Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

332 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO.

2, FEBRUARY 2010

Noise in Current-Commutating Passive FET Mixers


Saeed Chehrazi, Member, IEEE, Ahmad Mirzaei, Member, IEEE, and Asad A. Abidi, Fellow, IEEE

Abstract—Noise in the mixer of zero-IF receivers can compro-


mise the overall receiver sensitivity. The evolution of a passive
CMOS mixer based on the knowledge of the physical mechanisms
of noise in an active mixer is explained. Qualitative physical
models that simply explain the frequency translation of both the
flicker and white noise of different FETs in the mixer have been
developed. Derived equations have been verified by simulations,
and mixer optimization has been explained.
Index Terms—Flicker noise, passive mixer, physical mechanism,
white noise, zero-IF.

I. INTRODUCTION

IXER NOISE, particularly flicker noise in CMOS


M mixers, can be troublesome when receiving narrowband
wireless channels such as Global System for Mobile communi-
cations (GSM). In a direct-conversion receiver [1], where the
signal downconverts to baseband after only minimal amplifica-
tion at radio frequency (RF), mixer noise degrades signal-to-
noise ratio, and as a consequence, the overall noise figure of the
Fig. 1. CMOS Mixers. (a) Current-commutating double-balanced passive
receiver suffers. Therefore, employing the mixer which has the mixer. (b) Double-balanced active mixer.
lowest noise and optimizing it gains a lot of receiver sensitivity.
In the course of designing downconversion mixers for the first
prototype of a software-defined radio (SDR) for GSM and wire-
filtering. We describe the processes whereby both flicker and
less LAN [2] and having analyzed the noise in a voltage-driven
white noise appear at the output of a current-commutating pas-
passive mixer a few years before [3], we concluded that a cur-
sive CMOS downconversion mixer. In investigating these pro-
rent-commutating mixer [Fig. 1(a)], which only switches RF
cesses, simple intuitive models of noise in this mixer have been
current, can have a superior noise performance to other mixer ar-
developed whose predictions agree very well with accurate sim-
chitectures. Although the voltage-driven passive mixer is called
ulations of the mixer noise. Straightforward equations capture
passive because it does not provide any gain, a current-driven
the noise (both flicker and white) originating in different parts
passive mixer has the same gain as an active mixer. We call this
of the mixer. There are many noise sources in the mixer but
mixer passive because its current-commutating FET switches
only some of them are important. Section II explains the evolu-
always remain in the triode region. We believe this circuit was
tion of the current-commutating passive mixer and its different
first introduced in [4] but it was distinguished from a voltage-
modes of operation. Sections III and IV present the flicker and
driven passive mixer in [5]. Since then, it has been reported in
white-noise analyses of the mixer, respectively. Section V veri-
several publications from both industry and academia [6]–[17].
fies the accuracy of the derived equations with SPECTRE-RF
The general belief is that passive mixers are flicker noise
simulations, and Section VI presents the overall mixer noise
free because the dc of the switches is zero [5]. Our findings,
both at low frequencies, where flicker noise is dominant, and at
as reported in this paper, show that, without requiring dc flow,
higher baseband frequencies, where the white noise dominates,
flicker noise appears at baseband if undesired signals (blockers)
and points out the major noise contributors. Section VII con-
are present at certain frequencies. This is particularly important
cludes the paper.
when the passive mixer is used in a broadband front end such
as in SDR, where blockers can reach the mixer with minimum
II. CURRENT-COMMUTATING PASSIVE MIXER
Manuscript received July 28, 2008; revised February 02, 2009. First published In narrowband applications such as paging or GSM, the mixer
June 02, 2009. Current version published February 10, 2010. This paper was
recommended by Associate Editor S. Mirabbasi.
noise is very likely to overwhelm the downconverted signal
S. Chehrazi is with Marvell Semiconductor Inc., Santa Clara, CA 95054 USA and limit, as a consequence, the overall receiver noise figure.
(e-mail: chehrazi@ee.ucla.edu). Here, we show the evolution of a current-commutating passive
A. Mirzaei is with the Broadcom Corporation, Irvine, CA 92617 USA. mixer using the knowledge of the flicker-noise mechanisms in
A. A. Abidi is with the Electrical Engineering Department, University of Cal-
ifornia, Los Angeles, CA 90095 USA. an active mixer. We also discuss its different modes of operation
Digital Object Identifier 10.1109/TCSI.2009.2023762 and explain which one is well suited for downconversion.
1549-8328/$26.00 © 2010 IEEE

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.
CHEHRAZI et al.: NOISE IN CURRENT-COMMUTATING PASSIVE FET MIXERS 333

Fig. 2. Reducing the dc portion of the commutated current by injecting part of


the tail current into the common-source node lowers flicker noise.

A. Evolution of a Current-Driven Passive Mixer

Previous research [18] indicates that the output noise of an ac-


tive mixer [Fig. 1(b)] due to the flicker noise of the FET switches
is directly proportional to the dc which is commutated by the
switching pair. Therefore, reducing this dc results in less output
flicker noise. This is usually implemented by injecting part of
the transconductor bias current into the tail of the differential Fig. 3. Illustrating OFF and ON overlap modes in passive mixer. (a) OFF
pair, as in [19] (Fig. 2). This method has a few drawbacks such as overlap. (b) ON overlap.
increased white-noise level due to the additional current source
plus linearity degradation. Moreover, reducing the bias current
of the active FET switches of the differential pair results in a simple common-gate (CG) amplifier which was used as the cur-
smaller switch transconductance. This exacerbates the filtering rent-to-voltage buffer in [10].
of the high-frequency RF signal at the common source node
by the tail parasitic capacitance. These problems have inspired B. Modes of Operation
other innovative designs, such as [20], which reduces the bias The tail of a matched differential pair in the Gilbert cell-based
current only at the local-oscillator (LO) zero crossings. active mixer adapts its voltage to commutate the tail current ex-
A mixer which fundamentally has no flicker noise is simpler actly at the differential zero crossings of the voltage drive. This
and easier to design. Setting the commutated dc to zero, instead is not so in a passive mixer, whose terminal voltages are forced
of just reducing it, completely removes the output flicker noise by external voltage bias into one of the three regions of oper-
due to this mechanism. This leads us to a mixer which only com- ation [3], [4], [21], [22]: OFF overlap, zero overlap, and ON
mutates the ac signal currents. Since the mixer switches carry overlap (Fig. 3). Overlap refers to a window in time around the
no bias current, in order to have a small ON resistance, they LO zero crossings where two FETs with common input termi-
should turn on in the triode region and should be driven hard nals are in the same state; they are either both ON or both OFF.
by strong LO voltages. The capacitive coupling of the input Over the rest of the LO cycle, one FET is ON and the other
transconductance stage to the switching pair [ in Fig. 1(a)] is OFF. The current-commutating active mixer always operates
guarantees the triode operation. When the LO is high, the triode in ON overlap because, in the transition region, both FETs of
FET switch directly connects the input transconductor to the a differential pair are ON. The overlap type is determined by
output load through its ON resistance. This is in contrast with the relative value of the bias voltage at the source terminals to
the differential pair FETs in an active mixer which serve as cur- the bias (or voltage of differential zero crossing) at the gate. For
rent buffers at the same time when they route the signal and NMOSFETs, if , the mixer operates in OFF
the dc to one side or the other. Therefore, in order to minimize overlap, where is the gate bias voltage of the FET and
the effect of the circuits following the mixer on the operation is its threshold voltage. This assumes an amplitude of the LO
of the switching FETs, they should be isolated by a current voltage greater than . If , the
buffer [Fig. 1(a)]. This current buffer, which ideally has zero mixer works in the ON overlap.
input resistance, along with the hard-driven triode switches al- While the voltage-driven passive mixer can operate in any
leviates the filtering effect of the parasitic capacitance at the mode of operation [3], a current-driven passive mixer is best
input of the switching FETs and reduces the voltage swing at used in ON overlap. This is to avoid large swings at the input
the input node. If a voltage output is desirable, a transresistance while all switches are momentarily off in overlap time. In this
buffer can be used. The switched current can be converted to paper, it is assumed that the switches of the current-driven pas-
voltage while it is partially filtered by the output impedance of sive mixer are biased in ON overlap.
this buffer. In order to reject the blockers and unwanted chan-
nels to the level that the receiver ADC can withstand, further
III. FLICKER NOISE
filtering stages might be necessary after this buffer. The input
bias voltage of this buffer also sets the source/drain dc bias Flicker noise is present in all transistors making up a mixer:
voltage of the FET switches ( ). In this paper, we analyze the the input transconductor, the switching core, and the output

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.
334 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO. 2, FEBRUARY 2010

buffer. We have successfully used an empirical model devel-


oped some years ago [23] to predict flicker noise, with notable
accuracy, in time-invariant and time-varying circuits [3], [18].
This model assumes that the flicker noise in FET in any region
of its operation is specified by an input-referred gate voltage
, whose magnitude depends on the FET
dimensions and and an empirical parameter that is
technology dependent. is the normalized specific capac- Fig. 4. Single-balanced current-commutating active mixer.
itance of the gate insulator. When gate referred, the spectral
density is independent of bias voltage. Thus, to simulate the
output flicker noise, first, the periodic transfer function from indirect mechanism, which is because of the parasitic capaci-
the gate of each mixer FET to the output is found through tance at the tail of the differential pair. In the direct mechanism,
simulation, and then, this is multiplied by the input-referred the slowly varying gate-referred flicker noise randomly modu-
gate voltage calculated as previously mentioned. lates the commutation instant which is ideally located at the zero
In this section, we describe the process through which the crossings of the LO. This modulation results in a train of noise
flicker noise of the transconductor, the noise of the switches, and pulses which add to the ideal square-wave commutation wave-
the buffer appear at the mixer output. In this paper, we assume form, and as a consequence, flicker noise appears at the output.
square-wave-like LO voltages, with finite transition slopes, for The analysis in [18] is for a mixer with zero RF input. In a real
simplicity. This is a realistic assumption since many practical situation, when the mixer is downconverting, in addition to the
applications use CMOS-inverter-based LO drivers. dc, the RF current of the transconductor is also fed to the dif-
ferential pair. Thus, a more accurate approximation of the noise
A. Transconductor Flicker Noise pulse train should include both currents which, for the mixer
Although the transconductor current is contaminated by the shown in Fig. 4, is
FET flicker noise, the spectrum of the desired channel—which
is at RF—is far away from that of the flicker noise, which is at
baseband. After downconversion, flicker noise will be upcon-
verted away from the received channel to and to its odd
harmonics. This assumes that the FETs in the mixer switching
core are perfectly matched; however, if there are offsets and mis-
matches, the transconductor flicker noise can leak to the output.
The output flicker-noise current due to the input transconductor (2)
and the offset in one pair of the mixer switches is [18]
is the width of the error pulse which is equal to (Fig. 5).
(1) Taking the Fourier transform of the aforementioned equation
gives the output noise in frequency domain
where is the input-referred offset voltage of the switching
pair, is the slope of the single-ended LO during the tran-
sition and is its period, is the input-referred flicker-
noise voltage of the transconductor, and is its single-ended
transconductance. (3)
As mentioned in Section II-A, this mixer best operates if the
FET switches turn on in the triode region. In order to guarantee Equation (3) reveals that, in a practical situation, the direct
triode operation, the transconductor is capacitively coupled to switch noise goes to the output in two different ways: One is
the mixer core. This practically small coupling capacitor, along due to the switched dc, and the other is due to the commutated
with the series combination of the switch resistance and the RF current. Reference [18] explains the former completely;
input resistance of the buffer, forms a high-pass filter. The de- however, the latter can also be troublesome. When there is
sired RF channel lies in the passband of this filter and passes an input RF signal at , output flicker noise appears both at
through without attenuation, while the flicker noise—which is DC and . This says that, if the only input signal
in its stopband—is filtered out. Therefore, if the transconductor is the desired channel, in a zero-intermediate-frequency (IF)
is capacitively coupled to the mixer core, it will contribute no receiver—where —flicker noise will be upconverted
flicker noise to the mixer output, even in the presence of mis- to the LO frequency, and as result, it does not appear at the
matches in the switch FETs. baseband. However, if there is a blocker at ,
this mechanism will transfer the switch flicker noise to the
B. Switch Flicker Noise
baseband output—the same place where the desired channel
In an active mixer, the flicker noise of the FET switches ap- is after downconversion. Even if a low IF or a heterodyne
pears at the output through two different mechanisms [18]: di- architecture is used in order to avoid the switch flicker noise
rect mechanism, which is due to the commutated currents, and due to the dc commutation, it can still appear at IF if there is

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.
CHEHRAZI et al.: NOISE IN CURRENT-COMMUTATING PASSIVE FET MIXERS 335

Fig. 5. Process through which flicker noise appears at the output of a current-driven passive mixer. (a) Mixer circuit. (b) LO waveform. (c) Sinusoidal input
current. (d) Output differential current with noiseless and noisy switching. (e) Noise pulses.

to the output. If there is a blocker tone with an amplitude


located at , the baseband flicker noise in output
current will be

(5)
Fig. 6. Translation of switch FET flicker noise to IF due to the presence of a
blocker at f = f f . + We can show, with an example, the importance of this second
flicker-noise mechanism both in the active and the passive
a blocker at , where is the frequency of mixer. Assume that the tail bias current of the active mixer in
the desired channel (Fig. 6). This is particularly important in a Fig. 4 is 1 mA and that it is run by a LO whose differential
broadband front end such as in an SDR, which allows all the . SPECTRE-RF simulation shows that the switch
blockers to enter the receiver. flicker noise appears at the output with a gain of 63 dB due
The flicker noise of the current-commutating passive mixer to the dc commutation. Assuming that the mixer downconverts
can be analyzed in a similar way, except that the only commu- to zero IF, in the presence of a 10-dBm blocker at
tated current is the RF current. In an active mixer, the flicker and for a 15-mS single-ended transconductor, the flicker noise
noise of each switch affects the turnon/turnoff instant of both transfer gain is about 68 dB. This shows that baseband flicker
switches due to the floating source voltage of the differential noise due to the blocker can be comparable with that due to
pair. Here, since the source voltage is fixed by the buffer, the the dc in the active mixer. The flicker-noise transfer gain for
flicker noise of each FET only modulates the turnon or turnoff the current-driven passive mixer is also 68 dB. If the passive
instant of that FET and does not affect the other; hence, is the mixer is used in a narrowband receiver, where the faraway
slope of the single-ended LO, and the magnitude of the noise blockers are removed before reaching the mixer, the core of
pulse is the same as the input current. Therefore, the pulse train the current-driven passive mixer will not create any baseband
can be found as flicker noise. In this case, the mixer shows a significant noise
advantage over the Gilbert cell-based active mixer which
always has the flicker noise due to the dc. If the front end is
(4)
broadband, such as that for SDR [10], and large blockers appear
at specific frequencies, they will create baseband flicker noise
where is the single-ended input transconductance of the in both active and passive mixers, but the overall flicker noise
mixer. Using the same argument as in an active mixer, only the of the passive mixer will remain lower due to lack of dc.
blockers located at special frequencies determine the flicker The indirect mechanism of flicker noise [18] in active mixers
noise. Therefore, the switch flicker noise in a current-commu- is due to the parasitic capacitance at the tail of the differential
tating passive mixer appears at the output only in special cases pair. Flicker-noise voltage charges and discharges this tail
and is proportional to the amplitude of the transconductor ac capacitance through the ON switch which acts as a source
due to the blocker. This is, however, different from the expla- follower. This mechanism does not exist in the current-driven
nation in [4], where the current due to the coupling of the large passive mixer. The buffer maintains a virtual ground through
LO voltage to the drain circuit through overlap capacitance is on switches which always turn on in the triode, and as a
said to be responsible for transferring the switch flicker noise result, the flicker-noise source does not effectively charge or

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.
336 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO. 2, FEBRUARY 2010

discharge the tail parasitic capacitance. This has been verified


by SPECTRE-RF simulations.
C. Buffer Flicker Noise
The transresistance buffer which follows the mixer core in
Fig. 1 also adds its own flicker noise to the downconverted
channel. At first glance, it may seem that the flicker-noise con-
tribution of this buffer can be analyzed independently without
considering the preceding mixer switching core. However, we
will show that, although the flicker noise of some transistors ap- Fig. 7. CG amplifier as a transresistance buffer.
pears directly at the output, that of some others depends on the
operation of the switching core. For this analysis, we use the switches are replaced by their on conductances. Because the
transresistance buffer shown in Fig. 7, which was used in [10]. It FET switches are always in deep triode, the square-law model
is made up of CG transistors and , bias current sources is accurate enough for hand calculations [24]. We can write
, and a load circuit. and are the downconverted RF
signals, which come from the switching core of the mixer. If
made out of FET transistors, each of these building blocks con- (6)
tributes flicker noise to the output.
The flicker noise of the load circuit directly adds to the output.
The same is true for the bias current source which sees the CG (7)
transistor at its output. Nonideal switching makes the transfer
function of this noise smaller than one but this effect is negli- where and are the gate and source bias voltages, respec-
gible since the overlap time is much smaller than the LO period. tively, is the threshold voltage of the FETs, is the dc
Minimizing the flicker noise of these directly lowers the overall effective voltage of the switch, and . is the
flicker noise of the transresistance buffer. Since the FET flicker channel mobility, is the gate oxide capacitance, and and
noise is inversely proportional to its area, large FET transistors are the width and length of the switch. Therefore, the gate-re-
can be used for both the load and the bias current source. A large ferred flicker-noise voltage creates a differential noise cur-
parasitic capacitance of these FETs can be tolerated because the rent
desired channel is already downconverted to baseband, before
reaching this buffer. If the voltage headroom is not a limitation, (8)
resistive loads can be used, and the bias current source can be
replaced with polyresistors in order to completely eliminate the which is only present during the overlap time.
flicker-noise contribution of these blocks. The source degenera- is the total transconductance of the CG transistor, including
tion resistor along with the gate bias voltage of the CG transistor that due to body effect . This noise pulse repeats at every
determines the bias current of the buffer. transition of the LO [Fig. 8(b)]. The baseband flicker noise at
Considering Fig. 7, the output flicker noise due to the CG the output is the average of this periodic noise pulse
transistors themselves seems to be zero. These FETs see the
large impedance of the bias current source across the bandwidth (9)
of the desired channel,1 and as a result, the flicker-noise gain
from their gate to the output should be very small. However, where and are the edges of the overlap time. Taking the LO
this is true if the mixer switching core commutates instantly and zero crossing as the origin of time
does not have any tail parasitic capacitance. These two nonide-
alities result in the appearance of the flicker noise of the CG (10)
transistors at the output. In comparison with the noise processes
in the active mixer [18], we call the mechanism which is due to (11)
imperfect commutation direct mechanism and that which is due
(12)
to the parasitic capacitance indirect mechanism.
1) Buffer Flicker Noise, Direct: In contrast to the stand-alone where is time. is like a slowly varying offset voltage whose
buffer, where the CG transistors have close-to-zero flicker-noise period is much larger than , and it can be assumed constant
contribution, when it is connected to the mixer, CG noise goes over the transition time where the average integral is calculated.
to the output. This is due to the nonzero rise and fall times of Thus
the LO. As it was explained in Section II-B, switches should be
biased in ON overlap. As a result, during the transition time of
the LO, all four FETs in the mixer core are on. Fig. 8(a) shows
the equivalent circuit of the mixer during the ON overlap where
the input transconductor is considered an open circuit and the
1Parasitic capacitance lowers the output impedance at higher frequencies but (13)
that will be outside the bandwidth of the downconverted channel.

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.
CHEHRAZI et al.: NOISE IN CURRENT-COMMUTATING PASSIVE FET MIXERS 337

Fig. 10. Mixer followed by the CG buffer.

Fig. 8. Direct mechanism of buffer flicker noise. (a) Equivalent circuit of the
mixer during overlap time. (b) Gated flicker-noise current during overlap time.

Fig. 11. Equivalent circuit of the mixer followed by the CG buffer with ideal
switching used for indirect buffer noise calculation. (a) M is on. (b) M is on.

tribution can be made very small. The large parasitic capacitance


Fig. 9. Factor inside the bracket in (13). at the input of the buffer was ignored in this analysis. In the pres-
ence of that, the transfer function will increase a little bit since
this capacitance will sample the noise at the edge of the overlap
Fig. 9 shows the plot of the expression inside the bracket in time and will charge to or discharge to dc outside the overlap.
(13) as a function of , which is the ratio of the 2) Buffer Flicker Noise, Indirect: The analysis so far sug-
CG transconductance to the dc conductance of the switch at gests that the CG flicker noise may be eliminated if the LO
the LO zero crossing. As we expected intuitively, if the CG waveform is a perfect square wave with infinite slope at zero
transconductance is much larger than the switch conductance, crossings. However, as the LO slope rises, the CG flicker noise
the flicker-noise current of the FET only circulates inside the CG appears through another mechanism that depends on the LO fre-
transistor itself. On the other hand, if the switch conductance is quency and the parasitic capacitance at the input of the switches.
much larger, most of this current will leave the transistor and This is like the indirect mechanism in the active mixer [18].
appear at the output. In that case, the output noise simplifies to Fig. 10 shows the mixer followed by the CG buffer. The input
transconductor is replaced by its total output parasitic capaci-
(14)
tance, i.e., and . When the LO is high, the CG transistor
is connected to the tail parasitic capacitor through the
In practice, the parasitic capacitance of the switch will limit
ON resistance of the switch . As soon as the LO falls,
its size. Although, to the first order, the intrinsic time constant
turns off and , which is driven by , turns on and con-
of the switch is independent of its size,2 large switches have a
nects to the other CG transistor . Now, let us consider
large gate capacitance. Therefore, a strong but power-hungry
the flicker noise of the buffer transistor , which is repre-
LO buffer is needed to drive this capacitance. Therefore, since
sented by the gate series voltage source in Fig. 11. During
the power budget of the LO buffer is limited, there exists a min-
the ON period of the LO, charges up the tail capacitor
imum achievable switch resistance. On the other hand, the buffer
through the source follower transistor and the ON resis-
transconductance can virtually be very large because the band-
tance of . For simplicity, we assume that the tail band-
width requirement is not very stringent. Therefore, its noise con-
width of the mixer is much larger than the LO frequency ( 4
2Because C / W and R / 1=W . times) and can charge up to its maximum voltage which is

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.
338 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO. 2, FEBRUARY 2010

Similar to the load and the current source, larger FETs can
be used for the CG transistors in order to lower their contribu-
tion to the total output flicker noise. This is again possible as
long as their parasitic capacitances do not limit the bandwidth
at the input and output nodes of the buffer. It is also important
Fig. 12. Differential output of the buffer due to indirect CG noise.
to remember that, if same size or smaller FETs are used for the
tail current sources, they will contribute more flicker noise than
the CG transistors. The noise of the current source continuously
. The drain current of is a spike current, which is propor-
goes to the output, while the noise of the CG transistors only
tional to the derivative of the capacitor voltage. During the next
adds during the overlap time, which is a small fraction of the
half cycle, —which was previously charged to —connects
LO period.
to and starts discharging. At the same time, the already dis-
charged capacitor charges up to with the same process as
IV. WHITE NOISE
did during the previous half cycle. Therefore, for the ON
cycle of the LO, we can write So far, we have only talked about the flicker noise of the cur-
rent-commutating passive mixer. White noise is also present in
(15) all transistors and resistors making up the mixer. This noise
appears at the baseband output either directly or after folding
(16) due to frequency conversion. Using intuitive and simple circuits
and systems principles, the output noise originating in the white
where noise of the devices is analyzed. In this section, we start with
the white noise of the input transconductor; after which, we will
(17) explain the underlying physical mechanisms through which the
white noise of the switches and that of the buffer appear at the
is the on resistance of the switch, and is the total parasitic output.
capacitance at the mixer tail. Similar to the previous section,
A. Transconductance Noise
is the total transconductance of the CG tran-
sistor. As a result The white noise originating from the transconductor is indis-
tinguishable from the RF input signal. Therefore, as mixer com-
(18) mutation is assumed to be square wave like, the LO frequency
and its odd harmonics will downconvert the respective compo-
(19) nents of white noise to IF. This is the same process as in the ac-
tive mixer [18]. The single-sided power spectral density (PSD)
The differential output current consists of a train of noise spikes of the noise current at the mixer output is
which is equal to when the LO is high and when the
(21)
LO is low (Fig. 12). The average of these noise spikes will deter-
mine the flicker noise at baseband. Assuming that the tail band- where is the transconductance and is a noise factor which
width is large accounts for all the noise contribution of the transconductor.
This analysis ignores the filtering of the noise sidebands around
the harmonics of the LO by the tail capacitance of the mixer
and the series combination of the switch and the buffer input
resistance. The best case noise, when all the noise sidebands
(20) except the fundamental are filtered out, is only 0.9 dB lower
than what (21) predicts.
As we expected intuitively, minimizing the tail parasitic ca-
pacitance makes a more ideal signal current source at the input B. Switch Noise
of the mixer, which also results in smaller flicker-noise contri- The noise of a FET switch in deep triode is that of a resistor
bution by the CG transistors of the buffer. equal to the channel resistance of the switch. The single-sided
So far, we neglected the effect of the large parasitic capaci- PSD of this noise is
tance at the input of the buffer which lowers the impedance of
that node at higher frequencies. Since the input bandwidth of the (22)
buffer is larger than that of the downconverted channel, this par-
asitic capacitance is effectively an open circuit over the channel in which is the excess noise factor of the transistor and is
bandwidth, and as a result, it does not affect the analysis. As its on conductance in (6) and (7). The switches contribute noise
long as the tail capacitance charges and discharges completely, to the mixer output during the overlap time when they are both
the input capacitance of the buffer does not affect the dc transfer ON. If one switch is OFF, it obviously contributes no noise, and
function since the dc is the total transported charge over a half neither does the other switch that is ON because its current is
period which is constant. fixed by the RF input transconductance stage. For consistency,

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.
CHEHRAZI et al.: NOISE IN CURRENT-COMMUTATING PASSIVE FET MIXERS 339

Fig. 13. Switch direct white noise. (a) Single-balanced mixer (half mixer) high-
lighted for calculations. (b) Equivalent circuit for noise during the overlap time.

we call this the direct mechanism again. There is also another


mechanism related to the charge and discharge of the tail par-
asitic capacitance, which we call indirect mechanism. In this
section, we explain these two physical mechanisms and propose M
M
Fig. 14. White noise due to indirect switch noise. (a) Noise when is on.
simple models to analyze them. (b) Noise when is off. (c) Output noise.
1) Switch White Noise, Direct: Without loss of generality,
consider the highlighted half mixer in Fig. 13(a). 3 It is simpler LO frequency. The measured baseband PSD of this periodi-
to analyze the white noise of this switch pair first and then ex- cally time-varying white noise (cyclostationary) can be found
pand the results to the full mixer. Fig. 13(b) shows the equivalent by taking its average over a period [25]
circuit of the switching pair during the overlap time. The input
transconductor is considered an open circuit, and for simplicity,
the input resistance of the buffer is set to zero. FETs are replaced
with the time varying on the conductances of their channel (25)
and which were defined in (7) and (6). and are the noise
current sources associated with each switch.
Solving for the output differential noise current of the circuit where and are the edges of the overlap time in Fig. 11. After
in Fig. 13(b), , we have calculating the integral and some simplifications, the single-
sided baseband noise spectrum for the half mixer can be found
as
(23)
(26)
where is the effective dc bias voltage of the switch equal to
and . Substituting For a double-balanced mixer, the output noise power is twice
and from (22) and and from (7) and (6) and after some that of a single-balanced mixer since the noise sources are inde-
simplification pendent and uncorrelated.
2) Switch White Noise, Indirect: If the parasitic capacitor at
the tail of the mixer is considered, the switch noise appears at the
(24) output even outside the overlap time. Let us consider Fig. 14(a),
and for simplicity, let us assume that the LO is an ideal square
wave. When is on, it is represented by its on conductance
where is the LO voltage during the overlap. is only
. At low frequencies, the impedance of the tail parasitic ca-
present during the ON overlap which repeats with twice the
pacitance is larger than the on resistance of the switch ;
3This is also called a single-balanced mixer. therefore, the low-frequency noise sidebands circulate inside the

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.
340 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO. 2, FEBRUARY 2010

Fig. 16. Plot of the factor in parenthesis in (31).

Fig. 15. Block diagram of the system used for noise calculation due to the tail After doing the mathematical simplifications, the differential
cap. single-sided baseband noise PSD due to one switch is

switch itself and do not appear at the output. As the noise fre-
quency increases, short-circuits the tail to ground, and as (31)
a result, the high-frequency sidebands of go to the output
Fig. 14(a)]. Thus, during the on time of , the output noise
current is a high-pass version of its noise current . The factor inside the parenthesis is the transfer function of
As the noise current of goes to the output over its on the high-pass filter, evaluated at LO frequency and its odd
time, it creates a voltage across the tail capacitor which is harmonics. is an indication of the ratio between the
proportional to the integral of . When turns off, turns LO frequency and the corner frequency of this filter. Fig. 16
on instantly and shunts this tail capacitor to ground with its on shows the factor for different values of this ratio. If the LO
conductance . From this moment up to the end of the second frequency is much larger than the tail bandwidth, all the LO
LO half cycle, the noise charge of discharges through harmonics will lie inside the passband of the high-pass filter,
and creates a noise current in the other output. and therefore, the factor inside the parenthesis will approach
Because the switches have the same on conductance, it is ev- . If the LO frequency is smaller than the tail bandwidth,
ident from Fig. 14(a) and (b) that the topology of the circuit the factor is smaller than . Since, in a well-designed
in both LO half cycles remains the same; it consists of a se- zero-IF mixer, the tail bandwidth has to be larger than the
ries combination of a grounded capacitor and a grounded re- input RF frequency—which is equal to , the factor inside
sistor. The current source , however, is present in one LO half the parenthesis is always smaller than . As a result, the
cycle—during which, it is high-pass filtered and goes to the pos- white-noise contribution of the switch is
itive differential output—while it is absent in the other half cycle (32)
and the tail parasitic capacitor discharges in the other differen-
tial output. Therefore, for the purpose of this analysis, the half Since the noise sources of the four transistors are uncorrelated,
mixer can be modeled with the system in Fig. 15. The input cur- for a double-balanced mixer
rent is first multiplied by the gating function —which is
a replica of the LO—to model the ON and OFF states of . (33)
This gated noise passes through a high-pass filter
where is the on conductance of the MOS switch when the LO
is high
(27)
(34)

which represents the series switch on resistance and the tail ca-
pacitance. Finally, and are recovered by multiplying the C. Buffer Noise
filtered noise by and . is the delayed version of The white noise of the CG transistors, bias current sources,
by . and the load circuit in the transresistance buffer add to the total
Therefore noise of the mixer. Similar to the flicker noise, the white noise
of the load and that of the tail current sources directly appear
at the output but the white noise of the CG transistors depends
(28) on the LO waveform and the tail parasitic capacitance. Here, we
use methods similar to those we used to analyze the flicker noise
of the CG transistors.
(29) 1) Buffer White Noise, Direct: Similar to the case of flicker
noise, the white noise of the CG FETs of the buffer can go
(30) to the output during the overlap time. In order to make the

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.
CHEHRAZI et al.: NOISE IN CURRENT-COMMUTATING PASSIVE FET MIXERS 341

TABLE I
SUMMARY OF THE SIMULATION PARAMETERS

Fig. 17. White-noise sidebands that are around kf remain at kf at output.

analysis simpler, the broadband white noise of the FET can


be broken into several uncorrelated sidebands around dc and
LO harmonics as well as everywhere else. The sum of these
new narrowband sidebands gives the original broadband white
noise. Now, the output noise due to each sideband can be found.
Fig. 18. Leakage gain of the transconductor flicker noise to the output current.
A portion of the white noise which is close to dc goes to the
output with the same transfer function as that of the flicker
noise as in (13). At first glance, those sidebands that are lo-
cated at seem to fold to baseband; however, they will be
shorted to ground by the large parasitic capacitance at the input
of the buffer. These sidebands appear at the output at their orig-
inal frequency without downconversion and do not de-
grade the quality of the desired channel which is at baseband
before reaching the buffer. Therefore, if the switch conductance
is larger than the transconductance of the CG transistors, using
(13), the white-noise contribution of the two CG FETs to the
baseband output is

(35)

2) Buffer White Noise, Indirect: Similar to the case of the di-


Fig. 19. Transfer function of the switch flicker noise at the output of (a) an
rect white noise, the high-frequency sidebands of the CG white active mixer and (b) a passive mixer in the presence of a blocker at f=f +
noise appear at the output without frequency conversion but f .
the low-frequency components go to the output with the same
transfer function as the flicker noise in (20). Fig. 17 shows this
process graphically. Therefore, the output noise of the differen- Fig. 18 shows the leakage gain of the transconductor flicker
tial buffer due to the indirect mechanism of the CG white noise noise to the output current when the switching pair offset
is voltage changes. The simulated gain was found using a
SPECTRE-RF Periodic AC Analysis (PAC) simulation. Very
(36) close agreement is seen between the simulation and theory
plotted using (1). This is just to validate the theory, but in the
proposed current-driven passive mixer [Fig. 1(a)], the input
coupling capacitor completely removes the transconductor
V. VERIFYING THEORY BY SIMULATION flicker noise.
In order to validate this theory, a current-driven passive mixer Fig. 19(a) shows the transfer function of the switch flicker
was simulated using SPECTRE-RF. A generic BSIM was used noise at the output of an active mixer when a blocker with
for the FETs since its noise parameters could be set more conve- is present at the input of the mixer transconductor.
niently. A square-wave LO was used for both hand calculations The output flicker noise increases with blocker amplitude and
and simulations. Other parameters are shown in Table I unless can reach the noise transfer function due to the commutated
otherwise indicated for simulations. tail current (dotted line in the figure). Fig. 19(b) shows the re-

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.
342 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO. 2, FEBRUARY 2010

Fig. 21. (a) Direct switch white noise in passive mixer. (b) Indirect switch white
noise in passive mixer.

Fig. 21 shows the direct and indirect mechanisms of


the output white noise originating in the mixer switches.
SPECTRE-RF PNoise simulation was used. For accurate sim-
ulation results, enough number of harmonics should be used
in PNoise to account for white-noise folding due to the mixer
switching [26].
Simulations show that buffer white-noise transfer functions
are like that shown in Fig. 20.

Fig. 20. (a) Direct flicker-noise transfer function of the CG when VI. TOTAL MIXER NOISE AND DISCUSSION
g = 18 1: mS and switch size changes. (b) Direct flicker-noise transfer The expression for the low-frequency noise at the differential
function of the CG transistors in the buffer when CG transconductance g ( ) output of the mixer, which is dominated by the flicker noise, can
changes. (c) Indirect flicker-noise transfer function of the CG transistors when
LO frequency changes. be found as

sult of the same simulation for a current-driven passive mixer.


-
SPECTRE-RF PAC was also used for these simulations. (37)
Fig. 20(a) shows the direct transfer function from the gate
of the CG transistors in the buffer to the output current when where is the input transconductance, is the amplitude of
the of the CG FET is fixed to 18.1 mS and the switch size the special blocker, is the switch flicker noise, and
changes. As it was shown in Fig. 9, when the switch conduc- are the LO slope and period, is the flicker noise of the
tance is larger than , most of the CG flicker noise goes to buffer bias circuit, is the flicker noise of the buffer load
the output but, when it is smaller, it only circulates inside the circuit, is the buffer CG transconductance, - is the
FET itself. Fig. 20(b) shows the same transfer function when the switch dc effective voltage, is the parasitic capacitance at the
switch size is fixed but varies. Although the factor enclosed tail of the differential pair, and is the flicker noise of the
in square brackets in (13) decreases as increases, since the CG FET.
noise current itself grows faster (the transconductance factor In case of narrowband receivers, where there is no high-fre-
outside the bracket), the overall transfer function increases with quency blocker, or if the level of blockers in a wideband re-
; hence, Figs. 9 and 20(b) are in agreement. The parasitic ca- ceiver are small, the transresistance buffer following the mixer
pacitance at the input of the buffer is not accounted for in this dominates the flicker noise. If FETs are used in the bias cir-
simulation. If that capacitance is present, the simulated transfer cuit of the buffer or in its load, their flicker noise dominates the
function is larger by a few decibels. Fig. 20(c) shows the indi- low-frequency noise of the buffer. Therefore, large FETs should
rect transfer function. The simulated tail parasitic capacitance be used in order to lower the flicker noise at the output. Large
was 350 fF in this simulation, which includes the switch para- FETs are accompanied by large parasitic capacitances. The fil-
sitics, output capacitance of the transconductor, the bottom plate tering effect of these capacitances is not of concern since the
parasitic of the coupling capacitor, and routing parasitics. desired channel is already downconverted to baseband before

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.
CHEHRAZI et al.: NOISE IN CURRENT-COMMUTATING PASSIVE FET MIXERS 343

reaching the buffer. If voltage headroom is not a limitation, re- the white noise of the buffer is mostly due to its bias and load
sistors can be used both for biasing the buffer and in its load to circuits. Therefore, a low-noise buffer and transconductor are
compellably remove this source of low-frequency noise. When essential.
the receiver is broadband, switch flicker noise in the presence of Although an LO voltage with fast rise and fall times will re-
large blockers can contribute significant low-frequency noise to duce the noise mechanisms due to the overlap time, those mech-
the mixer output. This is also possible in an active mixer. Using anisms do not determine the noise of the circuit; therefore, it
a sharp LO voltage will minimize the flicker noise due to the may not benefit the noise performance. However, fast LO tran-
switch; however, power consumption in the LO buffer can be- sitions are needed for high linearity. Inverter-based LO drivers
come a limitation. Therefore, careful exploration of blockers is can be used to sharpen the LO transition. Moreover, the parasitic
necessary in order to guarantee optimum noise performance. If capacitance at the input of the switches should be minimized to
the flicker noise due to the blocker and the bias and load circuits lower the noise due to the second mechanism.
of the buffer are minimized, the flicker noise of the CG FET it-
self will determine the low-frequency noise. That can also be ACKNOWLEDGMENT
minimized by using a sharp LO and by minimizing the tail par-
The authors would like to thank A. Parsa and S. Sarhangian
asitic capacitance of the mixer.
of the Integrated Circuits and Systems Laboratory, University
At higher frequencies, where flicker noise is not dominant,
of California at Los Angeles (UCLA), for the very useful dis-
the differential output noise is
cussions and S. Moloudi, D. Murphy, M. Youssef, and A. Parsa
for reading this paper’s manuscript and for their valuable sug-
gestions.
-
- REFERENCES
[1] A. A. Abidi, “Direct-conversion radio transceivers for digital commu-
-
(38) nications,” IEEE J. Solid-State Circuits, vol. 30, no. 12, pp. 1399–1410,
Dec. 1995.
[2] R. Bagheri, A. Mirzaei, S. Chehrazi, M. Heidari, M. Lee, M. Mikhemar,
Depending on the design, the noise of the buffer—mostly from W. Tang, and A. Abidi, “An 800 MHz to 5 GHz software-defined radio
receiver in 90 nm CMOS,” in Proc. Int. Solid-State Circuits Conf., Feb.
its bias and load circuits—or that of the input transconductor 2006, pp. 1932–1941.
will dominate the output white noise. If the buffer is biased by [3] S. Chehrazi, R. Bagheri, and A. A. Abidi, “Noise in passive FET
resistors, they should be large; however, the large voltage drop mixers: A simple physical model,” in Proc. IEEE Custom Integr.
Circuits Conf., 2004, pp. 375–378.
across those resistors will limit the amount of voltage headroom [4] W. Redman-White, D. M. Leenaerts, and W. Leenaerts, “1=f noise
for signal swing. If biased by FETs, their current noise should in passive CMOS mixers for low and zero IF integrated receivers,” in
be minimized. This can be done by minimizing the transconduc- Proc. Eur. Solid-State Circuits Conf., Sep. 18–20, 2001, pp. 41–44.
[5] E. Sacchi, I. Bietti, S. Erba, L. Tee, P. Vilmercati, and R. Castello, “A
tance of those bias and load FETs. 15 mW, 70 kHz 1=f corner direct conversion CMOS receiver,” in Proc.
IEEE Custom Integr. Circuits Conf., Sep. 21–24, 2003, pp. 459–462.
[6] S. Zhou and M. C. F. Chang, “A CMOS passive mixer with low flicker
VII. CONCLUSION noise for low-power direct-conversion receiver,” IEEE J. Solid-State
Circuits, vol. 40, no. 5, pp. 1084–1093, May 2005.
This paper has presented simple but accurate models to pre- [7] M. A. T. Sanduleanu, M. Vidojkovic, V. Vidojkovic, A. H. M. van
dict flicker and white noise in current-driven passive mixers. Roermund, and A. Tasic, “Receiver front-end circuits for future gen-
erations of wireless communications,” IEEE Trans. Circuits Syst. II,
Flicker noise is modeled by gate-referred slowly varying voltage Exp. Briefs, vol. 55, no. 4, pp. 299–303, Apr. 2008.
sources for all FETs. Our analysis shows that, in contrast to the [8] S. Ayazian and R. Gharpurey, “Feedforward interference cancellation
general belief that passive mixers are flicker noise free, switch in radio receiver front-ends,” IEEE Trans. Circuits Syst. II, Exp. Briefs,
vol. 54, no. 10, pp. 902–906, Oct. 2007.
noise can appear at the output if there are blockers present at spe- [9] O. Boric-Lubecke, J. Lin, A. Verma, I. Lo, and V. M. Lubecke, “Multi-
cial frequencies. This is also true for active mixers. This flicker band 0.25-m CMOS base station chips for indirect and direct conver-
noise is proportional to the blocker amplitude and can be compa- sion receivers,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 55, no.
7, pp. 2106–2115, Aug. 2008.
rable to noise due to other flicker mechanisms in active mixers. [10] R. Bagheri, A. Mirzaei, S. Chehrazi, M. E. Heidari, M. Lee, M.
While this mechanism is important in broadband receivers, it Mikhemar, W. Tang, and A. A. Abidi, “An 800-MHz–6-GHz soft-
might not be an issue in narrowband receivers where the faraway ware-defined wireless receiver in 90-nm CMOS,” IEEE J. Solid-State
Circuits, vol. 41, no. 12, pp. 2860–2876, Dec. 2006.
blockers are filtered out before reaching the mixer. Analysis and [11] N. Poobuapheun, W.-H. Chen, Z. Boos, and A. M. Niknejad, “A
simulations show that, in case of a broadband receiver, either the 1.5-V 0.7–2.5-GHz CMOS quadrature demodulator for multiband
buffer flicker noise or that due to the switch noise in the presence direct-conversion receivers,” IEEE J. Solid-State Circuits, vol. 42, no.
8, pp. 1669–1677, Aug. 2007.
of blockers dominates. Therefore, large FETs should be used in [12] H. Darabi, “A blocker filtering technique for SAW-less wireless re-
the buffer, and blockers should be considered in order to have ceivers,” IEEE J. Solid-State Circuits, vol. 42, no. 12, pp. 2766–2773,
minimum low-frequency noise. Dec. 2007.
[13] B. Tenbroek, J. Strange, D. Nalbantis, C. Jones, P. Fowers, S. Brett, C.
The output white noise of the mixer was also analyzed using Beghein, and F. Beffa, “Single-chip tri-band WCDMA/HSDPA trans-
similar methods. The white noise of the FETs appear at the ceiver without external SAW filters and with integrated TX power con-
output of the mixer either directly or after frequency conversion. trol,” in Proc. Int. Solid-State Circuits Conf., Feb. 2008, pp. 202–203.
[14] A. Mirzaei and H. Darabi, “A low-power WCDMA transmitter with
Analysis shows that the white noise of the input transconductor an integrated notch filter,” in Proc. Int. Solid-State Circuits Conf., Feb.
or that of the buffer will dominate. Similar to its flicker noise, 2008, pp. 212–213.

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.
344 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO. 2, FEBRUARY 2010

[15] A. Mazzanti, M. Sosio, M. Repossi, and F. Svelto, “A 24 GHz sub- Ahmad Mirzaei (S’02–M’06) received the B.Sc.
harmonic receiver front-end with integrated multi-phase LO generation and M.Sc. degrees in electrical engineering from
in 65 nm CMOS,” in Proc. Int. Solid-State Circuits Conf., Feb. 2008, the Sharif University of Technology, Tehran, Iran, in
pp. 216–217. 2000 and 2002, respectively, and the Ph.D. degree
[16] L. Nathawad, M. Zargari, H. Samavati, S. Mehta, A. Kheirkhahi, P. in electrical engineering from the University of
Chen, K. Gong, B. Vakili-Amini, J. Hwang, M. Chen, M. Terrovitis, B. California, Los Angeles, in 2006.
Kaczynski, S. Limotyrakis, M. Mack, H. Gan, M. Lee, S. Abdollahi- He is currently a Sr. Staff Scientist with Broadcom
Alibeik, B. Baytekin, K. Onodera, S. Mendis, A. Chang, S. Jen, D. Corporation, Irvine, CA. His interests include analog
Su, and B. Wooley, “A dual-band CMOS MIMO radio SoC for IEEE and RF IC design for wireless communications.
802.11n wireless LAN,” in Proc. Int. Solid-State Circuits Conf., Feb.
2008, pp. 358–359.
[17] M. Camus, B. Butaye, L. Garcia, M. Sie, B. Pellat, and T. Parra, “A 5.4
mW 0.07 mm 2.4 GHz front-end receiver in 90 nm CMOS for IEEE
802.15.4 WPAN,” in Proc. Int. Solid-State Circuits Conf., Feb. 2008,
pp. 368–369. Asad A. Abidi (S’75–M’80–SM’95–F’96) received
[18] H. Darabi and A. A. Abidi, “Noise in RF-CMOS mixers: A simple the B.Sc. degree (with honors) from Imperial Col-
physical model,” IEEE J. Solid-State Circuits, vol. 35, no. 1, pp. 15–25, lege, London, U.K., in 1976, and the M.S. and Ph.D.
Jan. 2000. degrees in electrical engineering from the University
[19] D. Manstretta, R. Castello, and F. Svelto, “Low 1=f noise CMOS ac- of California, Berkeley, in 1978 and 1981, respec-
tive mixers for direct conversion,” IEEE Trans. Circuits Syst. II, Analog tively.
Digit. Signal Process., vol. 48, no. 9, pp. 846–850, Sep. 2001. From 1981 to 1984, he was with Bell Laboratories,
[20] H. Darabi and J. Chiu, “A noise cancellation technique in active Murray Hill, NJ, as a Member of Technical Staff with
RF-CMOS mixers,” in Proc. Int. Solid-State Circuits Conf., Feb. 6–10, the Advanced LSI Development Laboratory. Since
2005, pp. 544–616. 1985, he has been with the Electrical Engineering
[21] J. Pihl, K. T. Christensen, and E. Bruun, “Direct downconversion with Department, University of California, Los Angeles,
switching CMOS mixer,” in Proc. Int. Symp. Circuits Syst., Sydney, where he is currently a Professor. He was a Visiting Faculty Researcher with
Australia, May 6–9, 2001, vol. 1, pp. 117–120. Hewlett-Packard Laboratories in 1989. His research interests include RF
[22] D. Ham and A. Hajimiri, “Complete noise analysis for CMOS CMOS design, high-speed analog integrated-circuit design, data conversion,
switching mixers via stochastic differential equations,” in Proc. IEEE and other techniques of analog signal processing.
Custom Integr. Circuits Conf., Orlando, FL, May 21–24, 2000, pp. Dr. Abidi was the Program Secretary for the IEEE International Solid-State
439–442. Circuits Conference (ISSCC) from 1984 to 1990 and the General Chairman of
[23] J. Chang, A. A. Abidi, and C. R. Viswanathan, “Flicker noise in CMOS the Symposium on VLSI Circuits in 1992. He was the Secretary of the IEEE
transistors from subthreshold to strong inversion at various tempera- Solid-State Circuits Council from 1990 to 1991. From 1992 to 1995, he was the
tures,” IEEE Trans. Electron Devices, vol. 41, no. 11, pp. 1965–1971, Editor for the IEEE JOURNAL OF SOLID-STATE CIRCUITS. He received an IEEE
Nov. 1994. Millennium Medal, the 1988 TRW Award for Innovative Teaching, and the 1997
[24] P. R. Gray, P. J. Hurst, S. H. Lewis, and R. G. Meyer, Analysis and IEEE Donald G. Fink Award, the 2007 Lockheed-Martin Award for Excellence
Design of Analog Integrated Circuits, 4th ed. New York: Wiley, 2001. in Teaching, and the 2008 IEEE Solid-State Circuit Society Donald O. Pederson
[25] W. A. Gardner, Introduction to Random Processes With Applications Award. He was a corecipient of the Best Paper Award at the 1995 European
to Signals and Systems, 1st ed. Berlin, Germany: Springer-Verlag, Solid-State Circuits Conference, the Jack Kilby Best Student Paper Award at the
1989. 1996 ISSCC, and the Jack Raper Award for Outstanding Technology Directions
[26] “Simulating switched-capacitor filters with SpectreRF,” Ken Kundert, Paper at the 1997 ISSCC. He was named one of the top ten contributors to the
The Designer’s Guide Community, 2006. [Online]. Available: http:// ISSCC, and is a member of the National Academy of Engineering, as well as of
www.designers-guide.org the Third World Academy of Sciences.

Saeed Chehrazi (S’98–M’04) received the B.Sc. de-


gree in electrical engineering from Sharif University
of Technology, Tehran, Iran, in 2001, and the M.Sc.
and Ph.D. degrees from the University of California
at Los Angeles, in 2004 and 2008, respectively.
He is currently a Staff RFIC Design Engineer with
Marvell Semiconductor Inc., Santa Clara, CA. His re-
search interests are design and modeling of analog
and RF circuits.

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:423 UTC from IE Xplore. Restricon aply.

You might also like