IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 52, NO.

1, JANUARY 2004 159

A Robust Digital Baseband Predistorter Constructed

Using Memory Polynomials
Lei Ding, Student Member, IEEE, G. Tong Zhou, Senior Member, IEEE, Dennis R. Morgan, Senior Member, IEEE,
Zhengxiang Ma, J. Stevenson Kenney, Senior Member, IEEE, Jaehyeong Kim, and Charles R. Giardina

Abstract—Power amplifiers (PAs) are inherently nonlinear Of all linearization techniques, digital baseband predistortion
devices and are used in virtually all communications systems. is among the most cost effective. A predistorter is a functional
Digital baseband predistortion is a highly cost-effective way block that precedes the PA. It generally creates an expanding
to linearize PAs, but most existing architectures assume that
the PA has a memoryless nonlinearity. For wider bandwidth nonlinearity since the PA has a compressing characteristic. Ide-
applications such as wideband code-division multiple access ally, we would like the PA output to be a scalar multiple of the
(WCDMA) or wideband orthogonal frequency-division multi- input to the predistorter-PA chain. For a memoryless PA (i.e.,
plexing (W-OFDM), PA memory effects can no longer be ignored, the current output depends only on the current input), memory-
and memoryless predistortion has limited effectiveness. In this less predistortion is sufficient. There has been intensive research
paper, instead of focusing on a particular PA model and building
a corresponding predistorter, we focus directly on the predistorter on memoryless predistortion during the past decade [1].
structure. In particular, we propose a memory polynomial model For wider bandwidth applications such as wideband CDMA
for the predistorter and implement it using an indirect learning (WCDMA) or wideband OFDM (W-OFDM), PA memory ef-
architecture. Linearization performance is demonstrated on a fects can no longer be ignored. Moreover, higher power ampli-
three-carrier WCDMA signal. fiers such as those used in wireless base stations exhibit memory
Index Terms—Communication system nonlinearities, nonlinear effects. The cause of memory effects can be electrical or elec-
systems, polynomials, power amplifiers (PAs), Volterra series. trothermal as suggested in [2]. Memoryless predistortion for
a PA with memory often results in poor linearization perfor-
I. INTRODUCTION mance.
Most researchers approach the problem of PA linearization by

P OWER AMPLIFIERS (PAs) are indispensable com-

ponents in a communication system and are inherently
nonlinear. It is well known that there is an approximate inverse
first finding a good model for the PA. Volterra series is a general
nonlinear model with memory [3] and has been used to model
PAs with mild nonlinearities [4]. A serious drawback of the
relationship between the PA efficiency and its linearity. Hence, Volterra model is the large number of coefficients that must be
nonlinear PAs are desirable from an efficiency point of view. extracted. Predistortion of the Volterra model is usually imple-
The price paid for higher efficiency is that nonlinearity causes mented using the th-order inverse technique [3], which is com-
spectral regrowth (broadening), which leads to adjacent channel plicated. One thing to keep in mind is that the exact inverse of a
interference. It also causes in-band distortion, which degrades Volterra system is difficult to construct and the th-order inverse
the bit-error rate (BER) performance. Newer transmission is only an approximation. Recently, two special cases of the
formats, such as code-division multiple access (CDMA) and Volterra model have been proposed to capture the memory non-
orthogonal frequency-division multiplexing (OFDM), are linear effects in the PA associated with wideband signals. One
especially vulnerable to PA nonlinearities, due to their high is the Wiener model; i.e., a linear time-invariant (LTI) system
peak-to-average power ratio, corresponding to large fluctua- followed by a memoryless nonlinearity, proposed by Clark et
tions in their signal envelopes. In order to comply with spectral al. [5]. The advantage of Wiener modeling for the PA is that
masks imposed by regulatory bodies and to reduce BER, PA the corresponding predistorter is a Hammerstein system; i.e., a
linearization is necessary. memoryless nonlinearity followed by an LTI system, and it is
possible for the predistorter to be an exact inverse of the PA.
Another special case of the Volterra model is the memory poly-
Paper approved by Z. Kostic, the Editor for Wireless Communication of the nomial model proposed by Kim et al. [6]. Similar to the Volterra
IEEE Communications Society. Manuscript received July 17, 2002; revised model, an exact inverse of the memory polynomial is difficult
May 22, 2003. This work was supported in part by the National Science to obtain, but another memory polynomial can be constructed
Foundation under Grant 9703312 and Grant 0219262, and in part by the State
of Georgia’s Yamacraw Initiative. This paper was presented in part at the IEEE as an approximate inverse. A variety of other PA models exist,
Global Telecommunications Conference, Taipei, Taiwan, November, 2002. for example, the parallel Wiener model investigated in [7].
L. Ding, G. T. Zhou, and J. S. Kenney are with the School of Electrical It is difficult to judge which PA model is the best, since it
and Computer Engineering, Georgia Institute of Technology, Atlanta, GA
30332-0250 USA (e-mail: [email protected]). could depend on the type of the PA, the data format being trans-
D. R. Morgan and Z. Ma are with Bell Laboratories, Lucent Technologies, mitted, etc. Moreover, the most accurate PA model may not be
Murray Hill, NJ 07974 USA. the most amenable to predistortion. In applications that we are
J. Kim and C. R. Giardina are with Lucent Technologies, Whippany, NJ 07981
USA. interested in, predistortion linearization is the ultimate objec-
Digital Object Identifier 10.1109/TCOMM.2003.822188 tive, whereas accurate PA modeling is only a secondary concern.
0090-6778/04$20.00 © 2004 IEEE

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on November 30, 2009 at 09:53 from IEEE Xplore. Restrictions apply.
 160 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 52, NO. 1, JANUARY 2004

For a nonlinear PA with memory, its inverse must also be a

nonlinear system with memory. Our goal here is to find a good
model to approximate the inverse of the PA nonlinearity, in the
sense that parameter extraction and system implementation are
straightforward and the predistorter model is robust. In [6], it
was shown that a memory polynomial is a good model for the
PA, so it is then natural for us to consider it for the predistorter
as well.
In this paper, we impose a memory polynomial model on the
predistorter and implement it using the indirect learning archi-
tecture proposed in [8]. Simulation results show that this is an
improvement over our previous work [9] where a Hammerstein
predistorter is used. The improvements are mostly in terms of Fig. 1. Indirect learning architecture for the predistorter.
robustness and simplicity of predistorter parameter estimation.
We need to assume that the PA nonlinearity is invertible so that
II. MEMORY POLYNOMIAL MODEL corresponds to .
Here, we consider that the PA characteristics do not change
Let us consider, as an example, a baseband linear-cubic model
rapidly with time; changes in PA characteristics are often due to
with input and output
temperature drift, aging, etc., which have long time constants.
After gathering a block of and data samples, the
training branch (block A) can process the data offline, which
lowers the processing requirements of the predistortion system.
Once the predistorter identification algorithm has converged,
(1) the new set of parameters are plugged into the high-speed pre-
distorter, which can be readily implemented using application-
where the linear and cubic kernels and specific integrated circuits (ASICs) or field programmable gate
characterize the nonlinear system. If arrays (FPGAs). When the predistorter coefficients have been
, then (1) reduces to found and it is believed that the PA characteristics are hardly
changing, the setup in Fig. 1 can be run in open loop, i.e., we
(2) temporarily shut down the training branch, until changes in the
PA characteristics require a predistorter coefficient update.
which is a memoryless nonlinear system. If ,
except along the diagonal , then (1) becomes IV. PREDISTORTER IDENTIFICATION
Generalizing (3) and changing to , the
memory polynomial predistorter (similar to [6]) can be de-
scribed by
(3)
and we refer to (3) as a memory polynomial. We can view the
memory polynomial as a compromise between the memoryless (4)
nonlinearity and the full Volterra nonlinearity. It has the advan-
tage of capturing memory effects while keeping the number of
coefficients on the order of . This is contrasted with the where we have slightly generalized the model by including
full Volterra system, where the number of unknowns is on the even- as well as odd-order terms. Therefore, the predis-
order of , where is the nonlinearity order. torter has memory and highest nonlinearity order
. In most predistorter designs, only odd-order non-
linearities are included, i.e., , and the term
III. INDIRECT LEARNING ARCHITECTURE
is of order
Fig. 1 shows the indirect learning structure that is used for . Thus, the exponent of is , and the exponent
predistorter identification [8]. The feedback path labeled “Pre- of is —such designation follows the analysis of [10,
distorter Training” (block A) has as its input, where p. 69]. As we will show in Example 2 (Section V), by including
is the gain of the linearized PA, and as its output. The ac- even-order nonlinear terms in the predistorter [i.e., allow even
tual predistorter is an exact copy of the feedback path (copy of in (4)], spectral regrowth can be further reduced (by about
A); it has as its input and as its output. Ideally, we 3–5 dB in Example 2). Detailed investigation of the benefits of
would like , which renders and the even-order terms in the baseband model (4) is reported in [11].
error term . Given and , our task is to find the Note that the even-order term is different
parameters of block A, which yields the predistorter. The algo- from . The former contains phase information,
rithm converges when the error energy is minimized. whereas the latter is “phase blind.”

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on November 30, 2009 at 09:53 from IEEE Xplore. Restrictions apply.
 DING et al.: A ROBUST DIGITAL BASEBAND PREDISTORTER CONSTRUCTED USING MEMORY POLYNOMIALS 161

Fig. 2. W–H model diagram.

Since is linear in the parameters , the latter can be

estimated by a simple least-squares method. By defining a new
sequence

(5)

at convergence, we should have

(6)

where , Fig. 3. Effectiveness of predistortion in suppressing spectral regrowth

when the PA is modeled by a W–H system. (a) Output without predistortion.
, , and (b) Output with memoryless predistortion. (c) Output with memory polynomial
. The least-squares solution Q K
predistortion ( = 2, = 5). (d) Original input. (c) and (d) almost coincide.
for (6) is
The baseband input is a three-carrier WCDMA signal.
(7) Memory polynomial predistorter identification is carried out
based on 8000 data samples. Next, we compare the power spec-
where denotes complex conjugate transpose. tral density (PSD) of the input and output signals to evaluate the
effectiveness of the predistorter in reducing spectral regrowth.
V. SIMULATIONS Here, we used the predistorter (4) with two delay taps
In this section, we illustrate, through computer simulations, and fifth odd-order nonlinearity . Performance of the
performance of the memory polynomial predistorter identified predistorter is demonstrated in Fig. 3. Spectral regrowth is
using the indirect learning architecture. We will show that the almost fully suppressed with only two delay taps, even though
same memory polynomial structure can be used to linearize sev- the LTI portions of the W–H system have much longer impulse
eral different nonlinear models with memory, hence, demon- responses.
strating the robustness of memory polynomial predistortion. Example 2: Here, we assume that the PA obeys an
Example 1: Here, the nonlinearity to be compensated for is odd-order-only memory polynomial model
assumed to obey a Wiener–Hammerstein (W–H) model (see
Fig. 2), i.e., an LTI system followed by a memoryless nonlin- (11)
earity, which in turn is followed by another LTI system. Such
a configuration is commonly used for satellite communication
The coefficients
channels where the PA at the satellite transponder is driven near
saturation to exploit the maximum power efficiency [12]. The
LTI blocks before and after the memoryless nonlinearity, which
are denoted by and , respectively, are assumed to be

(8)
(12)
For the memoryless nonlinear portion of the W–H model were extracted from the same PA as in Example 1. Fig. 4 shows
the performance of various predistorters. The memory polyno-
mial predistorter with and was able to suppress
(9) most of the spectral regrowth. However, when both even- and
odd-order nonlinearities are included in the predistorter, an ad-
ditional 3–5 dB suppression can be achieved. Spectral regrowth
where and are, respectively, input and output of the can be further suppressed by increasing the memory of the pre-
memoryless nonlinear block. For the coefficients, we had distorter to .
Example 3: We start with a Wiener model whose
, , and .
Recall that the Wiener model can be regarded as a special case
(10) of the Volterra system. We added zero-mean complex Gaussian
noise with variance to the corresponding Volterra ker-
which were extracted from an actual Class AB PA. nels , , and used the resulting Volterra

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on November 30, 2009 at 09:53 from IEEE Xplore. Restrictions apply.
 162 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 52, NO. 1, JANUARY 2004

Fig. 6. Parallel Wiener model diagram. H (1) is an LTI block, and F (1) is a
memoryless nonlinear block.

Fig. 4. Effectiveness of predistortion in suppressing spectral regrowth,

when the PA itself is modeled by a memory polynomial. (a) Output without
predistortion. (b) Output with memoryless predistortion. (c) Output with
memory polynomial predistortion (Q = 2, K = 5, odd order). (d) Output
with memory polynomial predistortion (Q = 2, K = 5, even and odd orders).
(e) Original input.

Fig. 7. Effectiveness of predistortion in suppressing spectral regrowth,

when the PA is modeled by a parallel Wiener system. (a) Output without
predistortion. (b) Output with memoryless predistortion. (c) Output with
memory polynomial predistortion (Q = 2, K = 5). (d) Output with memory
polynomial predistortion (Q = 5, K = 5). (e) Original input.

The memoryless nonlinearity in the th branch has input/output

relationship

Fig. 5. Effectiveness of predistortion in suppressing spectral regrowth, when

the PA is modeled by a perturbed Wiener (hence Volterra) system. (a) Output (14)
without predistortion. (b) Output with memoryless predistortion. (c) Output with
memory polynomial predistortion (Q = 2, K = 5). (d) Output with memory
polynomial predistortion (Q = 10, K = 5). (e) Original input.
where and are the input and output of the nonlin-
earity , respectively. The coefficients used were
system (perturbed Wiener) as the model for the PA. The re-
sults are shown in Fig. 5. We still observe significant reduc-
tion in spectral regrowth with the memory polynomial predis-
torter ( , ). With the maximum delay increased to
, the predistorter almost fully suppressed the spectral
regrowth.
(15)
Example 4: The PA here is assumed to follow a three-branch
parallel Wiener model (sum of Wiener subsystems; see Fig. 6). Since , the first branch is actually a memoryless
The LTI blocks in the model are defined by nonlinearity here. This reflects some belief that the dominating
type of nonlinearity in a PA is memoryless. The second and third
branches both exhibit memory nonlinearity, with 10 dB and 13
dB less power than the first branch, respectively.
Fig. 7 shows the performance of our memory polynomial pre-
distorter in linearizing such a PA. With the memory polynomial
(13) predistorter ( , ), there is a significant decrease in

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on November 30, 2009 at 09:53 from IEEE Xplore. Restrictions apply.
 DING et al.: A ROBUST DIGITAL BASEBAND PREDISTORTER CONSTRUCTED USING MEMORY POLYNOMIALS 163

spectral regrowth, and the result is further improved when is

increased to five.
In all of the above cases, memoryless predistortion is not very
effective in suppressing spectral regrowth, which underscores
the notion that PA memory effects must be taken into account
when designing the predistorter.
The objective of PA linearization is two-fold: suppression
of spectral regrowth to reduce adjacent channel interference
and minimization of in-band distortion to improve BER.
Although only PSD plots are shown here, this does not mean
that in-band distortion is left unchecked. Recall that in the
indirect learning architecture, our convergence criterion re- Fig. 8. Parallel Hammerstein model diagram. F (1) is a memoryless nonlinear
quires the mean squared error between and to be block, and H (1) is an LTI block.
minimized. Therefore, at convergence, the PA is linearized,
which automatically ensures the suppression of both in-band is actually more straightforward [c.f., via the least-squares
and out-of-band distortions. The PSD plots are shown for solution (7)].
verification purposes. Because the PSD is phase blind, if one Let
were to define a linearization criterion solely in terms of the
(20)
PSD, the resulting predistorter may not be a true linearizer.
(21)
VI. DISCUSSION where , and denotes convolution. We can rewrite
The Volterra series is the most general polynomial type of the memory polynomial model as
nonlinearity with memory. In this paper, we have encountered
the memory polynomial, the Wiener, the Hammerstein, and the (22)
parallel Wiener models as special cases of the Volterra model.
Next, we would like to point out some interesting links among
these models. (23)
For the memory polynomial model (4), let us collect the co-
efficients in a matrix Therefore, a memory polynomial model is also a parallel Ham-
merstein model (see Fig. 8) where the memoryless nonlinear
block is a polynomial.
.. .. .. (16) In [13], the authors tried to linearize a Wiener system with a
..
. . . . Hammerstein predistorter using the indirect learning architec-
ture. They adopted the least-squares approach1 to solve for the
predistorter coefficients, although the parameters of the memo-
A Hammerstein model on the other hand, can be described by
ryless nonlinear and the LTI blocks of the Hammerstein model
are not explicitly recovered.
(17) Alternatively, we can also rewrite the memory polynomial
model as

(18) (24)

(25)
Let us collect the coefficients in (17) and (18) in vectors
, . Substitution of
(17) into (18) yields (26)

(19) where in (24) denotes convolution. Comparing with the par-

allel Wiener model (see Fig. 6), we observe that a memory poly-
nomial is also a special parallel Wiener model with
Comparing (19) with (4), we can see that the Hammerstein
system (17)–(18) is a special case of the memory polynomial or (27)
model (4) with . In other words, , and In summary, when considering the polynomial type of
hence, rank . This implies that as a predistorter, the nonlinearities, both the parallel Wiener and parallel Ham-
memory polynomial is expected to work well with a Wiener merstein models are special cases of the Volterra series. In
PA. On the other hand, the memory polynomial predistorter is
1There is a typo in [13, eq. (13)]: jx[n]j x [n 0 1] should be
expected to be more robust than the Hammerstein predistorter.
Interestingly, although the memory polynomial model is more
jx[n 0 1]j x [n 0 1]; jx[n]j x [n 0 2] should be jx[n 0 2]j x [n 0 2].
Moreover, we believe that the baseband expression (9) should be in terms of
general than the Hammerstein model, its parameter estimation x[n 0 i]jx[n 0 i]j instead of x [n 0 i].

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on November 30, 2009 at 09:53 from IEEE Xplore. Restrictions apply.
 164 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 52, NO. 1, JANUARY 2004

fact, it can be shown that the memory polynomial model is [12] S. Benedetto and E. Biglieri, “Nonlinear equalization of digital satellite
equivalent to the parallel Hammerstein model. We have also channels,” IEEE J. Select. Areas Commun., vol. SAC-1, pp. 57–62, Jan.
1983.
shown that a memory polynomial model is a special case of the [13] S. Chang and E. J. Powers, “A simplified predistorter for compen-
parallel Wiener model. Obviously, the parallel Hammerstein sation of nonlinear distortion in OFDM systems,” in Proc. IEEE
model includes the Hammerstein model as a special case, and Global Telecommunications Conf., San Antonio, TX, Nov. 2001, pp.
3080–3084.
the parallel Wiener model includes the Wiener model as a
special case. Hammerstein and Wiener models are the most
“specialized” with the least number of coefficients, but are
Lei Ding (S’01) received the B.S. degree in electrical
by no means the easiest to identify. The memory polynomial engineering from the Nanjing University of Aero-
model, however, offers a good compromise between generality nautics and Astronautics, Nanjing, China, in 1997,
and ease of parameter estimation and implementation. and the M.S. degree in biomedical engineering from
the University of Alabama at Birmingham in 2000.
He is currently working toward the Ph.D. degree in
VII. CONCLUSIONS the School of Electrical and Computer Engineering,
Georgia Institute of Technology, Atlanta.
In this paper, instead of trying to find a “good” memory non- During the summers of 2001 and 2002, he worked
as an intern at the Wireless Research Laboratory, Bell
linear PA model and then devising a suitable predistorter for that Laboratories, Lucent Technologies. His research in-
particular model, we have focused directly on predistorter de- terests are in the general areas of signal processing and communications.
sign. In particular, we adopted a memory polynomial structure
for the predistorter and implemented it using an indirect learning G. Tong Zhou (S’92–M’95–SM’00) received
architecture. The predistorter parameters are easy to extract, in- the B.Sc. degree in biomedical engineering and
instrumentation from the Tianjin University, Tianjin,
volving only linear least squares. We demonstrated that such a P.R. China, in July 1989. She was then with the
predistorter is not tied to a particular PA model and is therefore University of Virginia (UVA), Charlottesville, where
robust. The effectiveness of predistortion is demonstrated on a she obtained the M.Sc. degree in biophysics in 1992,
the M.Sc. degree in electrical engineering in 1993,
W–H system, a memory polynomial nonlinearity, a perturbed and the Ph.D. degree in electrical engineering in
Wiener (full Volterra) system, and a parallel Wiener model. 1995.
She has been with the School of Electrical
and Computer Engineering, Georgia Institute of
ACKNOWLEDGMENT Technology (Georgia Tech), Atlanta, since September 1995, and currently
holds the rank of Associate Professor. Her research interests are in the general
The authors also wish to thank R. Raich for many insightful areas of statistical signal processing and communications. Specific current
discussions on the topic of predistortion linearization of PAs interests include predistortion linearization of nonlinear power amplifiers for
wireless applications, communication channel identification and equalization,
with memory effects. and bioinformatics.
In 1997, Dr. Zhou received the National Science Foundation Faculty Early
Career Development (CAREER) Award. She is also recipient of the 2000 Mer-
REFERENCES itor Teaching Excellence Award at Georgia Tech. She was awarded the 1995
Allan Talbott Gwathmey Memorial Award for outstanding research in the phys-
[1] S. C. Cripps, RF Power Amplifiers for Wireless Communica-
ical sciences at UVA based on her Ph.D. dissertation.
tions. Norwood, MA: Artech House, 1999.
[2] J. H. K. Vuolevi, T. Rahkonen, and J. P. A. Manninen, “Measurement
technique for characterizing memory effects in RF power amplifiers,” Dennis R. Morgan (S’63–M’69–SM’92) was
IEEE Trans. Microwave Theory Tech., vol. 49, pp. 1383–1388, Aug. born in Cincinnati, OH, on February 19, 1942. He
2001. received the B.S. degree in 1965 from the University
[3] M. Schetzen, The Volterra and Wiener Theories of Nonlinear Sys- of Cincinnati, Cincinnati, OH, and the M.S. and
tems. New York: Wiley, 1980. Ph.D. degrees in 1968 and 1970, respectively, from
[4] S. A. Maas, Nonlinear Microwave Circuits. Piscataway, NJ: IEEE Syracuse University, Syracuse, NY, all in electrical
Press, 1997. engineering.
[5] C. J. Clark, G. Chrisikos, M. S. Muha, A. A. Moulthrop, and C. P. Silva, From 1965 to 1984, he was with the General Elec-
“Time-domain envelope measurement technique with application to tric Company, Electronics Laboratory, Syracuse, NY,
wideband power amplifier modeling,” IEEE Trans. Microwave Theory specializing in the analysis and design of signal pro-
Tech., vol. 46, pp. 2531–2540, Dec. 1998. cessing systems used in radar, sonar, and communi-
[6] J. Kim and K. Konstantinou, “Digital predistortion of wideband signals cations. He is now a Distinguished Member of Technical Staff with Bell Labo-
based on power amplifier model with memory,” Electron. Lett., vol. 37, ratories, Lucent Technologies, Murray Hill, NJ, where he has been since 1984.
no. 23, pp. 1417–1418, Nov. 2001. From 1984 to 1990, he was with the Special Systems Analysis Department,
[7] H. Ku, M. D. McKinley, and J. S. Kenney, “Extraction of accurate be- Whippany, NJ, where he was involved in the analysis and development of ad-
havior models for power amplifiers with memory effects using two-tone vanced signal processing techniques assiciated with communications, array pro-
measurements,” in IEEE MTT-S Int. Microwave Symp. Dig., Seattle, cessing, detection and estimation, and active noise control. From 1990 to 2002,
WA, June 2002, pp. 139–142. he was with the Acoustics Research Department, Murray Hill, NJ, where he was
[8] C. Eun and E. J. Powers, “A new Volterra predistorter based on the indi- engaged in research on adaptive signal processing techniques applied to elec-
rect learning architecture,” IEEE Trans. Signal Processing, vol. 45, pp. troacoustic systems, including adaptive microphones, echo cancellation, talker
223–227, Jan. 1997. direction finders, and blind source separation. Since 2002, he has been with
[9] L. Ding, R. Raich, and G. T. Zhou, “A Hammerstein predistorter de- the Wireless Research Laboratory, Murray Hill, NJ, where he is involved in re-
sign based on the indirect learning architecture,” in Proc. IEEE Int. search on adaptive signal processing applied to RF and optical communication
Conf. Acoustics, Speech, Signal Processing, Orlando, FL, May 2002, systems. He has authored numerous journal publications, and is coauthor of Ac-
pp. 2689–2692. tive Noise Control Systems: Algorithms and DSP Implementations (New York:
[10] S. Benedetto and E. Biglieri, Principles of Digital Transmission With Wiley, 1996) and Advances in Network and Acoustic Echo Cancellation (New
Wireless Applications. New York: Kluwer Academic/Plenum, 1999. York: Springer-Verlag, 2001).
[11] L. Ding and G. T. Zhou, “Effects of even-order nonlinear terms on pre- Dr. Morgan served as Associate Editor for the IEEE TRANSACTIONS ON
distortion,” in Proc. 10th IEEE DSP Workshop, Pine Mountain, GA, Oct. SPEECH AND AUDIO PROCESSING from 1995 to 2000, and has been Associate
2002, pp. 1–6. Editor for the IEEE TRANSACTIONS ON SIGNAL PROCESSING since 2001.

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on November 30, 2009 at 09:53 from IEEE Xplore. Restrictions apply.
 DING et al.: A ROBUST DIGITAL BASEBAND PREDISTORTER CONSTRUCTED USING MEMORY POLYNOMIALS 165

Zhengxiang Ma received the B.S. degree in physics Jaehyeong Kim received the BSEE and MSEE de-
from University of Science and Technology, Hefei, grees from Seoul National University, Seoul, Jorea,
China, in 1989, and the M.S. and Ph.D degrees in in 1988 and 1990, respectively, and the Ph.D. degree
applied physics from Stanford University, Stanford, in electrical engineering from the University of Cali-
CA, in 1991 and 1995, respectively. fornia, Los Angeles, in 1996.
He joined Bell Labs, Lucent Technologies in He is currently with Wireless Advanced Tech-
1995, where he is currently a Distinguished Member nology Laboratory, Lucent Technologies, Whippany,
of Technical Staff in the Wireless Research Labora- NJ. His interests are in channel coding and mod-
tory, Murray Hill, NJ. His research interests include ulation, signal processing for CDMA system,
novel digital signal processing in high-performance power amplifier predistortion, and wireless system
basestation radio, power amplifier linearization, and architecture for high-speed packet data service.
wireless networking infrastructure architecture. He has coauthored several
technical papers and holds six patents.

Charles R. Giardina was born in the Bronx, NY, on December 29, 1942. He
received the B.S. degree in mathematics from Fairleigh Dickinson University,
Rutherford, NJ, and the M.S. degree in mathematics from Carnegie Institute of
J. Stevenson Kenney (S’84-M’85-SM’01) was born
Technology, Pittsburgh, PA. He also received the M.E.E. degree in 1969, and the
in St. Louis, MO in 1962. He received the B.S.E.E.
Ph.D. degree in mathematics and electrical engineering in 1970 from Stevens
degree (Hons.) in 1985, the M.S.E.E. degree in 1990,
Institute of Technology, Hoboken, NJ.
and the Ph.D. degree in electrical engineering in
He was Professor of Mathematics, Electrical Engineering, and Computer Sci-
1994, all from the Georgia Institute of Technology
ence at Fairleigh Dickinson University from 1965 to 1982. From 1982 to 1986,
(Georgia Tech), Atlanta.
he was a Professor at the Stevens Institute of Technology. From 1986 to 1996,
In January 2000, he joined the Georgia Institute of
he was a Professor at the College of Staten Island, City University of New York.
Technology, where he is currently an Associate Pro-
Since 1996, he has been with Lucent Technologies, Whippany, NJ. His research
fessor in Electrical and Computer Engineering. He
interests include digital signal and image processing, pattern recognition, ar-
is currently teaching and conducting research in the
tificial intelligence, and the constructive theory of functions. He has authored
areas of power amplifier linearization, smart antenna
numerous papers in these areas, and several books on image processing, signal
design, and RFIC design. He also has over 14 years of industrial experience in
processing, and artificial intelligence.
wireless communications. He has held engineering and management positions
at Electromagnetic Sciences, Scientific Atlanta, and Pacific Monolithics. Prior
to returning to Georgia Tech, he was Director of Engineering at Spectrian Corp.,
Sunnyvale, CA. He has authored or co-authored more than 50 technical papers,
conference papers, and workshop presentations in the areas of acoustics, micro-
electronics, microwave design, and telecommunications. He is currently serving
as Cochair of the RF Components technical interest group of the National Elec-
tronics Manufacturing Initiative. In 2002, he was the Technical Program Com-
mittee Cochair for the 2002 Radio and Wireless Conference (RAWCON), and
is currently serving as the General Cochair for the 2003 RAWCON.
Dr. Kenney has been an active member of IEEE Microwave Theory and Tech-
niques Society for 20 years. He served as an officer on the Santa Clara Valley
chapter of MTT-S from 1996–2000. He is currently serving his second term in
the MTT-S AdCom, and was appointed to the office of Treasurer for 2001–2003.
He served on the IMS Steering Committee in 1993 and 1996. He has served on
the Editorial Board for the IEEE TRANSACTIONS ON MICROWAVE THEORY AND
TECHNIQUES and the Microwave and Wireless Components Letters, and serves
on the IMS Technical Program Committee.

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on November 30, 2009 at 09:53 from IEEE Xplore. Restrictions apply.