IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO.

2, MARCH 2004 647

Blind OFDM Channel Estimation Through

Simple Linear Precoding
Athina Petropulu, Ruifeng Zhang, Member, IEEE, and Rui Lin

Abstract—A novel approach of blind channel estimation for of the last IDFT samples, called cyclic prefix (CP), is ap-
orthogonal frequency-division multiplexing (OFDM) systems is pended in front of the IDFT block. The augmented blocks are
proposed. A linear transformation is applied on each block before sent one after the other through the communications channel.
it enters the OFDM system. The transform imposes a correla-
tion structure on the transmitted blocks, which is exploited at As long as the channel length is smaller than the length of the
the receiver to recover the channel via simple cross-correlation cyclic prefix, the spreading of the block, caused by the convo-
operations. The proposed approach is computationally simple and lution with the channel, affects less than symbols. Thus,
converges fast, which makes it a good candidate for estimation interference occurs between two adjacent blocks only. At the re-
of fast-varying channels. Its performance is tested analytically, ceiver, the cyclic prefix part is discarded, and an -point DFT is
through a mean-square error analysis, and also via simulations.
Results show that it compares favorably to the training-based performed on the remaining -symbol segment of the block. It
scheme used in the IEEE 802.11a wireless standard. can be shown that the received symbol at the th carrier equals
the transmitted one scaled by the channel frequency response
Index Terms—Blind channel estimation, orthogonal frequency-
division multiplexing (OFDM) systems, precoding. at frequency . Thus, multicarrier modulation armed with a
cyclic prefix effectively turns the wideband frequency-selective
channel into a number of parallel narrowband frequency-flat
I. INTRODUCTION subchannels.
Although ISI can be avoided, via the use of cyclic prefix,
I T IS well understood that the communication channel sets a
limit to the transmission rate of the communication system.
Signal pulses are broadened in time as they travel through the
the phase and gain of each subchannel is needed for coherent
symbol detection. An estimate of these parameters can be ob-
channel (multipath effect), leading to intersymbol interference tained with pilot/training symbols [8], [14], [19], [20], [26], at
(ISI). The pulse spread limits the speed at which adjacent data the expense of bandwidth. Blind channel estimation methods
pulses can be sent without overlap, thus limiting the maximum avoid the use of pilot symbols, which makes them good can-
information rate of the wireless system. One technique to avoid didates for achieving high spectral efficiency. Existing blind
the effect of multipath, without sacrificing the transmission channel estimation methods for OFDM systems can be classi-
rate, is multicarrier modulation [29]. This is a parallel trans- fied as statistical and deterministic. The statistical methods ex-
mission scheme, where the overall frequency band is divided plore the cylcostationarity that the cyclic prefix induces to the
into a number of subbands with separate subcarriers. On each transmitted signal. They recover the channel using cyclic sta-
subcarrier, the modulated symbol rate is low in comparison to tistics of the received signal [16] or subspace decomposition
the channel delay spread, thus ISI can be prevented. An increas- of the correlation matrix of the pre-DFT received blocks [11],
ingly popular multicarrier modulation technique is orthogonal [27], [32]. The deterministic methods process the post DFT re-
frequency division multiplexing (OFDM) [6], [9], [13]. It has ceived blocks and exploit the finite-alphabet property of the in-
been implemented in several wireline and wireless high-speed formation-bearing symbols. A maximum likelihood approach
data communications standards (ADSL [1], IEEE 802.11 [5], has been proposed in [12], and iterative Bayesian methods that
HiperLAN [2]) and has been adopted by the European digital alternate between channel estimation and symbol detection/de-
audio and video broadcasting standards (DAB [4] and DVB coding have been proposed in [22], [24], and [30] for the case
[3]). of coded OFDM systems. Taking into account specific proper-
OFDM modulation is performed using the following steps. ties of -PSK or QAM signals while utilizing an exhaustive
First, the information-bearing symbol, modulated via any type search, a blind method has been proposed in [31]. In compar-
of constellation (e.g., BPSK, QAM) is segmented into blocks of ison to the statistical methods, the deterministic ones converge
length . An -point inverse discrete Fourier transform (IDFT) much faster; however, they involve high complexity, which be-
is then performed on each block, and a preamble, consisting comes even higher as the constellation order increases. A deter-
ministic blind channel estimation method that takes advantage
Manuscript received July 9, 2002; revised December 6, 2002 and March 20, of receive diversity and does not require an exhaustive search
2003; accepted March 21, 2003. The editor coordinating the review of this paper
and approving it for publication is G. Leus. This work was supported by the
has been proposed in [28]. Receive diversity has also been pro-
ONR under Grant N00014-03-1-0123. posed for blind channel estimation in OFDM systems without
The authors are with the Department of Electrical and Computer Engineering, cyclic prefix in [7] and [18].
Drexel University, Philadelphia, PA 19104 USA (e-mail: [email protected].
edu). In this paper, we propose an approach that belongs to the
Digital Object Identifier 10.1109/TWC.2003.821140 statistical class. It consists of a simple linear transformation
1536-1276/04$20.00 © 2004 IEEE
648 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 2, MARCH 2004

applied to blocks of symbols before they enter the OFDM The coded block goes through
system, which enables blind channel estimation at the output the regular OFDM transmission steps. The th received OFDM
of the OFDM system via cross-correlation operations. Unlike block after removal of the CP and DFT is
other coded OFDM schemes [10], [22]–[25], [30], the proposed
one does not introduce redundancy to the block, nor does it
change the block length. The proposed channel estimation
method is computationally simpler than deterministic methods
and subspace approaches. (2)
The paper is organized as follows. Section II specifies the where is the complex gain of the th subcarrier, and
proposed OFDM system and Section III presents the proposed models the noise. In the following discussions, we will assume
blind channel estimation method. Section IV provides analytic that: 1) the baud-rate channel can be modeled as a finite im-
expressions for the normalized mean-squared error of the pulse response (FIR) filter of length with tap coefficients ,
channel estimates, which were obtained based on the proposed ; hence, ,
approach. Section V includes simulation results that test the ; 2) the channel stays the same for at least
performance of the proposed approach and also compares it to the duration of the block, and is quasistationary between blocks;
the subspace approach of [27]. Finally, concluding remarks are and 3) the noise is complex, circular Gaussian, zero-mean,
presented in Section VI. variance, white across subcarriers and blocks, and indepen-
dent of the information symbols.
II. SIGNAL PRECODING The channel frequency response , , or
Let us consider an OFDM system and apply at its input a equivalently, the channel impulse response ,
linear precoding block that performs the following task. It trans- , is needed in order to recover the transmitted signal ; con-
forms the th OFDM block of information symbols sequently, the information symbols , . In
according to Section III, it is shown how the correlation structure introduced
by the proposed precoding facilitates blind channel estimation.

III. BLIND CHANNEL ESTIMATION

(1) Consider the correlation of the signals on the th and th sub-
carriers, as in (3), shown at the bottom of the page, where is
where and are both predefined numbers, assumed to be the variance of the information symbols, and the expectation is
known to the transmitter and receiver, is an integer in taken over successive OFDM blocks. Based on (3), an estimate
, and is a pure imaginary number with . This pre- of the channel within the complex constant , can
coding has several properties. be obtained as in (4), shown at the bottom of the page. Since
1) It introduces no redundancy to the transmitted data, which is known to the receiver, the required scaling in (4) is feasible.
makes the approach bandwidth efficient. The channel estimateof (4) can be further improved, noting that
2) It preserves the transmission power on each subcarrier. , should be the DFT of the channel im-
We should note here that the requirement for to be pure pulse response , , where . The latter
imaginary is necessary in order to maintain the power of length constraint can be enforced by performing IDFT on ,
the th carrier. setting to zero the last samples of the IDFT output, then per-
3) It maintains the zero-mean of the signal transmitted on forming an -point DFT on the result. This procedure is referred
each subcarrier. to as denoising [8], and the improved channel estimate equals
4) It maintains zero DC offset in each block.
5) It introduces a correlation structure in signals transmitted
over different subcarriers, which can be explored for
channel estimation. (5)

, ,
(3)

, , (4)
PETROPULU et al.: BLIND OFDM CHANNEL ESTIMATION THROUGH SIMPLE LINEAR PRECODING 649

Examining (3) and (4), we can see that is the only Subsequently, the channel estimate obtained based on (7) can
sample of the channel response estimate that contains the effect be normalized with respect to .
of noise. To eliminate the noise contribution, one could com-
pletely discard that sample. There is enough redundancy within IV. PERFORMANCE ANALYSIS
the remaining s to recover the missing sample, via inter-
The expectation needed in (3) [or (6)] can be implemented as
polation, for example.
averaging over a finite number of OFDM symbols, i.e.,
A potential problem with the estimate of (4) might arise when
the th carrier is in deep fade, in which case is close to zero
for all s. However, it is interesting to note that, at the receiver, (10)
any subcarrier can play the same role as the th one. Let be an
integer in with . Then we have (6), shown where is referred to as the smoothing factor.
at the bottom of the page. Based on (6), the channel response We next investigate the effect of finite sample averaging on
can be estimated within the complex constant the performance of the channel estimation. We consider here
as the case in which the th subcarrier is used for the correlation
operation at the receiver. The case where the th subcarrier is
used can be treated similarly. As a performance metric we use
, (7)
the normalized mean square error (NMSE), which is defined as
.
Although there is no way of knowing in advance what the
locations of the channel nulls are, a criterion for selecting
can still be implemented at the receiver. Observing that (11)
scales the amplitude of , can be selected as

(8) where is the noise-smoothed channel estimate defined

in (5). Under the assumptions of independent and identically
This step would require the estimation of the entire correlation distributed information symbols and white Gaussian noise, the
matrix of the received blocks, thus introducing a small increase can be derived as
in complexity.

A. Resolving the Scalar Ambiguity

Scalar ambiguity is common to all blind channel estimation
methods. One way to resolve it is using a pilot symbol. In our
case, the pilot could be inserted in the block before or after pre-
coding. In the former case, a good place for inserting the pilot
would be the th location within the block. Then, the would
be the known pilot. An important property of , as defined in
(1), was that it varied from block to block, thus maintaining a
zero-mean for the symbol transmitted over each carrier. Thus, if
were to be set to some known value, this value would have
to be assigned for each by a pseudo-random number generator.
Perfect cooperation between transmitter and receiver would be
needed in this case for the synchronization of the pilot sequence.
The pilot could also be placed in the block after precoding.
Assuming that one pilot is placed at location in the precoded
block, then we can obtain an estimate of based on re-
ceived blocks as

(9) (12)

where is the pilot symbol used in block .

, , ,
, (6)
650 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 2, MARCH 2004

where is the kurtosis of the information divided into groups of blocks, where was selected so that
symbol and is the signal-to-noise ratio. The the channel did not change significantly during each -block
details of the derivation can be found in Appendix A. segment.
The following observations can be drawn from (12). Let denote the channel estimate obtained based on
1) decreases almost linearly with the SNR and the the th group of received blocks, or, equivalently, the blocks
smoothing factor . corresponding to
2) As , .
For the proposed method, was obtained based on

(17)

(13) where
As
(18)
(14)

and is a parameter in . Denoising was implemented ac-

cording to (5).
Thus, for NMSE to vanish, both and should ap- The obtained estimate was used to recover the transmitted
proach infinity. If is kept small, then as increases symbols, , for , via an MMSE
the will level off to some nonzero value. The error equalizer.
floor can be lowered by increasing . If a pilot is placed at location of the symbol before pre-
3) The larger the value of , the smaller the will coding, estimates of the symbols are obtained as
be. However, the effect of on bit-error rate (BER) is
different. Consider the equalized output for the th sub- (19)
carrier
On the other hand, when a pilot is used after precoding, esti-
mates of the symbols can be obtained as
(15)

where for the MMSE equalizer

(20)
(16) where is the MMSE equalizer as defined in (16). Both of
the above described pilots were tested in our simulations but
yielded no significant difference in terms of BER performance.
One can see that in case of imperfect channel estima- Estimation and symbol recovery for each case considered
tion, increases intercarrier interference which in turn in the sequel was performed for Monte Carlo runs,
increases the BER. Therefore, the choice of should bal- where between runs, independent realizations of noise, input,
ance out channel estimation error and the bit error rate. and channel were used.
The following channel performance metric was used:
V. SIMULATION RESULTS
In this section, we present simulation results of the proposed
approach. We simulated an OFDM system according to the
IEEE 802.11a standard [5], which calls for a block length of
and a cyclic prefix length 8.
The simulated channel was a three-tap FIR filter, (21)
each tap being independently generated according to the modi- where is the th realization of the average simulated
fied Jakes’ model [17]. The maximum Doppler frequency nor- channel that was experienced by blocks
malized with the Nyquist frequency was set to and corresponding to , and is a nor-
. At 5.2-GHz carrier frequency, 54-Mb/s data malization parameter, accounting for the scalar ambiguity in the
rate with 64-QAM modulation and 3/4 coding (802.11a stan- obtained channel estimate. In the following simulations, was
dard [5]), corresponds roughly to a vehicle speed estimated using a pilot symbol placed at the th location of each
0.01 m/s, while corresponds to a vehicle speed transmitted block after precoding.
of 1.5 m/s. The noise was taken as a complex, zero-mean white Equation (21) can be viewed as the empirical version of the
Gaussian variable with variance . defined in (11).
We simulated OFDM blocks that were transmitted As a benchmark, we also tested the performance of the
through the time-varying channel. The received blocks were training-based method used in the 802.11a standard, in which
PETROPULU et al.: BLIND OFDM CHANNEL ESTIMATION THROUGH SIMPLE LINEAR PRECODING 651

(a)
Fig. 1. Slowly varying channel: NMSE of proposed and training-based
method. Further reduction of NMSE of the proposed method by combining
(23) along with (7) for SNR > 15 dB (solid/star line). Result corresponding
to the estimate obtained based on (7) only is shown in dotted/star line.

the channel is estimated once only, based on two training

OFDM blocks, i.e.,

(22)

The obtained channel estimate is then used to recover all

transmitted blocks. For a fair comparison, denoising was also
implemented for the training-based estimation method.
We also conducted comparisons between the proposed
channel estimation method and the blind subspace method
proposed in [27]. This approach exploits the redundancy in-
duced to the received data by the cyclic prefix. In particular, the
(b)
autocorrelation matrix of pairs of received blocks (before cyclic
prefix removal) is estimated, and the channel is subsequently Fig. 2. Slowly varying channel: BER performance of proposed versus
training-based and blind subspace method. (a) BER performance for smoothing
computed by exploiting the orthogonality of the signal and J 0
factor = 300 with minimum channel gain 16.5 dB. (b) BER performance
noise subspaces of that matrix, through singular value decom- J 0
for smoothing factor = 300 with minimum channel gain 20 dB.
position. The autocorrelation matrix was updated along the
lines of (17). Denoising was applied as in the proposed method. additional information available in the autocorrelation of the re-
ceived signal. An additional channel magnitude estimate can be
A. Slowly Varying Channel
extracted as follows:
We first considered the case of a slowly varying channel
and simulated transmission of OFDM (23)
symbols. Fig. 1 shows the average NMSE obtained with the
proposed training-based methods for 64-QAM input signals.
The results of the proposed method were obtained with , Although contains the effect of noise, at high SNR it pro-
, and two different smoothing factors, i.e., duces a better magnitude estimate than (4) or (7). The reduc-
and . One can see that the proposed approach (star tion in NMSE is evident in Fig. 1 (solid/star lines), where for
dotted lines) achieves lower NMSE than the training-based one dB, (23) is used along with the phase obtained from
for low SNR values. However, while the training-based NMSE (4) or (7) to obtain the channel estimate.
continues to drop at higher SNR, the proposed one levels off. Fig. 2 shows the BER performance of two methods. It can be
The final NMSE level can reduce if the amount of averaging seen that the proposed method, for this slowly varying channel
increases. This can be seen by comparing the two sets of curves case, performs comparably to the training-based one. The per-
corresponding to and . formance of both methods, however, depends on the level of
For high SNR, i.e., higher than 15 db, the NMSE can be fur- fading. Fig. 2(a) and (b) shows the performance corresponding
ther reduced for fixed . This can be achieved by exploiting to minimum channel gain 16.5 and 20 dB, respectively.
652 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 2, MARCH 2004

(a)
Fig. 4. BER performance of the proposed approach and the training-based
method. Ideal value of BER for 64-QAM, when the channel is perfectly known,
for proposed approach is shown by dashed/triangle line.

(b)
Fig. 3. Fast varying channel: proposed versus training-based method.
(a) Channel estimation error per J = 100 blocks at SNR = 20 dB. (b) NMSE
of proposed approach and training-based method.
Fig. 5. Fast-varying channel: NMSE of proposed approach and that of the
blind subspace method of [27].
The BER performance of the subspace method of [27] is also
shown in Fig. 2(b), obtained based on the same parameters used
can be seen from Figs. 3 and 4, the proposed blind channel es-
for the proposed method. It can be seen that it is comparable to
timation approach not only achieves lower channel error than
that of the training method and slightly better than the proposed.
the training-based one, but also results in lower BER. This be-
It should be noted though that the complexity of the subspace
havior suggests that 600 blocks is too long for the channel to be
approach is significantly higher as it is based on singular value
assumed constant [which can also be seen from Fig. 3(a)] and
decomposition of the autocorrelation matrix.
that the training method would have to use additional training
data in order to guarantee low BER.
B. Fast-Varying Channel On the other hand, the proposed method tracks the channel
Next, we consider the case of a fast-varying channel with very well. To get an idea of how much the channel errors affect
and simulated transmission of the BER, we also provide in Fig. 4 the BER of 64-QAM cor-
OFDM symbols. Fig. 3(a) shows the versus , for the responding to the proposed approach operating under perfect
proposed and training-based methods, at dB. Here, channel estimation.
averaging was performed over blocks, and we took Figs. 5 and 6 compare the performance of the subspace
and . The minimum channel gain was 16.5 dB. method of [27] against the proposed one for the fast-varying
The average NMSE is shown in Fig. 3(b) as a function of the channels based on OFDM symbols, , and
SNR, while the corresponding BER is shown in Fig. 4. As it smoothing factor . The value of is still as above.
PETROPULU et al.: BLIND OFDM CHANNEL ESTIMATION THROUGH SIMPLE LINEAR PRECODING 653

APPENDIX

A. Proof of (12)
It is convenient to use matrix representations. The following
notation is used for special matrices and operations:
vector with 1 at even indexed entries and 1
at the odd indexed entries;
vector with 1 at the th entry and 0 at others;
identity matrix;
matrix composed of the first columns of
an DFT matrix, the th element being
;
matrix for vector transpose;
matrix, vector, or scalar conjugate;
diagonal matrix with the elements of the vector on
the diagonal.
We also define
Fig. 6. BER performance of the proposed approach and that of the blind
subspace method of [27].

One can see that the proposed approach outperforms the sub-
space method. It should also be noted that the performance of
the subspace method could not be improved by decreasing the
smoothing factor in anticipation of the fast-varying channel. In With the above conventions, we rewrite (2) in matrix form as
theory, it requires a smoothing factor of at least twice the block
length (i.e., ). Our simulations actually suggested that (A.1)
had to be greater than 130 for the method to produce mean- where is an vector with at each entry.
ingful results even for slowly varying channels. The channel estimate obtained using the proposed method and
based on finite samples can be expressed as [cf. (4)]

VI. CONCLUSION
(A.2)
We propose a computationally simple approach for blind
channel estimation in OFDM systems. At the transmitter, a where is an vector with 1 at the th entry and
simple preprocessing is performed at each block, which induces at the th entry. From
correlation structure to the block. This structure is exploited at (A.2), we have
the receiver to estimate the channel within a complex scalar.
Simulations showed that the proposed method compares favor- (A.3)
ably to training-based and subspace approaches.
The improvement of channel estimate by enforcing the channel
Although it might appear that adding a constant to the block memory length at time-domain can be described as [cf.(5)]
during the precoding step would result is a simpler channel esti-
mation method that would employ first-order statistics of the re- (A.4)
ceived signal, such an approach would introduce a bias to each
carrier. Such a bias is not desirable in a practical system. The Consequently, the defined in (11) becomes (A.5),
proposed precoding maintains the zero-mean of each carrier. as shown at the bottom of the page. Using (A.2) and

(A.5)
654 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 2, MARCH 2004

(A.6)

(A.3), we obtain (A.6), shown at the top of the page.

Furthermore (A.7)

Substituting (A.7) into (A.6) and then into (A.5), the desired
result of (12) follows.

REFERENCES
[1] The DWMT: A Multicarrier Transceiver for ADSL Using M-Band
Wavelets ANSI Standard, 1993.
[2] “Broadband Radio Access Networks (BRAN): HIPERLAN Type
2 technical specifications Part 1—Physical layer ETSI DTS/
BRAN030 003–1,”, 1999.
[3] Digital Video Broadcasting: Framing Structure, Channel Coding and
Modulation for Digital Terrestrial Television European Telecommuni-
cations Standard, 1997.
[4] Radio Boadting Systems: Digital Audio Broadcasting to Mobile,
Portable and Fixed Receivers Eurpoean Telecommunications Standard,
1995.
[5] Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY)
Specifications IEEE Standard 802.11a, Part 11, 1999.
[6] A. Akansu, P. Duhamel, X. Lin, and M. De Courville, “Orthogonal trans-
multiplexers in communications: A review,” IEEE Trans. Signal Pro-
cessing, vol. 463, pp. 979–995, Apr. 1998.
[7] H. Ali, J. H. Manton, and Y. Hua, “A sos subspace method for blind
channel identification and equalization in bandwidth efficient ofdm sys-
tems based on receive antenna diversity,” in Proc. SSAP, Singapore, Aug.
2001, pp. 401–404.
[8] J. van de Beek, O. Edfors, M. Sandell, S. K. Wilson, and P. O. Böjesson,
“On channel estimation in OFDM system,” in Proc. Vehicle Technology
Conf., vol. 2, Chicago, IL, July 1995, pp. 815–819.
[9] J. A. C. Bingham, “Multicarrier modulation for data transmission: An
idea whose time has come,” IEEE Commun. Mag., May 1990.
[10] R. S. Blum, Y. Li, J. H. Winters, and Q. Yan, “Improved space-time
coding for MIMO-OFDM wireless communictions,” IEEE Trans.
Commun., vol. 49, pp. 1873–1878, Nov. 2001.
[11] X. Cai and A. N. Akansu, “A subspace method for blind channel identi-
fication in OFDM systems,” in Proc. ICC, New Orleans, LA, July 2000,
pp. 929–933.
PETROPULU et al.: BLIND OFDM CHANNEL ESTIMATION THROUGH SIMPLE LINEAR PRECODING 655

[12] N. Chotikakamthorn and H. B. Suzuki, “On identifiablility of OFDM Athina Petropulu received the Diploma in electrical
blind channel estimation,” in Proc. VTC, Amsterdam, The Netherlands, engineering from the National Technical University
Sept. 1999, pp. 2358–2361. of Athens, Greece, in 1986 and the M.Sc. degree in
[13] L. J. Cimini Jr, “Analysis and simulation of a digital mobile channel electrical and computer engineering in 1988, and the
using orthogonal frequency division multiplexing,” IEEE Trans. Ph.D. degree in electrical and computer engineering
Commun., vol. 33, pp. 665–675, July 1985. in 1991, both from Northeastern University, Boston,
[14] O. Edfors, M. Sandell, J.-J. van de Beek, S. K. Wilson, and P. O. Bö- MA.
jesson, “OFDM channel estimation by singular value decomposition,” In 1992, she joined the Department of Electrical
IEEE Trans. Commun., vol. 46, pp. 931–939, July 1998. and Computer Engineering, Drexel University,
[15] G. B. Giannakis, “Filterbanks for blind channel identification and equal- Philadelphia, PA, where she is now a Professor.
ization,” IEEE Trans. Signal Processing, vol. 4, pp. 184–187, June 1997. During the academic year 1999/2000 she was an
[16] R. W. Heath and G. B. Giannakis, “Exploiting input cyclostationarity Associate Professor at Université Paris Sud, École Supérieure d’Electricité,
for blind channel identification in OFDM systems,” IEEE Trans. Signal Paris, France. Her research interests include the area of statistical signal pro-
Processing, vol. 47, pp. 848–856, Mar. 1999. cessing, communications, higher order statistics, fractional-order statistics, and
[17] Y. R. Zheng and C. Xiao, “Improved models for the generation of mul- ultrasound imaging. She is the co-author of the textbook entitled Higher-Order
tiple uncorrelated Rayleigh fading waveforms,” IEEE Commun. Lett., Spectra Analysis: A Nonlinear Signal Processing Framework (Englewood
vol. 6, pp. 256–258, June 2002. Cliffs, NJ: Prentice-Hall, 1993).
[18] Y. Li, “Pilot-symbol-aided channel estimation for OFDM in wireless Dr. Petropulu is the recipient of the 1995 Presidential Faculty Fellow
systems,” IEEE Trans. Veh. Technol., vol. 48, pp. 1207–1215, July 2000. Award. She has served as an Associate Editor for the IEEE TRANSACTIONS
[19] , “Simplified channel estimation for OFDM systems with multiple ON SIGNAL PROCESSING and the IEEE SIGNAL PROCESSING LETTERS. She is
transmit antennas,” IEEE Trans. Wireless Commun., vol. 1, pp. 67–75, a member-at-large of the IEEE Signal Processing Board of Governors and a
Jan. 2002. member of the IEEE Signal Processing Society Conference Board and the
[20] Y. Li, L. J. Cimini Jr, and N. R. Sollengerger, “Robust channel estimation Technical Committee on Signal Processing Theory and Methods. She is the
for OFDM systems with rapid dispersive fading channels,” IEEE Trans. general chair of the 2005 IEEE International Conference on Acoustics, Speech,
Commun., vol. 46, pp. 902–915, Jul. 1998. and Signal Processing (ICASSP).
[21] Y. Li, N. Seshadri, and S. Ariyavisitakul, “Channel estimation for
OFDM systems with transmitter diversity in mobile wireless channels,”
IEEE J. Select. Areas Commun., vol. 17, pp. 461–471, Mar. 1999.
[22] L. Lin, J. C.-I. Chuang, and Y. Li, “Near optimal joint channel estimation Ruifeng Zhang (M’97) received the B.S. degree
and data detection for COFDM systems,” in Proc. Globalcomm, San from Huazhong University of Science and Tech-
Francisco, CA, Nov. 2000. nology, Wuhan, China, in 1993, the M.E. degree
[23] B. Lu and X. Wang, “Bayesian blind turbo receiver for coded OFDM from Beijing Institute of Technology, Beijing, China,
systems with frequency offset and frequency-selective fading,” IEEE J. in 1996, and the Ph.D. degree from Stevens Insti-
Select. Areas Commun., vol. 19, pp. 2516–2527, Dec. 2001. tute of Technology, Hoboken, NJ, all in electrical
[24] B. Lu, X. Wang, and K. R. Narayanan, “LDPC-based spacetime coded engineering.
OFDM systems over correlated fading channels: Analysis and receiver He is an Assistant Professor of the Electrical and
design,” IEEE Trans. Commun., vol. 50, pp. 74–88, Jan. 2002. Computer Engineering Department, Drexel Univer-
[25] B. Lu, X. Wang, and Y. G. Li, “Iterative receivers for space-time sity, Philadelphia, PA, which he joined in 2000. He
block coded OFDM systems in fading channels,” IEEE Trans. Wireless was a Postdoctoral Associate and Research Assistant
Commun., vol. 1, pp. 213–225, Apr. 2002. at the Electrical and Computer Engineering Department, Stevens Institute of
[26] M. Morelli and U. Mengali, “A comparison of pilot-aided channel es- Technology from 1997 to 2000. His research interests include the areas of sta-
timation methods for OFDM systems,” IEEE Trans. Signal Processing, tistical signal processing and communications. The current focus of his work is
vol. 49, pp. 3065–3073, Dec. 2001. on signal processing methods for wireless networks.
[27] B. Muquet, M. de Courville, and P. Duhamel, “Subspace-based blind and Dr. Zhang has been Chair of the joint SP/BT/CE Chapter, IEEE Philadel-
semi-blind channel estimation for OFDM systems,” IEEE Trans. Signal phia Section since 2001. He serves on the organizing committee for the 2005
Processing, vol. 50, pp. 1699–1712, July 2002. ICASSP. He was a recipient of the 2002 IEEE SP Society Best Paper Award. He
[28] H. Wang, Y. Lin, and B. Chen, “Blind OFDM channel estimation also received the Peskin Award from Stevens Institute of Technology in 2000
using receiver diversity,” in Proc. Conf. Information Science Systems, and AT&T and the ACM Student Research Award in 1999.
Princeton, NJ, Mar. 2002.
[29] Z. Yang and G. B. Giannakis, “Wireless multicarrier communications:
Where Fourier meets shannon,” IEEE Signal Processing Mag., pp. Rui Lin was born in China in 1975. She received the
29–48, May 2000. B.S. degree in 1997 and the M.Sc. degree in 2000
[30] Z. Yang and X. Wang, “Blind detection of OFDM signals in multipath from the Huazhong University of Science and Tech-
fading channels via sequential Monte Carlo,” IEEE Trans. Signal Pro- nology, Wuhan, China, both in electrical engineering.
cessing, vol. 50, pp. 255–270, Feb. 2002. She is currently working toward the Ph.D. degree at
[31] S. Zhou and G. B. Giannakis, “Finite-Alphabet based channel estimation Drexel University, Philadelphia PA.
for OFDM and related multicarrier systems,” IEEE Trans. Commun., Her research interests include the areas of com-
vol. 49, pp. 1402–1414, Aug. 2001. munications and signal processing, including blind
[32] X. Zhuang, Z. Ding, and A. L. Swindlehurst, “A statistical subspace channel estimation and equalization algorithms and
method for blind channel identification in OFDM communications,” in wireless, multicarrier, spread-spectrum communica-
Proc. ICASSP, vol. 5, Istanbul, Turkey, June 2000, pp. 2493–2496. tion systems.