. Most circuits can be separated into lower-order circuits and
analyzed individually. Transfer-function numerators and denominators can be
38 Chapter 1
factored into first- and second-order factors that are separated by partial-fraction
expansion.
Previously, the effect of z on the response was examined. For a step response,
we are interested in how much above the step amplitude is the response for
complex poles. For accurate reproduction of a step, this overshoot should be
minimal. We also want to avoid the other extreme of a highly overdamped
response. The larger z becomes, the longer it takes for the response to approach
its asymptotic value. In other words, the risetime (or for a negative step, falltime)
is excessive. An obvious compromise occurs for critical damping, when z = 1.
This value of z is seldom chosen for wideband amplifiers because a much faster
step can be achieved with a small amount of overshoot.
When the step response overshoots, its peak occurs at time tp. This time is
derived by taking the derivative of the expression for vstep(t), setting it to zero,
and solving for t. The derivative is
d ν step (t ) α
= ⋅ e −α⋅t ⋅ sin (ωdt + φ ) − ωn ⋅ e −α⋅t ⋅ cos (ωdt + φ )
dt sin φ
This reduces to
tan (ωdt + φ ) = tan φ
which requires that
ωd ⋅ t = k ⋅ π , k = 0, 1, 2, . . .
The peak occurs at k = 1 for a damped response and is
π
tp =
ωd
The peak at this time is, from the expression for vstep(t),
v step (t p ) = 1 + e − π tanφ
Because the input is a unit step, this is the fractional peak. The overshoot is
defined as
M p = e − π tanφ
Mp is related to z through tan f and
ζ = cos φ
or
1−ζ2
tan φ =
ζ
Overshoots for several angles are tabulated below.
f Mp , % z
0° (0 rad) 0 1
π  0.433 0.866
30°  rad
6 
π  4.321 0.707
45°  rad
4 
π  16.30 0.500
60°  rad
3 
To find a compromise optimum between overshoot and risetime, we need to

define risetime. Furthermore, a third criterion is the time it takes the step to
settle to its steady-state value. This is the settling time ts; its definition depends on
the amount of settling that is adequate for the application. If we define Ms to
be the fractional settling amplitude, then
M s ≡ e −α⋅ts
40 Chapter 1
Solving for the settling time,
1 log 2 M s ln 2 0.7  1 
ts = − ⋅ ln M s = − =− ⋅ log 2 M s ≅ ⋅ log 2 
α α ⋅ log 2 e α α  M s 
where log2(1/Ms) is the number of bits of resolution of settling. For a second-

order response,
α = 1 (2 ⋅ τ )
and ts further reduces to
 1 
t s ≅ 1.4 ⋅ τ ⋅ log 2 
 M s 
A few values of ts in units of t are given in the following table.
ts, t log2(1/Ms), bits

5.6 4
8.4 6
11.2 8
13.9 10
16.6 12
19.4 14
In observing an underdamped step, an estimate of f or z can be based on

the number of cycles of oscillation, Ns, until the waveform settles. The oscillation
frequency is wd. Then,
ωd ⋅ t s ωd  1 tan φ  1 
Ns = = ⋅ − ⋅ ln M s  = ln 2 ⋅ ⋅ log 2 
2π 2π  α  2π  M s 
 1 
≅ (0.11) ⋅ tan φ ⋅ log 2 
 M s 
Some values of Ns for 8 bits of resolution (approximately the resolution of an

oscilloscope trace) are the following:
Ns f, deg z
0 0 1
0.5 30 0.866
0.9 45 0.707
1.5 60 0.500
3.3 75 0.259
8.8 84 0.100
Finally, the risetime tr could be defined as settling time or as a given number

of time constants. The most commonly used definition is the time it takes the
step to change from 10% to 90% of its final value. This definition is general in
that it does not assume a particular kind of response. For a first-order system
response,
t r ≡ t 90% − t10% = −τ ⋅ ln (1 − 0.9) + τ ⋅ ln (1 − 0.1) = τ ⋅ ln (9) ≅ 2.2 ⋅ τ
This risetime formula holds approximately for complex poles with small pole
angles.
THE S-PLANE FREQUENCY RESPONSE OF TRANSFER FUNCTIONS

An alternative to characterization of circuits by their step response is to input a
sinusoid or “sine wave.” Unlike the step, sinusoids are periodic and character-
ized by amplitude and phase, neither of which are time dependent. By exciting
the circuit input with sine waves over a range (or band) of frequencies, the
amplitude and phase as functions of frequency can be determined. This is the
frequency response of the circuit; it is the steady-state sinusoidal response. In prac-
tice, approximation of this procedure is accomplished by sweeping slowly enough
through a band of frequencies to let the transient response die out. This
approach to circuit characterization is frequency-domain analysis.
Time- and frequency-domain analyses reveal different aspects of the complex-
frequency domain. A cross-section of the transfer function M(s) along the jw
42 Chapter 1
axis is a function of w only and is the frequency response M(jw). The magnitude
and phase of M can be found for a particular jw1 by substituting into M(jw).
The magnitude and angle of the resulting complex number is the amplitude
and phase of the frequency response at w1. This can be done graphically on the
s-plane. (Zeros are marked on s-plane plots by an open circle.) M(jw1) can be
calculated from the graph by first finding the length and angle of each vector.
Then,
Π zeros( zero vector magnitudes )

M ( j ω1 ) = K ⋅
Π poles( pole vector maagnitudes )
∠M ( j ω1 ) = ∑ ( zero vector angles ) − ∑ ( pole vector angles )

zeros pole
es
For example, the s-domain plot shows an M(s) with three poles and one zero.
jw
−p2 = −4 + 2j
j2
−∠p2 = −14.04°

√17
j1

6)|| = √37
|| j 1 − (− 
√5
∠p1 = 9.46° 26.57°
−p1 = −6 −4 −z1 = −2 σ
36.87°
−j2
−p2*= −4 − 2j
If K = 100, then the frequency response at w = 1 is
(2.24 )
M ( j1) = (100 ) ⋅ = 1.78
(6.08)(4.12)(5.00 )
Each of the numbers in the fraction is the contribution of a critical frequency.

For example, the contribution of p2 is
[ j1 − (−4 + j 2)] = 4 − j 1 = (4 )2 + ( −1)2 ≅ 4.12
The phase angle of M at j1 is
∠M ( j1) = 26.57° − (9.46° −14.04° + 36.87°) = −5.73°
where, for example, the angle of p1 is
∠ ( j 1 − ( −6 )) = ∠ (6 + j 1) = tan−1 {}
1
6
≅ 9.46°
GRAPHICAL REPRESENTATION OF FREQUENCY RESPONSE

Several graphical methods of representing frequency response are found in
circuits and control systems literature:
1. Bode plot: two graphs, one of amplitude and the other of phase, against fre-
quency. The amplitude graph is of log-log scale; the phase angle graph is
semi-log (log frequency).
2. Reactance chart: similar to a Bode plot, a log-log plot of impedance magni-
tude, on which divider-type transfer functions can be constructed directly.
3. Nyquist diagram: graph of M(jw) with frequency as a parameter, on the
complex plane with axes jIm{M(jw)} and Re{M(jw)}. For feedback systems the
loop gain GH is plotted.
4. Root-locus plot: feedback amplifier plot on the s-plane of the movement of
the closed-loop poles with static gain K.
44 Chapter 1
5. Nichols chart: plot on rectangular coordinates of magnitude versus

phase (the gain-phase plane) of open-loop response of a unity-feedback
amplifier, with superimposed contours of constant magnitude and phase
of M(jw) of the resulting closed-loop magnitude and phase [for M(jw) =
G/(1 + G)].
6. Hall chart: complex plot of Im{G(jw)} versus Re{G(jw)} for unity-feedback
system with superimposed contours of constant ||M(jw)|| loci [for M =
G/(1 + G)].
We will survey the construction of Bode plots and later make extensive use of
them. For a transfer function in normalized form, the rational expression in s
is one at s = 0, and the constant K must be the static transmittance.
A transfer function, M(s), having one real pole p, evaluated at jw, is
1 1 −1
(ω p )
= ⋅ e − tan
( jω p + 1) (ω p ) + 1
2
The log-log plot of the magnitude of a real pole is
 1  1  ω  2 
log   = − ⋅ log   + 1
 ( )
ω p
2
+ 1  2  p  
For w/p << 1, this becomes
1  ω  2 
− ⋅ log   + 1 ≅ log (1) = 0
2  p   ω p1
For w/p >> 1,
1  ω  2 
− ⋅ log   + 1 ≅ − ( log ω − log p ) ≅ − log ω
2  p   ω p 1
For w >> p, log w − log p ≅ log w. The last two equations are the piecewise linear
asymptotic approximations of the exact pole magnitude.
An ideal Bode plot can easily be constructed from these straight-line approxima-
tions, as shown below.
log 1
j ω +1
p 1
−1
p log w
p
10 p 10 p log w
0°
−45°
−90°
−f
The magnitude graph is flat until it reaches a frequency of p, where the slope
changes to −1 (in log-log coordinates). Hence, p, the frequency of the pole, is
called the corner or break frequency.
The relationship between the real part of a pole or zero and Bode-plot break
frequencies is that the break frequency corresponds to −1/s for negative poles
or zeros. The convention is used here of indicating pole location on Bode plots
as p instead of −p. Because Bode-plot frequencies are always positive, no confu-
sion should result.
Often, magnitude is scaled in decibels (dB). For voltage or current, ||A||dB ≡
20 ⋅ log10 ||A||. (Note that the decibel, like the radian, is a pseudo-unit, a scaling
transformation.) A real pole “rolls off” (that is, decreases in magnitude) with a
slope of −1 on a log-log plot or −20 dB/dec ≅ −6 dB/oct. (A decade (dec) is a
10 to 1 frequency range; an octave (oct) is a 2 to 1 range.) The error in the
46 Chapter 1
asymptotic approximation is greatest at the break frequency, where the actual

curve is at 2 2 ≅ 0.707 , or −3 dB.
The phase angle plot is also subject to asymptotic approximation. On a
semilog plot, the exact phase is
ω
φ = − tan−1  
 p
For w/p << 1,
ω
− tan−1   ≅0
 p  ω p 1
and for w/p >> 1,
ω π
− tan−1   ≅ − tan−1 {∞} = − = −90°
 p  ω p1 2
At w = p, f = −p/4 = −45°. The semilog slope of f at p is
d  −1  ω   −1 ω ln 10
− tan    = ⋅ ln (10 ) ⋅ =−
d ( log ω )   p   ω = p 1 + (ω p )2
p ω=p
2
A line tangent to f at p intercepts the asymptotes at 0° and −90° (−p/2) at fre-

quencies logarithmically symmetric about p, at ap and p/a:
−π 2 − 0 ln (10 )
=−
log (a ⋅ p ) − log ( p a ) 2
Then a = e p/2·ln10 ≅ 2. This linear approximation to f does not minimize the

maximum error. Yet for a = 10, the maximum error is less than 6°. The phase
plot for a single pole is shown above.
Linear approximations of Bode plots for other cases are shown below. Because
frequency-response analysis is linear, these elemental plots can be combined
linearly to produce the total response plot. For complex critical frequencies,
decreasing z increases the magnitude peak and the slope of the phase near the
break frequency.
The maximum magnitude, Mm, for underdamped response occurs at fre-
quency wm. This is derived by setting the derivative of M(jw) to zero and solving.
For a quadratic pole factor
 
d 1 d  1 
=
dω [ j (ω ωn )]2 + j 2ζ (ω ω p ) + 1 d ω  1 − (ω ω )2 2 + (2ζω ω )2 
  n  n 
When set to zero, w is
2
ωm = ωn ⋅ 1 − 2ζ 2 , ζ <
2
and
1 2
Mm = , ζ<
2 ⋅ζ ⋅ 1 − ζ 2 2
For z << 1, Mm varies inversely with z. Construction of an approximate Bode

plot for z < 1 is aided by a few points around the break frequency. Besides the
peak at wm, the magnitude at wn is
1
M ( j ωn ) = =Q
2 ⋅ζ
M crosses 1 at
ω M =1 = 2 ⋅ ωm
48 Chapter 1
log log log

K 1
−1 +1
1
log w p log w z log w
f p
10 p 10p log ω f
0 0 90
f
−90 0
z z 10z log w
10
(a) (b) (c)
log log log

1 1 +1 1
−1
−2
1 log w 1 log w p log w
p
log w f 10 p 10p log w
90
0 0
f f
0 −180
−90
log w
(d) (e) (f)
log log log

1
+2 +1
1 1 +1
z log w p log w z log w
z
10 z 10z log w
180 90
f f 0
f
0 0
−90
z z 10z log w p p 10p log w
10 10
(g) (h) (i)
(a) constant, K. (b) real left half-plane pole, p. (c) real left half-plane zero, z. (d) pole at origin. (e)
zero at origin. (f) complex left half-plane pole pair. (g) complex left half-plane zero pair. (h) real
right half-plane pole, p. (i) real right half-plane zero, z.
The frequency at which M is 1 2 (or −3 dB) down from 1 is defined as the

bandwidth. This definition, like that for risetime, does not assume a particular
kind of response. The bandwidth of a single pole M is its break frequency. For
a quadratic pole, set M = 1 2 and solve for w. This simplifies to
2
  ω  2   2ζω  2
1 −  ω   +  ω  = 2
 n  n
Solving for w (the larger root) yields
ωbw = ωn 1 − 2ζ 2 + 4ζ 4 − 4ζ 2 + 2
Example: Phase-Lag Circuit with Capacitive Output Loading
R1 = 10 kΩ
+
R2 = 100 Ω
Vi C1 = 1 nF Vo
−
C2 = 0.1 µF
This is another passive RC circuit. It was constructed of 1% tolerance metal film

resistors, and 1 nF, 1% mica and 0.1 µF, 2% plastic film capacitors. A 6 V peak-
peak sinusoid for Vi was applied at various frequencies, and the corresponding
peak-peak voltages at Vo were recorded as follows:
frequency, Hz ||Vo||, V ||Vo/Vi||

10 6.00 1.00
20 6.00 1.00
50 5.95 0.99
100 5.10 0.85
200 3.74 0.62
500 1.80 0.30
1000 0.96 0.16
2000 0.47 0.078
5000 0.20 0.033
10,000 0.11 0.018
50 Chapter 1
frequency, Hz ||Vo||, V ||Vo/Vi||

20,000 0.075 0.013
50,000 0.064 0.011
100,000 0.064 0.011
200,000 0.064 0.011
500,000 0.060 0.010
6
10 0.050 0.008
2 * 106 0.035 0.006
From these data, asymptotic approximation of pole and zero frequencies results in
p1 = 150 Hz, z = 15.5 kHz, p2 = 1.6 MHz
The transfer function is
Vo( s ) s ⋅ R2C2 + 1
= 2
Vi( s ) s ⋅ [ R1C1R2C2 ] + s ⋅ [ R1(C1 + C2 ) + R2C2 ] + 1
When the circuit values are substituted, the exact theoretical poles and zero
are
p1 = 156 Hz, z = 15.9 kHz, p2 = 1.62 MHz
The error is largely a matter of parts tolerances, accurate asymptote plotting,

and chart reading.
LOCI OF QUADRATIC POLES

The quadratic equation
as 2 + bs + 1 = 0
has complex roots at
b 1  b 2
s=− ±j −
2a a  2a 
For complex poles,
σ = − 
b
 2a 
1  b 2
ω2 = −
a  2a 
and
b 1 b
α= , ωn = , ζ=
2 ⋅a a 2 a
The coefficients a and b are composed of circuit values when the quadratic
equation is the characteristic equation of a transfer function. Often, a circuit
element value appears in only one coefficient. This allows control of the poles
by varying the value of the element until desired pole placement is achieved.
Consider three cases in which constraints are placed on the loci of the poles
in the s-plane. The + and − markings on the loci associate a locus with the posi-
tive or negative second term in the equation for s. With these precalculated loci,
pole movement is based on variations in circuit element values.
a = constant, b is parameter
jw
b increasing
a>0
− +
s
1
−
b
=−
2 
√a
2a b
−
52 Chapter 1
Let b be the parameter. Solving for b in the equation for s gives

b = −2aσ
and substituting into the equation for w2 gives
1
ω2 + σ 2 =
a
This equation describes the locus of the poles for constant a as a circle with
radius 1 a and centered at the origin, as shown above.
jω
a increasing
b>0 +
1
a=0 b a=0
− +
s
− b =−2 −1
2a b b
b = constant
−
b = constant, a is parameter
From the equation for s, 1/a = −2s/b. Substituting into the equation for w2
results in
2σ  b  2  2σ  2
⇒ ω2 + σ 2 + σ  = 0
2
ω2 = − − −
b  2   b   b 
Adding (1/b)2 to both sides and completing the square,
1 2 1 2
ω2 + σ +  =  
 b b
This equation describes the locus as a circle centered at −1/b with a radius of
1/b, as shown above.
jω
b increasing
b
= constant a increasing
2a
b
>0
2a
+
_ +
s
b 2
− =−
2a b _
b/2a = constant
Whenever −s = b/2a, the locus is a vertical line at
b
σ =−
2a
This locus is shown above.
OPTIMIZATION OF TIME-DOMAIN AND FREQUENCY-DOMAIN RESPONSE

To achieve accurate step response, the major criteria are minimum risetime and
overshoot. For wideband amplifiers, the usual criterion of performance in the
frequency domain is constant magnitude (or flat response) out to a maximum
bandwidth. A wider bandwidth can be achieved at the expense of greater Mm
(or peaking). In the time domain, a faster risetime is achievable if more over-
shoot is allowed. As z and tr decrease, wbw, Mm, and Mp all increase.
Optimization of time and frequency response requires identification of rela-
tionships between the two domains. For an amplifier with one pole (or with a
dominant pole approximation), the relationship between risetime and band-
width is
54 Chapter 1
2.2 0.35
t r ≅ 2.2 ⋅ τ = ≅
2π ⋅ f bw f bw
This relationship is approximate for complex poles with z ≅ 0.7.

Of particular interest are the pole placements for f of 30°, 45°, and 60°.
A complex pole pair at 45° gives a maximally flat amplitude (MFA) response
over frequency (also called a Butterworth response). For f = 30°, the response has
maximally flat envelope or group delay (MFED) (also called a Bessel response).
The major characteristics of these responses are as follows:
Characteristic Critical Damping MFED MFA

f 0° 30° 45° 60°
z 1.000 0.866 0.707 0.500
 ωd  0 0.500 0.707 0.866

 
ωn 
 ω bw  0.644 0.786 1.000 1.272

 
ωn 
Mm – – – 1.155
 ωm  – – – 0.707
 
ωn 
tr⋅wn tr⋅a = 3.36 2.73 2.15 1.64

Mp 0% 0.433% 4.32% 16.3%
From the equation for Mm,
1 2
Mm = , ζ<
2 ⋅ζ ⋅ 1 − ζ 2 2
Mm = 1 when ζ = 2 2 ≅ 0.707 At this value of z, Mm is at the onset of

.
peaking.
Phase delay is defined as
φ
phase delay, τ p ≡ −
ω
This is the delay time of a sinusoid at frequency w with a phase lag of f. If phase
angle decreases linearly with frequency, each frequency component of a waveform
maintains its alignment in time with the others and no waveform distortion
occurs.
A related quantity, envelope or group delay, is defined as
dφ
τg ≡ −
dω
Group delay characterizes amplitude distortion in the time domain. If all fre-
quency components of a waveform are delayed the same amount, they remain
aligned in time and the waveshape remains unchanged. If not, components of
different frequencies are shifted in time, resulting in waveform distortion. A
pole angle of 30° results in a second-order Bessel response, with maximally flat
group delay. For a quadratic pole factor, the phase is
 2ζ (ω ωn ) 
φ = − tan−1  2
1 − (ω ωn ) 
From the tg equation, the group delay is
quadratic pole group delay
= (2ζ ωn )
(ω ωn ) + 1
2
(ω ωn )4 + 2 [2ζ 2 − 1](ω ωn )2 + 1
The denominator can be factored into poles of (w/wn)2, located at
1 − 2 ⋅ζ 2 ± 2 ⋅ζ ⋅ ζ 2 − 1
We can find z for the MFED just as we found Mm. To find the maximum tg,
set dtg/dw = 0 and solve for w. It is more convenient to find (w/wn)2 after the
derivative is taken, and it is
ω
2
max τ g at   = −1 ± 2 1 − ζ 2 , 0 ≤ ζ ≤ 1
 ωn 
56 Chapter 1
Substituting this into the quadratic-pole group delay, the maximum tg

results:
 2ζ  ± 1−ζ2
max τ g =  
(
 ωn  4 (1 + ζ 2 ) 1 B 1 − ζ 2 )
For MFED response, the maximum tg must equal tg at w = 0, or 2⋅z/wn. Setting
the maximum tg equal to this and solving for z yields ζ = 3 2 and a 30° pole
angle.
This subject has been developed further in the electronics specialty of filter
circuits. Higher-order responses are often characterized according to optimal
parameters:
• Butterworth filters have maximally flat amplitude response.
• Bessel filters have maximally flat group delay.
• Chebyshev filters optimize the trade-off between amplitude ripple and sharp
amplitude roll-off (or cutoff) with no ripple in the frequency response above
the cutoff frequency.
• Elliptic (or Cauer) filters have the maximum (or “sharpest”) cutoff for a given
order of filter but have ripple above the cutoff frequency.
The response of common transfer functions are shown below in three repre-
sentations: (1) s-domain pole-zero locations; (2) time-domain step response;
and (3) frequency response.
Right half-plane zeros cause preshoot in the step response. The frequency
response in figure (g) is independent of frequency and is an instance of an all-
pass filter. It is not a Bessel filter, however, because the distorted step response
has too much phase distortion.
Amplifier designs are often a trade-off between conflicting transient and fre-
quency response performance.
s-Domain Step response Frequency response

1 1
p p
(a) undershoot
−z −p z
p
z
z
p p
z overshoot z
(b) 1 p
−p −z
1
z
1
(c) 1
−p z preshoot
−1
1 1
p1 −1
(d)
−p2 −p1
p2 −2
1 1
−p
(e) p −2
2 poles
1 + Mp Mm
wn 1 1
f mm
(f) ringing −2
−f
ωn
tp
(g) 1
58 Chapter 1
Example: Parallel Resonant Circuit
L +
+
vi C R vo
−
−
This RLC circuit provides a way of generating the response of a quadratic pole.
Its transfer function is
Vo (s ) 1
= 2
Vi (s ) s LC + s (L R ) + 1
When L, C, and wn are set to values of 1, then
1 L 1Ω
ζ= ⋅ =
2⋅R C 2⋅R
or R = 1 Ω/(2⋅z). A SPICE simulation produced the response to a unit step input

for the following:
R, W z f, deg
1.000 0.500 60
0.707 0.707 45
0.577 0.866 30
0.500 1.000 0
The step response, frequency response (amplitude and phase), and group
delay are plotted from a SPICE-based simulation. Note that for z = 0.707, the
amplitude remains flat to the highest frequency without peaking (MFA) and
that the group delay for z = 0.866 similarly remains flat longest without
peaking.
1.2
1.0
0.8
vo
0.6
vi
0.4
0.2
0.0
0s 2s 4s 6s 8s 10 s
1 2 3 Time
z 1
2 2 2
1.2
1.0
0.8
Vo
0.6
Vi
0.4
0.2
0.0
1.0 mHz 10 mHz 100 mHz 1.0 Hz 10 Hz
1 2 3 Frequency
z 1
2 2 2
60 Chapter 1
−50
f, deg
−100
−150
−200
1.0 mHz 10 mHz 100 mHz 1.0 Hz 10 Hz
1 2 3 Frequency
z 1
2 2 2
2.5
2.0
1.5
tg , s
1.0
0.5
0.0
1.0 mHz 10 mHz 100 mHz 1.0 Hz 10 Hz
1 2 3 Frequency
z 1
2 2 2
REACTANCE CHART TRANSFER FUNCTIONS OF PASSIVE CIRCUITS

The reactance chart is a powerful aid for graphically determining the magni-
tudes of the transfer functions of passive divider circuits. Combinations of RL
and RC are plotted below.
Asymptotic approximations similar to those of Bode plots are used to con-
struct the ||Z(jw)|| on the next page. The impedance of more complicated cir-
cuits can be built from these basic combinations. These impedance plots follow
directly from the behavior of the elements. For the series RC, at low frequencies
the reactance of C dominates the series RC combination, but at high frequen-
cies, the impedance is dominated by R. The frequency of equal impedance
magnitudes is at the pole frequency 1/RC, where the impedance plot shows an
asymptotic break. The other basic combinations are analyzed similarly. The
break frequency in all cases is 1/t, where t, the time constant, is RC for RC
combinations and L/R for RL combinations.
Transfer functions of voltage dividers are constructed by first plotting the
divider input impedance magnitude ||Zin|| and the impedance ||ZL|| across which
the output voltage is developed. Then ||M(jw)|| is plotted as ||ZL||/||Zin|| by visually
dividing the two impedances. The inverse of ||Zin|| for a linear segment is a
segment with the opposite slope. Division is accomplished by subtraction
because the reactance chart is a log-log graph.
The figure on the page after the next shows examples of transfer functions
constructed with reactance charts for the RC integrator and differentiator.
When the plots of ||Zin|| and ||ZL|| track in slope, ||Vo/Vi|| is constant (a flat, zero-
slope plot). For the integrator, ||ZL|| and ||Zin|| track in slope until w = 1/RC, at
which ||Zin|| becomes flat, causing the transfer function to roll off with ||ZL||. For
the RC differentiator, it is ||Zin|| that slopes below w = 1/RC and causes ||Vo/Vi|| to
slope in the opposite direction. The s-domain transfer functions are given with
the circuit for comparison.
62 Chapter 1
logZ
C
R
R
1 log ω
RC
(a)
logZ
R
L L
1 log ω
L/R
(b)
logZ
C
R C
R
1 log ω
RC
(c)
logZ
L
R L R
1 log ω
L/R
(d)
logZ
C
Z in
R
R Z L
+ +
Vi C Vo 1 log ω
RC
– –  Vo 
log Vi 
1
s RC + 1
1
1 log ω
RC
(a)
logZ
C
Z in
R
C Z L
+ +
1 log ω
Vi R Vo RC
– –  Vo 
log V 
i
s RC
s RC + 1
1
1 log ω
RC
(b)
64 Chapter 1
The basic combinations of the first figure of this section are part of the
dividers shown below.
logZ
C Z in
R1 + R2
R1 Z L
R1 R2
+ +
Vi R2 V 1 1 log ω
o
(R1 + R 2) C R2 C
– C –
 Vo 
logV 
i
s R2 C + 1
s (R1 + R 2) C + 1 1
R2
R1 + R2
1 1 log ω
(R1 + R 2) C R 2 C
The impedance magnitudes for Zin and ZL are plotted. ||Zin|| has a break fre-
quency where the line for C intersects the line for R1 + R2 at w = 1/(R1 + R2)C.
||ZL||| decreases (or “rolls off ”) with C until it reaches R2, where it breaks and is
flat. This break frequency is at w = 1/R2C.
Below wp = 1/(R1 + R2)C, ||Zin|| and ||ZL|| track, and ||Vo/Vi|| is flat. Along this
segment, the input and load impedances are equal, and the transmittance is 1.
At wp, ||Zin|| becomes flat while ||ZL|| continues to roll off (with a slope of −1).
This causes ||Vo/Vi|| to roll off until ||ZL|| breaks at w2 = 1/R2C. Above w2, both
||ZL|| and ||Zin|| are flat, and ||Vo/Vi|| is flat at the ratio of ||ZL||/||Zin||, or R2/(R1 +
R2). As on Bode plots, the phase of Vo/Vi decreases (or lags) whenever the mag-
nitude decreases. This circuit causes a phase lag for frequencies between wp and
wz.
logZ
C Z in
R1 + R2
R1
R2
R1 Z L
R1 R2
+ +
1 1 log ω
Vi C R2 Vo R 2 C (R  R ) C
1 2
– –  Vo 
log V 
i
s (R1  R2 ) C + 1
Zin = (R1 + R 2) R2
s R2 C + 1
R1 + R2
1 log ω
(R1  R2 ) C
In the above circuit, ZL is a parallel RC. Parallel combinations introduce a

slight complication in identifying a break frequency location. First, ||ZL|| is domi-
nated by R2 until wp = 1/R2C, at which the shunt reactance of C dominates and
||ZL|| rolls off with C, as shown. ||Zin|| is ||ZL|| added to R1. On the reactance plot,
this is accomplished by shifting ||ZL|| upward until, at low frequencies, ||Zin|| =
R1 + R2. The break frequency due to ||ZL|| is present in ||Zin||, which rolls off until
it reaches R1, where it again breaks flat. This frequency is found by going down
vertically to the curve for C and reading the resistance from the vertical axis. It
is R1||R2. The associated break frequency is therefore at wz = 1/(R1||R2) ⋅ C.
That the equivalent resistance for wz is R1||R2 can be demonstrated by taking
into account the log-log scaling of the reactance chart axes. The impedance of
C, ||XC ||, rolls off at a (log-log) slope of −1 as does ||Zin|| between wp and wz. By
calculating the slopes of ||XC || and ||Zin|| between wp and wz and equating,
∆ log Z log R1 − log (R1 + R 2 ) log X C (ω z ) − log X C (ω p )

= =
∆ log ω log ω z − log ω p log ω z − log ω p
66 Chapter 1
At wz, ||XC || equals the equivalent resistance we are seeking. This equation
reduces to
log X C (ω z ) = log X C (ω p ) + log R1 − log (R1 + R 2 )
At wp, ||XC(wp)|| = R2. Substituting and simplifying gives
log X C (ω z ) = log R 2 + log R1 − log (R1 + R 2 )

 RR 
= log  1 2  = log (R1 R 2 )
 R1 + R 2 
Therefore, the value of resistance that is read off the graph where wz intersects
the line for C is R1||R2.
For ||Vo/Vi||, ||Zin|| and ||ZL|| are flat to wp and ||Vo/Vi|| is also flat with a value of
R2/(R1 + R2). At wp, both ||Zin|| and ||ZL|| roll off, maintaining a flat ||Vo/Vi|| until
wz, where ||Zin|| flattens. As ||ZL|| continues to roll off, so does ||Vo/Vi||. Conse-
quently, ||Vo/Vi|| has a pole at wz.
A third example is shown below.
logZ
Z in
R1 + R2
R1 Z L
R1 C R2
C
+ +
Vi R2 Vo 1 log ω
(R1 + R 2) C
– –  Vo 
log V 
s R2 C i
s (R1 + R 2) C + 1 R2
R1 + R2
1 log ω
(R1 + R 2) C
And finally, the circuit below is a phase-lead circuit. It causes an increase in

phase as ||Vo/Vi|| increases. This circuit also demonstrates how to handle a paral-
lel resistance in the time constant.
logZ C
R1 + R2
Z in
R1 Z L
R2
R1
R1 R2
+ +
1 1 log ω
Vi Vo R1 C (R  R ) C
R2 1 2
C  Vo 
– – log V 
i
( ) R2
R1+ R2
s R1 C + 1
s (R1  R2 ) C + 1
1
R2
R1 + R2
1 1 log ω
R1 C (R  R ) C
1 2
Its input impedance is
s (R 1 R 2 ) ⋅ C + 1
Z in = (R1 + R 2 ) ⋅
sR1C + 1
68 Chapter 1
Example: Wien-Bridge Filter (continued)
logZ
C Z in
Z L
R
1 log w
RC
 Vo 
log V 
i
1
2
1 log w
RC
The reactance-chart transfer function of the Wien-bridge filter rolls up to

1/RC and then rolls off. Whenever a slope change of 2 or more occurs within
a narrow band (less than a decade) on a reactance chart, the asymptotic approxi-
mation is degraded. The actual poles are at
−0.38 RC
p1,2 = −
1
2RC
(3 B 5) ≅ 
−2.62 RC
These poles are centered at −1/RC on a logw scale.
Example: Inverse of Wien-Bridge Filter (continued)

The reactance chart method applied to the inverse Wien-bridge filter produces
a flat transfer function of 1 through 1/RC. ||ZL|| dominates ||Zin|| below 1/RC by
C and above by R, making ||Zin|| = ||ZL||. If C2 is made large, the transfer function
approaches that of the phase-lead circuit above, with a zero at 1/R1C1 and a pole
at 1/(R1||R2)C1. At low frequencies it is R2/(R1 + R2).
Example: Phase-Lag Circuit with Capacitive Output Loading
R1 = 10 kΩ
+
R2 = 100 Ω
Vi C1 = 1 nF Vo
−
C2 = 0.1 µF
Reactance chart determination of the transfer function of this circuit is simpli-

fied by the wide separation of break frequencies. The low-frequency pole occurs
at about 1/R1C2, or 159 Hz.
The zero is at 1/R2C2, or 15.9 kHz, and the high-frequency pole is at 1/R2C1,
or 1.59 MHz. These frequencies are not exact but are derived from the reac-
tance chart. For example, the exact high-frequency pole is 1.6232 MHz. This
difference is insignificant if the accuracy of a graphical method is considered
adequate.
Example: Cascaded RC Integrators
R1 R2 30
+
+ +
Vi C1 V1 C2 Vo
– –
–
Two RC integrators are cascaded to form a passive filter. Its transfer function
can be found using the reactance chart.
70 Chapter 1
logZ C2 C1 Cs
Cp
R2
Z 1
1 1 log ω
R2 C2 R 2 Cs
The first step is to find V1(s) by loading the first stage with the second. At the
input port, ||Zin|| is constructed on the reactance chart by beginning with the
graph of R2C2. It follows C2 until it intersects R2 at wz = 1/R2C2. C1 shunts this
impedance, with C1 and C2 in parallel at low frequencies (Cp = C1 + C2). The
combined Z1 decreases along Cp until it reaches the break frequency wz. It then
flattens, following the R2C2 curve (following R2), but at a lower resistance. The
curve again breaks where C1 dominates, at wp = 1/R2Cs, where Cs is the series
combination of C1 and C2, and ||Z1|| rolls off.
The situation here is similar to that of the phase-lead circuit above, in which
a curve is shifted from its original location by the addition of another imped-
ance. In it, the R2C curve was shifted upward when the series resistance R1 was
added to it. This caused the capacitive roll-off of ||Zin|| to be shifted to the right
so that its break frequency at R2 was at the same frequency as C when combined
with R1||R2. A similar effect occurs in this example, except that it is due to the
addition of shunt C instead of series R. Because Cp dominates Z1 at low frequen-
cies instead of C2, it reaches wz at a resistance of
 C2 
R2 ⋅ 
 C1 + C 2 
When C1 dominates Z1 at wp, the break in ||Z1|| occurs where this resistance
intersects C1. As the upward arrow on the plot shows, the capacitance that would
result when combined with R2 is Cs. To construct ||Zin||, R1 is added to ||Z1||; it
shifts the graph upward. Then the transfer function with first-stage output Vi
can be constructed from ||Zin|| and ||Zi||. Similar construction for the second stage
(R2C2) and a combination of reactance chart transfer functions produces the
desired transfer function magnitude.
The expression for Z1(s) is found by writing a voltage-divider formula from
the schematic diagram above:
sR 2C 2 + 1 CC
Z 1(s ) = ; C p = C1 + C 2 , C s = 1 2 = C1 C 2
sC p ⋅ (sR 2C s + 1) C1 + C 2
where || is the “parallel” math operator, not a topological descriptor. This expres-
sion is consistent with the reactance graph shown above.
The first-stage transfer function is
V1 s ⋅ R 2C 2 + 1
= 2
Vi s ⋅ [R1R 2C1C 2 ] + s ⋅ [R1C p + R 2C 2 ] + 1
And the overall transfer function is
V o V1 1 1
= ⋅ = 2
Vi Vi s ⋅ R 2C 2 + 1 s ⋅ [R1R 2C1C 2 ] + s ⋅ [R1C p + R 2C 2 ] + 1
It is of interest to note that a double pole at −1/RC does not occur when the
resistors and capacitors are of the same values. Under these conditions,
−0.382 RC
R1 = R 2 = R , C1 = C 2 = C ⇒ ζ = 1.5, p1,2 = 
−2.618 RC
This attempt to design a two-pole filter at −1/RC fails because the second-
stage loading causes the poles to shift. To achieve a two-pole roll-off at a specified
frequency, circuit values must be chosen to make the denominator of the trans-
fer function a perfect square. The minimum value of z = 1 is approached when
the second-stage loading is minimized by making R2 >> R1 and C2 << C1. For
either R1 = R2 or C1 = C2, minimum ζ = 2 .
R1 R2
40 60
+ + + +
Vi C1 V1 V1 C2 Vo
– – – –
72 Chapter 1
An alternative two-pole filter without interstage loading (shown above) has a

voltage buffer separating the stages. The frequency response simulations of the
previous integrator and this one show the difference. For this one, both poles
are at −1/RC for the same RC values. The simulated characteristics are shown
below.
3.0 s
2.0 s
tg
1.0 s
0.0 s
vg(30) vg(60)
1.0 V
0.5 V
0.0 V
1.0 mHz 10 mHz 100 mHz 1.0 Hz 10 Hz
Frequency
vg(30) vg(60)
1.0 V
0.8 V
0.6 V
0.4 V
0.2 V
0.0 V
0s 2s 4s 6s 8s 10 s
Time
v(30) v(40) v(60)
CLOSURE
This survey of linear dynamic response is the foundation for analysis of active
circuits with reactive elements. We will return to the amplifiers in Designing
Amplifier Circuits and extend their analyses to the complex-frequency domain
using the methods presented here in Designing High-Performance Amplifiers. In
practical circuits, the assumption of linearity applies for small-signal amplifiers.
The extensive analysis done here of second-order circuits does not readily apply
to higher-order circuits, and the formulas for tp, Mp, Mm, and z are not neces-
sarily valid when zeros or additional poles are present.
2
Dynamic Response Compensation
PASSIVE COMPENSATION: VOLTAGE DIVIDER
R1 C1
Vi(s)
+
R2 C2 Vo(s)
−
The familiar resistive voltage divider, shown above, illustrates the idea of com-
pensation. When a capacitive load C2 shunts R2, the step response is overdamped
and bandwidth is reduced. To compensate for C2, C1 is added in parallel with
R1. The transfer function of this divider is
V o (s )  R 2  s ⋅ R1C1 + 1
=  ⋅
Vi (s )  R1 + R 2  s ⋅ (R1 R 2 ) ⋅ (C1 + C 2 ) + 1
The addition of C1 introduces a finite zero and makes N(s) and D(s) of the
same degree in s, a condition for an all-pass filter. When the pole and zero are
equated, the (all-pass) compensation condition is
R1⋅C1 = R 2 ⋅C 2
A similar technique can be used with the current-divider dual, in which series
load inductance is compensated by placing series inductance in the other
76 Chapter 2
branch of the divider. To compensate, the L/R time constants of the two
branches are set equal.
1+a
1−a
Now suppose that this divider, or a circuit with a similar transfer function, is
not properly compensated and has a step response like that shown above, in
which the fractional overshoot or undershoot is a. This time response is
 sτ + 1 1   τ 1  −1 τz 1 −τ p 
L−1 z ⋅  = L−1 z + = L  + + 
 sτ p + 1 s   sτ p + 1 s (sτ p + 1)   sτ p + 1 s sτ p + 1 
τ 
= 1 +  z − 1 ⋅e −t τ p
τ p 
At t = 0, the exponential is 1. Its coefficient therefore is a, and the relationship

between the pole and zero time constant is
τ z = (1 + a ) ⋅ τ p
An additional cascaded compensation network with a pole time constant

tpc = tz and a zero time constant tzc = tp results in a flat frequency and step
response. The value of tp can be estimated by observing the transient decay of
the step response. The settling time is 4 to 5 times tp as observed on an oscillo-
scope screen. With this estimate for tp and from measurement of a from the
step response, tz can be calculated from the above equation.
Example: Shunt-Series All-Pass Circuit

This circuit has a terminal impedance of
Dynamic Response Compensation 77
L C
Z
R1 R2
Z = R1 ⋅
(sL R1 + 1) (sR 2C + 1) = R ⋅ s 2(LCR 2 R1 ) + s (L R1 + R 2C ) + 1
s 2LC + s (R1 + R 2 )C + 1 s 2LC + s (R1 + R 2 )C + 1
1
Z has two poles and two zeros. If the poles and zeros cancel, the input resistance
is merely R1 and is independent of frequency. This is achieved when
R  L
LC ⋅  2  = LC , + R 2 ⋅C = (R1 + R 2 ) ⋅C
 R1  R1
or
L
R1 = R 2 = R , = R ⋅C
R
This circuit suggests frequency compensation schemes. A series RC can be com-

pensated with a series RL and vice versa.
Example: Series-Shunt All-Pass Circuit

This is the dual of the previous example, for which
sL R2 s 2LC + s [(L R1 ) + (L R 2 )] + 1
Z = + = R2 ⋅ 2
sL R1 + 1 sR 2C + 1 s [ LCR 2 R1 ] + s [ L R1 + R 2C ] + 1
The all-pass conditions are found by equating N(s) and D(s) and then equating
coefficients:
78 Chapter 2
R1 L
R2 C
L ⋅C ⋅ R 2 L L L
= L ⋅C , + = + R 2 ⋅C
R1 R1 R 2 R1
This reduces to the all-pass conditions:
L
R1 = R 2 = R , = R ⋅C
R
Note that they are the same as for the previous circuit example. For both,
R = Zn.
OP-AMP TRANSFER FUNCTIONS FROM REACTANCE CHARTS

The reactance-chart method can be applied to operational amplifier (op-amp)
circuits to find their transfer functions. With this capability, we can more easily
attend to their compensation. To begin, consider the voltage gain of ideal invert-
ing op-amps:
V o (s ) Z f (s )
=−
V i (s ) Z i (s )
This is an s-domain extension of the inverting-op-amp gain equation (from Design-

ing Amplifier Circuits) and assumes an infinite op-amp bandwidth and gain.
logAv
C 1
R RC
−1
Vi −
Vo
+ 1 log w
(a) (b)
logAv
+1
R 1
C RC
Vi −
Vo
+ 1 log w
(c) (d)
Under these simplifying conditions, the op-amp integrator and differentiator

shown above have the following gains.
Op-amp integrator:
V o (s ) 1
=−
V i (s ) sRC
Op-amp differentiator:
V o (s )
= −sRC
V i (s )
The transfer functions are shown on the Bode plots of figures (b) and (d) above,
to the right of their respective circuits.
For the op-amp integrator, a finite-gain op-amp cannot supply adequate gain
as the input frequency approaches zero. At 0 Hz (dc), the op-amp circuit is open
loop and subject to static drift from offset errors. To stabilize the closed-loop
gain (at some high value and at a low frequency), the feedback capacitor is
shunted by a large resistor, as shown below.
80 Chapter 2
Ri Rf
Vi −
Vo
+
Rf 1
Av = −
Ri · sRf C + 1
The static gain is then −Rf/Ri and the output, though not exactly the integral,
is predictable and stable.
logZ
Rf
Zf
Ri
1 log w
Rf C
logAv
Rf
Ri
1 log w
Rf C
The reactance plots for ||Zf || and ||Zi || are shown above. The ratio, ||Zf ||/||Zi ||, is
the magnitude of the gain ||Av||. At frequencies below wp = 1/RfC, C is effectively
an open circuit, and the gain is determined by the resistors. Above wp, C domi-
nates Rf, and integration occurs; the −1 slope (single-pole roll-off) is character-
istic of time-domain integration.
The op-amp differentiator has similar limitations but at high frequencies. To
limit high-frequency gain, Ri is added in series with C, as shown below.
Rf
Ri C
Vi −
Vo
+
Av = −Rf C s
sRi C + 1
The differentiator is statically stable because of the resistive feedback. Above

wp = 1/RiC, the circuit fails to accurately differentiate, and the gain is deter-
mined by the resistors. The transfer function plot is derived from the reactance
chart as before, shown below.
logZ
Rf
Zi
Ri
1 log w
Ri C
logAv
Rf
Ri
Zi
1 log w
Ri C
The reactance-chart method is not limited to these simpler examples. The

following figure shows a more involved circuit.
Ci Cf
Vi –
Rs Ri Vo
+
sR C + 1
Av = – —
— ————
— —
—— i i
–—————————— ——
s ( Rs + Ri ) Cf (s ( RsRi ) Ci + 1)
82 Chapter 2
For this circuit, ||Zf || = 1/w ⋅ Cf, and ||Zi || is shown on the log ||Z|| plot.
logZ Cf logAv
Ci
Ri + Rs Z i
Ri 1
Rs (Ri + Rs)Cf
RiRs
1 1 log ω 1 1 log ω
Ri Ci (RiRs) Ci Ri Ci ( R Rs)Ci
i
As shown for a similar passive circuit, the addition of Rs to the RiCi plot shifts it
upward to Ri + Rs at 0 Hz. This ||Zi || plot rolls off and intersects Rs at a break fre-
quency that, if it were caused by Ci, would be due to an equivalent resistance of
Rs||Ri. This is shown by the dotted lines with arrows. The upward-shifted ||Zi || plot
rolls off at a capacitive value less than Ci. Since the circuit has no capacitor of
this value, the zero of ||Zi || is referred to the Ci curve (following the dotted lines)
so that its resulting expression is readily interpretable in terms of the circuit
topology. The ||Av|| plot follows, as in previous examples, from the plots of ||Zf ||
and ||Zi ||.
Rc Cc
Ci
Rf
Vi –
Rs Ri Vo
+
Rf ( sRc Cc + 1) ( sRi Ci + 1)
Av = −
R i + Rs (s [ RiRs] Ci + 1) (s [Rf + Rc] Cc + 1)
The op-amp circuit shown above does not have a unique transfer function
plot but depends on the relative values of its poles and zeros. The reactance-
chart method is limited in generality (compared with the s-domain transfer

function Av) because only one case can be plotted. All possible orderings of pole
and zero values have to be considered by generating separate plots. In practice,
the relative (if not actual) values of the elements are known because they are
determined by the functional requirements of the circuit.
logZ Cc logZ Ci
Z i
Rf + Rc Z f Ri + Rs
Rf Ri
Rc Rs
RfRc RiRs
1 1 log ω 1 1 log ω
(Rf + Rc)Cc Rc Cc Ri Ci ( R Rs)Ci
i
The reactance plot of ||Zf || is shifted from the plot of RcCc because ||Zf || = Rf ||Rc
at high frequencies. At 0 Hz, ||Zf || must be Rf. The zero of ||Zf || is set by the RcCc
plot, and ||Zf || has a −1 slope between resistances of Rf ||Rc and Rf. This slope
represents a capacitance greater than Cc but not an actual circuit element value.
Therefore, the break frequency at Rf is found by referring the resistance to the
Cc plot (the dotted line with arrow pointing upward). The resistance at Cc is
Rf + Rc, and the pole of ||Zf || is at 1/(Rf + Rc)Cc. This technique of scaling the
impedance at a given frequency by referring to a reactive circuit element (such
as Cc here) to find the associated resistance is also used to find ||Zi||.
When ||Zf || and ||Zi|| are combined to form ||Av||, the transfer function shown
below results. In the particular plots shown above, Rf > Rc, Ri > Rs, and the order-
ing of poles and zeros as shown is assumed. Again, this frequency response is
not unique but depends on the placement of poles and zeros. Some ordering
limitations are imposed by basic circuits laws. The pole at 1/(Ri||Rs)⋅Ci must
always be higher in frequency than the zero at 1/Ri⋅Ci, and the zero at 1/Rc⋅Cc
must be greater than the pole at 1/(Rf + Rc)⋅Cc. Furthermore, depending on
circuit values, complex poles and zeros are possible for this circuit, and the
reactance chart asymptotic approximations may not be adequate for lightly
damped response.
84 Chapter 2
logAv
RfRc
Rs
Rf
Ri + Rs
1 log ω
1 Ri Ci
1
( Rf + Rc )Cc
1 (RiRs)Ci
RcCc
Noninverting op-amp frequency response is determined with the reactance

chart method in the same way that passive dividers were treated previously. The
difference is that for the op-amp, the closed-loop response is the reciprocal of
the divider H, or
Z f + Zi Z
Av = = Hin
Zi Zi
where ZHin is the impedance of the feedback network from the op-amp output.
On a reactance chart, ||ZHin|| is plotted by adding ||Zf || and ||Zi|| on the chart. Since
asymptotic approximations are used,
log Z 1 + Z 2 = log Z1 2 + Z 2 2
) = log Z
log Z 1 , Z1 Z 2
= ⋅ log ( Z 1 2 + Z 2
1 2
2  2 , Z 2 >> Z 1
Consequently, ||Z1 + Z2|| = ||Z1|| + ||Z2|| under the given constraints, and reactance
chart impedance magnitudes can be asymptotically combined by addition of
individual impedance magnitudes.
FEEDBACK CIRCUIT RESPONSE REPRESENTATION

The feedback techniques in Designing Amplifier Circuits derived closed-loop
response from loop gain. The closed-loop gain Av(s) is similarly determined
from the loop gain GH(s). Feedback in the s-domain is the subject of control
theory, found typically in control and circuits textbooks, and will not be system-
atically developed here. Instead, basic aspects of amplifier stability and desirable
dynamic response are explained, leading to methods for compensation of ampli-
fiers that have undesirable responses.
Of the representations of Av(s), the Bode, Nyquist, and root-locus plots are
the most commonly used. Bode plots are already familiar and present the fre-
quency and phase response. Nyquist plots of the imaginary (jw-axis) and real
(s-axis) components of GH with w as the parameter are an alternative repre-
sentation. For each of these, closed-loop performance is determined by the
loop-gain characteristics.
Root-locus diagrams are s-plane plots of the loci of closed-loop poles with
open-loop gain K as a parameter. As K increases from zero, the closed-loop poles
begin at open-loop poles and proceed toward open-loop zeros (some of which
may be at infinity). When these poles leave the left half-plane, the feedback
circuit becomes unstable. The pole loci can be found by setting the denomina-
tor of Av(s) to zero. Then,
1 + G (s ) ⋅ H (s ) = 0
or
GH = −1 = 1⋅ e ±π
In polar form, the locus conditions are
GH = 1, ∠ (GH ) = ±180°
Locating the loci in the s-plane is simplified by root-locus rules. These rules are
constraints imposed on the location of the closed-loop poles by these constraint
equations. Some of the more commonly used (and easily remembered) rules
are the following:
1. The root loci start at the poles of GH (for K = 0).
2. The root loci terminate at the zeros of GH.
3. There are as many separate root loci as poles of GH.
86 Chapter 2
4. The loci are symmetrical about the real axis.

5. The root loci are on the real axis to the left of an odd number of real poles
and zeros of GH.
6. The sum of the closed-loop poles is constant. (The centroid of the loci
remains constant.)
Other rules can also be constructed from the constraints.
The Bode magnitude plot for an amplifier with a frequency-independent H
and a single, real pole −p is shown below.
log
GH
K
1
H
p (KH + 1)p log w
The amplifier gain is
K
G (s ) =
s p +1
The closed-loop gain for positive K and H is then
=
G K  1
Av (s ) = ⋅
1 + GH  
KH + 1 s (KH + 1) p + 1
The closed-loop response is also that of a single, real pole, but at the fre-
quency of wbw = (KH + 1) ⋅ p. The bandwidth has been extended by KH + 1. This
response is unconditionally stable. (Whenever steady-state frequency response
(that is, jw-axis response) is related to pole locations in s, it is assumed that a
positive value of the real component of the pole location is used for the steady-
state frequency. To be precise, wbw = (KH + 1) ⋅ |−p| for real poles. Frequency
response involves only positive frequencies, and p > 0 for negative poles, so no
confusion should result.) The root-locus plot is shown below.
The open-loop pole at −p moves toward and terminates as the closed-loop

pole −(KH + 1)p.
jw
−(KH + 1)p -p s
Next, consider an amplifier with two poles:
K
G (s ) =
(s p1 + 1) ⋅ (s p2 + 1)
For H constant with frequency, the closed-loop response is
Av (s ) = 
KH  1
⋅
 KH + 1 s 2 (KH + 1) ωn2 + 2ζ s (KH + 1) ωn + 1
Av is also a quadratic pole response. The closed-loop parameters are
ωnc = ωn ⋅ KH + 1 = p1⋅ p2 ⋅ (KH + 1)
and
ζ p + p2
ζc = = 1
KH + 1 2ωnc
88 Chapter 2
For complex poles, both pole angle and magnitude depend on the static loop
gain, as did the single-pole response. That is why static loop gain is the param-
eter of closed-loop pole movement for root-locus plots. For both first- and
second-order loop gain, stability is unconditional. Response can become unac-
ceptably underdamped in Av(s) for excessive loop gain, but the poles remain in
the left half-plane. The Bode magnitude and root-locus plots are shown for
second-order loop gain below.
GH
p1 p2 log w
(a)
jω
−p2 −p1 s
(b)
The Bode plot of ||G|| and ||1/H ||, for the single-pole G and constant H, is shown
below.
log
K G
1
H
p wbw log w
Because the magnitude axis is logarithmic, the difference between the ||G|| and
1/H plots is the loop gain. That is,
log G − log (1 H ) = log GH
These Bode plots are an alternative to calculation and plotting of ||GH|| to

determine response characteristics. We need only plot ||G|| and 1/H separately
and then use 1/H as the unity-gain axis. This applies also for ||H(jw)||. In the
above plot, the open- and closed-loop gains intersect at wbw of Av(s). ||Av|| rolls
off with ||GH|| above this closed-loop bandwidth.
The closed-loop bandwidth can be calculated from the plot. The static
gain magnitude of G is K and since 1/H is constant, the difference between
them is KH on a Bode plot. The slope of ||G|| due to the pole at p is −1. The
w-axis is also logarithmic, a logarithmic frequency difference is a ratio, and
wbw/p = KH + 1. The bandwidth is then
ωbw = ( KH + 1) ⋅ p
and is the same as for the plot of ||GH|| in the first Bode plot.
The figures below show some Bode and root-locus plots for circuits with up
to three poles and two zeros. Bode plots show the gain-phase relationship with
frequency directly and are most useful for compensating fixed-gain amplifiers.
Root-locus plots show the closed-loop poles in the s-plane and how these poles
vary with loop gain. For circuits with three or more poles, the closed-loop poles
can leave the left half-plane with increasing K. The addition of zeros tends
to “bend” the loci back from the jw-axis. This effect is a basis for response
compensation.
90 Chapter 2
−z −p −p −z
p z z p
(a) (b)
−z −p2 −p1 −p2 −z −p1
p1 p2 z p1 z p2
(c) (d)
−p2 −p1 −z −p3 −p2 −p1
z p1 p2 p1 p2 p3
(e) (f)
−p2
−z −p3 −p2 −p1 −p3 −z −p1
p1 p2 p3 z p1 p2 z p3
(g) (h)
−p3 −z −p2 −p1 −z2 −p3 −p1

−z1 −p2
p1 p2 z p3 p1 z1 z2
p2 p3
(i) (j)
FEEDBACK CIRCUIT STABILITY

Circuits with no RHP poles or zeros are minimum-phase circuits. Most circuits are
of this kind. The stability of a minimum-phase circuit can be determined from
a Bode plot. When loop gain, G(jw)H( jw) ≤ −1 (or G(−H) ≥ 1), the feedback
is in phase with (and thus reinforces) the error input with a loop gain magni-
tude ≥ 1, enough to sustain oscillation. In other words, the phase lead or lag
around the GH loop is large enough to invert the waveform and to cause it to
come back to the input in phase. This is positive feedback. When GH(jw) = −1,
then ||GH|| = 1 and f = ±180°. On a Bode plot, when f crosses −180°, stability
requires ||GH|| < 1. Or, when ||GH|| ≥ 1, −180° < f < 180° for stability. This stabil-
ity condition is called the Nyquist criterion.
For minimum-phase circuits, stability can be determined from a Nyquist plot
of GH(jw) by observing whether GH encloses the (−1, j0) point. By traversing GH
as w goes from 0+ to infinity, if (−1, j0) remains to the left of the curve, it is not
enclosed and the circuit is stable (that is, has no closed-loop RHP poles). The
following figure shows some examples of non-enclosing curves.
Im{GH( jw)}
w = 0+
−1 w = 0+ Re{GH( jw)}
b
c
w = 0+
Stability is not as easy to determine for nonminimum-phase circuits – that is,

those with right half-plane poles or zeros. Circuits with RHP zeros can be condi-
tionally stable within a loop-gain range. For minimum-phase circuits, a decrease
in static loop gain K increases the relative stability. But for a conditionally stable
circuit, a decrease in gain can decrease stability instead. The reason for this can
be seen graphically below.
92 Chapter 2
Im{GH}
w=∝ w = 0+
−1 f Re{GH}
GH
The plot of GH(jw) extends above f = −180° with a magnitude exceeding

unity. As magnitude decreases, the phase reverts to the stable side of −180° (to
quadrant III) and skirts around (−1, j0), not enclosing it. The phase again lags
beyond −180° at a loop-gain magnitude of less than unity. Because the plot
crosses −180° on both sides of −1, too great an increase or decrease of K could
cause it to enclose (−1, j0).
Im
B A Re
Enclosed,
Encircled, encircled
not enclosed
The figure above shows a nonminimum-phase circuit polar plot in which GH

encircles points A and B but encloses only A. The complete locus is needed to see
the encirclements and includes GH for w = 0− to w = −∞. The negative frequency

range locus of GH is symmetric with the positive range locus relative to the real
axis. The GH locus in the s-plane closes at infinity (that is, from w = −∞ to
w = +∞) with a counterclockwise path at infinity, enclosing the stable LHP. The
Nyquist criterion must be generalized to include the nonminimum-phase case.
The number of RHP poles must be zero for stability; their number is as
follows:
number of closed-loop RHP poles = number of poles of GH in RHP − net number of
counterclockwise encirclements of (−1, j0) by GH
For nonminimum-phase circuits, stability cannot be determined by enclosure;
the Nyquist criterion requires encirclements instead. For stability the net number
of CCW encirclements of (−1, j0) must equal the number of positive poles
of GH.
Bode plots cover only the positive frequency range of GH and for
nonminimum-phase circuits can be misleading. But for minimum-phase cir-
cuits, stability and (to some extent) major response characteristics can be readily
determined from them. Since most circuits are minimum phase, Bode plots are
usually applicable.
Relative stability is measured by gain and phase margins. The gain margin is
the difference between one and the gain at f = −180°. The phase margin (PM)
is the difference between the phase at a gain of one and −180° and is the amount
of additional phase lag that will make the circuit unstable. Although second-
order circuits are unconditionally stable, phase margin still describes relative
stability whereas gain margin is infinite. Therefore, phase margin is usually more
meaningful in circuits than gain margin.
Gain and phase margins are related to second-order response parameters
such as z, Mp, and Mm. As the margins decrease, the closed-loop damping ratio
zc decreases, and Mpc and Mmc increase. For second-order feedback circuits with
no finite zeros, the relationship between PM and zc is approximately
PM
ζc ≅ , PM in deg, 0 < ζ c < 0.7, 0 < PM < 64°, 2nd-order
100
94 Chapter 2
Because overshoot is a function of zc, by combining equations for Mp and pole

angle, PM can be expressed in terms of overshoot:
M pc ≅ 75 − PM, M pc in %, PM in deg, PM > 20°, 2nd-order
From the Mm equation, Mmc is also a function of zc and can be related similarly to
PM. Pole angle is cos−1(zc); a phase margin of 50° corresponds to a pole angle of 60°
and an overshoot of 25%. This is greater than the 16% of an open-loop second-
order circuit. The exact relationship between PM and zc is found by choosing
1
G(s ) = , H =1
 s   s 
  ⋅ + 2 ⋅ζc 
ω nc   ω nc 
This choice results in a closed-loop transfer function with only a quadratic pole
factor. Solve for the unity-gain crossover frequency wT. Then solve for the phase
margin, and substitute wT. The result, in radians, is
π  2 1 
PM = − tan −1 ⋅ + 1 − 1
2  2 4 ⋅ζc
4

Based on this result, the error of PM in the above approximation of zc is calcu-

lated to be less than about ±5 degrees. The plot below is this function when
converted to units of degrees.
90
70
PM, deg
50
30
10
0 0.2 0.4 0.6 0.8 1.0 1.2 zc

Example: Two-Pole Feedback Amplifier Stability

A feedback amplifier has two poles in G, none in H, and a closed-loop step
response that has 45% overshoot. The static loop gain is 20. What is its phase
margin and damping ratio?
The loop gain has only a quadratic pole factor, so the previous formulas apply.
(If G has zeros or H has poles, they become zeros of the closed-loop gain, and
the quadratic-pole analysis does not apply.) The closed-loop transfer function
is similar to Av(s), where KH is the static loop gain. For overshoot, zc is calculated
from Mp (or found in the overshoot table) on page 39 and is zc ≅ 0.25. From
the prior zc equation, the open-loop z is
ζ = KH + 1 ⋅ζc = 21 ⋅ ( 0.25) = 1.13
z can also be calculated by using the approximations of zc and Mpc with the zc
equation:
ζ≅
(75 − M pc ) ⋅ KH + 1 = 1.37
100
This overdamped open-loop response becomes underdamped when the loop is

closed. From zc, PM ≅ 100·zc = 25°.
Example: Transimpedance Amplifier with Input Capacitance
G(s) Vo
Ii C
96 Chapter 2
This amplifier consists of a voltage amplifier with voltage gain G(s). The feed-
back blocks are
1 R 1
Ti = R = , H =−
sC sRC + 1 sRC + 1
The closed-loop transimpedance is
G (s ) R G (s )
Z m (s ) = Ti (s ) ⋅ = ⋅
1 + G (s ) H (s ) sRC + 1 1 − G (s ) (sRC + 1)
G (s )
= R⋅
sRC + 1 − G (s )
For a single-pole amplifier,
−K
G (s ) =
sτG + 1
Then
Z m (s ) = −R ⋅ 
K  1
⋅
 K + 1 s 2 ⋅[RC τG (K + 1)] + s ⋅[(RC + τG ) (K + 1)] + 1
To find PM, because of the zero in G/(1 + GH) due to H, we cannot use the
second-order approximations even though Zm(s) has no finite zeros. Instead we
apply a more general approach using the Bode plot.
For K = 999, R = 1 MΩ, tG = 159 µs (pG = 1 kHz), and C = 5 pF, then at 0 Hz,
Zm(0) = −999 kΩ, and the closed-loop poles are at
(1000)
(1kHz)⋅ (1000) = 1MHz, = (31.8 kHz ) ⋅ (1000 ) = 31.8 MHz
2π RC
These poles are not too close, but the static loop gain is high. The damping
ratio is low:
b
ζc = = 0.092
2 a
logGH
1000
100
31.4
10
178 kHz
1
1 kHz 100 kHz 1 MHz log f
3.18 kHz 31.8 kHz

f 1 kHz 10 kHz 178 kHz
log f
−45°
−90°
−135°
−159°
From the ideal Bode plot, the PM is about 11°, which is nearly unstable.
PM is found as follows. First, find the gain at the higher pole. It is 31.8 kHz/
1 kHz = 31.8 times less than the first pole. The gain slope between poles
is −1, so the ratio of gains is also 31.8. The open-loop gain at 31.8 kHz is thus
1,000/31.8 = 31.4. The magnitude then decreases at a slope of −2 and crosses
unity gain at
fT = (31.8 kHz ) ⋅ 31.4 = (31.8 kHz ) ⋅ ( 5.60 ) = 178 kHz
We now know fT and proceed to plot the phase. The phase lags of the two
poles overlap. The 1 kHz pole phase range extends from 100 Hz to 10 kHz, and
the 31.8 kHz pole range is from 3.18 kHz to 318 kHz. In the overlap (between
98 Chapter 2
3.18 kHz and 10 kHz), the phase slope is twice that due to a single pole. For a
single pole, phase changes −90° in two decades for a −45°/dec slope. In the
overlap, it is −90°/dec. At the higher pole, f = −135°. To find the additional
phase lag to fT, calculate the number of decades and multiply by the phase slope,
then add −135°:
 178 kHz 
log  ⋅ ( −45° dec ) − 135° = −169°
 31.8 kHz 
Then PM = −169° − (−180°) = 11°. The result from the exact Bode plot is also 11°.
Most feedback circuits have more than two poles and are capable of instabil-
ity. Feedback circuit compensation relies on an intuitive understanding of how
pole and zero placement affects stability. Although optimal compensation tech-
niques exist, they are rarely the most expedient, cost-effective, or reliable ways
to compensate most feedback circuits.
Consider a loop gain with n poles at frequency w = p. The Bode plots of mag-
nitude and phase are shown below. The magnitude rolls off at p with a slope of
−n. The phase lags by −45°⋅n at p and rolls off at −45°⋅n/dec. The frequency at
which the asymptotic approximation for phase crosses −180° is
ω φ = p ⋅10(4 −n ) n
Similarly, for the unity-gain crossover frequency wT of a magnitude with static

gain of K,
ωT = p ⋅ K 1 n
For stability, wT < wf or wf − wT > 0. Subtracting log wT from log wf yields
1  104 −n 
log ω φ − log ωT = ⋅ log  >0
n  K 
logGH 
−n
1
p wΤ log w
f p/10 p 10p wf log w

0
n=1
−45°n
−90°n (−45°)n/decade
−135°n
−180°n
As n increases, f rolls off toward −180° faster than ||GH|| does toward one. As
this frequency difference approaches zero, the maximum K for stability is
approximately
K < 104 −n
For n = 4, K < 1, which is hardly a useful feedback circuit. If the poles are
complex, the situation is worse; f rolls off even faster. Frequency plots for a
three-pole circuit are shown in root-locus plot (f) on page 90.
Now consider the effect on stability of separating the poles. For a two-pole Av,
zc increases with open-loop z, which can be expressed in p1 and p2 by multiply-
ing the factors of G. The coefficients yield
1 p +p
ωn = p1⋅ p2 , ζ = ⋅ 1 2
2 p1 p2
100 Chapter 2
wn is the geometric mean of the two poles and lies midway between them on a
Bode (log w) plot. Relate the poles by a constant g :
p1 = γ ⋅ p2, γ ≥ 0
Then
(γ + 1)
ζ=
2 γ
Minimum z = 1 when g = 1 for real poles. For maximum pole separation of

g = 0 or ∞, z = ∞. For the two-pole case, maximum pole separation increases
stability.
A root-locus plot of two real poles, maximally separated, shows that they
must travel a maximum distance along the real axis before meeting and
becoming complex. This can be generalized from inspection of the root-
locus plot (f) for three poles. A heuristic stability rule suggested by these obser-
vations is
Pole separation increases stability.
COMPENSATION TECHNIQUES
Compensation is often necessitated by circuit imperfections. Parasitic circuit
elements, unavoidable reactive input and output loading, and undesirable
amplifier frequency response are the major reasons. Some of these are shown
for the following op-amp circuit.
The power supply leads to the op-amp terminals contribute series inductance
(L1 and L2). Stray capacitance from the supply terminals to the inputs is signifi-
cant if appreciable high-frequency dynamic voltage is present at the supply ter-
minals. The op-amp inputs have some internal capacitance to ground, causing
a shunt RC with Ri. The op-amp output is an equivalent shunt RL in series with
a voltage source. The inductance is due to gain roll-off above the op-amp band-
width. (See Chapter 3.) This output impedance can resonate with a capacitive
Rf +V
L1
C1
Ri −
+
Cin + Rout
Vi Vo
Rin
− Cin − Lout
C2 CL RL
L2
−V
load. Furthermore, the op-amp usually has several poles. All of this amounts to
a “naturally occurring” unstable circuit requiring response compensation.
A clue to compensation comes from studying root-locus plots on which mul-
tiple poles cause instability (with sufficient gain), as shown below. The inclusion
of a zero in the loop gain causes poles that would head to the right to be “pulled
back” from their course toward the jw axis.
-p3 -p2 -p1
p1 p2 p3
(f )
102 Chapter 2
Root-locus rule g (p. 90) is intuitively powerful for envisioning where the
poles of separate loci will move. They maintain a fixed centroid on the real axis
so that pole movement – say, to the left – is accompanied by corresponding pole
movement to the right.
−z −p3 −p2 −p1
p1 p2 p3 z
(g)
When a zero terminates the movement of a pole to the left (as in root locus
(g), shown above), the poles moving right also cease moving in that direction.
Depending on the order of poles and zero, various loci occur but always act
according to this rule (as seen in root loci (h–j)).
Adding LHP zeros to the loop gain enables the response to be compensated.
So another heuristic stability rule is as follows:
LHP zeros tend to increase stability.
−p2
−z −p3 −p2 −p1 −p3 −z −p1
p1 p2 p3 z p1 p2 z p3
(g) (h)
−p3 −z −p2 −p1 −z2 −p3 −p1

−z1 −p2
p1 p2 z p3 p1 z1 z2
(i) p2 p3 (j)
Now, consider how to apply these heuristic guidelines more specifically as

compensation techniques.
log GH Uncompensated jω

response
1
−4 −3 −2 −1 s
1 2 3 4 log m
Pole-zero cancellation
Pole-zero cancellation places the compensator zero on an offending pole of

GH. If the compensator pole is far removed from its zero, then the offending
pole is effectively shifted far away. Pole-zero cancellation is demonstrated
above.
1 −2.5
−4 −3 −2 −1
1 2 3 4
Phase-lead
Phase-lead compensation places the zero near wf, where f = −180°. This prolongs
a stable phase while magnitude continues to roll off toward one. The compensa-
tor pole is an implementation side effect that must be put somewhere. Since
the zero is placed where phase lead is needed, the pole should be placed at a
higher frequency, well beyond wT, where the additional phase lag it contributes
will occur beyond where the magnitude crosses one. Phase-lead compensation
is demonstrated in the above figure. Because phase-lead compensation occurs
at high frequencies, it mainly affects the transient response.
104 Chapter 2
1
−4 −3 −2 −1.5 0
1.5 2 3 4
Phase-lag
Phase-lag compensation places the compensator pole at a lower frequency than

the zero. The idea is to introduce the pole at a frequency below the poles of
the loop, where the magnitude is flat. By decreasing the magnitude while the
phase lag is still small and then correcting it with a zero, the magnitude is
reduced while contributing little phase lag. This technique allows higher static
loop gain and consequently smaller steady-state error.
Phase-lag compensation mainly affects low-frequency response error because
the compensating pole and zero are placed at low frequencies, relatively near
0 Hz. The step response can have a long-lasting exponential decay (or “tail”)
(see pole-zero cancellation above) before settling to the steady-state value.
Whenever a low-frequency pole and zero are meant to cancel but are mis-
aligned, a dipole is created with a time-domain response showing a long-lasting
exponential.
1
−7 −6 −5 −4 −3 −2 −1
1 2 3 4 5 7
Lag-lead
Compensation
Lag-lead compensation is a combination of lag and lead compensation, in which

two poles and two zeros are introduced into the loop. Both techniques may be
required to stabilize amplifiers with many close poles.
Pole separation can itself be a technique. If the poles are far enough apart, the
magnitude, starting from the lower-frequency poles, has enough frequency
range to decrease to a gain of one before excessive phase lag accumulates. An
important instance of pole separation is dominant-pole compensation, in which one
pole is placed at a frequency much lower than the others (and thus dominates
the response).
Another pole separation technique is pole-splitting, in which a low-frequency
zero is introduced to pull an adjacent pole toward it; all the while the next
higher frequency pole increases in frequency. The effect is just the opposite
of what is usually expected on a root-locus plot; the poles separate instead of
moving toward each other (while satisfying root-locus rules).
One of the simplest of all compensation techniques is static loop gain reduction.
This may not be desirable in many applications due to its reduction in the
beneficial effects of feedback. But for circuits with abundant loop gain (as many
op-amp circuits have), this can be an effective technique.
Although these techniques are usually sufficient to achieve desirable response,
combinations of them may be necessary for highly unstable amplifiers. In addi-
tion to stabilization of the loop with compensators, stages in GH can be individ-
ually compensated. Sometimes a transistor causes an oscillation and must be
stabilized before overall loop response can even be considered. Therefore, good
design practice is to start with an evaluation of stage responses before consider-
ing loop response.
The techniques described in this section have various realizations in analog
circuitry. But a technique and its various realizations (and how to design them)
are different considerations, just as filter types (such as Butterworth, Bessel)
have corresponding circuit realizations (such as state-variable, negative imped-
ance converter, Sallen-Key). The limitations on circuit topology can affect the
choice of technique (bottom-up design) though ideally the nature of the
problem determines the best choice of technique (top-down design).
COMPENSATOR DESIGN: COMPENSATING WITH ZEROS IN H

How can zeros be realized in circuitry, and at what frequencies should they be
placed? Realizable circuits have no fewer poles than zeros. This complicates
compensation because we must be careful where the added poles are placed. If
106 Chapter 2
the pole is less than the zero, the response of (a) below results; if the zero is
less than the pole, (b) results.
−z −p −p −z
p z z p
(a) (b)
Some passive compensator circuit realizations are shown below.
R1
jw
+ +
Vi Vo −p −z s
C
R2
− −
z = 1/R1C Phase-lead
p = 1/(R1R2)C
(a)
jw
+ R1 +
R2
Vi Vo −z −p s
C
− −
p = 1/(R1 + R2)C Phase-lag
z = 1/R2C
(b)
R1
jw
+ +
Vi C1 −p2 −z2 −z1 −p1 s

Vo
R2
C2
− −
z1 = 1/R2C2 Lag-lead
z2 = 1/R1C1
p1 << p2
(c)
For the phase-lead (a) and phase-lag (b) compensators, the separation of pole
and zero depends on the ratio of R2 to R1 + R2. For effective compensation, this
separation must be significant; therefore, the values of R1 and R2 must be sig-
nificantly different. The lag-lead compensator of (c) has transfer function
T (s ) =
(sR1C1 + 1) ⋅ (sR 2C 2 + 1)
s ⋅ R1C1R 2C 2 + s ⋅ (R1C1 + R1C 2 + R 2C 2 ) + 1
2
From (c), the conditions on pole and zero placement are
p1 < z1 << z 2 < p2
and R1C1 << R2C2. The wide separation of these critical frequencies is desirable.
By choosing the separation of the zeros, we can determine the pole and zero
pair separations. A trade-off between these separations must be based on the
particular amplifier requirements.
Cf
Rf
Ri
−
+ Vo
Ci +
Vi
−
(a)
+
+ Vo
−
Vi Rf
−
Cf
Ci Ri
(b)
108 Chapter 2
These are not the only compensator realizations. The uncompensated ampli-
fier topology affects choice of design, especially if the compensator can be syn-
thesized from part of the given topology. What follows are particular amplifier
compensation techniques.
The op-amp shown above has one op-amp pole p in G(s). Another pole is due
to the input capacitance Ci. If the poles are too close, compensation might be
needed. Inserting a phase-lead compensator in cascade with either the input or
output of the op-amp is undesirable because it decreases input resistance or
increases output resistance. Because H consists of a voltage divider, it can be
modified to form a phase-lead compensator. The topology of H is familiar; it is
an uncompensated voltage divider. A compensation capacitor Cf is placed in
parallel with Rf. Then
 Ri  sR f C f + 1
H = −  ⋅
 R f + Ri  s (R f Ri ) ⋅ (C f + C i ) + 1
H is equivalent to the compensated divider formula. If the pole is set to equal

the zero, H becomes an all-pass network, and the pole due to Ci is cancelled. In
this case,
R 
Cf =  i  ⋅C i
 Rf
Example: Op-Amp Input Capacitance Compensation

The amplifiers of (a) and (b) above have values of Ri = 47 kΩ, Rf = 220 kΩ,
Ci = 100 pF, K = 100 k, and p = 10 Hz. The uncompensated loop has an op-amp
pole at 10 Hz and a pole due to Ci at 1/(220 kΩ || 47 kΩ)(100 pF) or
41.1 kHz.
From the simulated open-loop Bode plot of GH, shown below, fT ≅ 80 kHz
and PM ≅ 26°.
84.867 −5.7066
43.334 −40.560
Magnitude, dB
1.8000 −75.414
Phase, log
0
−39.733 −110.27
−81.288 −146.12
−154
−122.80 −179.98
1.0000 100.00 10000.0 +1.00E + 0.6
Frequency, Hz
(zc or Mpc cannot be applied here because H has a pole. Poles of H become
zeros of the closed-loop gain.) The closed-loop frequency response is shown
below.
28.413 27.934
14.732 4.3472
Magnitude, dB
−19.240
Phase, deg
1.0513
−12.630 −42.826
−26.311 −66.413
−39.992 −90.000
1.0000 100.00 10000.0 +1.00E + 0.6
Frequency, Hz
110 Chapter 2
The closed-loop step response is also simulated, as shown below.

6.2444
5.8654
Amplitude, V 5.4865
5.1076
4.7286
4.3497
0.0000 +2.00E−05 +4.00E−05 +6.00E−05
Time, s
The addition of Cf creates a zero at 1/(220 kΩ)⋅Cf. According to the design

formula, Cf = 22 pF. The compensated response has a single pole at 10 Hz, with
a phase lag approaching −90° and a 90° PM.
log GH
p comp.
pH comp.
57 Hz
10 Hz 100 Hz 1 kHz 10 kHz 100 kHz

log f
We could have chosen to cancel p with the zero instead. If p were cancelled, Cf
would be 72.3 nF, and the magnitude plot would be flat to the compensator pole,
now at only 56.8 Hz, with single-pole roll-off from this pole. This alternative pro-
vides greater loop bandwidth. For the inverting op-amp, the larger Cf causes a
lower-frequency pole in Ti, reducing closed-loop bandwidth considerably.
The maximum amount of phase lead that a phase-lead compensator intro-
duces into a loop depends on the separation of its pole and zero. From the
asymptotic approximation for phase, both pole and zero linearly affect phase
for one decade on each side of them. If they are separated by two decades, the
zero achieves a full 90° of phase lead before the pole begins to take effect.
Consequently, maximum phase lead equals
(45°) ⋅ log  p  , 1 ≤  p  ≤ 100

 z z

90°,   > 100
 p
 z
The frequency of maximum phase lead is at p ⋅ z or about p/10, where the

pole begins to cancel the effect of the zero. The frequency range over which
phase-lead compensation occurs is
log ( p z ) dec, 1 ≤ p z ≤ 100


 2 dec, p z > 100
Example: Op-Amp Phase-Lead Compensation

Cf
10 kΩ
Rf
Ri
+ −
1 kΩ Vo
+
Vi
A fast op-amp has a gain of 2.2 M and poles at 100 Hz and 1 MHz.
105.98 −5.7051
73.987 −40.449
Magnitude, dB
−75.194
Phase, deg
41.998
10.009 −109.94
−21.981 −144.68
−53.970 −179.43
10.000 1000.0 +1.00E + 05 +1.00E + 07
Frequency, Hz
112 Chapter 2
The above Bode plot of GH (without Cf) shows that phase lag approaches −180°,
causing oscillatory response. Cf is introduced to phase-lead compensate the
loop. The closed-loop response is shown below.
33.624 −1.4489
22.962 −36.909
Magnitude, dB
−72.370
Phase, deg
12.300
1.6374 −107.83
−9.0249 −143.29
−19.687 −178.75
+1.00E + 06 +1.00E + 07
Frequency, Hz
Bode plot gain at the second pole frequency and the unity-gain frequency fT
are of interest. On a log-log plot, a line with slope n relates changes in magni-
tude and frequency according to
n
 y2   f 2 
 y  =  f 
1 1
The loop gain is 2.2 M/11 = 200 k. The pole separation of p1 and p2 is
1 MHz /100 Hz or 4 decades. With a negative slope, the loop gain is reduced
over 4 decades at p2, to 200 k/104 = 20. For fT, the slope is −2 and
fT = (1 MHz ) ⋅ 20 1 ≅ 4.47 MHz
The phase plot (dashed curve) shows a phase lag of −169° at fT, and PM =
11°. At p2, the phase lag is −135°, and the phase rolls off at −45°/dec. At
4.47 MHz, then over
log (4.47 MHz 1 MHz ) = 0.65 dec
this results in an additional 29°, or a total of −164°. This linear approximation

is +5° in error. The phase plot value results in z ≅ 0.11, a pole angle of about
84°, and a step-response overshoot Mp, of 71%. This response is too under-
damped for many applications.
The maximum phase lead that can be introduced is
 ( R R )C 
(45°) ⋅ log   = (45°) log  f i f  = (45°) ⋅ log  f + 1
p R
 
z  R fC f    Ri
or (45°)(log 11) ≅ 47°. The phase lead of the zero occurs over a frequency range
of log (11) ≅ 1.04 dec. If the high end of this range is placed at the compensated
fT, or fTc, then phase lag is held constant from fTc/(p/z) to fTc. This placement of
phase-lead range is accomplished by noting that p begins to affect phase at p/10.
So we set
p 10 ⋅ fTc
fTc = , z=
10 (p z)
An increase in z from this placement fails to use the full range of phase lead. A
decrease in z increases fTc until the break to a −2 slope occurs at fTc.
logGH
z decreasing
fTc2
z2 z3 f T fTc1 log w
114 Chapter 2
Then fTc remains fixed as z continues to decrease. When a decreasing z

increases fTc, the magnitude slope at fTc is −1, and phase is decreasing. For
maximum PM, phase should begin to decrease again at fTc. The phase-lead range
is then placed with the high end at fTc.
We still must relate fTc to the uncompensated plot. At z the two plots roll off
at their respective slopes to fT and fTc. Because their gain change is the same,
their locations depend on the slope differences. Consequently, using the slope
equation results in
−1 −2
 fTc  =  fT  ⇒ f = fTc ⋅ z
 z   z  T
Combining with the fTc equation,
p ⋅z
p = fT ⋅ 10 ⋅  
10 p
fT = , z = fT ⋅ ,
10 p z z
For the example, fT = 4.47 MHz and p/z = 11. Then
p
z = 4.26 MHz, p = 11⋅ z = 46.9 MHz, fTc = = 4.69 MHz
10
At z/10 = 426 kHz, the phase is
 426 kHz 
log  ⋅ ( 45°) − 135° = −118°
 1 MHz 
This is an improvement of about 46° (as calculated from the maximum-phase

equation) over the uncompensated amplifier. The compensated PM ≅ 57°.
Phase-lag compensation of the amplifiers in (a) and (b) above can be imple-
mented in H by connecting a series RC between the op-amp input terminals, as
shown below.
Rf
Ri
−
+ Vo
Ci +
Vi Rc
−
Cc
(a)
+
+ Vo
Cc −
Vi
Rf
− Rc
Ri Ci
(b)
Given an op-amp with dominant pole at p in G(s), then
 Ri  sRcC c + 1
H (s ) = −   ⋅ 2
 R f + Ri  s [RcC c R pC i ] + s [R p (C c + C i ) + RcC c ] + 1
where Rp = Rf ||Ri. For Rc << Rp and Cc >> (Rp/Rc)·Ci,
 Ri  sRcC c + 1
H (s ) ≅ −   ⋅
 R f + Ri  (sR pC c + 1) ⋅ (sRcC i + 1)
The effect of this compensation is to add a pole and zero. Because Rc << Rp, the
pole frequency, 1/Rp ⋅ Cc, is less than the zero frequency, 1/Rc ⋅ Cc. For Cc >>
(Rp/Rc)·Ci, this zero is less than the second pole, 1/Rc ⋅ Ci. This results in a pole
and zero ordering of
1 1 1 1
p< < < <
R pC c RcC c R pC i RcC i
116 Chapter 2
The effect of the compensation network is shown below, which demonstrates

phase-lag compensation.
logGH
−1 Uncompensated
response
−2
1 1
−1 Rp C i Rc C i
p 1 1 log w
Rp Cc Rc C c
−2
Above 1/Rp ⋅ Cc, the magnitude decreases at a steeper slope at small phase
angles. Then, as w approaches a gain of one, the zero is introduced (at 1/Rc ⋅ Cc)
to reduce the phase-angle slope and to increase phase margin. The root-locus
plot is shown below.
jω
–1 –1 –1 –p σ
Rp C i Rc C c Rp Cc
Another way to add a zero to H when a feedback capacitor is used for

compensation is to add a resistor in series with it. The effect of this Rc is to
add a zero at 1/Rc ⋅ Cc and to move the pole at 1/Rf ⋅ Cc down in frequency to
1/(Rf + Rc) ⋅ Cc.
Example: Phase-Lag Compensation
Rf
Ri 10 k Ω
1 kΩ
–
+ Vo
+
Vi Rc
–
Cc
We want to phase-lag compensate the amplifier for maximum PM. The op-amp
has poles at p1 = 100 kHz and p2 = 1 MHz and a static gain of 2200. A phase-lag
compensator is realized by adding a series RC from the inverting input of the
op-amp to ground with elements Rc and Cc. Then
 Ri  s ⋅ RcC c + 1
H (s ) = −   ⋅
 R f + Ri  s ⋅[Rc + (R f Ri )]⋅C c + 1
Substituting values, H(0) = 1/11 = 0.0909 and GH(0) = 200. To achieve as much
PM as possible, the pole and zero of H must be widely separated. The ratio of
the H zero to pole is
z R f Ri
= 1+
p Rc
and is large when Rc << Rf ||Ri. Rf and Ri are given, allowing us to both place
and separate z and p by selecting Rc and Cc. If the high end of the phase range
of z is placed two decades below p1, then phase is restored to zero at p1/10,
but the gain will have rolled off by a decade to 20. By placing z at 1 kHz,
RcCc = 159 µs. For a full 90° of phase lag, z/p = 100. Substituting into the
above equation gives Rc = 91.8 Ω. The closest 5% resistor value is 9.1 Ω. Cc is
159 µs/9.18 Ω = 17 µF, or a 5% tolerance value of 18 µF.
118 Chapter 2
Estimate the compensated PM as follows. The gain at p1 is 20, and f must be

−45°. The gain rolls off another decade to p2, where it is 2. Then
fTc = 2 ⋅ (1 MHz ) = 1.41 MHz
This is log (1.41), or 0.15, decades above p2. At p2, f = −90°. At −90°/dec, the
additional phase lag at fTc is 59° or −149° total, and the PM = 31°.
Example: Transimpedance Amplifier with Input Capacitance (continued)

In the transimpedance amplifier circuit on page 95, z was unacceptably low. We
can apply phase-lead compensation in H by shunting R with a compensation
capacitor Cc.
The open-loop poles are at frequencies of 1/R ⋅C and 1/tG = pG. The addition
of Cc shifts the pole in H to pH = 1/R ⋅ (C + Cc) and adds a zero at z = 1/R ⋅ Cc.
With phase-lead compensation, z < pH. The root locus plot is shown in (c) on
page 90. The complex pole locus bends left in a circle and rejoins the real axis
above the zero. Because the static loop gain is given, pole placement for the
desired response largely depends on the placement of z.
The addition of Cc results in a closed-loop transimpedance of
sR (C + Cc ) + 1
= −
Vo K 
⋅ 2
Ii  
K + 1 s [τG R (C + Cc ) (K + 1)] + s [τG (K + 1) + RC (K + 1) + RCc ] + 1
The closed-loop z of the poles is derived as
b 1 (RC + τG ) (K + 1) + RCc
ζc = = ⋅
2 a 2 τG R (C + Cc ) (K + 1)
For a maximally flat envelope or group delay (MFED) pole placement,

ζ c = 3 2 . Substituting into zc and solving for Cc it is 2.3 pF.
COMPENSATOR DESIGN: REDUCING STATIC LOOP GAIN

Phase-lag compensation reduces loop gain except at low frequencies. The
simpler technique of reducing K does not require a compensation capacitor,
Cc, and does not appreciably degrade static performance for amplifiers (such
as op-amps) with high loop gains. Neither does it introduce compensation
poles and zeros that must be readjusted when closed-loop gain is adjusted.
The Bode magnitude plot is shifted downward. With Cc shorted, Av of the
inverting op-amp, shown on the next page in (a), is not affected by Rc but
the loop gain is. The inverting op-amp has input attenuation Ti = 1 + H;
both Ti and H are affected by Rc such that Av remains unchanged. Consequently,
GH can be adjusted by adjusting H with Rc without affecting closed-loop
gain.
The noninverting op-amp configuration can also be compensated by Rc. To
achieve the same result, Rc must be placed across the op-amp inputs in (b). The
flow graph for this topology has a transmittance of Ti in front of the feedback
loop. The feedback equations are
Vo = K ⋅ E , E = Ti ⋅Vi − H ⋅Vo = Vi − V −
The transmittances are
Rc
Ti =
Ri R f + Rc
Rc Ri
H =
Rc Ri + R f
G =K
Combining these equations yields the closed-loop gain
 R f + Ri  [ K (1 + K )]⋅[ Rc Ri ]  R f + Ri 
Av =  ⋅ =
 Ri  Rc Ri + R f (1 + K )  Ri  K →∞
This is the familiar noninverting op-amp gain formula when K → ∞. Av is not

affected by Rc, but the loop gain is. Because GH = KH, the effect on loop gain
is to attenuate H by Rc shunting Ri. An apparent disadvantage of this topology
is that Rc reduces the input resistance. But the effect is minimal with large K
because Rc is across E, a small voltage, and is bootstrapped.
120 Chapter 2
Rf
Ri
E
Vi −
Vo
+
Rc
(a)
Vi
+ +
Vo
E −
Rc −
Rf
Ri
(b)
Ri and Rf can be generalized to impedances. The frequency-dependent closed-

loop gain is unaffected by Rc whereas the static loop gain is reduced. This shifts
the Bode magnitude plot downward, causing it to cross a gain of one at a lower
frequency, where the phase lag is less.
COMPENSATOR DESIGN: POLE SEPARATION AND PARAMETER VARIATION

One of the simplest compensation techniques is pole separation by dominant-pole
compensation. If one pole is introduced into a feedback loop at a much lower
frequency than the other loop poles, it causes the gain to roll off at −20 dB/dec
(−1 slope) over a large frequency range until the next pole is encountered.
If the range is large enough, the unity-gain frequency is less than the remaining
poles, so that their influence is inconsequential. An existing pole can often be
reduced in frequency by modifying the value of its associated circuit
elements.
R
Vo
+
Vi
−K
− s/pG + 1
Another pole separation technique, pole-splitting, is commonly applied by

placing a capacitor from input to output across an inverting amplifier stage with
a dominant pole. The diagram above shows a feedback amplifier with dominant
forward-path pole at −pG and a transfer function of
1
G = −K ⋅
s pG + 1
The feedback path is that of an RC differentiator. The resulting loop gain is
K sRC
GH = ⋅
s pG + 1 sRC + 1
The closed-loop root locus is plotted below.
jw
−1 −pG s
RC
122 Chapter 2
Because of the zero at the origin, the pole at −pG migrates to the right while the
pole at −1/RC increases in frequency. This is pole-splitting. The poles split apart
instead of coming together and thereby achieve pole separation. The closed-
loop transfer function is
V o (s ) 1
= −K ⋅ 2
V i (s ) s (RC pG ) + s [( K + 1) RC + 1 pG ] + 1
Because of the Miller effect, evident in the linear term of the denominator, the
integrator pole moves away from pG as K increases, separating the poles.
Zf
Vi
−Gm Vo
Ii Zi ZL
To analyze the shunt capacitive realization of pole-splitting in more detail, the

figure above shows an inverting amplifier stage with its flow graph, shown below.
ZfZL
G1 =
−Rm
ZiZf Vi
Ii Vo
ZL
G2 =
Zf + ZL
Zi
Zf + Zi
The active forward path is a transadmittance amplifier with a gain of −Gm =

−1/Rm < 0. The input and output are both loaded by generalized impedances
Zi and ZL, and the transadmittance amplifier is shunted by a feedback imped-

ance Zf. From the flow graph, the closed-loop transimpedance is
Vo
= (Z f Z i ) ⋅
(Z f Z L ) (Z f −Rm )
Ii 1 + [(Z f Z L ) (Z f −Rm )]⋅[Z i (Z f + Z i )]
and Zin is
Zf
Z in = Z i
1−G
For the simpler case of no loading, Zi and ZL are removed, and the transimped-
ance is
Vo
= −Z f + Rm
Ii Z i ,Z L →∞
Because the output quantity is a voltage, a low-impedance output is desirable

(to approximate a voltage source). In this case, ZL << Zf, and G2, the passive path
in G, is negligible. Then
Z f ZL Z
G1 = ≅ L
− Rm − Rm
and
 Z   Zi 
G1H =  L  ⋅ 
 Rm   Z f + Z i 
To make the circuit more specific, let ZL contribute a single pole due to a
parallel RC load:
RL
ZL =
sR LC L + 1
124 Chapter 2
Similarly, let the input loading of the Gm amplifier be a parallel RC:
Ri
Zi =
sRiC i + 1
and let the feedback impedance be a capacitance:
1
Zf =
sC f
These choices of impedances simulate a common-emitter (CE) (or common-

source [CS]) stage with collector-base (or drain-gate) capacitance. Substituting
these impedance expressions,
RL sRiC f
G1H = ⋅
Rm [sRi (C f + C i ) + 1]⋅ (sR LC L + 1)
The frequency response and root loci are plotted below.
logG1H
−1 −1 log w
Ri (Cf + Ci) RL CL
jw
−1 −1 s
RL CL Ri (Cf + Ci)
The zero at the origin splits the poles. Because the zero is not positive, there
is no danger of instability with too much static loop gain, Gm. As gain increases,
however, the lower-frequency pole decreases in frequency, and the bandwidth
is correspondingly decreased.
The above assumed that G2, the passive forward path through Zf, was negligi-
ble. If we extend this analysis to include it, we get some interesting and impor-
tant results. The complete G is
Z f ZL ZL Z f ZL
G = G1 + G 2 = − + =
Rm Z f + Z L Z f − Rm
Specifically, G1 and G2 are
RL 1
G1 = − ⋅
Rm sR L ⋅ (C f + C L ) + 1
ZL sR LC f
G2 = =
Z f + Z L sR L ⋅ (C f + C L ) + 1
Then G becomes
RL −sRmC f + 1
G =− ⋅
Rm sR L ⋅ (C f + C L ) + 1
This more complete expression for G has an additional RHP zero at +1/Rm⋅Cf.
The loop gain is
GH =
RL
⋅
( −sRmC f + 1) ⋅ (sRiC f )
Rm [sR L (C f + C L ) + 1]⋅[sRi (C f + C i ) + 1]
The root locus of GH is not directly obtainable as before because the RHP
zero varies with Rm, and RL/Rm is the static loop gain. Root-locus plotting is based
on fixed open-loop poles and zeros that are independent of static gain. In GH,
both elements that affect static loop gain also affect a pole and zero. This sug-
gests limits on the applicability of the root-locus technique.
126 Chapter 2
Consequently, we proceed directly to the closed-loop voltage gain Vo/Vi :
Vo R ( −sRmC f + 1) ⋅[sRi (C f + C i ) + 1]
=− L⋅
Vi Rm s 2a + sb + 1
where
a = Ri R L ⋅ (C iC f + C LC f + C iC L )
 R 
b = RiC f ⋅  1 + L  + RiC i + R LC f + R LC L
 Rm 
Finally, the closed-loop transimpedance is
Vo V V R −sRmC f + 1
= Z f Z i ⋅ o = Z f ( − H ) ⋅ o = −Ri . L ⋅ 2
Ii Vi Vi Rm s a + sb + 1
This expression has two LHP poles and one RHP zero. Its characteristic equa-
tion is the same as the voltage gain equation.
The Miller effect is evident in the linear coefficient b in the first term, where
the Miller capacitance Cf is multiplied by the static voltage gain plus 1. As Rm
decreases, b increases while a remains constant. This causes the poles to move
apart with decreasing Rm. It also causes the RHP zero to increase in frequency.
The LHP zero of Vo/Vi remains fixed as Rm varies. The movement of its poles
and zero with decreasing Rm (or increasing Gm) is shown below. The poles move
to zero and −∞; the RHP zero goes to +∞.
jw
Rm 0
−p2 −p1 ∞
+1 s
−1 Rm C f
Ri (Cf + Ci)
A root-locus plot with a circuit element as parameter is a root contour plot. If

KGH can be reformulated as X⋅F(s), where F(s) is independent of X, then X is
a constant relative to F and can be varied in the root-locus equation, KGH =
XF = −1. The root-locus rules then apply, and the pole loci are mapped as a
parameter of X. If a formulation compatible with the root-locus technique is
not feasible, then movement of the poles and zeros of the closed-loop transfer
function with variation of a circuit element can still be investigated. The result-
ing closed-loop parameter variation technique is often quite useful in determining
the effect of a circuit element on the dynamic response.
If Cf is made the parameter instead of Rm, the movement of the poles with Cf
can be plotted. As Cf increases in Vo/Ii, the RHP zero decreases toward zero. For
the poles, Cf is in both a and b. We can estimate pole location from extreme
values of Cf. When Cf = 0, the poles are located at −1/Ri⋅Ci and −1/RL⋅CL. These
are also the open-loop poles for Cf = 0. As Cf increases slightly from zero, the
poles will decrease until Cf dominates (that is, Cf >> Ci, CL). Then the quadratic
pole factor becomes
  R  
s 2C f ⋅[Ri R L ⋅ (C i + C L )] + sC f ⋅  Ri ⋅  1 + L  + R L  + 1
  Rm  
In solving for the poles, we find that the b/2a term is independent of Cf. Also
the (Ci + CL) factor in a is not found in b. If it is varied (as long as Cf continues
to dominate), the pole loci will move together and form a complex circular arc
centered at the origin. (See “Loci of Quadratic Poles” for a description of root
loci due to parameter variation.)
As Cf → ∞ in Vo/Ii, both poles and zero move toward the origin. One pole and
the RHP zero reach the origin and cancel, leaving a single pole at
−1
(R L Ri Rm ) ⋅ (Ci + C L )
An infinite Cf is a short between input and output such that Vo = Vi. The substi-
tution theorem applies to the transadmittance current source, transforming it
into a resistance across Vi of value Rm. All circuit components are in parallel.
128 Chapter 2
Therefore, a very large Cf couples input and output so that their separate poles
are effectively merged into one.
From the open-loop gain, GH, if CL >> Cf, then Cf has negligible effect on
the open-loop pole at −1/RL(Cf + CL); it remains relatively fixed as Cf increases.
For the other open-loop pole, if Ci is not much greater than Cf, it moves
appreciably to the right. Under these conditions, variation in Cf causes pole
separation.
Now analyze the closed-loop transimpedance of this amplifier for other
choices of circuit impedances. Consider new circuit conditions, in which Zi is
removed and the load is only capacitive:
1 1
ZL = , Z i → ∞, Z f =
sC L sC f
Then
Vo 1 −sRmC f + 1
= ⋅
I i sC f sRmC L + 1
G has the RHP zero and a pole at the origin with coefficient Rm(Cf + CL). H = −1
and GH = −G. As Rm decreases, closed-loop pole and zero separate.
jw
ZL = 1/sCL
−1 +1 s
RmCL RmCf
The pole and zero can be adjusted independently by varying CL or Cf,

respectively.
If a resistive Zi is added, the modified conditions are
1 1
ZL = , Z i = Ri , Z f =
sC L sC f
Then
−sRmC f + 1
G =−
sRm (C f + C L )
and
Ri sRiC f
H =− =−
1 sC f + Ri sRiC f + 1
Vo − Ri −sRmC f + 1
= ⋅
I i Rm (C f + C L ) + RiC f s ⋅ (s ⋅ ([Ri (C f C L )] RmC L ) + 1)
The closed-loop poles and zero are plotted below.

jw
ZL = 1/sCL
Zi = Ri
−1 1 s
[Ri (CfCL)]RmCL R m Cf
The static factor, a pole, and the RHP zero vary with decreasing Rm. Again, pole
and zero move outward, away from the origin. This circuit behaves as an integra-
tor due to the fixed pole at the origin. The effect of adding Ri is only to shift the
non-zero pole.
130 Chapter 2
Ri is now removed and RL added. The conditions are
1 RL 1
Z L = RL = , Z i → ∞, Z f =
sC L sR LC L + 1 sC f
For these conditions, H = −1 and the open-loop gain is
RL −sRmC f + 1
GH = ⋅
Rm sR L (C f + C L ) + 1
Vo  RL  1 −sRmC f + 1
= −  ⋅ ⋅
Ii  R L + Rm  sC f s (Rm R L ) ⋅C L + 1
Again, the s-plane plot is similar to the previous two cases, with the pole location
modified due to RL.
jw
ZL = RL1/sCL
−1 1 s
(RmRL)CL Rm Cf
Finally, consider the addition of a resistive Ri to the circuit of the above trans-
conductance equation. The conditions are then
1 RL 1
Z L = RL = , Z i = Ri , Z f =
sC L sR LC L + 1 sC f

Vo R −sRmC f + 1
= − Ri ⋅ L ⋅ 2
Ii Rm s (R L RiC LC f ) + s [R L (C f + C L ) + RiC f (1 + R L Rm )] + 1
The effect of resistances at both input and output is to move the low-frequency
pole off the origin. The circuit no longer behaves as a pure integrator.
jw
ZL = RL1/sCL
Zi = Ri
−p2 −p1 1 s
Rm Cf
When Rm decreases, the poles split in the usual way; one goes to the origin
and the other to −∞. Again, this locus of poles is due to variation in the linear-
term coefficient of the denominator of Vo/Ii. The quadratic coefficient remains
constant with Rm.
This extended analysis of the generic transconductance amplifier demon-
strates the conditions for pole-splitting due to variation in static loop gain Gm
and in Cf. The limitation of the root-locus technique was largely overcome by
closed-loop parameter variation. This circuit is representative of CE and CS
amplifiers and wideband amplifiers in general.
Example: Transimpedance Amplifier Pole-Splitting
Cf
−1
Rm Vo
Ii Ri Ci RL
CL
132 Chapter 2
This amplifier is an idealized form of inverting transistor amplifier with a feed-

back capacitance. Let
R L = 1 kΩ, Rm = 100 Ω, Ri = 10 kΩ, C f = 10 pF, Ci = 10 pF, C L = 90 pF
From Vo/Ii, solve for the zero and poles. They are
z = 159 MHz, p1,2 = 124 kHz, 10.8 MHz
The Bode plot confirms the pole and zero values.
−20
−40
Magnitude, dB
−60
−80
−100
−120
10 kHz 100 kHz 1.0 MHz 10 MHz 100 MHz 1.0 GHz
Frequency
200
150
100
Phase, deg
50
−50
−100
10 kHz 100 kHz 1.0 MHz 10 MHz 100 MHz 1.0 GHz
Frequency
As b remains constant with Rm, the poles should move toward each other as
Rm is increased. Calculating again for Rm = 1 kΩ,
z = 15.9 MHz, p1,2 = 461 kHz, 2.89 MHz
As predicted, the poles are now closer. With a further increase in Rm, they will
eventually become complex.
TWO-POLE COMPENSATION
High-performance feedback amplifiers require high loop gain over a wide fre-
quency range. Dominant-pole (single-pole) compensation reduces gain appre-
ciably at higher frequencies because the pole must be placed at a relatively
low frequency to decrease loop gain to one at a desirable phase margin. The
time response might then be acceptable, but the side effect is that, except at
low frequencies, the loop gain rolls off, and high-frequency amplifier perfor-
mance suffers. The benefits of feedback are retained by keeping the loop gain
high over the frequency range of interest. If loop gain is too low at higher fre-
quencies, then distortion (or nonlinearity) is high and noise rejection low. For
an analog-to-digital converter (ADC) interface, bits of accuracy and signal-to-
noise ratio (SNR) will be lost at higher frequencies, and for the audiophile, the
cymbals will sound “tinny.”
The two-pole compensation technique sustains high gain to a higher break fre-
quency, where it then rolls off at −40 dB/dec (−2 slope) followed by a zero that
restores the magnitude to that of dominant-pole compensation. The difference
from dominant-single-pole compensation is shown below.
The high loop gain is extended from the dominant-pole break frequency at
pd to p, where two poles reside. The gain then decreases with a −2 slope to z,
the frequency of the zero. Above z, the response follows the dominant-
pole response, with a −1 slope. The zero restores the phase margin lost by the
additional pole.
134 Chapter 2
logAv
K Two-pole
compensation
-2
Dominant-pole
compensation
-1
pd p z log w
The above Bode plot can be expressed algebraically by the generic voltage-
gain transfer function:
Vo s z +1 s z +1 s z +1
= 2 = 2 =
Vi as + bs + 1 (s p + 1)  s   2 ⋅ζ 
2
  +   ⋅s + 1
ωn ωn 
where wn is the pole radius and z is the damping ratio. For z = 1, the pole pair
is critically damped and the two poles are equal, at p. The pole angle is cos−1(z)
for z ≤ 1. We will assume identical, real poles for now.
Two-Pole Compensator Circuit

A two-pole compensation circuit is shown below. For the ideal op-amp, static
(dc) gain, K, is infinite, and the compensation poles reside at the origin – a
dominant-pole amplifier. In a typical feedback amplifier loop, however,
K can be finite. From the Bode plot above, as frequency increases, the reactance
of the capacitors decreases relative to R until XC << R. Then the equivalent
circuit consists of the two capacitors in series, shunting the op-amp. The
series capacitance forms an op-amp integrator with a dominant-pole response.
The zero of the circuit is located at the frequency for which R becomes
negligible relative to XC. At the zero, the frequency response breaks from a slope
of −2 to −1.
C2 C1
Vi
−K Vo
Ii Ri
The above circuit has a loop gain of
s 2RRiC1C 2
GH = K ⋅
s 2RRiC1C 2 + s (RC1 + RC 2 + RiC 2 ) + 1
where G = −K.
The closed-loop voltage gain includes a preloop transfer function of
sR (C1 + C 2 ) + 1
Ti = Ri ⋅
s RRiC1C 2 + s (RC1 + RC 2 + RiC 2 ) + 1
2
and is
Vo G sR (C1 + C 2 ) + 1
= Ti ⋅ = −K ⋅ 2
Vi 1 + GH s RRiC1C 2( K + 1) + s (RC1 + RC 2 + RiC 2 ) + 1
where Vi = RiIi. For an ideal op-amp, K → ∞, and the voltage gain becomes
Vo sR (C1 + C 2 ) + 1
=−
Vi K →∞ s 2RRiC1C 2
As K increases, the quadratic term in the closed-loop voltage gain dominates,

shifting the poles to the origin. This results in loop-gain rolloff from 0 Hz,
136 Chapter 2
defeating the two-pole compensator. Consequently, the above circuit can be

used as a two-pole compensator under the condition that K remain finite. The
two-pole compensator, with its poles far removed from the origin, cannot use
the open-loop gain of an op-amp for G.
Another simplification of the circuit is to let Ri → ∞. This is the case of a
transimpedance amplifier with input Ii. Its transresistance is
Vo sR (C1 + C 2 ) + 1
= −K
Ii Ri →∞ sC 2[s ( K + 1) RC1 + 1]
The pole dependent on Ri moves to the origin. Without a finite Ri, the poles
cannot be equal and two-pole compensation is not realized under this condition
either.
For an op-amp transimpedance amplifier, both Ri and K → ∞. In this case,
Vo sR ⋅ (C1 + C 2 ) + 1
=−
Vi Ri ,K →∞ s 2R ⋅C1 ⋅ C 2
This amplifier behaves as a dual integrator with a finite zero. For the special
case of C1 = C2 = C:
Vo s ⋅ 2 ⋅ R ⋅C + 1
=−
Vi Ri ,K →∞ ,C1 =C 2 =C s 2R ⋅ C 2
As R → ∞, Vo/Vi becomes
Vo 1
= −K ⋅
Vi R →∞ sRi (C1 C 2 ) ⋅ ( K + 1) + 1
With R open, the circuit defaults to dominant-pole compensation. (The factor

C1||C2 is the series combination of C1 and C2; || is a mathematical operator, not
a topological descriptor.) In the case of an ideal op-amp, as K → ∞, the pole
approaches the origin and the gain is
Vo 1
=−
Vi R ,K →∞ sRi (C1 C 2 )
This is a dominant single-pole amplifier response with an amplifier shunt capaci-

tance of C1 and C2 in series. For comparison, the plots of Vo/Vi for K → ∞ and
R, K → ∞ are shown below.
log
R finite
R→∞
z= 1 log f
2π R(C1 + C2)
Two-Pole Compensator Design Constraints

The above algebraic expressions for closed-loop gain do not of themselves satisfy
the requirements for two-pole compensation. The following conditions must
also hold:
1. The poles must be equal (or close): p1 = p2 = p.
2. The poles must be less than the zero: z/p > 1.
The first condition is satisfied for real poles when the coefficients of the qua-
dratic pole factor of the closed-loop voltage gain have the relation
b 2
a = 
2 
where a is the quadratic coefficient and b is the linear coefficient. Under the
above condition, the quadratic polynomial factors into a perfect square. Because
the K + 1 factor is in a only, its variation produces the loci of poles for a constant
b, shown below.
138 Chapter 2
jω
a increasing
b>0 +
1
a=0 b a=0
− +
s
− b =−2 −1
2a b b
b = constant
−
The poles are equal when their value is −b/2a, and the corresponding gain
is found by setting the discriminant, b2 − 4a, to zero and solving for K + 1:
R 2(C1 + C 2 ) + 2Ri R (C1 + C 2 )C 2 + Ri2C 22

2
K + 1 p1 = p2 =
4Ri RC1C 2
K is a finite amplifier open-loop gain that can be implemented as an op-amp

inner-loop fixed-gain stage. But this requires additional circuitry and the feed-
back network that sets the gain of the op-amp must not interfere with the two-
pole feedback network. The two-pole circuit is usually made to be one of the
stages in the forward path (G) of a feedback amplifier, within a larger loop.
Instead of setting the compensator by adjusting K, solve the K + 1 equation
above for one of the compensator elements, R:
RiC 2
R= ⋅ ( 2K + 1) ⋅C1 − C 2 ± 2 ⋅ C1⋅ ( K + 1) ⋅ (KC1 − C 2 ) 
(C1 + C 2 )2 
where p1 = p2 and, of course, R is positive and real, requiring that KC1 > C2. This
equation for R is rather involved and can be simplified by approximation to
 4 ⋅ K ⋅ (C1 C 2 ) 
R ≅ Ri ⋅   , K >> 1, K ⋅C1 >> C 2, p1 = p2
 C1 + C 2 
The second constraint on two-pole compensator realization is that z > p. From

Vo/Vi,
1 1
z= =
R (C1 + C 2 ) τ z
and the positive value of the two poles is
b R (C1 + C 2 ) + RiC 2 1
p= = =
2a 2Ri RC1C 2( K + 1) τ p
Because the poles are equal, a = (b/2)2, and they are located on the real axis
at
b 2
=
2a b
Then the condition z > p becomes
1 2 1
> =
τz b τ p
From the voltage-gain expression, b = tz + RiC2. Substituting
1 2 1
> =
τ z τ z + RiC 2 τ p
or
RiC 2 > τ z = R (C1 + C 2 )
Solve for RiC2 in terms of (z/p) from the inequality for 1/tz, and
 z 
RiC 2 = τ z ⋅  2 ⋅ − 1
 p 
140 Chapter 2
Then solving z > p, using b/2a instead of 2/b results in
z τz z  C1 
RiC 2 = ⋅ , < 2 ⋅ (K + 1) ⋅ 
p 2 ⋅ (K + 1) ⋅[C1 (C1 + C 2 )] − z p p  C1 + C 2 
The additional constraint on z/p is weak for large K but suggests that C1 be made
larger than C2 for maximum pole-zero separation. A special case of RiC2 is
z  C1 
≅ 2 ⋅ (K + 1) ⋅  , RiC 2 >> τ z
p  C1 + C 2 
When RiC2 dominates b, the pole-zero separation is pushed to the limits of

RiC2. In this case, with large K,
 C 
R ≅ Ri ⋅  2  , RiC 2 >> τ z , K >> 1, p1 = pz
 4KC1 
Finally, from 1/tz, the constraint on the capacitors is
C 1 Ri
< −1
C2 R
These formulas can now be used to design two-pole compensators with real
and equal poles, and gain values typical of either op-amps or low-gain
amplifiers.
Example: Two-Pole Compensation

A two-pole amplifier has the following circuit values:
K = 100, Ri = 33 kΩ, C1 = 10 pF, C 2 = 100 pF
For this amplifier,

 100 pF 
R = 33 kΩ ⋅   = 825 Ω
 4 (100 ) (10 pF ) 
τ z = R ⋅ (C1 + C 2 ) = (825 Ω ) ⋅ (110 pF ) = 90.75 ns
The conditions for application of the equation for R are satisfied:
 C 
R ≅ Ri ⋅  2  , RiC 2 >> τ z , K >> 1, p1 = p2
 4KC1 
All element values are determined, and the natural frequency of the pole factor,
which is the break frequency of the two poles, is found either from 1/tz or
directly from a:
1 1
fn = = = 96.5 kHz
2π a 2π Ri RC1C 2( K + 1)
As a check, when the poles of a quadratic factor are equal, damping ratio,
z = 1. From the pole factor of the closed-loop voltage gain, ζ = b 2 a = 1.03.
The zero is located at
1 2π ⋅ R ⋅ (C1 + C 2 ) = 1.75 MHz
C2 C1
100 pF 10 pF
Ri
−100 Vo
33 kΩ
+
Vi
−
142 Chapter 2
From the SPICE simulation, the phase is −90° at 100 kHz, where the poles
should be. The Bode plot from circuit simulation is shown below.
40
20
Magnitude, dB
−20
−40
10 kHz 30 kHz 100 kHz 300 kHz 1.0 MHz 3.0 MHz 10 MHz
Frequency
−50
f −180°, deg
−100
−150
10 kHz 30 kHz 100 kHz 300 kHz 1.0 MHz 3.0 MHz 10 MHz
Frequency
The zero is located at 1/2pR(C1 + C2) = 1.75 MHz. From the SPICE simula-
tion, the phase is −90° at 100 kHz, where the poles should be. As a check, the
magnitude will be down −6 dB (for two poles) at the break frequency. At 34 dB
(down from a static gain of 40 dB), it is 91 kHz. The maximum closed-loop
phase lag occurs at 631 kHz and is −142°. The nonmonotonic phase plot, which
dips down and comes back up due to the zero, is characteristic of two-pole-
compensated amplifiers. The magnitude plot rolls off with a −2 slope at the pole
frequency to the zero frequency at about 1.75 MHz. (Because the amplifier is
inverting, the Bode plot phase is offset by −180°.)
Generalized Two-Pole Compensator

The previous development assumed real poles. What happens if the circuit is
modified to allow for complex poles? The benefit in doing this, if it can be done,
is that for amplifiers with other poles and zeros in the loop gain, complex pole-
pair compensation can be achieved by the compensator while maintaining high
loop gain over bandwidth.
A more general set of design formulas takes the given design constraints as
independent variables and yields two-pole compensator element values. Starting
with Ri, K, wn, and z of the pole factor (now no longer necessarily one), a useful
design parameter, the pole-zero separation, will be given its own symbol as
defined:
z z
= =γ
p ωn
Noting that the location of the zero, z = 1/tz, and that the quadratic pole is of
the form
 s   2ζ 
2
as 2 + bs + 1 =   +   ⋅ s + 1
 ωn   ωn 
then z can be expressed in terms of design parameters as follows:
1
z = γ ⋅ ωn ⇒ = γ ⋅ τ z ⇒ a = γ 2 ⋅ τ z2
ωn
⇒ ( K + 1) ⋅ RRiC1C 2 = γ 2 ⋅ R 2 ⋅ (C1 + C 2 )
2
where a is taken from the voltage gain of the compensator circuit. Solving for
R,
144 Chapter 2
(K + 1) ⋅ (C1 C 2 )
R = Ri ⋅
γ 2 ⋅ (C1 + C 2 )
z is now brought in as
b b τ + RiC 2
ζ= = ⋅ ωn = z ⋅ ωn
2 a 2 2
The compensator element C2 results from solving the equation for z and is
C2 =
(2 ⋅ζ ⋅ γ − 1)⋅ τ z , ζ > 1
Ri 2 ⋅γ
C2 is expressed entirely in given parameters and is thereby determined. Next,

the equation for z,
1 1
z= =
R (C1 + C 2 ) τ z
is solved for C1:
τz
C1 = − C2
R
It is then substituted into R. This results in an expression for R in given param-

eters and C2, which is known;
τz   γ 2   τz    γ2 
R= ⋅ 1 −   ⋅  , R C >   ⋅τz
C 2   K + 1  RiC 2   K + 1
i 2
Finally, since R is now known, it is substituted into C1 to yield
 1 
C1 = C 2 ⋅  − 1
 1 − [γ ( K + 1)]⋅[τ z RiC 2 ] 
2
Example: Two-Pole Compensator Design

An amplifier has a gain of K = 10 k, Ri = 10 kΩ and is to be two-pole compen-
sated to have a zero at 500 kHz and begin its roll-off a decade lower, at
50 kHz. Furthermore, an MFED pole response (30° pole angle) is desired,
where z ≅ 0.866. Component tolerances are 5%. The required parameters
are
1 1
τz = = = 318 ns
2π f z 2π (500 kHz )
z 500 kHz
γ = = = 10
ωn 50 kHz
First, calculate C2; it is 519 pF. The closest 5% part is
C 2 = 520 pF
Next, calculate R, using the calculated value for C2 (instead of the 5% value)
to keep the calculations accurate. (This is important when C1 is calculated,
because the difference of two large numbers is taken.) Then R = 613 Ω. The
closest value is
R = 620 Ω
Finally, C1 is calculated from its equation, or from
τz
C1 = − C2
R
if care is taken to retain numerical consistency. It is 0.32 pF. This is a very small
discrete capacitor value and suggests that it might be difficult to realize this
reliably as a discrete circuit in manufacture because this value is on the order
of parasitic capacitances. The circuit-board layout between the output node and
R must minimize stray capacitance.
146 Chapter 2
One way to implement C1 is with a small trimmer capacitor of about 1 pF

maximum value. If such a small C1 is not feasible, then the given parameters
must be adjusted to result in larger capacitance. C1 increases if R increases due
to a decrease in C2. And C2 decreases when z, g, or tz decrease or Ri increases.
The amplifier design is shown below.
C2 C1
520 pF 0.33 pF
R 620 Ω
Ri
−104 Vo
10 kΩ
+
Vi
−
To check these results, we turn from synthesis to analysis and calculate a and
b of the pole factor:
a = RRiC1C 2( K + 1) = 1.06 × 10 −11 s2 ⇒ f n = 48.8 kHz ≅ 50 kHz

b = R (C1 + C 2 ) + RiC 2 = 318 ns + 5.19 µs = 5.51 µs
⇒ ζ = 0.85 ≅ 0.87
Both fn and z are within the 5% tolerance of the components.

Finally, check these results against the constraints in the formulas for C2 and
R:
1 1
ζ = 0.87 > = = 0.05 ( checks )
2γ 20
γ 2τ z
C 2 = 0.33 pF > = 0.32 pF ( checks )
Ri ( K + 1)
The lower limit of C2 is approached because C1 and C2 are so widely

separated.
The final check of this example is made from the SPICE frequency-response
simulation, shown below.
80
60
Magnitude, dB
40
20
0
1.0 kHz 10 kHz 100 kHz 1.0 MHz 10 MHz
Frequency
−50
f−180°, deg
−100
−150
1.0 kHz 10 kHz 100 kHz 1.0 MHz 10 MHz
Frequency
148 Chapter 2
Can the Compensator Circuit Be Statically Stabilized?

The above amplifier has no static (dc) feedback and behaves like an integrator
at 0 Hz. Unless it is within a larger feedback loop, the output drifts out of its
linear range due to offset errors. For standalone applications, Rf must be
included for static stabilization, as shown below. (The resulting compensator
has the topology of a bridge-T filter.)
Rf
C2 C1
R
Vi
−K Vo
Ii Ri
The transfer function, with Rf included, is approximately the same as before

under the conditions that
 Ri
R f >> 
1 sC 2 + (1 sC1 R )
Under these conditions, the static-path feedback through Rf is small compared

with the capacitive path (yet enough to statically stabilize the amplifier), and Rf
negligibly shunts Ri and does not affect the transfer function of the capacitive
path. Two-pole compensation can be achieved with limited, but often adequate,
static feedback, and all the theory developed thus far can be applied.
What happens if an op-amp is used? The voltage gain for K → ∞ is
Vo Rf sR (C1 + C 2 ) + 1
=− ⋅ 2
Vi K →∞ Ri s R f RC1C 2 + sR (C1 + C 2 ) + 1
As Rf → ∞, the voltage gain approaches Vo/Vi for K → ∞, as it must. This gain

differs from the generalized Vo/Vi in that b = tz and does not have the extra
degree of freedom that Vo/Vi does, with the RiC2 term. Consequently, z and g are
not independent but are related by
b τω 1 1
ζ = ωn = z n ⇒ ζ = ⇒γ =
2 2 2γ 2ζ
Proceeding similar to the derivation of C1,
τz  τ  τ
R= ⋅1 − 2 z  , C2 > 2 z
C2  γ R f C2  γ Rf
C2 is chosen to satisfy the above constraint that R > 0. This choice depends on
Rf and interacts with it. The pole locus of Vo/Vi was varied by (K + 1) since it
was in a but not b. For this compensator, variation with constant b is due to Rf
instead. To achieve g > 1, as required for a two-pole compensator, the poles
must be complex and have a pole angle greater than 60°, as plotted below.
jω
Rf increasing
ωn
60°
−z
s
150 Chapter 2
Although the pole-zero separation is zero, 60° establishes the minimum Mm as

1.15 (or 1.25 dB) and minimum Mp as 16%. For a useful compensator with one
octave of pole-zero separation, g = 2. Then z = 0.25 (f = 76°), Mm = 2.97 (or
6.3 dB), and Mp = 44%.
Infinite (op-amp) forward path gain drives the circuit poles to the origin,
defeating the two-pole scheme. With nonlarge Rf, the poles again become finite,
but because of the unavoidable underdamped response that accompanies ade-
quate pole-zero separation, the bridge-T two-pole op-amp compensator is very
limited for two-pole compensation. It functions better as a notch filter, which is
a typical application for bridge-T networks.
The design equations and a design example of the two-pole compensator
circuit have been presented. With the math worked out, use of this design
procedure is not difficult and can result in better feedback amplifier
accuracy and linearity at higher frequencies than dominant-single-pole
compensation. Remember: two-pole compensation is used to increase not
amplifier stability but upper-frequency loop gain. Two-pole compensation
tends to decrease stability and must be applied carefully, making sure that
no uncompensated poles exist in the loop below the two-pole break
frequency.
Second, this compensation technique, when implemented using the given
circuit, is best placed within a larger feedback loop or else static errors will cause
it to drift out of range. This problem can usually be corrected by simply placing
a large-value feedback resistor around the finite-gain amplifier. However, if an
op-amp is used, the pole-zero placement for two-pole compensation is con-
strained excessively, rendering the attempt a failure. Not every “good idea”
results in something useful.
OUTPUT LOAD ISOLATION

In some feedback amplifier applications, the load impedance is highly reactive,
and the amplifier has a significant output resistance Ro. This combination can
add a load-dependent output pole to the loop. A method for isolating capacitive
loads is shown below, where CL is the load, and Ro is the amplifier open-loop
output resistance.
Rf
Cf
Ri
V−
+ − V1 Ro Vh R
Vo
Vi +
− CL
The noninverting version is show below.
+ V1 Ro Vh R
+ Vo
−
Vi Cf
CL
− V−
Rf
Ri
The compensation scheme has two feedback paths, an accurate low-frequency

path and a load-isolated high-frequency path. The feedback compensation
capacitor Cf is isolated from the load by output decoupling resistor R. The low-
frequency feedback through Rf is taken at the output to eliminate static error
due to Ro and R.
With no load isolation or compensation,
R =Cf = 0
The load introduces a pole in the loop at −1/(Rf ||Ro) ⋅ CL. If Cf is then added to
compensate for this pole, the loop gain becomes
 R f Ro   Ri 
GH = −  ⋅ ⋅
{s [R f ( −Rm )]⋅C f + 1}⋅ (sR f C f + 1)
 
 R f ( −Rm )   R f + Ri  [s (R f Ro ) ⋅ (C f + C L ) + 1]⋅[s (R f Ri ) ⋅C f + 1]
152 Chapter 2
where Rm = Ro/(−K). When CL = 0 and Ri >> Ro, the poles are well separated. As
CL increases, pole separation decreases as the higher pole moves down in fre-
quency, reducing stability. Cf introduces a feedback zero and pole as a phase-lead
network. The zero can be placed to cancel the amplifier output pole by
setting
Ro
Cf = ⋅C L
Rf
jw
R=0
G H H Cf increasing G
× ×
−1 −1 −1 −1 σ
(Rf || Ro)(CL +Cf) (RI || Rf)Cf R1C1 (Rf ||(−Rm))Cf
From the parameter-variation plot above, as Cf increases, all poles and zeros shift
toward the origin. For CL >> Cf, the load pole shifts little, and the pole and zero
in H move together and away from the load pole.
When R is added and the topology is redrawn, the output network forms a
bridge.
V1 Ro Vh
Cf R
−KV−
V− Rf Vo
Ri CL
The exact solution for this circuit, a nontrivial exercise, can be found by revert-
ing to KCL, applied at the nodes with voltages: V−, V1, Vh, and Vo. This results in
the flow graph for the inverting amplifier.
c b
g d
k h f
a V− −K V1 g Vh d
Vi Vo
c
a f
g
b
a
where
1 1 1
a= , b= , c = sC f , d =
(Ri R f 1 sC f ) Rf (Z L R f R )
1 1 1 1
f = , g= , h= , k=
R (Ro 1 sC f R ) Ro Ri
The amplifier gain is −K and V1 = −KV−. Some simplifying assumptions can be

made that reduce the complexity of the flow graph (such as removing b/d, c/d,
and-or f/g), but the remaining circuit analysis is still unwieldy. We need a more
functionally oriented approach.
The low- and high-frequency (lf and hf) feedback paths have been approxi-
mated below, along with their Bode plots. The lf path has transmittance,
V−  Ri   1   1 
= ⋅  ⋅ ,
V1 lf  R f + Ri   s (Ro + R )C L + 1  s (R f Ri )C f + 1

1
Ro << , Ro + R << R f + Ri
sC f
where the simplifying assumptions are that Cf and Rf + Ri do not load the smaller
output resistances Ro and R.
154 Chapter 2
Ro R Rf
V−
+
V1 CL Ri Cf
log
Ri
Rf+Ri
1 1 logw
(Ro +R)CL (Rf ||Ri)Cf
The hf path is
V−  sRC L + 1   s (R f Ri )C f 
= ⋅ ,
V1 hf  s (Ro + R )C L + 1  s (R f Ri )C f + 1
1 1
Ro , R << , << R f
sC f sC L
The hf-path approximations are similar to those of the lf path. At high frequen-
cies, this feedback transmittance approaches
R
Ro + R
The composite feedback transmittance is the sum of the two paths, or
V −  Ri  s 2RC L R f C f + sR f C f + 1
= ⋅
V1  R f + Ri  [s (Ro + R )C L + 1]⋅[s (R f Ri )C f + 1]
Without R, there is one less LHP zero, as in GH. The feedback path is an all-pass
network when the coefficients of the pole and zero terms are equated. This
results in
Ro
V−
+ Cf
R Rf ||Ri
V1
− CL
log
1 1 1 logw
(Ro +R)CL RCL (Rf ||Ri)Cf
R   Ro + R   R f + Ri 
R = i  ⋅ Ro , C f =  R  ⋅  R  ⋅C L
 Rf f f
The load capacitance pole is removed from the loop gain. The closed-loop
response, however, is still affected by CL. The constraints of V−/V1 for lf and hf
paths require that 1/RoCf be checked after applying these formulas for R and
Cf, to make sure that this pole is well above fT.
Example: Load Capacitance Compensation

A fast op-amp with K = 105 and poles at 100 Hz and 4 MHz is used in the invert-
ing configuration to drive a 10 nF load with a voltage gain of −3. Rf = 30 kΩ and
Ri = 10 kΩ. The open-loop output resistance is 10 Ω. The feedback capacitor Cf
and decoupling resistor R are calculated from the R, Cf formulas:
R = 3 .3 Ω
and
C f = 5.9 pF
For the uncompensated amplifier, the step response shows obvious ringing.
156 Chapter 2
5.0
4.0
Uncompensated -vo’ V
3.0
2.0
1.0
0.0
0.0 µs 0.4 µs 0.8 µs 1.2 µs 1.6 µs 2.0 µs
Time
Peaking is evident in the frequency response and group delay plots.
20.0
Uncompensated
0.0
magnitude, dB
−40.0
−80.0
500.0
Uncompensated
group delay, ns
400.0
200.0
0.0
10 Hz 100 Hz 1.0 kHz 10 kHz 100 kHz 1.0 MHz 10 MHz 100 MHz
Frequency
Compensation is now applied, and the response improves. No ringing is evident

in the step response
3.0
2.5
Compensated -vo’ V
2.0
1.5
1.0
0.5
0.0
0.0 µs 0.2 µs 0.4 µs 0.6 µs 0.8 µs 1.0 µs
Time
No peaking appears in the group delay or frequency response (below).

Compensated magnitude, dB
0.0
−50.0
−100.0
250.0
group delay, ns
Compensated
200.0
0.0
10 Hz 100 Hz 1.0 kHz 10 kHz 100 kHz 1.0 MHz 10 MHz 100 MHz
Frequency
The schematic diagram of the compensated amplifier shows how the op-amp
is modeled, using RC integrators and buffers to create the poles.
158 Chapter 2
Rf
30 kΩ Cf
40
100 Hz 4 MHz 5.9 pF
10 Ri 15 50 RA 60 70 RB 80 30 Ro R 20
−105 1 1 Vo
+ 10 kΩ 10 kΩ 1 kΩ 10 Ω 3.3 Ω
vi
CA 0.159 µF CB 30.8 pF CL
− 10 nF
0
The SPICE program used for the simulation is listed below.
Load-Compensated Amplifier
.OPT NOMOD OPTS NOPAGE
.AC DEC 30 10 100MEG
.TRAN 2n 2u
VI 10 0 AC 1V PULSE (0 1V)
RI 10 15 10k; amplifier with pole at 100Hz and 4MHz
EA 50 0 15 0 -1E5
RA 50 60 10k
CA 60 0 0.159uF
EB 70 0 60 0 1
RB 70 80 1k
CB 80 0 39.8pF
EC 30 0 80 0 1
RO 30 40 10
R 40 20 3.3
CL 20 0 10nF
CF 40 15 5.9pF
RF 20 15 30k
.PROBE
.END
Capacitive loads can also be isolated by placing a shunt RL in series with

the amplifier output. At 0 Hz, the transmittance to the load is one. At high-
frequencies, L appears open, leaving R as isolation. For excessive inductive
loading, the load is often characterized by a series RL. If it is fixed, a series RC
in parallel with it can form a constant-impedance network.
COMPLEX POLE COMPENSATION

Previous compensation techniques involved real poles and zeros. Open-loop
complex poles appear as resonances that can sometimes be damped by identify-
ing the circuit elements involved in the undesired resonance. This identification
is not always successful, especially when the circuit has many possible parasitic
reactances.
jw
(a)
jw
(b)
From root-locus criteria, pole angle can be reduced by the addition of a real
pole at a lower frequency, as shown above in (a). As the pole increases in fre-
quency due to static loop gain, K, the complex pole radius (wn) decreases, but
so does the pole angle f. This decrease in f is slight in a narrow range of K,
making this a marginally useful technique.
A real zero, placed at a higher real frequency than that of the complex poles,
draws them out to a larger pole radius and lower pole angle with increasing
gain, as shown in (b).
160 Chapter 2
Complex poles can be compensated directly by complex pole-zero cancella-

tion. Complex zeros are realized by a series RLC network at the output of a
transconductance amplifier. Typically, the network is placed in parallel with a
transistor load resistor.
Vo
Gm Vi RL R
(a)
jw
(b)
The transfer function of this circuit is
Vo s 2LC + sRC + 1
= Zo = RL ⋅ 2
Io s LC + s ( R + R L )C + 1
The pole-zero placement is shown in (b). In addition to the desired zeros, there
is another pair of poles with a larger linear coefficient (due to RL). This is similar
to phase-lead compensation; the added poles are at a decreased pole angle from
the poles canceled by the zeros, with no loss of pole radius.
Empirical compensation of hidden complex pole-pairs begins by observing
the ring frequency fr (which is the damped frequency fd) and the time constant
of its decay, tr, from a step response. The value of tr can be calculated from the
peak overshoot Mp. The relationship among tr, fr, and Mp is
1 1
τr = =
2 ⋅ α 4 ⋅ f r ⋅ ln (1 M p )
Compensator element values for an MFED response are calculated by using

these empirical parameters:
2R L
R= −2
12 (π ⋅ τ r ⋅ f r ) + 3
2
2τ r
C=
R 4 (π ⋅ τ r ⋅ f r ) + 1
2
Rτr
L=
2
It is usually easier to measure the peak overshoot Mp than to estimate tr. By using
the formula for Mp, the compensator values for a pole angle of cos−1z are
R L ⋅ ln (1 M p )
R=
ζ π 2 + [ ln (1 M p )] − ln (1 M p )
2
ζ π 2 + [ ln (1 M p )] − ln (1 M p )
2
C=
π 2R L f r
L=
(
π + [ ln (1 M p )] R L
2 2
)
4π 2 f r ⋅ζ π 2 + [ ln (1 M p )] − ln (1 M p )
2
Example: Compensation by Complex Pole-Zero Cancellation

An amplifier has an excessively underdamped response. A maximally flat ampli-
tude (MFA) response is desired. A transadmittance stage in the amplifier is free
to be compensated and has a load resistance of 1 kΩ. The response to a step
shows a ring frequency of 30 MHz and a peak overshoot of Mp = 0.25.
162 Chapter 2
By substituting these values into the formulas for R, L, and C, and noting that
an MFA response has a pole angle of 45° and that ζ = 2 2, the compensator
element values are
R = 1.33 kΩ C = 3.52 pF, L = 4.10 µH
COMPENSATION BY THE DIRECT (TRUXAL’S) METHOD

The direct approach to calculation of G and H for a given closed-loop response
M is to begin by specifying what M should be. Neglecting Ti and To, if any, since
they are cascaded with the feedback loop, we describe M as
G
M=
1 + GH
Solving for the loop gain, we get
HM NHNM
GH = =
1 − HM DH DM − N H N M
where N and D are numerators and denominators of H and M. For a given G

we can solve directly for H:
G −M 1 1 N G DM − N M DG
H = = − =
GM M G NGN M
Not only M must be chosen to satisfy the system requirements, but also
the resulting H must be physically realizable. For a high-order system, M
must be high order for a realizable H. The familiar criteria of amplifier per-
formance are consequently more difficult to express in M. Therefore, this
method is of limited use. If the amount of calculation were the limitation, a
computer solution would be feasible, but creative design judgment is required
in selecting M.
POWER SUPPLY BYPASSING

The elimination or compensation of parasitic capacitance and inductance is not
a dynamic response compensation method in itself but is related. Before a feed-
back amplifier loop can be compensated, the individual stages of amplification
must be stable. Parasitic elements arise from both circuit components and
layout.
Rf +V
L1
C1
Ri −
+
Cin + Rout
Vi Vc
Rin
− Cin − Lout
C2 CL RL
L2
−V
The diagram shows some of the more common parasitic elements for op-amp
circuits. The connections of the amplifier to the power supplies involves conduc-
tors (wire or circuit-board traces) with a corresponding inductance. Circuit-
board trace inductance is difficult to estimate accurately but is roughly 10 nH/cm
for a rectangular trace that is much longer than its cross-sectional dimensions
(Ruehli 1972). Capacitive bypassing of trace inductance shortens the loop length
and decreases the characteristic impedance Zn to a low value.
164 Chapter 2
Oscillation can commonly occur due to the series LC resonance formed by

the stray power-supply inductance and stray capacitance from the local supply
node to an amplifier input. As L increases, Zn becomes sizable with a small C.
Consequently, the series resistance required to damp the resonance also must
be large. By adding bypassing, L is decreased and C greatly increased. Both
effects reduce Zn so that a smaller input-node resistance damps the series
resonance.
Example: Damping Oscillation through Bypassing
+V
6 cm
5 nH
10 nF
Ii Rs
50 Ω
The circuit has a source resistance of 50 Ω shunting the relatively high-resistance

base input. Cbc = 3 pF. The collector supply connection is about 6 cm long.
The inductance is about (6 cm) ⋅ (10 nH/cm) = 60 nH and the characteristic
impedance of the series resonance is
60 nH
Zn = = 141 Ω
3 pF
Rs is smaller than Zn and oscillation is likely. If a 10 nF bypass capacitor with

5 nH of parasitic series inductance is connected to the collector, then
5 nH
Zn = = 0 .7 Ω
10 nF
and the series LC resonance is well damped.

3
High-Frequency Impedance
Transformations
ACTIVE DEVICE BEHAVIOR ABOVE BANDWIDTH

Active devices such as transistors or op-amps have bandwidth limitations that
can result in instability or undesirable dynamic response in feedback circuits.
As gain decreases with frequency, feedback-amplifier port-resistance values also
change and effectively become reactive. These bandwidth-related reactances
can resonate with other circuit elements.
Many linear circuits might not appear to involve “high frequencies” or “wide
bandwidths,” but these concepts are relative to the bandwidth limitations of the
active devices. The frequency range over which bandwidth-limited instability
occurs lies between the bandwidth, fbw and the unity-gain frequency, fT. The two
frequency ranges of interest are
• low-frequency (lf) region, 0 ≤ f < fbw (below bandwidth)
• high-frequency (hf) region, fbw ≤ f ≤ fT
The reduction theorem can be extended to the complex-frequency domain.
In its more general form, reactances are transformed, or gyrated. The first task
is to develop an s-domain expression for b.
The first step is to find b(s) for bipolar junction transistors (BJTs) and apply
it using the b transform. This leads to some interesting impedance gyrations
that can create high-frequency (hf) resonances. To distinguish between the low-
frequency (quasistatic) b and frequency-dependent b, denote
low-frequency β ≡ βo
168 Chapter 3
For BJTs, current-gain bandwidth is denoted by fb and unity-current-gain fre-

quency by fT. They are related by
fT = βo ⋅ f β
Some liberty is taken with the symbol fT by letting it more generally denote the
unity-gain frequency (or gain-bandwidth product) of any active device. The inter-
pretation of fT depends on the kind of device. For BJTs, it is the frequency at
which b is one; for voltage amplifiers, it is a unity-voltage-gain frequency.
For a BJT with fT = 300 MHz and bo = 100, high-frequency behavior occurs
between fb = fT /bo and fT, or in the range from 3 MHz to 300 MHz. For power
BJTs, fb can be as low as several hundred kilohertz. The open-loop bandwidth
of typical op-amps is less than 100 Hz, and unity-gain frequency is 1 MHz. This
range of rather low frequencies is the op-amp hf region.
BJT HIGH-FREQUENCY MODEL

The hybrid-p BJT model is shown below.
r´b Cm
b´
b c
+
V
rp Vbé´ Cp
—
bé´
mr
− e´
ré
In actual transistors, Cm is distributed across r b′, but here it is shown connected

entirely to the internal base node b′.
This more complete model is simplified, as shown below, without Cm or
“ohmic” base and emitter resistances.
High-Frequency Impedance Transformations 169
b c
+
Vbe
rπ Vbe Cp
—
rm
−
This model is valid for both the low-frequency (lf ) and hf regions. The hf model
equations are developed using it.
The idea of the hf BJT model is that as the BJT input frequency increases
above 1/rp ⋅ Cp, a decreasing proportion of Ib flows through rp as the reactance
of Cp decreases. The base impedance is
1 rπ
Z π = rπ =
sC π srπC π + 1
As it decreases with frequency, Vbe also decreases, resulting in decreased collector

current. Consequently, b also decreases with frequency. The break frequency of
b is wb = 1/tb, where
τ β = rπ ⋅ C π
The frequency-dependent form of b is Ic/Ib, or
 1 sC π  βo
β (s ) = βo ⋅   =
 1 sC π + rπ  sτ β + 1
and
sα o τ T + 1
β (s ) + 1 = (βo + 1) ⋅
sτ β + 1
170 Chapter 3
where
τ β = βo ⋅ τT
The Bode plot of b(s) + 1 is shown below for s = jw.
 b ( jw)+1
bo+1
fb fT/ao f
In the lf region, the transistor model does not require reactive elements (as is
assumed in quasistatic analysis). In the hf region, b rolls off with frequency, and
above fT the device has essentially lost its gain (though power gain under the
right circuit conditions takes place up to the unity-power-gain frequency fMAX).
Other significant factors not accounted for in this model (such as base transit
time) cause its error to increase as fT is approached. The model predicts less
phase shift in b than actually occurs due to other transistor delays, yet it is accu-
rate enough to be quite useful.
A simplified model, valid only for the hf region, can be derived from b(s) by
letting bo → ∞. Then
1
lim β (s ) = β hf =
βo →∞ sτT
The expressions for b from b(s) and this hf b can now be used in circuit
analysis.
IMPEDANCE TRANSFORMATIONS IN THE HIGH-FREQUENCY REGION

The b transform as applied to quasistatic BJT circuits can be generalized using
b(s) in reactive circuits. The impedance at the base node due to impedance ZE
at the emitter node is
Z b = [β (s ) + 1]⋅ Z E
From the emitter, the impedance in the base circuit is
ZB
Ze =
β (s ) + 1
From the equivalent circuit shown below, we can derive Zb and verify the above
equation for it.
+
Vbe
Vbe Zp —
rm
+
−
Vi
Zb ⇒
−
+
Vo ZE
Applying Kirchhoff’s current law (KCL) to the emitter node,
Vo Vo − Vi Vbe
+ =
ZE Zπ rm
(Vbe = Vi − Vo) Solving for Zb gives
 βo 
Z b = Z π + Z E ⋅ 1 +
 srπC π + 1
172 Chapter 3
Above fb, Zp becomes negligible, and
 βo  1
Z b ≅ Z E ⋅ 1 +  = ZE +
 srπC π + 1 srπC π Z E βo + 1 Z E βo
Because tb = bo ⋅ tT, Zb can be rewritten as a continued fraction, which makes the

topology explicit in equation form:
1
Zb ≅ Z E + , τT = rm ⋅C π
1 1
+
Z E sτT βo Z E
The corresponding circuit topology is shown below.
ZE
Z
⇒
E
Zb —
st
boZE
T
Here, bo ⋅ ZE is the lf contribution to Zb, ZE/stT is the hf contribution, and the

series ZE is common to both. Below fb, ZE/stT approaches an open circuit and
Zb is consistent with the lf model. In the hf region, ZE/stT dominates Zb; dividing
ZE by s gyrates the impedance of ZE by −90° so that
R →C
C → −R
L →R
For the three cases of ZE (R, L, and C) the transformed impedances are shown
below.
RE
tT
tT −—
—
RE
boRE CE CE
−aoCE
(a) (b)
LE
L
E boLE
—
tT
(c)
Because this analysis is linear, combinations of the three elements in ZE can be

individually transformed and combined to produce the transformed ZE.
Example: Shunt RC-Loaded CC Amplifier

A common collector (CC) amplifier has a shunt RC load for which RE = 470 Ω
and CE = 10 pF. The BJT has a bo = 150 and fT = 300 MHz at IE = 10 mA (typical
of a 2N3904). What is Zb?
The combination of transformed impedances (a) and (b) is shown below.
RE tT
−—
CE
Zb ⇒ CE
tT
boRE —
RE −aoCE
To find the element values, calculate tT = 1/2pfT = 531 ps and ao = 0.993 ≅ 1.

Then tT/RE = 1.13 pF and −tT/CE = −53.1 Ω. Furthermore, bo ⋅ RE = 70.5 kΩ and
−ao ⋅ CE ≅ −10 pF. A hf equivalent circuit omits bo ⋅ RE. Whether Zp is negligible
174 Chapter 3
depends on the other elements in the circuit. If base reactance creates a reso-
nance with the emitter impedance near fb, then Zp is probably significant. For
this circuit, re ≅ 2.6 Ω and Cp = tT/re = 204 pF.
−
Vbe Zp
+
⇒
Ie Vbe
Ze —
rm
ZB
A similar circuit derivation for Ze, based on the equivalent circuit shown above,
results in
Zπ + ZB
Ze =
1 + βo (srπC π + 1)
Again approximating Zp ≅ 0, we obtain the continued fraction:
1
Ze ≅
1 1
+
Z B sτT Z B + Z B βo
The topology is shown below and is the dual of that above for Zb.
stTZB
Ze ⇒ ZB
BZ
—
b o
Below fb, stTZB approaches a short circuit and the circuit becomes the lf model.
The hf contribution of stTZB gyrates Ze by +90° so that
R →L
L → −R
C →R
These expressions for Zb and Ze are valid from 0 Hz to near fT.

The three cases of ZB are worked out below.
LB
tTRB tT
RB CB —
C −aoLB LB
−—
B tT
RB
—
bo boCB
A simpler model, applicable only in the hf region, is derived by using bhf. The
topologies of Zb reduce to those shown below for Zb(hf): (a) ZE; (b) RE; (c) CE;
and (d) LE.
ZE
RE
Z E tT
—
stT —
RE
(a) (b)
tT LE
−—
CE CE
L
−CE —E
tT
(c) (d)
176 Chapter 3
The equivalent circuits of Ze reduce to those below for Ze(hf): (a) ZB; (b) RB;
(c) CB; and (d) LB.
ZB stTZB RB tTRB
(a) (b)
T t
CB —
C B
(c)
LB
LB
−LB −—
tT
(d)
To derive these hf models, let
βo → ∞, Z π = 0
Then
 1 
Z b (hf ) = Z E ⋅  1 + 
 sτT 
ZB
Z e (hf ) =
1 + 1 sτT
The hf models are based on removal of the break frequency of b(s) + 1 at fb so that
b(s) rolls off from infinity at the origin. The expression for b + 1 becomes
1 sτT + 1
β hf + 1 = +1=
sτT sτT
This expression has a pole at the origin which breaks at fT.
REACTANCE CHART REPRESENTATION OF b-GYRATED CIRCUITS

The two representations of hf circuits used thus far, equations and circuit dia-
grams, are supplemented by a graphic representation using the reactance chart.
Reactance plots of Zb hf equivalent circuits are shown below, for ZE equal to (a)
RE and (b) LE.
The break frequencies of these plots are fT, and the nonzero slopes are ±1
in the hf region. Above fT the impedance gyration stops, once b(s) + 1 stops
decreasing with frequency.
log Zb
t
T
—
RE
-1
RE
fT log f
(a)
log Zb
LE
+1
L
E
—
tT
fT log f
(b)
178 Chapter 3
Two of the six gyrated circuits involve negative resistances. The reactance
chart has a log-log scale, and the logarithm of negative numbers is undefined.
Consequently, it might appear that reactance-chart representation of b-gyrated
impedances has limited use. Happily, this is not so. The equation for Zb when
ZE = 1/sCE can be reformulated as
sτT + 1
Z b (C E ) =
s 2τ T C E
log Ze
RB
tTRB
+1
fT log f
(a)
log Ze
tT
—
C B CB
−1
fT log f
(b)
Reactance plots of Ze hf equivalent circuits are shown above, for ZB equal to

(a) RB and (b) CB. For the case of Ze (LB),
s 2τ T L B
Z e (LB ) =
sτT + 1
These equations can be plotted on a reactance chart and are shown on the
next page for (a) Zb(CE), and (b) Ze(LB). Below fT, the plot of tTCE has
a slope of −2 and that of tTLB has a slope of +2. Above fT, the plots are of CE
and LB.
log Zb tTCE
−2
RB
CE
−1
fT log f
(a)
log Ze
+1
LB
+2
tTLB
fT log f
(b)
REACTANCE CHART STABILITY CRITERIA FOR RESONANCES

The two common resonant circuits are the series (a) and parallel (b) LC circuits,
shown below.
log Zs
−1 +1
C L
Zs ⇒ L
Zn
C
fn log f
(a)
log Zp
Zn
Zp ⇒ L C
L C
+1 −1
fn log f
(b)
180 Chapter 3
Define the intersection of the asymptotic L and C plots as the resonant point, at
which
1
fn =
2π LC
The characteristic impedance of the resonance is
L
Zn =
C
Zn is the reactance of each resonating element at the resonant point.

For all of these reactance plots, asymptotic approximations are used. Exact
plots require that a vertical asymptote at fn be approached on each side by a curve
tending to ±∞. For a series resonance, Zs = 0, which is at −∞ on the reactance
chart. For a parallel resonance, Zp → ∞, which is at +∞ on the reactance chart.
A characteristic feature of resonance is a ±2 change of slope on the chart: +2
for a series and −2 for a parallel resonance. In the case of LC circuits, this is a
change from ±1 to − +1. For the cases of Zb(CE) and Ze(LB) however, a change from
±2 to zero or zero to ±2 also cause resonances and possible oscillation if not
sufficiently damped.
The amount of damping of an RLC resonance is characterized by the damping
ratio z. For parallel resonances, this is
Zn
parallel resonance ζ =
2R
And for series resonances, it is
R
series resonance ζ =
2Z n
For critical damping, z = 1. Then Rp = Zn/2 and Rs = 2 ⋅ Zn. In both cases, critical
damping is achieved by a resistance equal to the combined reactances of the L
and C at resonance.
An estimate can therefore be made on the reactance plot as to how well a

resonance is damped. For a parallel resonance, the parallel resistance must be
below the LC resonant point to be well-damped; for series resonance, it must
be above it.
EMITTER-FOLLOWER REACTANCE-PLOT STABILITY ANALYSIS

CC stages are commonly used to drive capacitive loads such as transmission
lines. The high current gain of the CC configuration allows it to supply high
transient currents required to quickly charge the capacitive load. When capaci-
tive loading is combined with base resistance, a hf resonance can occur.
+V
RB
fT , bo
+
⇒
Vi CB Zb ⇒ Ze
− CE
−V
This CC amplifier has the hf equivalent circuit shown below. Ordinarily, we

would combine the parallel Rs, and Cs, before plotting. Here, sections a and b
of the hf model are plotted separately because the elements within them are
interdependent. Section a of Ze is resistive up to fT and is capacitive with value
CB above fT. Section b is inductive up to fT, above which it is resistive with
value RB.
t
Ze(hf ) ⇒ —
C
T
B
CB RB tTRB CE
a b
182 Chapter 3
The general equivalent circuit of the emitter node is shown below.
t RB
—T —
b
⇒ C o
Ze B
CB RB CE
boCB tTRB
The resulting reactance plot for the hf circuit is shown below.
log Ze
CE CB
RB
b
a r
tT
—
C B
tTRB
fb fT log f
On the reactance chart, RB and CB have been chosen so that RB ⋅ CB = tb. Their
plots intersect at fb. CE is much greater than CB, and it intersects the plot for
section b at resonant point r. From the reactance chart, this is a parallel reso-
nance; a resistance less than the impedance, Zn, at r is required to damp it. The
plot of the impedance from section a is resistive and less than Zn at fn; it damps
the resonance. This resistance decreases as CB increases, and increasing base
capacitance tends to stabilize a capacitively-loaded emitter-follower.
From the above reactance plot, we can see what effect changes in the values
of circuit elements have. For section a, increasing CB causes the resistive segment
of the a plot to move downward and thus provides more damping at r. At the
same time, the diagonal line representing CB moves to the left. The break fre-
quency does not move but remains constant at fT. Curve a moves downward as
CB increases. Similarly, an increase in RB increases the inductance below fT in
curve b while break frequency fT remains fixed. That is, RBtT always intersects RB
at fT. Increasing RB moves curve b upward.
From reactance chart analysis, we can observe that a decrease in CE or an

increase in RB or CB tends to stabilize the circuit because Zn increases relative to
the damping resistance. CE has a range in which instability can occur. As CE
increases, its plot moves to the left, and r moves with it and downward until it
intersects curve a at fb. Then 1/sCE = tT/CB, and r is eliminated because imped-
ance gyration does not occur below fb. This is also true above fT. If CE decreases
until it crosses RB at fT, r vanishes. In addition, transistor gain above fT may be
insufficient to sustain oscillation anyway. Since the reactance plots are asymp-
totic approximations, hf effects extend somewhat above and below the hf
region.
This analysis assumes that CB << CE and tT/CB << RB. More generally, as CB
increases, two effects occur: (1) Its b-gyrated resistance, tT/CB, decreases (increas-
ing damping); and (2) CB also adds to CE, decreasing Zn. Adequate damping can
occur only when CE dominates. If CE becomes negligible relative to CB, then the
resonance cannot be damped better than z = 0.5. (See next section.) For
maximum circuit speed, a minimum CB is desirable to minimize the base input
time constant.
EMITTER-FOLLOWER HIGH-FREQUENCY EQUIVALENT CIRCUIT
t
Ze(hf ) ⇒ —
C
T
B
CB RB tTRB CE
a b
An explanation of the CC stage based on the hf equivalent circuit topology is

that RB is gyrated at the emitter to become inductive with value tT ⋅ RB. CE forms
a parallel resonant circuit with tT ⋅ RB. As RB increases or CE decreases, Zn increases,
and the resonance can be damped by higher values of shunt resistance. Although
RB itself provides damping, it is not small enough to be adequate. For a parallel
resonance z, ignoring CB,
184 Chapter 3
1 τT
ζ≅ ⋅
2 R BC E
In the hf region, RB ⋅ CE > tT so that z < 1/2; the resonance is underdamped.

z could be increased by decreasing RB or CE. (Also, a slower transistor, with
increased tT, increases z.) As RB ⋅ CE approaches tb however, Zp becomes signifi-
cant and adds further damping. Also, as RB ⋅ CE approaches tT, Zn approaches RB
and is damped by it.
The addition of CB damps the resonance because CB is gyrated at the emitter
to a resistance of tT/CB. This resistance can be set as low as needed by increas-
ing CB. The damping ratio is then
Zn τT R B (C B + C E ) τT + R BC B
ζ= ≅ =
2R 2 (τT C B ) R B 2 τT R B (C B + C E )
An acceptable value of z can now be obtained by adjusting CB. The value of

CB can be found directly for a given z by solving for CB:
τ   CE 
C B =  T  ⋅ (2ζ 2 − 1) + 2ζ ⋅ ζ 2 + − 1
 RB   τT R B 
(Only one root of CB is possible for CB > 0.) This simplifies, for RBCB >> tT, to
2ζ 2τT  R BC E 
CB ≅
RB  1 + 1 + τ  , R BC B >> τT
T
Furthermore, if CE >> CB, then
τT ⋅C E
C B ≅ 2ζ ⋅ , R BC B >> τT , C E >> C B
RB
An alternative compensation technique is to adjust RB for a desired z. Increas-

ing RB increases Zn so that tT/CB more effectively damps the resonance. When
RB = tb/CB, no value of CE can cause a resonance with tT ⋅ RB.
Example: CC Stabilization Using Shunt Base RC
+Vcc
1 kΩ bo = 150
+ fT = 300 MHz
CB
Vi
Vo
− 10 pF
10 mA
−VEE
The CC has a 2N3904 BJT and is to be stabilized with CB, if necessary, so that
ζ = 2 2 for an MFA response. The transistor parameters are tT = 531 ps and
ao ≅ 1. The hf resonance is at
1
fn = = 69.1 MHz
2π (531 ps)(1 kΩ)(10 pF )
and the characteristic impedance is
Zn = (531 ps)(1 kΩ) (10 pF ) = 230 Ω

The resonant frequency lies within the hf range of 2 MHz to 300 MHz. Without
CB, z = 0.115 (Mp ≅ 0.7), a very underdamped resonance. The SPICE simulation
uses a BJT model with
.MODEL BJT1 NPN (BF = 150 TF = 531 ps )
and results in Mp = 0.65 and z = 0.136. This greater damping is partly attribut-
able to RB /bo and Zp.
To achieve the desired z, substitute and solve for CB to get 3.2 pF. Then
RB ⋅ CB = 32 ns >> tT. If the approximate formulas for CB are used, CB calculates
to be 2.9 pF, and the even more approximate CB yields 3.3 pF. The conditions
for the latter CB approximation are met fairly well, and the approximations are
valid for the accuracy required for parts selection.
186 Chapter 3
Example: CC Amplifier Series R Compensation

To eliminate hf resonance from the CC of the previous example, RB could be
increased instead of increasing CB. Let CB = Cµ = 5 pF, and the damping is criti-
cal (z = 1). Solving CB for RB,
4ζ 2τT (C E + C B )
RB =
C B2
For the example, RB must be greater than 1.27 kΩ; an additional 270 Ω is
needed.
This example illustrates why the addition of a small (10 Ω to 1 kΩ) series base
resistor usually damps an oscillating CC. The addition of RB might, however,
damp a series LC resonance with the collector circuit parasitic inductance
instead. With hf resonance analysis, one possible cause of oscillation can be
assessed from circuit element parameters.
EMITTER-FOLLOWER HIGH-FREQUENCY COMPENSATION

The hf BJT model is adequate for analyzing hf resonances but not for determin-
ing optimal compensation schemes. For wideband amplifiers, compensation
involves both lf and hf regions. The approximation of z ignored Zp, bo /CB, and
RB/bo. These additional elements cause further damping so that the actual z is
greater than what the hf z predicts.
If we use the general model instead, shown below, the emitter impedance,
when derived without CE, is
RB sτ β + 1
Ze = ⋅
βo + 1 (sαo τT + 1) ⋅ (sR BC B + 1)
t RB
—T —
b
⇒ C o
Ze B
CB RB CE
boCB tTRB
This result is obtained by substituting ZB into Ze
1
Ze ≅
1 1
+
Z B sτT Z B + Z B βo
or by dividing ZB by b(s) + 1. If RB ⋅ CB = tb, then the zero is cancelled, and Ze

simplifies to
RB 1
Ze = ⋅ , R BC B = τ β
βo + 1 (sαo τT + 1)
Under this condition, no value of CE can cause a resonance with a resistive ZB.
From the equivalent circuit, the series RL and RC branches have the form of
a constant-resistance network with resistance RB/bo when the elements have the
relationships
R B τT R B ⋅ τT
= =
βo C B βo ⋅C B
or RB ⋅ CB = tb. In view of (RB /bo )||RB = RB /(bo + 1), then
 R  1
Ze =  B  =
[R B (βo + 1)] (1 sC B ) = R B ⋅ 1
 βo + 1 sC B R B (βo + 1) + 1 sC B βo + 1 s (R BC B (βo + 1)) + 1
188 Chapter 3
The pole time constant can be expressed as
R B ⋅C B τ
= B = α o ⋅ τT
βo + 1 βo + 1
and Ze is equivalent to Ze for RB ⋅ CB = tb.

The condition RB ⋅ CB = tb achieves a resistive base or emitter impedance out
to near fT. This eliminates hf resonances but does not always result in maximum
circuit speed (minimum risetime or maximum bandwidth). It is feasible for CC
amplifiers when CE varies greatly or is unknown. The zero at fb can introduce
phase-lead out to 1/RB ⋅ CB if hf peaking is needed. For maximum circuit speed,
RB ⋅ CB is set so that the hf resonance is damped to the desired extent. In this
case, RB ⋅ CB < tb. The hf BJT model equations provide easily obtained estimates
for element values when CE and tT are within known bounds.
Example: Power Amplifier CC Output Stage
+V
⇒
RB
RE
ZL
RE
−V
This is a typical amplifier bipolar CC output stage. As a symmetrical class AB

amplifier, only the top or bottom half is active for most of the output voltage
range (except around zero). For analysis, assume that only the upper transistor
is conducting. The simplified equivalent circuit has ZL = 1/sCE.
li RB
RE
CE
The emitter-node hf circuit is shown below.
RE
RB tTRB
CE
The reactance plots (below) are for negligible RE.
log Z
CE
RB
tTRB
RB
—
bo r
b f fb fT log f
—
b o
RE causes a +1 slope change in the CE plot at RE. Here it is assumed that this
zero is above the resonant frequency. The intersection of tT ⋅RB and CE is the
resonant point r.
190 Chapter 3
RB is set by the preceding driver stage. No hf resonance can occur if it is large

enough to cause fn < fb, or RB ⋅CE ≥ bo·tb, as shown above. For unconditional stability,
f n < f β ⇒ R B ⋅ τT ⋅C E > βo ⋅ τT
or
1
RB >
( ω β βo ) ⋅C E
Alternatively, a sufficiently small RB places the resonance above the hf region.

Or the series RE, if large enough, causes RB ⋅CE < tn and damps the series reso-
nance. (Both series and parallel resonant modes are damped under the same
conditions. Any zeros of the series resonance are poles of the parallel reso-
nance.) Depending on re of the BJT, Zp may provide adequate series damping.
EMITTER-FOLLOWER RESONANCE ANALYSIS FROM THE BASE CIRCUIT

The previous CC analyses were performed at the emitter node. The analysis
from the base node illustrates the plotting of negative impedances.
tT
-—
CE
Zb(hf ) ⇒ RB CB CE
-CE
tTCE
(a)
log Zb CB tTCE r

RB
+2
CE
fn fT log f
(b)
The figure shows the hf equivalent base-circuit impedance and reactance

plots of RB, CB, and the section marked tT ⋅CE. The plot of this circuit section is
the same kind as for Zb(CE) and has a value of tT ⋅CE in the hf region. Above fT,
the curve becomes capacitive with the value of CE. At resonance point r, the
slope of tT ⋅CE intersects RB with a slope change from −2 to zero indicating a
resonance. Unlike the emitter-node analysis, this is a series resonance (because
the change in slope across r is positive).
CB damps the resonance by intersecting RB at a lower frequency than fn. Then
Zb rolls off at this intersection along CB at a −1 slope. CB intersects tT ⋅CE where
the slope changes from −1 to −2, a change of −1. Therefore, no slope change
greater than ±1 occurs, and the resonance at r is damped. If RB or CB increases,
the break frequency 1/RB ⋅CB decreases, and damping is increased. This result
is consistent with the analysis from the emitter node.
From (a) and Zb(CE),
 s τ T + 1
Z b (hf ) =  2 (R B C B )
 s τT 
sτT + 1
= RB ⋅
s [τT R B (C B + C E )] + s (τT + R BC B ) + 1
2
From this,
b τT + R BC B
ζ= =
2 a 2 τT R B (C B + C E )
This is the same z as when derived from the emitter, as it must be.
EMITTER-FOLLOWER COMPENSATION WITH A BASE SERIES RC

The tT ⋅CE section of Zb(hf) can be all-pass compensated by the addition of a
series RC branch with positive corresponding element values. The two branches
null each other, leaving an open circuit. This leaves CE shunting RB and CB, a
nonresonant circuit. This compensation technique is shown below.
The compensation conditions are
τT
C = αo ⋅C E , R =
CE
192 Chapter 3
+V
Zb
⇒
RB
fT´,bo
+
Vi
− R
CE
C
−V
tT
-—
CE
Zb(hf )⇒ RB CB CE
- CE
tTCE
The equivalent compensated circuit is shown below.
tT
-—
CE
Zb(hf ) ⇒ RB
R
CE
C - CE
tTCE
The reactance chart (below) shows how this works.
log Z tTCE r

RB
C = CE
τT
—
CE RC network
R
CE
fn fT log f
At fT, both the RC and the tTCE curves break. The RC curve rolls off until fT,
where it becomes flat with value R. This curve dominates Zb below fT. At fT, the
tTCE curve breaks to CE and dominates as the reactance of CE falls below that of
R. The curve for C remains unbroken when C = CE. The result is that
 1 
Zb = RB 
 sC E 
As compensating components have tolerances, an exact cancellation between

the two series RC branches does not occur. The effect can be seen on the reac-
tance chart, however. Below fT, C is the stabilizing influence since it causes Zb to
roll off without a resonant change in slope. The resonance at r is damped when
C intersects RB at a frequency below fn. The condition for this is
R B ⋅C > τT ⋅C E ⋅ R B
or
C2 
R B ⋅   > τT
 CE 
For unconditional damping (independent of CE), then RB ⋅C ≤ tb. When R

and C are set to cancel tT ⋅CE, their values depend only on tT and CE. This leaves
RB free to be set to ensure adequate damping under variation of tT and CE.
BJT AMPLIFIER WITH BASE INDUCTANCE

The dual of the circuit with emitter capacitance and base resistance is one with
base inductance and emitter resistance (a), equivalent circuit (b), and reactance
chart (c).
From the base, the emitter resistance gyrates −90° and produces capacitance
tT/RE, which series-resonates with LB. As shown, RE is not large enough to damp
it. The circuit can be stabilized in the following ways:
1. Decrease RE until fn < fb. For this,
194 Chapter 3
LB LB
RE
RE τ
T
—
RE
(a) (b)
log Zb τT
—
RE
r LB
RE
fn fT log f
(c)
LB ⋅ τ B L ωβ
> βo ⋅ τT → R E < 2 B = LB ⋅
RE βo ⋅ τT βo
This extreme measure eliminates any hf resonance but slows the circuit more
than necessary.
2. Decrease LB until fn > fT. Then LB/RE < tT or LB < RE · tT. This approach also
eliminates hf resonance by moving the resonance above the high end of the
hf region. If LB is parasitic, this might not be possible.
3. Increase LB until fn < fb. Then, from method 1,
RE
LB >
ω β βo
4. Either increase or decrease RE until hf resonance is eliminated or sufficient

damping occurs.
5. Add a series L in the emitter. It is gyrated −90° at the base and provides series
damping resistance.
The last approach is the dual of shunt CB damping.

In the common-base (CB) stage of cascode amplifiers, base inductance com-
bines with emitter capacitance. CE is the output capacitance, mainly Cm of the
common-emitter (CE) transistor. At fn, LB crosses tT ⋅ CE with a −3 slope change.
By using the hf model, we can derive Zb:
 sτ + 1  sLB (sτT + 1)
Zb =  2 T  sLB = 3
 s τT C E  s τT LBC E + sτT + 1
This circuit can be stabilized several ways:

1. Decrease either LB or CE until LB crosses tT ⋅ CE above fT. Then, fn > fT.
2. Increase either LB or CE until fn < fb; then damp LB ⋅ CE with a series RE.
3. Shunt LB with CB > CE and damp LB ⋅ CB with a series RB.
4. Add a series emitter resistance RE > LB/tT; then RE + tT ⋅ CE crosses LB above
fT.
Combinations of these techniques are also possible. Methods 1 and 3 achieve
maximum speed.
THE EFFECT OF rb′ ON STABILITY
r´b Cm
b´
b c
+
Vbé´
rp Vbé´ Cp
—
r m
− e´
ré
Beginning with the BJT hybrid-p model, the effect of ohmic base resistance
rb′ on stability will be examined. A more accurate model would distribute Cµ
across rb′. This distributed-parameter RC is approximately modeled with lumped-
196 Chapter 3
parameter Cµ and rb′. The base resistance can be divided into several resistances
with portions of Cµ connected between them. The two extreme cases are to
connect Cµ to either the internal node (as shown) or to the (external) base ter-
minal node. Actual transistor performance lies within a range bounded by the
behavior of these extremes.
For circuits in which RB << rb′, then rb′ appears as rb′·tT from the emitter and
can resonate with external emitter capacitance CE. As rb′ is inaccessible, a stabiliz-
ing C cannot be shunted across the b′ node.
If Cµ is internal, it helps damp rb′·tT, whereas an external Cµ does not. If the
internal Cµ is insufficient, an additional base resistance RB will increase the total
base resistance so that
( rb′+ R B ) ⋅C µ ≥ τ β
If adding RB is infeasible and if rb′ is not enough damping, then RE can be added
in series with the emitter. This could, however, slow the response. A shunt RC,
having a time constant of tT, added in series with the emitter, forms an all-pass
network with the base impedance and preserves speed, as shown below (a) and
with resulting impedance (b).
τΤ
C =—
r´b
CE
r´b τTr´b
R = r´b
(a)
r´b CE
(b)
FIELD-EFFECT TRANSISTOR HIGH-FREQUENCY ANALYSIS

The BJT hf model can be generalized to apply to other kinds of active devices.
If the BJT-to-FET terminal mapping is used, then the major difference between
the BJT and FET models is Zp. The FET model is configured as a source
follower.
CGD
g VGS
+ —rm
CGS
Vi
− s
Vo
Zs
The expression for Zg is derived from the corresponding BJT expression for
Zb. By applying rm = rp/bo,
 1 
Z b = Z π + Z E ⋅ 1 + 
 s ( rπ βo )C π + 1 βo 
Let bo → ∞. Then for a fixed rm, Zp → 1/sCp and
1  1 
Zb ≅ + Z E ⋅ 1 + 
sC π  srmC π 
By analogy, Cp = CGS and ZE = ZS. Substituting yields
1  1 
Zg = + Z S ⋅ 1 + 
sCGS  srmCGS 
198 Chapter 3
For the source impedance,
Z G + 1 sCGS
Zs =
1 + 1 srmCGS
CGD is analogous to Cµ, and the time constant rm ⋅ CGS is the tT of a FET. The
BJT equivalent circuits can be applied to FETs with these BJT to FET
correspondences.
OUTPUT IMPEDANCE OF A FEEDBACK AMPLIFIER

Finally, hf modeling can be extended to amplifiers having a dominant single-
pole response. Consider an amplifier with an open-loop voltage gain of
K
G=
sτ bw + 1
where 1/tbw is the small-signal open-loop bandwidth. In the lf region, the open-
loop output resistance rout is reduced by the feedback by 1 + GH. The resulting
closed-loop output impedance is
rout r sτ bw + 1
Z out (cl) = = out ⋅
1 + GH 1 + KH s (τ bw (1 + KH )) + 1
where KH is constant. This equation can be expressed as
1
Z out (cl) =
1 1
+
rout sτ bw (rout KH ) + (rout KH )
In continued-fraction form, the corresponding topology of Zout(cl) is

explicit.
tbw
rout —
KH
Zout(cl) ⇒ rout
rout
—
KH
(a)
tbw
Zout(hf ) ⇒ rout rout —
KH
(b)
log Zout
rout
rout
1+ KH
fbw fT log f
(c)
The lf closed-loop resistance, rout/(1 + KH), gyrates +90° at the open-loop

bandwidth to appear inductive out to the unity-gain frequency,
f bw (1 + KH )
Above this unity-gain frequency, Zout(cl) reverts to rout. By analogy, fbw corresponds
to the BJT fb, the unity-gain frequency to fT, and KH to bo . The simplified hf
equivalent output is derived by letting KH → ∞ as f → 0, with resulting output
impedance corresponding to Ze(RB). When rout is generalized to Zout, the corre-
sponding BJT models readily apply. The hf equivalent circuit of Zout(cl), as with
the BJT model, is only valid above fbw.
200 Chapter 3
Example: Feedback Amplifier Output Resonance
Rf
100 kΩ
R1 rout
+ Vo
100 kΩ 1 kΩ
1 nF CL
Vi −200
s (1 µs) + 1
−
This amplifier has a static gain of −200 and a pole at 1 Ms−1 (tbw = 1 µs). It has
an output resistance of 1 kΩ and a load capacitance of 1 nF.
The loop gain KH = 100 and fT ≅ 15.9 MHz. Then the gyrated resistance is
(10 ns) ⋅ (1 kΩ) = 10 µH and Zn = 100 Ω. For a parallel resonance, z = Zn/2rout
= 100 Ω/2 kΩ = 0.05. Then Mp = 0.85. The simulated circuit Mp ≅ 0.71, indicat-
ing that the hf model estimate of z is low. The amplifier simulations for loop
gains of 10, 100, and 1000 are tabulated as follows:
KH 1000 100 10
Mp(SPICE) 0.87 0.71 0.25
L 1 µH 10 µH 100 µH
Zn 31.6 Ω 100 Ω 316 Ω
rout /KH 1Ω 10 Ω 100 Ω
z(hf) 0.0158 0.050 0.158 = Zn/2rout = (rout/KH)/2Zn
Mp(hf) 0.95 0.85 0.60
z 0.0316 0.10 0.316 = 1 KH
Mp 0.91 0.73 0.35
The table reveals that the predicted z using either pure parallel or series reso-
nance is always low and the error increases as KH decreases. If rout/KH and the
load capacitance CL are included in the model (as in (a) above), then the expres-
sion for Zout(cl) is
sτ bw + 1
Z out (cl) =  out  ⋅
r
 1 + KH  {s [τ bw (1 + KH )] + 1} ⋅ (sroutC L + 1)
and
1
ζ=
KH
With this more accurate z (the lower entry in the above table), the agreement
with simulation results is much better in the corresponding Mp values. This
expression for resonant Zout(cl) has the same form as Ze, from which analogies
can be made.
Even with the more exact z, the error grows with decreasing KH. This is due
to the growing error in the asymptotic approximations of the impedance plot.
For KH = 10, the error in Mp is quite apparent (40%), but for KH = 100, is much
reduced (3%). In this example also, the resonant frequency is near the center
of the hf range, thus reducing error due to linear approximation. Near either
fbw or fT, this error becomes large; the approximate calculations of z should be
used only as a worst-case lower bound.
CLOSURE
By deriving the complex-frequency expression for b from the BJT hybrid-p
model, we can write the b-dependent impedance transformations at base and
emitter nodes in a general form. When these impedances are represented
graphically on a reactance chart, the effects of circuit element variations on
circuit behavior, especially hf resonances, become evident. The frequency-
dependent b transform is also applicable to FETs and single-pole feedback
amplifiers. They show the same impedance gyrations as the BJT, so the BJT
results can be easily extended to them also.
References
Archer, J. A., Gibbons J. F. and Purnayia M. (1968, March). Use of Transistor-Simulated

Inductance as an Interstage Element in Broadband Amplifiers. IEEE Journal of Solid-
State Circuits SC-3, 12–21.
Barna, A. (1973, June). On the Transient Response of Emitter Followers. IEEE Journal
of Solid-State Circuits SC-8, 233–235.
Battjes, C. (ca. 1970). “Amplifier Frequency and Transient Response” Course Notes.
Tektronix Education Program, Tektronix, Inc., Beaverton, OR.
Bilotti, A. (1967, September). Common-Collector Impedance Transformation Sensitiv-
ity. IEEE Transactions of Circuit Theory CT-14, 364–366.
Cheng, D. K. (1959). Analysis of Linear Systems. Boston: Addison-Wesley.
Chessman, M. and Sokol, N. (1976, June). Prevent Emitter-Follower Oscillation. Elec-
tronic Design 13, 110–113.
DeBella, G. B. (1966, April). Stability of Capacitively-Loaded Emitter Followers – a
Simplified Approach. Hewlett-Packard Journal 17, 15–16.
Gough, R. G. (1982, August). High-Frequency Transistor Modeling for Circuit Simula-
tion. IEEE Journal of Solid-State Circuits SC-17, 666–670.
Gupta, S. and Hasdorff, L. (1970). Fundamentals of Automatic Control. New York: Wiley.
Kozikowski, J. L. (1964, March). Analysis and Design of Emitter Followers at High Fre-
quencies. IEEE Transactions on Circuit Theory CT-11, 129–136.
Lindmayer, J. and North, W. (1965). The Inductive Effect in Transistors. Solid-State Elec-
tronics 8, 409–415.
Madhu, S. (1988). Linear Circuit Analysis. Upper Saddle River, NJ: Prentice-Hall.
Nilsson, J. W. (1986). Electric Circuits. Boston: Addison-Wesley.
Ollins, R. I. and Ratner, S. J. (1972, December). Computer-Aided Design and Optimiza-
tion of a Broad-Band High-Frequency Monolithic Amplifier. IEEE Journal of Solid-State
Circuits SC-7, 487–492.
204 References
Pettit, J. and McWhorter, M. (1961). Electronic Amplifier Circuits: Theory and Design. New
York: McGraw-Hill.
Roberge, J. K. (1975). Operational Amplifiers: Theory and Practice. New York: Wiley.
Ruehli, A. E. (1972, September). Inductance Calculations in a Complex Integrated
Circuit Environment. IBM Journal of Research and Development, 470ff.
Saucedo, R. and Schiring, E. (1968). Introduction to Continuous and Digital Control Systems.
Macmillan, New York.
Street, M. A. (ca. 1975). Passive A.C. Circuit Analysis Using Reactance-Bode Charts. Portland,
OR: Bonneville Power Administration.
Street, M. A. (1976, May 5). “Simplify AC Circuit Analysis with Reactance-Bode Charts”.
EDN, pp. 83–89.
Starič, P. and Margan, E. (2006) Wideband Amplifiers, Springer, Dordrecht, Nether-
lands, Part 3.
INDEX
Index Terms Links
all-pass 66 75 77 78
108 154 191 196
asymptotic approximations 45 61 84 180
183 201
bandwidth 53
Bode plot 47
break frequency 48
bridge-T 148 150
bypassing 164
capacitively-loaded emitter-follower 182

cascade amplifier 195
cascaded compensation network 76
characteristic equation 10 17 51 126
characteristic impedance 15 163 164 180
185
Chebyshev filters 56
circuit-referred flux 1
class AB amplifier 189
complex conjugate 12 32
This page has been reformatted by Knovel to provide easier navigation.

Index Terms Links
complex pole-pairs 33 160

complex-frequency domain 7 9 73 167
composite 154
conditionally stable 12 91
conductivity 2
constant-resistance network 187
continued fraction 172 174
critical frequencies 18 47 107
critically damped 14 15 134
CRT deflection circuit 36
damped frequency 14 160

damped sinusoid 13
damping factor 9
damping ratio 9 93 95 96
134 141 180 184
dipole 104
dominant pole 53
dominant-pole compensation 105 120 133 136
duals 2
envelope 13 54 55
exponential decay 6 7 14 104
frequency response 47
frequency-domain analysis 1 41

Index Terms Links
gain margin 93
geometric mean 100
group delay 54 58 156 157
hf equivalent circuit 173 177 178 181

183 199
high-frequency path 151
high-frequency region 167 170
hybrid-π BJT model 168
impedance gyration 167 177 183 201

impulse function 21 29
inverse Laplace transform 22 29 34
Laplace transform 22 23 29 30
33
LC circuits 179 180
load capacitance 155 200
load isolation 151
loop gain reduction 105
low-frequency path 151

Index Terms Links
maximum magnitude 47
MFA response 54 162 185
MFED response 56 161
Miller effect 122 126
minimum-phase 91
natural frequency 9 11 14 141

natural response 5 6 12
noise 133
nonlinear 13
nonminimum-phase 91
normalized form 18 44
Nyquist criterion 91 93
ohmic base resistance 195

oscillation 40 91 105 164
165 180 183 186
203
overshoot 38 39 53 76
94 95 113 161
parameter variation 127 131

parasitic reactances 109

Index Terms Links
partial-fraction expansion 29 32 33 35
37 38
passive path 123
peaking 53 54 58 156
157 188
permeability 2
permittivity 2
phase margin 93 116 133
phase-lag compensation 118
phase-lead compensation 103 111 118 160
phase-lead compensator 108 110
polar form 7 9 33 85
pole movement 51 88 102
pole separation 100 105 112 120
128 152
pole-splitting 105 121 122 131
pole-zero cancellation 104 161
pole-zero separation 140 143 150
power-supply leads 100
preshoot 56
quadratic formula 11
quadratic pole response 87
quartz crystal 28
RC differentiator 18 19 25 31
61 121

Index Terms Links
RC integrator 4 25 30 61
69 157
reactance chart 43
reactive circuit elements 7
rectangular form 7
reduction theorem 167
resistivity 2
resonant point 180 189
response compensation 89 101 163
ringing 14 37 57 155
risetime 48
RLC circuit 15 31 34 58
root loci 85 102 124 127
root-contour plot 127
root-locus plot 43
Sallen-Key 105
second-order loop gain 88
sensitivity 203
settling time 41 76
single-pole response 7 88 198
single-pole roll-off 80 110
SNR 133
source resistance 164
steady-state response 16 18
step function 21 29 37
substitution theorem 127
symmetric 46 86

Index Terms Links
time constant 6 14 31 41
61 67 76 160
183 188 196 198
time-domain analysis 1 29
trace inductance 163
transconductance 130 131 160
transformed impedances 172 173
transient response 4 6 10 11
16 18 19 21
29 41 103 203
transimpedance amplifier 118 136
transmission lines 181
two-pole compensation 137 140 148 150
unity-gain crossover frequency 94 98

unity-gain frequency 167 168
Wien-bridge 26 27 68

Designing Dynamic Circuit Response, Volume 2 - Analog Circuit Design Series (PDFDrive)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Designing Dynamic Circuit Response, Volume 2 - Analog Circuit Design Series (PDFDrive)

Uploaded by

Copyright:

Available Formats

Analog Circuit Design Series Volume 2

Copyright © 2010 by Dennis Feucht. All rights reserved.

Editor: Dudley R. Kay

Printed in the United States of America

Library of Congress Cataloging-in-Publication Data

Chapter 1 Transient and Frequency Response . . . . . . . . . . . . . . . . . . . . . . . 1

Chapter 2 Dynamic Response Compensation . . . . . . . . . . . . . . . . . . . . . . . 75

Two-Pole Compensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

Chapter 3 High-Frequency Impedance Transformations . . . . . . . . . . . . . .167

REACTIVE CIRCUIT ELEMENTS

where s is conductivity of the conductive material. This equation is the recipro-

where r is resistivity = 1/s, and R = 1/G.

The energy W stored by an inductor or capacitor can be derived by noting

and that power is, by deﬁnition,

Combining and solving for W,

For linear elements, the deﬁnitions of L and C can be simpliﬁed to

Substituting these expressions into W, the energy stored in a linear L or C is

The energy stored in an inductor is proportional to the square of the current

where t is a local (dummy) variable of integration, and t is time.

FIRST-ORDER TIME-DOMAIN TRANSIENT RESPONSE

or, arranging into the form of a standard differential equation,

This is an ordinary, linear, constant-coefﬁcient differential equation and can be

Substituting for vo gives an algebraic equation:

The exponential cannot be zero and V ≠ 0. Consequently, the middle factor

This algebraic equation in s is the characteristic equation of the differential equa-

and substituting for s, the natural response is

If the capacitor has an initial charge of vo(t) = V0 and vi = 0, then

COMPLEX POLES AND THE COMPLEX-FREQUENCY DOMAIN

The complication is in solving the characteristic equation. Its degree is equal to

and the Pythagorean theorem

The rectangular components are expressed in polar components as

Dividing w by s and solving for f results in

With these expressions we arrive at the general form of a quadratic characteristic

The general solution of the quadratic equation is

tan θ = − tan (180° − θ )

This pole pair is shown plotted below.

SECOND-ORDER TIME-DOMAIN RESPONSE: RLC CIRCUIT

Using KCL, its differential equation is

These poles are of the form

Each pole contributes a solution to the differential equation when substituted

For the RLC circuit, n = 2, and the response is

i (t ) = c1 ⋅ e (−α + jωd )⋅t + c 2 ⋅ e (−α − jωd )⋅t

This can be expressed as

i (t ) = e −α⋅t ⋅ (c1 ⋅ cos ωd ⋅ t + jc1 ⋅ sin ωd ⋅ t + c 2 ⋅ cos ωd ⋅ t − jc 2 ⋅ sin ωd ⋅ t )

= e −α⋅t ⋅ [(c1 + c 2 ) ⋅ cos ωd ⋅ t + j (c1 − c 2 ) ⋅ sin ωd ⋅ t ]

If c1 and c2 are complex conjugate constants, and

then the response is

i (t ) = e −α⋅t ⋅ (I 1 ⋅ cos ωdt + I 2 ⋅ sin ωdt )

i (t ) = I e −α⋅t ⋅ sin (ωd ⋅ t + ϑ )

This response is a damped sinusoid (shown below).

Poles split off w

0 < z < 1: Underdamped

This is the characteristic impedance of the LC elements. L and C also

FORCED RESPONSE AND TRANSFER FUNCTIONS IN THE S-DOMAIN

When the circuit is driven by a source, it continues to respond indeﬁnitely to

The transient response is found by setting N1 = 0 and xo = Xo(s) ⋅ est, where Xo is

The exponential can be factored out of D, leaving the characteristic equation

If xi is also made a complex exponential, then a similar factorization on the

This is an extremely important result. Whenever the input is a complex expo-

To illustrate these general concepts by example, consider the RC differentiator

Using KCL at the output node,

This can be rearranged as

s functions as a differentiation operator for complex exponentials and, more