IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 51, NO.

9, SEPTEMBER 2004 1823

Complex Signal Processing is Not Complex

Kenneth W. Martin, Fellow, IEEE

Abstract—Wireless systems often make use of the quadrature re-

lationship between pairs of signals to effectively cancel out-of-band
and interfering in-band signal components. The understanding of
these systems is often simplified by considering both the signals
and system transfer functions as “complex” quantities. The com-
plex approach is especially useful in highly integrated multistan-
dard receivers where the use of narrow-band fixed-coefficient fil-
ters at the RF and high IF must be minimized. This paper first
presents a tutorial review of complex signal processing for wire-
less applications. The review emphasizes a graphical and picto-
rial description rather than an equation-based approach. Next, a
number of classical modulation architectures are described using
this formulation. Finally, more recent developments such as com- Fig. 1. SFG of an quadrature mixer using (a) a real SFG and (b) a complex
plex filters, image-reject mixers, low-IF receivers, and oversam- SFG.
pling analog–digital converters are discussed.
Index Terms—Communication systems, complex signal pro- As an example, consider the popular “quadrature” mixer with
cessing, filters, frequency modulation, mixers, wireless receivers. a real input as shown in Fig. 1. In the complex SFG, two real
multiplications have been replaced by a single complex multi-
plication. Furthermore, since in the time domain, we have
I. INTRODUCTION

T HIS paper is intended to be used as a tutorial introduction

to complex signal processing, and also as an illustration of
its wide use in a number of different applications, both analog
(1)

and digital, in recent communication systems. then, taking the Fourier transform, we have
A complex signal consists of a pair of real signals at an in- (2)
stant in time. If one denotes the complex signal , as
where ; then, a Hilbert space can be de- and we see that an ideal quadrature mixer results in a simple
fined using appropriate definitions for addition, multiplication, frequency shift of the input signal with no images occurring, a
and an inner product and norm. In an actual physical system, conclusion that is obvious from Fig. 1(b), but is certainly not
the signals are both real (but are called the real and imaginary obvious from Fig. 1(a). If this mixer is followed by a low-pass
parts) and are found in two distinct signal paths. The multiplier filter in each path, then, it can be used to directly demodulate
“ “ is used to help define operations between different complex carrier-frequency signals without images as is discussed in Sec-
signals, and is to some degree imaginary; i.e., it does not actu- tion III-A. For this reason, it often called an image-reject mixer
ally exist in actual systems. Often, the dependence of signals on although if one is being accurate, the term should denote the
time is not shown explicitly. combination of the quadrature mixer and low-pass filters.
The use of complex signal processing to describe wireless sys- The use of complex signal processing for wireless appli-
tems is increasingly important and ubiquitous for the following cations has blossomed recently. This is especially true for
reasons: it often allows for image-reject architectures to be de- high-bit-rate standards, such as local-area networks (LANs),
scribed more compactly and simply; it leads to a graphical or and for multistandard transceivers. Evidence of this prolifera-
signal-flow graph (SFG) description of signal-processing sys- tion is that in 2001, 2002, and 2003, the majority of the papers
tems providing insight, and it often leads to the development of in IEEE International Solid-State Circuits Conference sessions
new systems where the use of high-frequency highly selective on wireless transceivers for LAN applications employed com-
image-reject filters is minimized. The result is more highly in- plex signal processing.
tegrated systems using less power and requiring less physical Much, but not all of this paper deals with the already men-
space. tioned tutorial review of complex signal processing, especially
as it is applicable to wireless systems. A great deal of the tutorial
material on complex analog signal processing is directly based
Manuscript received November 25, 2003; revised March 2, 2004. This on the research of Snelgrove, as described in [1], [6], [7], and in
work was supported in part by the Natural Sciences and Engineering Research
Council (NSERC) and in part by MICRONET, Canada. This paper was some theses by his graduate students [9], [11]. Unfortunately, a
recommended by Associate Editor U.-K. Moon. journal submission describing this early work was rejected, but
The author is with the Department of Electrical and Computer Engi- a reprint of the originally submitted paper can be obtained from
neering, University of Toronto, Toronto, ON M5S 3G4, Canada (e-mail:
[email protected]). [8]. Some of the early published work on complex digital filters
Digital Object Identifier 10.1109/TCSI.2004.834522 is described in [2] and [5].
1057-7122/04$20.00 © 2004 IEEE
1824 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 51, NO. 9, SEPTEMBER 2004

In the course of the review of the fundamentals of com-

plex signal processing, some important identities are given.
Examples are given using complex SFG representations. In
Section III, a number of classical wireless modulations systems
are described using the complex formulation. These systems
include Weaver and Hartley modulators and polyphase filters. Fig. 2. (a) Real SFG for adding two complex signals. (b) Equivalent complex
SFG.
Errors due to nonideal coefficients in mixers are quantified. In
Section IV, complex analog and digital filters are discussed.
Section III describes more-recent complex signal-processing
applications. These include, for example, complex adaptive
image-reject mixers, and complex low-IF receiver architectures
including a new modulation approach that doubles the data rate
for a given analog–digital (A/D) sampling frequency, and com-
plex bandpass oversampling A/D converters. These systems are
Fig. 3. (a) Real SFG of a complex multiplication. (b) Equivalent complex SFG.
especially useful in multistandard wireless applications as they
are inherently wide-band (without excessive imaging errors)
and allow the systems to be adapted to individual standards
using digital programming after the A/D converters.

II. COMPLEX SIGNAL PROCESSING

Fig. 4. Multiplying a signal by the imaginary constant j .
Complex signal processing systems operate on ordered pairs
of real signals. A possible definition for a complex inner product
is

(3)

Combining this definition with the typical definitions for com-

plex inversion, addition, and multiplication, and with the proper-
ties of commutation and association, constitutes a Hilbert space.
It is assumed that all signals are periodic and frequency limited,
and therefore, can be expressed as a finite series of “complex- Fig. 5. (a) Digital delay operator. (b) Analog integration operator for complex
oids”1 signals.

(4) Slightly more complicated is a complex multiplication; con-

sider multiplying a complexoid by a complex co-
efficient . We have
where is, in general, complex [5]. In this paper, we will limit
our consideration normally to a single (or a few) complexoids
in order to simplify explanations. (5)

The real SFG of this operation is shown in Fig. 3(a). Once again
A. Complex Operations
the complex SFG of a multiplication operation, as shown in
Most complex systems are realized using only four basic op- Fig. 3(b), is considerably simpler. Note that the real SFG does
erations: complex addition, complex multiplication, and com- not include the constant anywhere, whereas, in the complex
plex integration for continuous-time filters or complex delay for SFG, it is admissible.
discrete time filters. In addition, the operation of complex con- A special case is the multiplication of an input signal by the
jugation is required to describe imaging errors due to mismatch constant . This is shown using Fig. 4(a) and a real SFG in
and quadrature imbalance in real physical systems. Fig. 4(b). This operation is simply an interchanging of the real
A complex addition of two complex signals simply adds the and inverted imaginary parts; this operation is the basis of real-
two real parts and the two imaginary parts independently. Con- izing imaginary components in complex networks; that is, imag-
sider the real SFG that represents adding a complex signal to inary gains are realized by cross coupling in the two real net-
a second complex signal, as is shown in Fig. 2(a). Its complex works.
SFG is shown in Fig. 2(b). Two additional operations are required for realizing complex
filters, namely, the delay operator for digital filters and the inte-
1The author has taken the liberty of coining the word “complexoid” to rep-
gration operator for analog filters. These operations simply con-
resent the complex single-frequency signal component e = cos(!t) + sist of independently applying the corresponding real operation
j sin(!t) to each part of the complex signal, as shown in Fig. 5.
MARTIN: COMPLEX SIGNAL PROCESSING IS NOT COMPLEX 1825

and

(13)

Fig. 6. Complex conjugation operation. This important duality, along with some simple trigometric
identities, is useful in quantifying many of the imaging nonide-
alities of communication systems, as we shall see later.
Another operation, which is important for quantifying
imaging caused by component mismatch errors, is complex
conjugation. Consider C. Hilbert Transforms and Positive-Pass Filters
An important example of a complex signal processing system
(6) is that of a positive-pass filter (PPF) (or alternatively that of a
negative-pass filter or NPF). This is a filter that can take a pos-
with real. We have sibly complex but quadrature-imbalanced input signal2 and pro-
duce a necessarily complex quadrature-balanced output signal
(7) that has positive frequency components only. A signal with pos-
itive-frequency components only is called an analytic signal. It
and we see that besides being phase rotated, has been trans- has the property that, for every real spectral component, there
formed from a positive-frequency complexoid into a negative- is a corresponding imaginary spectral component that is equal
frequency complexoid. Conjugation also changes negative-fre- in magnitude but delayed in time by 90 [1]. A complex SFG
quencies complexoids to positive frequencies. The conjugation of a system that produces analytic signals from possibly com-
operation is shown in the real SFG of Fig. 6(a). Alternative com- plex signals is shown in Fig. 7(a), and the equivalent real SFG
plex SFGs are shown in Fig. 6(b) and (c). is shown in Fig. 7(b). As shown, Hilbert transform filters are
required to produce the signals delayed in time by 90 . For neg-
B. Rectangular to Polar Duality ative frequencies, which have phases continually changing to
Consider two parallel real systems having components at a more negative values, this means the outputs are phase shifted
single frequency in each path, where the signals do not have by , as opposed to for positive frequencies, which
quadrature balance; that is, the amplitudes do not match, and have phases that increase in time.3 Consider the system in Fig. 7
the phases are not different by exactly . In this general for a positive frequency input of . We have
case, assuming the signals in the second path are always to be
multiplied by , we can express the signal in rectangular (14)
form as
and the input signal is unchanged. However, for a negative fre-
(8) quency input of , we have
If the signals had quadrature balance, then, we would have
and ; in the quadrature-imbalanced case, this is not (15)
true. Using simple trigometric identities, it is straightforward to
show can equivalently be expressed in polar form as The PPF of Fig. 7, which is based on using the Hilbert trans-
form, can be considered a particular case of a complex filter. The
(9) concept of a PPF (or an NPF) can be generalized to that of com-
plex filters which have passbands at positive frequencies only
where and are possibly complex coefficients and given by (or negative frequencies only). Complex filters are discussed in
Section IV.
(10)
III. COMPLEX MODULATION AND CLASSICAL
and WIRELESS SYSTEMS
We have seen in the Introduction that a real input signal can be
(11) frequency shifted without imaging errors, using two real multi-
pliers as shown in Fig. 1. This operation ideally shifts both pos-
Thus, quadrature-imbalanced signals, at a single frequency, can itive and negative frequency components in the same direction
be expressed as two complexoids, one at positive frequencies, by the same amount. This operation, when combined with an ap-
and one at negative frequencies. propriate low-pass filter, can be used to frequency demodulate
Alternatively, given a positive and negative complexoid at a an RF input to a complex IF frequency or directly to baseband
single frequency expressed as in (9), one can equivalently rep- [12]–[14].
resent them in the rectangular form of (8) where 2Most texts [5] only consider the special case of real input signals.
3This is inherent in the conjugate-symmetry property of real transfer func-
(12) tions.
1826 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 51, NO. 9, SEPTEMBER 2004

Fig. 7. One possibility for realizing a PPF: (a) a complex SFG, and (b) an
equivalent real SFG.

Fig. 9. Image-reject up converter. (a) Complex SFG. (b) Example spectra. (c)
Equivalent real SFG.

Fig. 8. Image-reject down converter. (a) Complex SFG. (b) Example spectra.
(c) Equivalent real SFG.

A. Direct-Conversion Receiver
The direct-conversion receiver is a good choice of architec-
ture for less demanding but highly integrated applications [43],
[44]. A simplified receiver architecture, using both a complex
SFG (and a real SFG), is shown in Fig. 8. Example spectra of
various locations of Fig. 8(a) are shown in Fig. 8(b).

B. Direct-Conversion Transmitter
In a direct-conversion transmitter, a complex input signal is
directly modulated to RF using a complex mixer, and then con- Fig. 10. Weaver method for SSB generation [15].
verted to a real RF signal by simply ignoring the imaginary
output of the complex mixer; this process allows us to sim-
up so it is centered at . The lower sideband is then extracted
plify the complex mixer to two real multipliers, as is shown in
using two real low-pass filters. The complex signal is then fur-
Fig. 9(c).
ther frequency shifted to the desired RF frequency, in this case
C. Single-Sideband (SSB) Generation , and the imaginary part is discarded; the result is the final
SSB signal at RF. The spectra of various locations of the com-
One of the first uses of complex processing in wireless sys- plex SFG of Fig. 10(b) are shown in Fig. 10(c).
tems was for SSB generation. For example, consider the Weaver An alternative architecture for SSB generation, called
architecture, as is shown in Fig. 10.4 In this approach, the base- Weaver’s second method, replaces the first complex mixer and
band signal, which is symmetric about dc, is frequency shifted the low-pass filters with a complex PPF (or NPF), in order to
4This is a good example of how complex SFGs can be used to describe sys- extract the desired sideband directly before frequency shifting
tems without using equations. to the desired RF frequency.
MARTIN: COMPLEX SIGNAL PROCESSING IS NOT COMPLEX 1827

Fig. 12. (a) PPF used in the Hartley SSB receiver. (b) Polyphase approach for
realizing a PPF [17].
Fig. 11. Demodulating a SSB signal using the Hartley architecture [16].

D. SSB Receivers
An architecture closely related to SSB transmitters is an alter-
native for an SSB receiver based on using a PPF after a complex
demodulation from RF. The PPF is used to attenuate the unde-
sired sideband, as is shown in the real SFG of Fig. 11(a) and the
equivalent complex SFG of Fig. 11(b).5 From the plots of the
spectra at different locations of the complex SFG, it is seen that
the first complex mix frequency shifts the spectra so that the only
significant components at positive frequencies are in the de-
sired sideband. The sideband is extracted using a PPF. The final
real demodulated output signal is obtained by simply taking the
real output of the PPF and discarding its imaginary output. This
approach was originally proposed by Hartley [16], where the
PPF was implemented as two low-pass filters cascaded with
simple RC phase-shift networks, as is shown in Fig. 12(a). An
alternative means of realizing a wide-band passive PPF, that is Fig. 13. Imaging due to a nonideal complexoid.
less sensitive to component tolerances and is currently popular,
uses the polyphase network shown in Fig. 12(b) [17]–[19]. The
quadrature imbalance, and is discussed in Section IV.C. An
polyphase network is well analyzed in [19], where it is shown
in-depth and relatively complete analysis of imaging errors in
that it is an example of a complex filter obtained from frequency
complex mixers and quadrature samplers can be found in [20],
shifting a high-pass filter, as is discussed in Section IV.
which deals primarily with demodulating analog signals from
E. Imaging Errors in Complex Mixers a low-IF frequency to baseband. In this section, we discuss an
example of demodulating a real carrier-frequency signal (or RF
Perhaps the major motivation for using image-reject or signal) to a low-IF frequency.
quadrature mixers is that they allow for frequency translation Consider the case of demodulating an SSB input signal,
while minimizing the necessity for highly selective and sen- where the lower sideband is desired, for example. This case
sitive image-reject filters at RF and IF frequencies. This is is similar to that in Fig. 11, but now the ideal demodulating
only true to the extent that quadrature mixers operate ideally. complexoid, , is at a positive frequency. Assume that it is
For example, if the complexoid that is modulating the input not ideal; we will show that this case results in there being a
signal is quadrature imbalanced and has negative frequency small modulating complexoid at a negative frequency as well,
components, then interferers can be imaged to fall in-band. as shown in Fig. 13(a). The assumed input signal is shown in
The case of imaging errors due to quadrature imbalance in Fig. 13(b). The desired frequency shift takes the the input signal
the channel is equivalent to a complex gain constant having and modulates it to more positive frequencies so that the lower
5This is a good example of using complex filtering to eliminate interfering sideband is just above dc, as is shown in Fig. 13(c). However,
images. the smaller negative-frequency modulating complexoid in
1828 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 51, NO. 9, SEPTEMBER 2004

addition, shifts the input signal to more negative frequencies so

that the upper sideband, although attenuated, is sitting directly
at the same frequency band as the desired lower sideband, as is
shown in Fig. 13(d). The subsequent filtering by the PPF will
output both the desired lower sideband and the interfering upper
sideband image. For this reason, it is important to quantify the
magnitude of the imaging complexoid (in this case at a negative
frequency) as a function of the magnitude and phase errors of
the modulating waveform.
Consider an ideal quadrature mixer having a multiplying
complexoid given (using rectangular coordinates) as

(16)
Fig. 14. Realizing a complex filter H (s) = R(s) + jQ(s) by real SFG
If we assume there are phase and magnitude errors, we can ex- Rs Qs
transfer filters ( ) and ( ).

press the modulating waveform in a simplified form of (8) as

Complex analog transfer functions can be realized by either
(17) frequency-translating real filters [1], [2], [10], or by employing
a direct method for deriving transfer complex functions to
where is the magnitude error and is the phase error. This match asymmetrical specifications using procedures similar to
assumes that we have normalized the complexoid so that the those used for real filters [1], [8], [21]. Similar choices exist for
magnitude of the term is unity. We can now directly apply obtaining suitable transfer functions for digital filters [2]–[4],
(10) and (11) to give [22]–[26]. In this tutorial introduction, complex filters derived
(18) using real-to-complex transformations are emphasized6, for
reasons of brevity. For the same reason, we limit our discussions
where to polyphase filters, active-RC filters, integrator-based filters,
and delay-based digital filters.
(19) A complex transfer function can be realized by using four real
SFG filters [1], [12], as is shown in Fig. 14 for analog filters.
and An identical realization is possible for digital filters where the
transfer functions are functions of . For Fig. 14, we have
(20)
(22)
For , , we see that the positive frequency term is slightly
changed in magnitude and phase. These errors are normally and
small and easy to compensate; however, the negative frequency
complexoid, ideally zero, now has a multiplier given by (23)

When the filters leading from the real and imaginary branches of
Fig. 14 do not exactly match, imaging errors (plus less harmful
(21) magnitude response errors) occur as we shall see

which is also the multiplier for the final resulting images. It A. Frequency-Translated Analog Filters
might be mentioned that (21) is the basis for a proposed adap-
One method for deriving complex transfer functions is to em-
tive quadrature mixer to be described later.
ploy a complex substitution for the Laplace variable . For ex-
ample, to effect a positive frequency shift, one uses the substi-
IV. COMPLEX FILTERS tution
In addition to image-reject mixers, complex filters are im-
portant and widely used signal processing blocks in modern (24)
wireless transceivers [47]. Complex filters use cross coupling
The effect of this substitution on a low-pass filter is shown in
between the real and imaginary signal paths in order to realize
Fig. 15. Thus, one can replace the procedure for realizing a PPF
asymmetrical (in the frequency domain) filters having transfer
filter via a Hilbert transform, as is shown in Fig. 7, with the more
functions that do not have the conjugate symmetry of real
general procedure for designing a complex filter that has a pass-
filters. This implies that their transfer functions have complex
band restricted to the desired (possibly positive) frequencies.
coefficients. The PPFs described previously can be consid-
ered a special case of complex filters. Complex filters can be 6The author has developed a set of Matlab routines for the direct approxima-

realized using the basic operations of addition and multipli- tion of complex filters, both analog and digital, without having to first derive a
real prototype. These routines are described in a separate publication which has
cation, and the delay operator for discrete-time digital filters been submitted for publication. The routines may be obtained by e-mailing the
or the integrator operator for continuous-time analog filters. author at [email protected].
MARTIN: COMPLEX SIGNAL PROCESSING IS NOT COMPLEX 1829

Fig. 15. (a) Baseband prototype magnitude filter response. (b) Magnitude
response of a complex filter obtained by the substitution s ! 0
s j! .

Fig. 17. Using a complex frequency translation to realize a PPF. (a)

High-pass prototype. (b) Complex band-stop having a complex admittance. (c)
Fully-differential version of (b). (d) Replacing the complex admittances with
cross-coupled real admittances between the real and imaginary paths.

results in transforming the capacitor to an admittance of a ca-

pacitor (of the same size) in parallel with an imaginary resistor
of size according to the substitution
(26)
This is shown in Fig. 16(b) where
(27)

Fig. 16. (a) Capacitor in the real prototype filter. (b) Transformed capacitor in In order to realize the complex filter using real components, two
the complex filter. (c) Real-component realization of the complex capacitor. identical real filters are used with cross-coupling, as is shown in
Fig. 16(c). The negative cross-coupled resistor can be realized
using fully differential circuits.
A method for realizing complex analog filters is to start with
An example of the use of this complex frequency translation
an active RC prototype filter and then transform it to a complex
is a derivation of a polyphase network.7 Consider the passive
frequency-translated filter; the procedure to do this is simple.
RC high-pass filter shown in Fig. 17(a). In this example, the
1) First, use an active RC prototype filter that is fully-dif- transmission zero at dc will be frequency shifted to negative
ferential. This is used to make the realization of negative frequencies, so we use the substitution . Letting
components trivial.
2) Secondly, replicate exactly the prototype filter in order to (28)
realize the signal path for the imaginary signal compo-
nents. gives the required substitution as
3) Finally, add cross-coupled resistors for each capacitor.
(29)
The size and signs of these resistors determine the fre-
quency translation. Thus, the complex frequency-shifted high-pass is now as
To understand why this procedure works, consider first that 2) shown in Fig. 17(b). A fully-differential version of the circuit
transforms the real prototype filter to a complex conjugate sym- of Fig. 17(b) is shown in Fig. 17(c). This complex filter can be
metric filter at the same frequencies as the prototype filter. To realized by two parallel real filters with cross-coupled resistors,
understand how the cross-coupling resistors result in the fre- as is shown in Fig. 17(d). Upon examination, this circuit is seen
quency translation, consider Fig. 16. Fig. 16(a) shows one of to be same as the polyphase network of Fig. 12(b) except for the
the capacitors in a complex filter at the same frequencies as the inclusion of load resistors in Fig. 17(d). Taking
real prototype filter. Applying the frequency translation would result in very narrow notches. Normally, polyphase
networks are intended to be cascaded [17], [19], and the load
(25) 7The derivation presented here is an alternative to that presented in [19].
1830 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 51, NO. 9, SEPTEMBER 2004

Fig. 19. Frequency shifting a real delay operator to a complex positive

frequency.
Fig. 18. Frequency shifting a real integrator to a complex positive-frequency
integrator.
We have
(32)
on each section is the succeeding section, as is discussed in
depth in [19]. Thus, discrete-time filters can be transformed by once again
An alternative procedure for realizing continuous-time com- replicating the filter for the imaginary path and then placing a
plex filters is based on frequency-translating SFG prototype fil- complex-coefficient multiplier directly after every delay oper-
ters which are based on integrators [1]. This procedure is illus- ator, as is shown in Fig. 19. It is obvious that this approach can
trated by applying it to the single integrator shown in Fig. 18(a). be used to realize complex filters that have their frequency-shifts
In Fig. 18(b) and (c) are shown complex SFGs that realize the programmable. As an additional note, the complex rotation im-
frequency-shifted integrator. The complex SFG of Fig. 18(c) plied by the multiplication by is called a Cordic operation,
is converted to the equivalent real SFG of Fig. 18(d), where and its realization has been well studied [27], [28].
the necessary cross-coupling is explicit. The procedure for fre- Some special cases of frequency-shifted discrete-time filters
quency shifting any real SFG integrator-based filter is simply: 1) are worth mentioning. For example, some discrete-time filters
replicate the real SFG filter for the imaginary path; and 2) cross are based on using as fundamental building blocks Euler for-
couple at each integrator, as is shown in Fig. 18(d). The strength ward and backward-difference integrators [29], which are often
of the cross-couplings determine the frequency of the shift. As called lossless digital integrators [30]; these are bases for many
always, this cross coupling is simplified by using fully-differen- switched-capacitor (SC) filter realizations. The filters before fre-
tial circuits. quency translation have unity-gain feedback around the delay
operators. After translation, it is desirable to preserve this unity-
B. Frequency-Translated Digital Filters gain feedback in order to facilitate their realization using SC in-
tegrators as the fundamental building blocks. This is possible as
Complex discrete-time and digital filters are as essential to shown in the original and transformed discrete-time integrators
modern wireless systems as the complex analog blocks already
in Fig. 20. The Realization of the complex SFGs in Fig. 20(b)
described. For example, digital complex filters are used as dec- and (d) using real SFGs is straightforward.
imation filters for complex A/Ds, and also as band-limiting fil- Another interesting special case is when the frequency shift
ters at baseband for spectral shaping (i.e., Nyquist filters, such is by , where is equal to some multiple
as root raised-cosine filters, designed to minimize inter-symbol
times the sampling frequency. For this case, we have
interference). An approach can be used for frequency shifting
real discrete-time delay-based digital filters to complex asym- (33)
metrical filters [2] that is similar to the one that was used for Using Fig. 4. with (33), we see that frequency shifting by
analog filters. Consider the real delay operator times the sampling frequency can be realized without multi-
pliers, just by using a path interchange and a signal inversion;
(30) this is a considerable simplification. As an example of using this
undergoing a frequency shift by applying the variable substitu- transform, consider using as a prototype filter a first-order dig-
tion ital differentiator that has a notch or transmission zero at dc.
This prototype filter has a real transfer function given by
(31) (34)
MARTIN: COMPLEX SIGNAL PROCESSING IS NOT COMPLEX 1831

Fig. 22. (a) Generalized multiplication where a , b are not necessarily equal
to a , b . (b) Equivalent complex SFG.

example of realizing a simplified complex notch filter at 1/4 the

sampling frequency, with a transfer function given by (35), is
described in [20]. In this case, a discrete-time complex analog
filter/mixer combination was realized at the input of a pair of
baseband oversampling A/D converters by time multiplexing
some switched capacitors using a modified switch arrangement,
a simple yet effective architecture. Other examples of frequency
Fig. 20. Frequency translating (a) Forward-difference integrator. (c)
Backward- difference integrators to complex discrete-time integrators.
translated complex filters are filter banks. This area has been
extensively researched and described in the literature and is
outside the scope of this publication. However, one possible
approach developed by the author that has good step-band
offered performance is described in [32] and mentioned as an
example.

C. Imaging Errors in Complex Multipliers and Transfer

Functions
The analysis of imaging errors due to real coefficient errors
in complex multipliers and transfer functions is similar to that
presented in Section III.E for complex mixers. Consider a gen-
eralized multiplication function as shown in Fig. 22(a) where,
ideally for a complex multiplication and , but,
in the case of mismatch and quadrature imbalance, this is not
true. Assuming

(36)
and
Fig. 21. (a) Prototype filter with a notch at dc. (b) Complex frequency-shifted
filter with a notch at f =4. (c) Its realization using real SFGs. (37)

then it is straightforward to show that Fig. 22(b) is equivalent

Its SFG is shown in Fig. 21. If this prototype filter, with a notch to Fig. 22(a). Using (6) and (7), we see that real coefficient er-
at dc is frequency shifted to a complex filter with a notch at , rors result in the additional complexoids in the signal path at
then, its new transfer function is given by the opposite frequencies of the multiplier input complexoids;
(35) positive frequency signal components are both transmitted and
also shifted to negative frequencies, and vice versa, with the
The complex SFG of this transfer function is shown in magnitude of these imaging components being proportional to
Fig. 21(b), and its realization using real SFGs is shown in the differences of the real coefficients. For example, in-band
Fig. 21(c). For practical purposes, this filter is the same as signal components can result in additional complexoids in the
that proposed in [31], where a number of these filters were stopband. Also, out-of-band interferers can result in additional
cascaded in order to give wider notch bandwidths. Another components in the passband. If these coefficient errors occur
1832 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 51, NO. 9, SEPTEMBER 2004

at multipliers inside digital filters, then these imaging compo-

nents are generated at the output of each nonideal multiplier.
The imaging components are then filtered by the transfer func-
tions from the multiplier outputs to the filter outputs and sub-
sequently appear as imaging errors at the filter output. To min-
imize signal-to-stopband errors, one should design in order to
have out-of-band notches in the transfer functions from the mul-
tiplier outputs to the filter output [33]. In addition, to minimize
interferer-to-signal errors, one should design to have out-of-
band notches in the transfer function from the filter input to
the inputs of the multipliers. These situations can be quantified
using the intermediate-function analysis approach described in
[1], this is a good area for future research. An example of the ap- Fig. 23. On-line system for obtaining the magnitude and phase mismatch
plication of these considerations is in the filter design described errors of a quadrature mixer in a low-IF receiver.
in [52].
A similar situation occurs when realizing complex fil-
ters using the approach in Fig. 14(b), where we now have
nonmatched filters , , , and . The
nontranslated frequency components are filtered by the average
transfer functions

(38)
whereas imaging components are produced (both in-band and
out-of-band) by first conjugating the input signals, which
changes the signs of the frequencies of the spectral compo- Fig. 24. Digital adaptive filter approach for adapting a quadrature mixer [38].
nents, and then filtering these images by the difference transfer
functions [1].
A possible approach for adapting a quadrature mixer is based
(39) on (21). A block diagram showing how estimates for the mag-
nitude and phase errors are found is shown in Fig. 23. In this
In this case, the images occur directly at the filter outputs, and approach, assume a quadrature-imbalanced mixer is used to de-
there is less freedom to minimize images by careful choice of modulate an RF signal to a low IF. A complex PPF can be used
the intermediate transfer functions, as compared to the situation to extract the desired signal. In addition, assume a NPF is used to
where filters are cross coupled internally. extract the undesired signal-induced image at negative frequen-
cies, which is caused by the magnitude and phase mis-match
V. RECENT APPLICATIONS OF COMPLEX SIGNAL PROCESSING errors in the RF mixer. If the PPF and NPF complex outputs are
A. Adaptive Quadrature Mixers correlated, that is, multiplied together and averaged (or low-pass
filtered), then the approximate result is a complex dc signal
Many recent wireless receivers tune or adapt the systems to
given by (21). Thus, the real output of the multiplication can be
eliminate image errors due to nonideal quadrature mixers. This
used to drive the gain error to zero, and the imaginary part can
tuning can be done in either the analog or the digital domain.
be used to drive the phase errors to zero. This assumes the errors
As a preliminary to discussing adaptive algorithms, note that
are wide band and frequency independent. In normal practice,
most complex adaptive systems require the output of a sensi-
the signal processing for the filtering and correlation would be
tivity filter, , which is correlated with an error , in order
done in the digital domain, but the tuning would be done in the
to drive the error to zero. A possible update algorithm is given
analog domain at the mixers. Interferers would thus be elim-
by
inated before the A/D conversion, and this would help prevent
(40) overloading the A/D and preceding preamplifiers. The approach
just proposed is similar to that described in [36], except that in
where is a small constant controlling the update rate. Often, [36], the adaptation was done with an interferer only used as a
this update is approximated using the stochastic instantaneous reference input, the error functions minimized were different,
update to give and the adaptation update was performed in the analog domain.
Alternative approaches, where the signal processing and the
(41)
tuning are both done in the digital domain, are described in [37],
Note that is conjugated for complex signals. Alternatively, [38], with a realization based on [38] described in [39]. This ap-
could have been conjugated. The required sensitivity filter, proach is shown in Fig. 24. As in the previous approach, a com-
, is often trivial for FIR systems, and often not very com- plex PPF is used to extract the desired signal, and a second NPF
plicated even for IIR systems [34], [35] (and can be found using is used to extract the images. The fundamental idea on which
Tellegen’s Theorem [51]). this approach is based is that a symmetric decorrelating filter is
MARTIN: COMPLEX SIGNAL PROCESSING IS NOT COMPLEX 1833

Fig. 25. Low-IF architecture for a receiver.

used to minimize image components in and signal compo-

nents in [40]. This helps to minimize biases. Another advan-
tage of this approach is that frequency-dependent imaging er-
rors are cancelled as well. A recently proposed algorithm [41],
although completely implemented in the digital domain, does
include some features of both of the above algorithms. Another
interesting adaptive image-reject approach, that, again, is com-
pletely in the digital domain, is based on blind source separation
compensation [42]. Fig. 26. Complex 16 oversampling A/D converter architecture.
The alternatives described are just the beginning of what is
expected to be a fruitful area of research. One future possibility be programmed to accommodate the requirements of the dif-
is to use to image cancellers, one in the analog domain and a ferent standards. The low-IF architectures greatly minimize dc
second canceller in the digital domain. Tuning the mixer in- offset errors due to feedback of the local-oscillator (LO) and the
puts in the analog domain is useful in preventing overload in large noise of MOS transistors. The low-IF architectures do
the IF stages and in helping to attain timing acquisition, but suffer from images when the IF frequency is low, but these can
is useful only for correcting frequency-independent errors. A be significantly minimized by using even relatively simple com-
second adaptive fine tuning in the digital domain similar to [38] plex filters at the IF [47], [20].
can be used to further eliminate frequency-dependent errors of, In general, when the A/D conversion occurs before the
for example, complex oversampling A/Ds used at the IF. channel-selection filters, then the A/D conversion must neces-
sarily be high dynamic range, and the A/D converter is almost
B. Low-IF Receivers always the limiting factor determining the maximum achievable
A currently popular architecture for realizing highly in- data rate of the receiver. Assuming the A/D converter is the
tegrated, moderately wide-band, wireless transceivers is the limiting factor on the maximum achievable data rate, then it is
low-IF architecture [45]–[47]. An example of a low-IF receiver possible to improve significantly the maximum-achievable data
is shown in Fig. 25. In a low-IF receiver, the RF signal is de- rate using the modulation approach proposed in [49]. A typical
modulated to baseband in two stages; the first stage is a standard low-IF architecture has IF spectral components at positive
quadrature mixer with a real input signal that requires two real or negative frequencies only; this wastes half the available
analog multipliers. This translates the RF input signal down to bandwidth of the A/Ds.8 It is possible to send additional data
a low-IF frequency. The high-IF component after the first shift in the unused frequency band. In the simplest implementation,
is eliminated by low-pass filters; at that point a second complex one channel is implemented at positive frequencies, and one
mixer is used to shift the low-IF signal to baseband. The second channel is implemented at negative frequencies (as proposed
complex mixer requires four real multipliers. These can be in [49]); in more advanced realizations, multiple channels
realized using analog multipliers as in [46]. Alternatively, the are implemented at both positive and negative frequencies
A/D conversion can be done at the IF stage, as is shown in using orthogonal-frequency-division-multiplexing (OFDM)
Fig. 25, and then the second complex frequency shift can be techniques. In both cases, the physical hardware is the same as
realized in the digital domain [47]. This allows the second com- that used for typical low-IF receivers (as shown in Fig. 25); the
plex modulation to be almost ideal. Also, if the A/D sampling increase in complexity is all in the digital domain and is quite
frequency is chosen to be a small multiple of the IF frequency, feasible for current data rates.
then the second mix is significantly simplified. For example, if For example, consider the approach for two channels only,
the sampling frequency is four times the IF frequency, then the one at positive and one at negative frequencies. At the trans-
oscillator inputs to the second complex digital mixer are simply mitter, two simultaneous quadrature input channels are used;
or 0. This digital modulation alternative is often preferable. one is modulated to positive frequencies, and one is modulated
Low-IF architectures are especially suitable for multistan- 8This assumes that the A/D converters are not complex oversampling A/Ds; if
dard receivers when the channel-selection filters are realized they are, the bandwidth is not necessarily wasted. This is one of the motivations
after the A/D converters, using digital circuits, and, thus, can for using complex oversampling A/Ds.
1834 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 51, NO. 9, SEPTEMBER 2004

Fig. 27. Complex SFG of fourth-order quadrature 16 modulator from [52].

to negative frequencies using quadrature mixers in the digital
domain. A guardband is left at dc. The combined signal is then
modulated up to the RF and transmitted. The conversion to the
analog domain can occur just before or after the last modulation.
This approach is much like a double SSB system where both an
upper and a lower sideband are transmitted together (sometimes
called a suppressed-carrier modulation system). At the receiver,
the RF signal is demodulated so that one channel is again at
a low positive IF, and the second channel is at a low negative
IF. The IF signal is then digitized using two A/D converters in a
manner similar to that used in a typical low-IF receiver. After the
A/D conversion, the individual channels can be independently
demodulated using complex signal processing in the digital do-
main. This architecture results in a guardband at dc which allows Fig. 28. Using a quadrature 16 modulator to digitize an IF signal [52], [54].
the IF signal to be ac coupled, and this minimizes errors due to
the local feedback of the RF oscillator (LO feedback) and
and the A/D conversion are effectively combined in a single
noise, similar to the traditional low-IF architecture, yet it sup-
unit. The basic architecture of the complex oversampling
ports almost twice the data rate for a given sampling frequency
A/D converter is shown in Fig. 26.
at the A/D converter with no increase in hardware complexity.
In [52], a fourth-order complex, that is eighth-order real, over-
C. Complex and Quadrature Oversampling A/D Converters sampling A/D, that used 1-bit internal converters, was reported.
A complex SFG of the architecture showing the complex res-
All signal processing blocks preceding the channel-selection
onators is shown in Fig. 27. It should be noted that each of the
filters in a receiver must be high dynamic range. Typical specifi-
coefficients is complex, thus each block having a transfer func-
cations are linearities equivalent to 14-bit ideal A/D converters
tion
or more, that is, in excess of 86-dB SNDR. Typical specifica-
tions for the analog bandwidths are now in excess of 2 MHz and
as high as 5 MHz. Realizing A/D converters that can meet these (42)
demanding specifications is a challenging task. A possible ar-
chitecture that is gaining in popularity for this application is the realizes a complex resonator close to the desired IF frequency.
complex oversampling A/D converter [8]. Complex oversam- The design of the internal complex resonator-based filter was
pling A/D converters combine the complex filtering ideas done to maximize SNDR by careful choice of the NTF and STF
of [1] and the real bandpass oversampling A/D converters transmission zeros [33]. The converter was implemented using
of [50]. By using a complex filter in the loop, the effec- SC circuits. This first-integrated quadrature converter had
tive signal bandwidth is cut in half, and this doubles the over- a measured SNDR of 65 dB for a 100-kHz signal bandwidth.
sampling ratio (OSR). The result, ideally, is a much-improved The converter was also verified to have better stability properties
signal-to-noise distortion ratio (SNDR) [51], [52], depending than real oversampling converters.
on the order of the oversampling loop and the resolution of the Recently, examples of continuous-time complex A/D
D/As embedded in the A/D. Also, it is possible to choose the lo- converters have also been reported [53], [54]. For example,
cations of the zeros in the noise-transfer function (NTF) in order a converter intended to be used to digitize an IF signal of
to minimize errors due to quantization effects [33]. Another re- a wireless receiver, similar to [52] is reported in [54]. This
sult is that the zeros of the signal-transfer function (STF) can be application is shown in Fig. 28. A functional block diagram of
independently chosen to filter out images from an imperfect pre- the converter of [54] is shown in Fig. 29. The A/D converter
ceding quadrature mixer. Thus, complex image-reject filtering exhibited an SNDR of 76 dB for a 1-MHz signal bandwidth,
MARTIN: COMPLEX SIGNAL PROCESSING IS NOT COMPLEX 1835

[10] A. Sedra, W. Snelgrove, and R. Allen, “Complex analog bandpass fil-

ters designed by linearly shifting real low-pass prototypes,” in Proc. Int.
Symp. Circuits and Systems, vol. 3, 1985, pp. 1223–1226.
[11] Q. Liu, “Switched-capacitor complex filters,” M.A.Sc. thesis, Univ.
Toronto, Toronto, ON, Canada, 1985.
[12] E. A. Lee and D. G. Messerschmitt, Digital Communication. Boston,
MA: Kluwer Academic, 1988.
[13] J. Crols and M. Steyaert, CMOS Wireless Transceiver Design. Boston,
MA: Kluwer Academic, 1997.
[14] B. Razavi, RF Microelectronics. Englewood Cliffs, NJ: Prentice-Hall,
1998.
[15] D. K. Weaver, Jr., “A third method of generation and detection of single-
sideband signals,” Proc. IRE, pp. 1703–1705, June 1956.
[16] R. Hartley, “Modulation System,” U.S. Patent 1 666 206, Apr. 1928.
[17] M. J. Gingell, “Single-Sideband modulation using sequence asymmetric
polyphase networks,” Elect. Commun., vol. 48, no. 1-2, pp. 21–25, 1973.
[18] J. Crols and M. Steyaert, “An analog integrated polyphase filter for a
high-performance low-IF receiver,” in Proc. VLSI Circuits Symp., Kyoto,
Fig. 29. Block diagram of complex 16 modulator of [54]. Japan, June 1995, pp. 87–88.
[19] F. Behbahani, Y. Kishigami, J. Leete, and A. Abidi, “CMOS mixers and
polyphase filters for large image rejection,” IEEE J. Solid-State Circuits,
vol. 36, pp. 873–887, June 2001.
yet dissipated only 4.4 mW. These results are excellent at the [20] K.-P. Pun, J. E. Franca, C. Azeredo-Leme, and R. Reis, “Quadrature
time of publication. sampling schemes with improved image rejection,” IEEE Trans. Circuits
Currently, there has been little published on how best to im- Syst. II, vol. 50, pp. 641–648, Sept. 2003.
[21] C. Cuypers, N. Voo, M. Teplechuk, and J. Sewell, “The general synthesis
plement the complex digital filters required that must be in- of complex analogue filters,” in Proc. Int. Conf. Electricity, Circuits, Sys-
cluded following the analog modulators in order to attenuate tems, vol. 1, 2002, pp. 153–156.
quantization noise and allow the signals to be decimated. These [22] A. Fettweis, “Principles of complex wave digital filters,” Int. J. Circuit
Theory Applicat., vol. 9, pp. 119–134, 1981.
can be realized directly at the IF frequencies, or the IF signals
[23] X. Chen and T. W. Parks, “Design of FIR filters in the complex domain,”
can first be demodulated to baseband using a complex mixer, IEEE Trans. Acoust., Speech, Signal Processing, vol. 35, pp. 144–153,
and then the decimation filters can be realized at baseband. This Feb. 1987.
is expected to be an active area for future research. [24] P. Regalia, S. Mitra, and J. Fadavi-Ardekani, “Implementation of real co-
efficient digital filters using complex arithmetic,” IEEE Trans. Circuits
Syst., vol. CAS-34, pp. 345–353, Apr. 1987.
VI. CONCLUSION [25] X. Chen and T. Parks, “Design of IIR filters in the complex domain,”
IEEE Trans. Acoust., Speech, Signal Processing, vol. 38, pp. 910–920,
A tutorial description of the use of complex signal processing June 1990.
[26] M. Komodromos, S. Russell, and P. T. P. Tang, “Design of FIR filters
in wireless systems has been presented. First, the fundamentals with complex desired frequency response using a generalized remez al-
have been explained and then their use in a number of applica- gorithm,” IEEE Trans. Circuits Syst. II, vol. 42, pp. 428–435, Apr. 1995.
tions has been described. A highly graphical SFG approach has [27] J. Volder, “The CORDIC trigometric computing technique,” IEEE
Trans. Comput., vol. EC-8, pp. 330–334, Sept. 1959.
been used throughout. Example applications of complex mixers, [28] A. Madisetti, A. Kwentus, and A. N. Willson, Jr., “A 100-MHz 16-b di-
filters, low-IF receivers, and complex oversampling A/D con- rect digital frequency synthesizer with a 100-dBc spurious-free dynamic
verters have been described. range,” IEEE J. Solid-State Circuits, vol. 34, pp. 1399–1410, Aug. 1999.
[29] L. R. Rabiner and B. Gold, Theory and Application of Digital Signal
Processing. Englewood Cliffs, NJ: Prentice-Hall, 1975.
REFERENCES [30] L. T. Bruton, “Low-sensitivity digital ladder filters,” IEEE Trans. Cir-
cuits Syst., vol. CAS-22, pp. 168–176, Mar. 1975.
[1] W. M. Snelgrove, “Intermediate function synthesis,” Ph.D. dissertation, [31] S. Levantino, C. Samori, M. Banu, J. Glas, and V. Boccuzzi, “A CMOS
Univ. Toronto, Toronto, ON, Canada, 1982. GSM IF-sampling circuit with reduced in-channel aliasing,” IEEE J.
[2] T. H. Crystal and L. Ehrman, “The design and applications of digital fil- Solid-State Circuits, vol. 38, pp. 895–904, June 2003.
ters with complex coefficients,” IEEE Trans. Audio Electroacoust., vol. [32] K. W. Martin, “Small side-lobe filter design for data communication ap-
AU-16, no. 3, pp. 315–320, Sept. 1968. plications,” IEEE Trans. Circuits Syst. II, vol. 45, pp. 1155–1161, Aug.
[3] R. Boite and H. Leich, “On digital filters with complex coefficients,” 1998.
in Network and Signal Theory, J. K. Skwirzynski and J. O. Scanlan, [33] S. Jantzi, K. Martin, and A. S. Sedra, “The effects of mismatch in com-
Eds. London, U.K.: Peter Peregrinus, 1973, pp. 344–351. plex bandpass 61 modulators,” in Proc. IEEE Int. Symp. Circuits and
[4] M. T. McCallig, “Design of digital FIR filters with complex conju- Systems (ISCAS’96), May 1996, pp. 227–230.
gate pulse responses,” IEEE Trans. Circuits Syst., vol. CAS-25, pp. [34] K. Martin and M. T. Sun, “Adaptive filters suitable for real-time spectral
1103–1105, Dec. 1978. analysis,” IEEE Trans. Circuits Syst., vol. CAS-33, pp. 218–229, Feb.
[5] A. V. Oppenheim and R. W. Schafer, Digital Signal Processing. En- 1986.
glewood Cliffs, NJ: Prentice-Hall, 1975. [35] T. Kwan and K. Martin, “Adaptive detection and enhancement of mul-
[6] W. M. Snelgrove and A. S. Sedra, “State-space synthesis of complex tiple sinusoids using a cascade IIR filter,” IEEE Trans. Circuits Syst.,
analog filters,” in Proc. Eur Conf. Circuit Theory and Design, 1981, pp. vol. 36, pp. 937–947, July 1989.
420–424. [36] L. Der and B. Razavi, “A 2-GHz CMOS image-reject receiver with LMS
[7] G. R. Lang and P. O. Brackett, “Complex analogue filters,” in Proc. Eur calibration,” IEEE J. Solid-State Circuits, vol. 38, pp. 167–175, Feb.
Conf. Circuit Theory and Design, 1981, pp. 412–419. 2003.
[8] W. M. Snelgrove, A. S. Sedra, G. R. Lang, and P. O. Brackett. (1981) [37] J. Paez-Borrallo and F. J. C. Quiros, “Self adjusting digital image re-
Complex analog filters. Univ. Toronto, Toronto, ON, Canada. [On- jection receiver for mobile communications,” in Proc. IEEE Vehicular
line]. Available: ftp://ftp.eecg.toronto.edu/pub/software/martin/snel- Technology Conf., vol. 2, May 1997, pp. 686–690.
grove_cmplx.pdf [38] L. Yu and W. M. Snelgrove, “A novel adaptive mismatch cancellation
[9] R. Allen, “Complex analog filters obtained from shifted lowpass proto- system for quadrature IF radio receivers,” IEEE Trans. Circuits Syst. II,
types,” M.A.Sc. thesis, Univ. Toronto, Toronto, ON, Canada, 1985. vol. 46, pp. 789–801, June 1999.
1836 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 51, NO. 9, SEPTEMBER 2004

[39] K. P. Pun, J. Franca, and C. Azeredo-Leme, “The correction of fre- [54] K. Philips, “A 4.4 mW 76 dB complex 61 ADC for bluetooth re-
quency-dependent I=Q mismatches in quadrature receivers by adap- ceivers,” in Proc. IEEE Int. Solid-State Circuits Conf., Feb. 2003, pp.
tive signal separation,” in Proc. IEEE Int. Symp. Circuits and Systems 64–65.
(ISCAS’01), 2001, pp. 424–427.
[40] D. V. Compernolle and S. V. Gerven, “Signal separation in a symmetric
adaptive noise canceler by output decorrelation,” IEEE Trans. Signal
Processing, vol. 43, pp. 1602–1612, July 1995.
[41] C. C. Chen and C. C. Huang, “On the architecture and performance of
a hybrid image rejection receiver,” IEEE J. Select. Areas Commun., vol.
19, pp. 1029–1040, June 2001.
[42] M. Valkama and M. Renfors, “Advanced methods for I=Q imbalance Kenneth W. Martin (S’75–M’80–SM’89–F’91) re-
compensation in communication receivers,” IEEE Trans. Signal Pro- ceived the B.A.Sc., M.A.Sc., and Ph.D. degrees from
cessing, vol. 49, pp. 2335–2344, Oct. 2001. the University of Toronto, Toronto, ON, Canada, in
[43] A. A. Abidi, “Direct-conversion radio transceivers for digital commu- 1975, 1977, and 1980, respectively.
nications,” IEEE J. Solid-State Circuits, vol. 30, pp. 1399–1410, Dec. From 1977 to 1978, he was a member of the Sci-
1995. entific Research Staff at Bell Northern Research, Ot-
[44] B. Razavi, “Design considerations for direct-conversion receivers,” tawa, ON, Canada, where he did some of the early
IEEE Trans. Circuits Syst. II, vol. 44, pp. 428–435, June 1997. research in integrated, switched-capacitor networks.
[45] J. Crols and M. S. J. Steyaert, “A single-chip 900 MHz CMOS receiver Between 1980 and 1992, he was consecutively an As-
front-end with a high performance low-IF topology,” IEEE J. Solid-State sistant, Associate, and Full Professor at the Univer-
Circuits, vol. 30, pp. 1483–1492, Dec. 1995. sity of California at Los Angeles. In 1991, he ac-
[46] J. C. Rudell, J. Ou, T. B. Cho, G. Chien, F. Brianti, J. A. Weldon, and P. cepted the endowed Stanley Ho Professorship in Microelectronics at the Univer-
R. Gray, “A 1.9-GHz wide-band IF double conversion CMOS receiver sity of Toronto. In 1998, he co-founded Snowbush Microelectronics along with
for cordless telephone applications,” IEEE J. Solid-State Circuits, vol. Prof. David Johns where he is currently President while on a temporary leave
32, pp. 2071–2088, Dec. 1997. of absence from the Univeristy of Toronto. He has authored or coauthored two
[47] J. Crols and M. Steyaert, “Low-IF topologies for high-performance textbooks entitled Analog Integrated Circuit Design (New York: Wiley, 1997)
analog front ends of fully integrated receivers,” IEEE Trans. Circuits and Digital Integrated Circuit Design (New York: Oxford, 2000) in addition to
Syst. II, vol. 45, pp. 269–282, Mar. 1998. three research books coauthored with former Ph.D. students. He has published
[48] K. P. Pun, J. E. Franca, and C. Azeredo-Leme, “A quadrature sam- well over 100 papers and holds five patents.
pling scheme with improved image rejection for complex-IF receivers,” Dr. Martin was selected by the IEEE Circuits and Systems (CAS) Society for
in Proc. IEEE Int. Symp. Circuits Systems (ISCAS’01), May 2001, pp. the Outstanding Young Engineer Award that was presented at the IEEE Cen-
45–48. tennial Keys to the Future Program in 1984. He was awarded a National Sci-
[49] S. Mirabbasi and K. Martin, “Hierarchical QAM: A spectrally efficient ence Foundation Presidential Young Investigator Award (1985–1990) and the
DC-free modulation scheme,” IEEE Commun. Mag., vol. 38, pp. 1999 CAS Golden Jubilee Medal of the IEEE CAS Society. He was a co-recip-
140–146, Nov. 2000. ient of the Beatrice Winner Award at the 1993 International Solid-State Circuits
[50] S. A. Jantzi, W. M. Snelgrove, and P. F. Ferguson, Jr., “A fourth-order Conference and of the 1999 IEEE Darlington Best-Paper Award for the IEEE
bandpass sigma-delta modulator,” IEEE J. Solid-State Circuits, vol. 28, TRANSACTIONS ON CIRCUITS AND SYSTEMS. He was appointed as the IEEE CAS
pp. 282–291, Mar. 1993. Press Representative (1985–1986). He was elected by the CAS Society mem-
[51] S. Jantzi, “Quadrature bandpass delta-sigma modulation for digital bers to their administrative committee (ADCOM 1985–1987), and as a member
radio,” Ph.D. dissertation, Univ. Toronto, Toronto, ON, Canada, 1997. of the CAS Board of Governors (1995–1997). He served as an Associate Ed-
[52] S. A. Jantzi, K. Martin, and A. S. Sedra, “Quadrature bandpass 16 itor of the IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS from 1985 to 1987
modulation for digital radio,” IEEE J. Solid-State Circuits, vol. 32, pp. and as an Associate Editor of PROCEEDINGS OF THE IEEE (1995–1997) and has
1935–1949, Dec. 1997. served on the technical committee for many International Symposia on CAS
[53] F. Henkel et al., “A 1 MHz-bandwidth second-order continuous-time and on the Analog Program Committee of the International Solid-State Circuits
quadrature bandpass sigma-delta modulator for low-IF receivers,” in Committee. He gave an invited Plenary talk at the 2003 European Solid-State
Proc. IEEE Int. Solid-State Circuits Conf., Feb. 2002, pp. 214–215. Circuits Conference (ESSCIRC).