IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 50, NO.

5, MAY 2002 799

A Reduced Complexity Channel Estimation for

OFDM Systems With Transmit Diversity in
Mobile Wireless Channels
Hlaing Minn, Student Member, IEEE, Dong In Kim, Member, IEEE, and Vijay K. Bhargava, Fellow, IEEE

Abstract—A reduced complexity channel estimation for OFDM The channel estimation for OFDM systems without transmit
systems with transmit diversity is proposed by exploiting the corre- diversity has been studied by many researchers (e.g., [8]–[10]).
lation of the adjacent subchannel responses. The sizes of the matrix However, for systems with transmit diversity, the received signal
inverse and the FFTs required in the channel estimation at every
OFDM data symbol are reduced by half of the existing method for is a superposition of the different transmitted signals from all
OFDM systems with nonconstant modulus subcarrier symbols or transmit antennas and consequently, the channel estimation be-
constant modulus subcarrier symbols with some guard tones. The comes more complicated. Recently, the channel estimation for
complexity reduction of half FFTs size and some matrix multipli- OFDM systems with transmit diversity using space–time coding
cations is still achieved for constant modulus subcarrier symbols has been proposed in [1]. In order to reduce the complexity as-
with no guard tones. The price for the complexity reduction is a
slight BER degradation and for the channels with small relative sociated with the matrix inverse operation, [1] also proposed a
delay spreads, the BER performance of the reduced complexity simplified approach. In this paper, we focus on reduced com-
method becomes quite comparable to the existing method. An al- plexity channel estimation for OFDM systems with transmit di-
ternative approach for the number of significant taps required in versity.
the channel estimation is described which achieves a comparable
We propose a channel estimation method which has less com-
performance to the case with the known suitable number of signif-
icant taps. A simple modification which reduces the lost leakage of plexity than the simplified method of [1]. By decoupling the ef-
the nonsample-spaced channel paths is also proposed. This modi- fect of different transmit antennas, the sizes of the matrix in-
fication achieves a substantial performance improvement over the verse and FFTs required in the channel estimation for every
existing method without any added complexity. OFDM data symbol are reduced by half. The significant tap
Index Terms—OFDM, reduced complexity channel estimation, catching approach of [1] requires the knowledge of the number
transmit diversity. of significant taps. In this paper, we describe an alternative ap-
proach which adaptively finds this number. Moreover, a simple
I. INTRODUCTION modification which reduces the channel energy leakage lost in
the channel estimation of both [1] and the reduced complexity

O NE OF the desirable features of OFDM is its robustness

to the multipath induced intersymbol interference. On the
other hand due to the frequency selective fading of the disper-
method is also proposed. This modified approach achieves a
substantial performance improvement.
The rest of the paper is organized as follows. In Section II, the
sive wireless channel, some subchannels may face deep fades system and the channel considered are described. In Section III,
and degrade the overall system performance. In order to com- the reduced complexity channel estimation is derived and adap-
pensate the frequency selectivity, techniques such as error cor- tively finding the number of significant taps is described. The
recting code and diversity have to be used [1]–[3]. Transmit di- performance in terms of the channel estimation mean square
versity for wireless systems has been studied in research litera- error (MSE) and the complexity comparison are presented in
ture (e.g., [4], [5]). For OFDM systems, transmit diversity com- Section IV. The energy leakage of the channel path with non-
bined with Reed–Solomon code has been proposed for clustered sample-spaced delay is investigated in Section V. Based on this,
OFDM in [2]. Recently, space–time coding [6] has been shown a modified approach is proposed for further improvement. Sim-
to give high code efficiency and good performance. Application ulation results are discussed in Section VI and conclusions are
of space–time coding in OFDM systems has been studied in [7] given in Section VII.
with perfect channel knowledge at the receiver.

Paper approved by Y. Li, the Editor for Wireless Communications Theory of

the IEEE Communications Society. Manuscript received July 13, 2000; revised II. SYSTEM AND CHANNEL DESCRIPTION
July 31, 2001. This work was supported in part by a Strategic Project Grant from
the Natural Sciences and Engineering Research Council (NSERC) of Canada The considered OFDM system is the same as in [1, Fig. 1]
and in part by Telus Mobility, Canada. The work of D. I. Kim was supported by with two-branch transmit diversity and two-branch receive
the Research Fund of University of Seoul, Korea.
H. Minn and V. K. Bhargava are with the Department of Electrical and diversity. At a transmission time , a binary data block
Computer Engineering, University of Victoria, Victoria, BC V8W 3P6, Canada is space–time-coded (in fact, space–fre-
(e-mail: [email protected]). quency-coded) into two blocks of frequency-domain subcarrier
D. I. Kim is with the University of Seoul, Department of Electrical Engi-
neering, Dong Dae Moon-gu, Seoul 130-743, Korea. symbols, for , which are
Publisher Item Identifier S 0090-6778(02)05103-6. simultaneously transmitted from the two antennas.
0090-6778/02$17.00 © 2002 IEEE
800 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 50, NO. 5, MAY 2002

The DFT output frequency-domain subcarrier symbols from dently performed for each receive antenna. The receive antenna
receive antenna can be expressed as index will be omitted in the following. The frequency-domain
subcarrier symbols from each receive antenna can be expressed
(1) as

(6)
where is the additive complex Gaussian noise with zero
mean and variance , on the th receive antenna, that is uncor-
related for different s, s, or s, and is the channel Due to the limited delay spread of the channel, the channel
frequency response for the th tone at time , corresponding subcarrier responses are correlated. Generally, the smaller the
to the th transmit antenna and the th receive antenna. For a ratio of the channel delay spread to the OFDM symbol interval,
system with -transmit diversity, the average SNR at the re- the more correlated the adjacent subcarrier responses are. In this
ceiver is defined as paper, we attempt to find a reduced complexity channel estima-
tion by exploiting the correlation of the subcarrier responses. In
particular, we will assume in the derivation of the reduced com-
SNR (2) plexity channel estimation that

(7)
where is assumed. In our simulation, the channels
are modeled such that . The above assumption would be a good approximation for the
The channel impulse response of the mobile wireless channel case with a very small ratio of channel delay spread to OFDM
can be given by symbol interval. We will also investigate the applicability of this
assumption for different channels with different delay spreads.
(3) Let us define the following:

where is the delay of the th path, is the corresponding (8)

complex gain, is the Dirac Delta function, and the antenna
indexes are omitted for simplicity. The frequency response
(9)
at time is

(10)

(4)
(11)

The path gains s are modeled as independent wide-sense With (7), we obtain
stationary (WSS), narrowband complex Gaussian processes
with the time-domain correlation defined by the classical
Doppler spectrum. The frequency-domain correlation of the (12)
channel is defined by the channel power delay profile.
With tolerable leakage, the channel frequency response can where
be expressed as [1]
(13)
(5)
(14)
where , , is the
From (12), it can be observed that the channel responses
total number of sample-spaced channel taps, is the number of
corresponding to different transmit antennas are decoupled.
tones (including guard tones), and are the total OFDM
Hence, the channel estimation can be performed independently
symbol interval and the tone spacing of the OFDM system, re-
for each transmit–receive antenna pair by minimizing the
spectively, and .
following mean square error (MSE) cost function:
The considered channel power delay profile models are
two-ray (equal average power on each ray), GSMs typical
urban (TU) (6 taps), GSMs hilly terrain (HT) (6 taps), and JTCs
indoor office channel model A (3 taps) [11].
(15)
III. REDUCED COMPLEXITY CHANNEL ESTIMATION
Since the channels considered have independent responses to where , and are trial values for the MMSE
different receive antennas, the channel estimation is indepen- estimates of .
MINN et al.: CHANNEL ESTIMATION FOR OFDM SYSTEMS 801

Solving the following degree for different time instants. This prompts a question on
the applicability of this significant tap catching approach in a
fast time-varying channel. However, the results in [1] and also
in this paper show that even with a Doppler frequency of 200
Hz for the OFDM system with 200 s symbol interval and 10%
training, the significant tap catching approach works well.
(16)
An alternative approach in the significant tap catching rather
than using fixed value of is described in the following. Sup-
where and denote the real and imaginary part of a pose the maximum value of for which the complexity can be
complex number, respectively, and , re- afforded be . Let be the sorted version of
sults in in descending order, and be the number of taps satisfying the
following condition:

(26)
(17)
where are the MMSE estimates. Then, the value of is chosen as the minimum of and .
Define In the above equation, is a threshold value. Although
is applicable, it was observed in our investigation that a little
(18) larger value (say, 3 or 4) gives more robustness to the noise.
The threshold value depends on the SNR value and is chosen
according to
(19) SNR
(27)
SNR
Then, (17) can be expressed as
where is for SNR . We observed that the value of about the
range of 0.01 to 0.04 works well for a SNR value of 10 dB. This
(20)
threshold value can be set according to the designed received
SNR or by means of the sync detection metric such as received
for and . It can be expressed in power measurement. Due to the varying value of at every
matrix form as training symbol, this approach will be denoted by “adaptive” in
(21) contrast to the fixed value of . For channel environments where
a suitable fixed value of is unknown, this adaptive approach
where would be an alternative. It is noted that in the adaptive approach,
the value of need not be the same for the different transmit
antennas.
.. .. .. ..
. . . . IV. PERFORMANCE ANALYSIS
A. MSE Performance
(22) In the previous section, a reduced complexity channel estima-
(23) tion for OFDM systems with transmit diversity is proposed by
assuming that (7) holds. In this section, the MSE performance
(24) of the proposed method in a multipath fading channel not satis-
fying (7) is analyzed. For simplicity, the time index is omitted
Hence, the channel impulse response can be estimated by
in the following analysis. Define
(25)
Following [1], the complexity involved in can be further
reduced and the channel estimation performance can be further (28)
improved by using only significant channel taps whose in-
dexes are denoted by : ;
. The value of generally depends on (29)
the number of paths and the path delays, both of which are typi-
cally defined by the terrain environment. In [1], it is reported that Then,
the value of typically ranges between 5 and 9 for 2-ray, TU, (30)
and HT channels. Due to the time-varying nature of the mobile
wireless channel, the effective number of (sample-spaced) sig- (31)
nificant channel taps and/or their tap indexes can vary to some
802 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 50, NO. 5, MAY 2002

where By defining

(32)
.. .. .. ..
. . . .
(33)
(42)
In matrix form,
we can express that
(34)
(43)
where
For constant modulus subcarrier symbols ( are
constant), if are constant, and
become scaled identity matrices and the diagonal elements
of become :
where is the
effective sample-spaced power delay profile of the channel
corresponding to the transmit antenna . One criterion that
The channel impulse response estimate corresponding to satisfies the above condition is given by
transmit antenna 1 is given by which is the same as the optimal training design for [1]. Under
this condition, the MSE (36) of the reduced complexity method
(35)
becomes
The MSE of the channel impulse response estimate can be
given by MSE (44)

MSE where the index from is omitted for simplicity since gen-
erally all channels have the same power delay profile.
Although the above MSE analysis is derived for , it also
holds for other channels and hence is the average MSE. The
second term in (44) is the same as the MSE for [1] at the op-
(36) timal condition. Hence, the reduced complexity method has a
larger channel estimation MSE than [1] due to the assumption
of (7). The extra term in the MSE expression also indicates the
The elements of are given by
dependence of the reduced complexity channel estimation MSE
performance on the channel power delay profile. Particularly,
the larger delay taps have more impact on the MSE. This fact is
intuitively justified since the larger delay taps would cause more
frequency selectivity of the channel.
It should be mentioned that the above MSE expressions for
(37) [1] and the reduced complexity method are derived assuming
that the channel gains remain constant over one OFDM symbol,
with and all subchannels are used (i.e., without guard tones). If the
channel gain varies during one OFDM symbol, there may be
(38) some degradation due to the inter-subchannel interference (ICI).
If guard tones are used, of [1] would not be exactly a diag-
onal matrix, and there may be some degradation, too.
(39) From (44), the channel estimation MSE mainly depends on
which in turn indicates that the channel estimation
MSE performance evaluation by computer simulation depends
The elements of are given by on how to model . For a SNR value and fixed
(40) , if are modeled very small,
will be very small and consequently, the obtained channel esti-
where mation MSE will be very small. Hence, a suitable simulation
model for channel estimation performance evaluation should
be adopted. There are two possible approaches: the first one
is to model , and the second is to
(41) model . Due to the similar reasoning of
MINN et al.: CHANNEL ESTIMATION FOR OFDM SYSTEMS 803

TABLE I V. FURTHER IMPROVEMENT ON CHANNEL ESTIMATION

CHANNEL ESTIMATION COMPLEXITY
In previous sections, the channel response is esti-
mated based on the sample-spaced impulse response
. Generally, the channel path delays
are not sample-spaced which causes the channel energy leakage
to other sample-spaced taps besides the adjacent sample-spaced
taps. In the following, the effect of nonsample-spaced path
delay on the channel estimation is investigated.
By using (4), the sample-spaced channel tap gains can be
given by

(45)

where , are the contribution of the unity

the MSE’s dependence on channel power gain and to be com-
gain channel path with delay over the sampled-spaced taps
parable with different number of diversity branches, the second
and given by
approach of is adopted in this paper.

B. Channel Estimation Complexity

For channel estimation based on the training symbol, the re- if integer
quired matrix inverse can be pre-computed. Hence, the discus-
(46)
sion on the complexity will be focused on the channel estimation
based on the decision-directed reference symbol. The number of
otherwise
complex multiplications will be used as a performance measure.
Operations such as -point FFT and matrix inver- The plots of : for
sion have complexity order of and , respectively. and are shown in Fig. 1(a) and (b), respectively.
However, the exact complexity can vary depending on their im- Firstly, the effect of different values on a sample-spaced tap
plementation. Hence, we will use FFT and , respec- will be discussed. Consider the range with
tively in the complexity expression rather than the number of being an integer. As can be expected, the energy distribution
complex multiplications for them. For a two transmit antenna to tap is larger if is closer to . The same holds for tap
space–time coded OFDM system with total subcarriers and . To the other taps away from these two adjacent taps and
used subcarriers, Table I lists the complexity measures of , the largest energy distribution (energy leakage) occurs at
[1] and the reduced complexity method (denoted by RC), both . In other words, the energy leakage to not-nearby
using significant taps. The significant tap catching is indepen- taps is maximum when .
dently performed at each receive antenna and its complexity is Secondly, the energy distribution of a unity gain path with
not included in the complexity expression. It is noted that if the delay over the sample-spaced taps
significant tap catching is performed by averaging over different is discussed. As can be expected, the larger energy is dis-
receive antennas, the number of matrix inverse for the two re- tributed to the nearer tap. However, since ,
ceive antenna case will be halved for all methods. the path with near 0 will also cause leakage to the taps
For constant modulus subcarrier symbols with no guard where tap will get the largest leakage
tones, the matrix inversion required in [1] can be among them. Consequently, the channel estimation based on
reduced to a matrix inversion and some matrix multi- will lose some leakage energy es-
plications (see [1, eq. 24a]). Due to the spectrum requirement, pecially from taps . From Fig. 1(a) and (b),
OFDM systems generally use some guard tones. In this case, it can be observed that if can be intentionally increased by
the above simplification [1, eq. 24a] does not hold. From some amount, the leakage to those taps
Table I, it can be observed that RC method achieves complexity can be substantially reduced and the channel estimation can be
reduction for all conditions. More complexity reduction is improved. A simple way of establishing this is to use a pre-ad-
observed for the cases of nonconst modulus subcarrier symbols vancement of the timing point. A timing pre-advancement of
and constant modulus subcarrier symbols with guard tones. It samples will effectively increase to . The value
is also noted that the size of FFT for RC is while that for of should not only be less than the ISI-free guard interval but
[1] is for all conditions. Moreover, the matrix inverse size also be small enough that most channel energy is contained in
for RC is half of that for [1] for all conditions except constant the . If necessary, the value of
modulus subcarrier symbols with no guard tones. may be increased in account for the timing pre-advancement. By
804 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 50, NO. 5, MAY 2002

(a)

Fig. 2. The worst-case leakage of a unity gain channel path with

nonsample-spaced path delay.

(b)
Fig. 1. The leakage of a unity gain channel path with nonsample-spaced path
delay. l is the index of sample-spaced channel taps,  is the normalized delay
of the k th path.

Fig. 3. Performance of the channel estimation methods in OFDM with

using (46), a worst-case leakage may be found over . 4-PSK, 16 states space–time code. The rms delay spread is 1.06 s, the
Doppler frequency is 40 Hz.
Fig. 2 shows this worst-case leakage for an OFDM system with
subcarriers. Based on this, may be chosen for a de-
signed SNR value. For example, the use of will include In the following, the performance degradation associated
all leakage down to approximately 23 dB of the total energy with the complexity reduction is investigated. For the channel
of the path with in the channel estimation. estimation based on the training symbol, [1] is also used in
the reduced complexity method. The different methods are
applied only for the channel estimation based on the decision
VI. SIMULATION RESULTS AND DISCUSSIONS
directed reference symbols. Figs. 3–7 show the BER and MSE
The OFDM system parameters used in the simulation are performance for 4-PSK, 16 states space–time code in different
the same as [1]: 128 subchannels, 4 guard-subchannels on each channel models with different delay spreads and Doppler
end, 1/160 - s subchannel spacing, and 40 - s guard interval. frequencies. For the same delay spread, two-ray model shows
The channel models considered are two-ray and TU channel a better performance due to the less channel energy leakage.
models with delay spread of 1.06 s, two-ray and HT channel The channels with larger delay spread may not necessarily
models with delay spread of 5.04 s, and JTC channel model show a larger MSE. For example, the MSE of [1] for two-ray
with delay spread of 34.8 ns. The classical Doppler spectrum model with delay spread of 1.06 s is larger than that with
with the maximum frequency of 40 Hz and 200 Hz are consid- delay spread of 5.04 s. The reason can be ascribed to the
ered. The space–time codes used are of 16 states with 4-PSK and much larger channel energy leakage of the former to the taps
16-QAM [6]. The number of significant taps used in all channel ( ) (see previous Section). By comparing
estimation methods is 7. the MSE in the JTC, TU and HT models, it is observed that
MINN et al.: CHANNEL ESTIMATION FOR OFDM SYSTEMS 805

Fig. 4. Performance of the channel estimation methods in OFDM with Fig. 6. Performance of the channel estimation methods in OFDM with
4-PSK, 16 states space–time code. The rms delay spread is 5.04 s, the 4-PSK, 16 states space–time code. The rms delay spread is 5.04 s, the
Doppler frequency is 40 Hz. Doppler frequency is 200 Hz.

Fig. 5. Performance of the channel estimation methods in OFDM with Fig. 7. Performance of the channel estimation methods in OFDM with 4-PSK,
4-PSK, 16 states space–time code. The rms delay spread is 1.06 s, the 16 states space–time code. The rms delay spread is 34.8 ns.
Doppler frequency is 200 Hz.

mation errors on BER performance becomes more dominant at

at the same MSE level, RC method has more degradation in higher SNR values.
the channel with larger delay spread. This is due to the fact Another observation is that when comparing performance be-
that the larger delay spread channel is more deviated from the tween JTC and other models, although JTC model has a larger
assumption (7). For the same delay spread, RC method has MSE gap between [1] and RC than the other models at the same
more degradation in two-ray, especially if compared with HT SNR value, it has a smaller BER gap between [1] and RC than
model. The reason can be deduced from (44). Although the the other models. This indicates that if the MSE is much smaller
maximum delay spread of HT is larger than that of two-ray, the than the SNR value, (say, a difference of 6 dB or more), provided
last path’s power is larger in two-ray than HT. This causes a that the ICI is negligible, then further reduction in MSE would
larger value of the first term in (44), hence a larger MSE. not improve the BER performance very much.
When comparing for different Doppler frequencies, (more or From the above investigation, we can observe that the cost of
less) a larger MSE degradation for the higher Doppler frequency RC method’s complexity reduction is just a slight BER degrada-
is observed at lower SNR values in all channel models. This tion for all SNR values in a channel with very small delay spread
may be ascribed to a larger reference symbol error caused by such as JTC model. For channels with larger delay spreads such
the larger time selectivity of the channel with a larger Doppler as TU and HT, a small BER degradation is observed for low
spread, in addition to the low SNR condition and ICI. However, to moderate SNR values. Hence, if the complexity is afford-
on the contrary, a larger BER degradation is observed at higher able, [1] is a better solution. If the complexity is of concern,
SNR values. This indicates that the effect of the channel esti- RC method can be an alternative.
806 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 50, NO. 5, MAY 2002

Fig. 8. Performance of the channel estimation methods in OFDM with Fig. 10. Performance of the modified channel estimation methods in TU
16-QAM, 16 states space–time code. The rms delay spread is 1.06 s for channel with the Doppler frequency of 40 Hz for OFDM with 4-PSK, 16 states
two-ray and TU, 34.8 ns for JTC, the Doppler frequency is 40 Hz. space–time code.

are and . This adaptive approach has almost

the same performance as the case with the knowledge of
suitable fixed number of significant taps. For low SNR value,
the adaptive approach has a slight performance improvement
since the adaptive selection would quite often choose a smaller
number of taps and hence would suppress the effect of the
dominant noise to a larger extent. The results also show that
threshold value around the range of 0.01 to 0.04 at a SNR of
10 dB works well. It is also noted that if in the approach with
fixed value and in the approach with adaptive value are
the same, the adaptive approach saves some complexity.
As discussed in the previous section, both [1] and RC
methods can be modified by introducing some timing pre-ad-
vancement of samples. The results of this modified approach
with as an example are presented in Fig. 10 for TU
channel model with a Doppler frequency of 40 Hz. The results
clearly show that for moderate to high SNR values, the modi-
Fig. 9. Performance of the channel estimation methods with adaptive fied approach achieves quite substantial improvement without
significant tap selection (J= 7;  = 4) in two-ray and TU channels with any added complexity. This is a result of less unused leakage
delay spread of 1.06 s, and the Doppler frequency of 40 Hz for OFDM with
4-PSK, 16 states space–time code. energy of the modified approach in the channel estimation.

In Fig. 8, the performances for 16-QAM, 16 states space–time VII. CONCLUSION

coded OFDM system in two-ray, TU, and JTC channel models In this paper, we have investigated a reduced complexity
are presented. A higher MSE gap between [1] and RC is ob- channel estimation for an OFDM system with space–time
served for 16-QAM than for 4-PSK case. The reason is that due coding in time-varying, dispersive multipath fading channels.
to the nonconstant modulus subcarrier symbols, the MMSE cri- In particular, the motivation was to reduce the complexity in
terion in RC method loses its optimality in the maximum likeli- the matrix inversion needed for every OFDM data symbols.
hood (ML) sense while [1] still holds the ML optimality. How- The method is developed based on a channel with relatively
ever, in terms of BER performance, only a small BER gap is ob- small delay spread. By decoupling the channel responses
served. This indicates that RC method can be an alternative ap- from different transmit antennas, much complexity reduction
proach also for nonconstant modulus subcarrier symbols when is achieved. In particular, the sizes of the matrix inverse and
the complexity is an issue. the FFTs required in the channel estimation for every OFDM
In Fig. 9, the performance of the adaptive selection of the data symbol is reduced by half. The simulation results show
number of significant taps is presented for two-ray and TU that the price for the complexity reduction is just a slight
channel models with delay spread of 1.06 s, and the Doppler BER degradation for channels with relatively small delay
frequency of 40 Hz. The threshold values used at a SNR of 10 spreads. For cases where the complexity is of a major concern,
dB, denoted by , are 0.04 and 0.01. The other parameters the reduced complexity method can be considered as a good
MINN et al.: CHANNEL ESTIMATION FOR OFDM SYSTEMS 807

alternative. Adaptively finding the number of significant Dong In Kim (S’89–M’91) received the B.S. and
taps based on a threshold decision is also shown to have a M.S. degrees in electronics engineering from Seoul
National University, Seoul, Korea, in 1980 and
comparable performance to the case where the suitable number 1984, respectively, and the M.S. and Ph.D. degrees
of significant taps is known. We have also investigated the in electrical engineering from the University of
effect of the channel path with nonsample-spaced delay on the Southern California (USC), Los Angeles, in 1987
and 1990, respectively.
channel estimation based on the sample-spaced channel taps. A From 1984 to 1985, he was with the Korea
simple modification by means of the timing pre-advancement Telecommunication Research Center as a Re-
is proposed in order to reduce the energy leakage of the paths searcher. During 1986–1988, he was a Korean
Government Graduate Fellow with the Department
with nonsample-spaced delays lost in the channel estimation. of Electrical Engineering, USC. From 1988 to 1990, he was a Research
This modified approach brings about a substantial improvement Assistant with the USC Communication Sciences Institute. Since 1991, he
without any added complexity. has been with the University of Seoul, Seoul, Korea, where he is currently
an Associate Professor with the Department of Electrical and Computer
REFERENCES Engineering, leading the Wireless Communications Research Group. He was
a Visiting Professor with the University of Victoria, Victoria, BC, Canada,
[1] Y. Li, N. Seshadri, and S. Ariyavisitakul, “Channel estimation for during 1999–2000. He has given many short courses and lectures on the topics
OFDM systems with transmitter diversity in mobile wireless channels,” of spread-spectrum and wireless communications at several companies. He has
IEEE J. Select. Areas Commun., vol. 17, pp. 461–470, Mar. 1999. performed research in the areas of packet radio networks and spread-spectrum
[2] L. J. Cimini, Jr., B. Daneshrad, and N. R. Sollenberger, “Clus- systems since 1988. His current research interests include spread-spectrum
tered OFDM with transmitter diversity and coding,” IEEE Global systems, cellular mobile communications, indoor wireless communications,
Telecommun. Conf., pp. 703–707, Nov. 1996. and wireless multimedia networks.
[3] L. J. Cimini and N. R. Sollenberger, “OFDM with diversity and coding Dr. Kim has served as an Editor for the IEEE JOURNAL ON SELECTED
for high-bit-rate mobile data applications,” in IEEE Global Telecommun. AREAS IN COMMUNICATIONS —Wireless Communications Series, and also a
Conf., Phoenix, AZ, Nov. 1997, pp. 305–309. Division Editor for the Journal of Communications and Networks. He serves
[4] A. Wittneben, “A new bandwidth efficient transmit antenna modula- as an Editor for the IEEE TRANSACTIONS ON COMMUNICATIONS the the IEEE
tion diversity scheme for linear digital modulation,” in IEEE Int. Conf. TRANSACTIONS ON WIRELESS COMMUNICATIONS.
Commun., Geneva, Switerland, June 1993, pp. 1630–1634.
[5] J. H. Winters, “The diversity gain of transmit diversity in wireless sys-
tems with Rayleigh fading,” in IEEE Int. Conf. Commun., Chicago, IL,
June 1994, pp. 1121–1125.
[6] V. Tarokh, N. Seshadri, and A. R. Calderbank, “Space–time codes for
high data rate wireless communication: Performance analysis and code
construction,” IEEE Trans. Inform. Theory, vol. 44, pp. 744–765, Mar.
1998.
[7] D. Agarwal, V. Tarokh, A. Naguib, and N. Seshadri, “Space–time coded
OFDM for high data rate wireless communication over wideband chan-
nels,” in IEEE Vehic. Technol. Conf., Ottawa, Canada, May 1998, pp.
2232–2236.
[8] J.-J. van de Beek, O. Edfors, M. Sandell, S. K. Wilson, and P. O.
Borjesson, “On channel estimation in OFDM systems,” in IEEE Vehic.
Technol. Conf., Chicago, IL, July 1995, pp. 815–819.
[9] O. Edfors, M. Sandell, J.-J. van de Beek, S. K. Wilson, and P. O. Bor-
jesson, “OFDM channel estimation by singular value decomposition,” Vijay K. Bhargava (S’70–M’74–SM’82–F’92)
IEEE Trans. Commun., vol. 46, pp. 931–939, July 1998. received the B.Sc., M.Sc., and Ph.D. degrees from
[10] Y. Li, L. J. Cimini, Jr., and N. R. Sollenberger, “Robust channel estima- Queen’s University, Kingston, Canada in 1970,
tion for OFDM systems with rapid dispersive fading channels,” IEEE 1972, and 1974, respectively.
Trans. Commun., vol. 46, pp. 902–915, July 1998. Currently, he is a Professor of Electrical and
[11] K. Pahlavan and A. H. Levesque, Wireless Information Networks. New Computer Engineering with the University of
York: Wiley, 1995, ch. 6. Victoria and holds a Canada Research Chair in
Wireless Communications. He is a coauthor of the
book Digital Communications by Satellite (New
York: Wiley, 1981) and coeditor of Reed–Solomon
Hlaing Minn (S’99) received the B.E. (electronics) Codes and Their Applications (New York: IEEE
degree from Yangon Institute of Technology, Press). His research interests are in multimedia wireless communications.
Myanmar, in 1995, and the M.Eng. (telecommuni- Dr. Bhargava is a Fellow of the B.C. Advanced Systems Institute, Engineering
cations) degree from Asian Institute of Technology Institute of Canada (EIC), the IEEE and the Royal Society of Canada. He is a
(AIT), Thailand, in 1997. recipient of the IEEE Centennial Medal (1984), IEEE Canada’s McNaughton
During 1998, he was with the Telecommunica- Gold Medal (1995), the IEEE Haraden Pratt Award (1999), the IEEE Third Mil-
tions Program of AIT as a laboratory supervisor. lennium Medal (2000), and the IEEE Graduate Teaching Award (2002). He is
Since 1999, he has been a research assistant with the very active in the IEEE and has been nominated by the IEEE Board of Director
Department of Electrical and Computer Engineering, for the Office of IEEE President-Elect in this year’s election. Currently he serves
University of Victoria, Canada. Currently, he is on the Board of the IEEE Information Theory and Communications Societies.
completing the Ph.D. degree in electrical engi- He is an Editor for the IEEE TRANSACTIONS ON COMMUNICATIONS and the IEEE
neering at the University of Victoria. His research interests include wireless TRANSACTIONS ON WIRELESS COMMUNICATIONS. He is a Past President of the
communications, signal processing, and error control coding. IEEE Information Theory Society.