《Taiwan-Japan Joint Conference on Communications Technology 2007》

Joint Channel Estimation and Data Detection for Space-Frequency Transformed MIMO OFDM System
Kun-Chien Hung and David W. Lin
Department of Electronics Engineering and Center for Telecommunications Research National Chiao Tung University, Hsinchu, Taiwan 30010, ROC Emails: hkc.ee90g@nctu.edu.tw, dwlin@mail.nctu.edu.tw
Abstract—We recently proposed a design of coded multi-input multi-outout (MIMO) orthgonal frequency-division multiplexing (OFDM) systems that uses simple space-frequency transform (SFT) to achieve nearly optimal performance. The SFT combines an orthogonal transform and a space-frequency interleaver. The receiver employs a multi-stage turbo decision feedback equalizer (TDFE) for data detection. But the issue of channel estimation has not been addressed. In this work, we present a technique for joint channel estimation and data detection (CE-DD) in the receiver. We assume that a frame consists of a number of bursts, where a burst is made of a short preamble followed by some MIMO OFDM symbols containing data only. We use an adaptive channel predictor, together with iterative joint channel estimation and data detection, to track the varying channel. We conduct simulations to investigate the performance of the proposed technique and compare it with that of conventional MIMO OFDM and MIMO block single carrier with cyclic prefixing. The results show that the proposed joint CE-DD technique has good channel tracking ability, which can be used to reduce pilot usage and increase spectrum efficiency. They also show that SFT MIMO OFDM has the best performance among the compared transmission techniques. Index Terms—Block gradient descent algorithm, channel estimation, channel prediction, MIMO, OFDM, space-frequency transform, turbo DFE

I. I NTRODUCTION Multi-input multi-output (MIMO) orthogonal frequencydivision multiplexing (OFDM) has been a favored subject in recent research on high-speed wireless digital transmission. To exploit more fully the inherent diversity over multipath fading channels, one usually needs to employ space-time coding [1], [2] or space-time or space-frequency processing [3], [4]. Towards a good design of the coding or processing scheme, it has been shown that a key factor influencing the performance of space-time/frequency coded transmission is the determinant of the correlation matrix of pairwise codeword differences [5], [6]. In a recent paper [7], we showed that the determinant was maximized when the correlation matrix was a multiple of the identity matrix, or in other words, when the codeword differences were “white.” Moreover, we proposed a design of the transmitter and the receiver to exploit the available diversity and to achieve near-optimal code condition while keeping the complexity under guard. At the transmitter, a space-frequency transform (SFT) that combined an orthogonal transform and a space-frequency interleaver (SFI) was used

before the MIMO OFDM modulation to quasi-whiten the signal. At the receiver, a shaped block-based turbo decisionfeedback equalizer (TDFE) [7], [8], [9] was adopted for data detection. The TDFE had a particular multi-stage design to achieve low complexity. In this paper, we consider the issue of channel estimation for the system and propose a technique that keeps the use of preambles and pilots to a minimum. Specifically, we consider a frame structure that is made of bursts, where each burst consists of a short preamble followed by some data-only MIMO OFDM symbols. The all-pilot preamble symbol is used to derive an initial channel estimate by, for example, a least-square (LS) method [10]. For each subsequent data symbol, then, an adaptive channel predictor is first used to predict the channel. Then a jointly iterative channel estimation and data detection (CE-DD) procedure, based on a projected gradient descent technique [11], is enacted. This joint CE-DD method is a main contribution of this paper. As later simulation results will show, the SFT MIMO OFDM, together with the proposed joint CE-DD receiver technique, yields much better performance than the coded MIMO OFDM [12], [13] and the coded MIMO block single carrier with cyclic prefixing (CPBSC) [12], [14]. This paper is organized as follows. Section II describes the transmission system model, including the transceiver structure and the channel model. Section III describes the receiver design, with emphasis on jointly iterative CE-DD. Section IV presents and discusses some simulated results. And Section V is the conclusion. II. T RANSMISSION S YSTEM M ODEL A. Transceiver System Fig. 1 illustrates the proposed transmitter and receiver structures. The transmitted data first undergo forward-errorcontrol (FEC) encoding. The result is whitened by SFT before being fed to OFDM modulators. As mentioned, the SFT consists of an orthogonal transform and an SFI. The channel coding is not restricted to any particular type. For example, it could be bit-interleaved coded modulation (BICM) employing convolutional coding, trellis coded modulation, or turbo coding. The use of the SFT is a major distinguishing feature of the proposed transmission scheme compared to more

36

《Taiwan-Japan Joint Conference on Communications Technology 2007》

the inverse SFT, the FEC decoder and the SFT form the equalization-detection loop of every iteration. When separable SFI is adopted, computational complexity of the loop can be reduced by rearranging some functions of the blocks [7]. Whatever the equalization method, some form of channel estimation is needed. The present paper considers joint CE-DD in a turbo manner; more details are provided later. B. Channel Model Let there be transmit and receive antennas. Let be the -vector of transmitted signals from the transmit antennas at time and the -vector of received signals from the receive antennas at time . Assume perfect synchronization. Then by the equivalent lowpass discrete-time representation of the MIMO multipath channel, we have [15]
Fig. 1. Proposed transmitter and receiver structures.

Fig. 2.

Illustration of separable space-frequency interleaving.

where is an matrix of zero-mean complex Gaussian is the matrix summarizing the pairwise entries, correlations between the channel responses associated with different transmit antennas at delay path , and is the matrix summarizing the pairwise correlations between the channel responses associated with different receive antennas at delay path . Alternatively, the signal samples in one OFDM symbol transmitted or received by each antenna, excluding CP, can be put in a column vector form. Let be the vector transmitted by antenna and be the vector received by antenna . Then the entire transmitted signal can be represented as and the signal propagation behavior can be modeled as (3) where noise, and , is the vector of additive

. . .

. . .

..

.

. . .

(4)

37

a

%

D

q 2x ˆ ƒ q "H ƒ ‡†††… q ƒ

C

¦A¦ VU A T¦ S§ % IHRF" IHGF" ) %Q P ) % Q% ¤% E£ % P D

ihphh H xƒ "xƒ ) ‚ hiihphphhh HRIH"H I"„""H £ ƒƒ ƒƒ € u teryx be f(phihphf H be f " be w£ b c ) uv5 Y £ b f terseq Y f(ihhpihfeH Y g"e Y d£ Y c a

`Y

b`

C

%

§A XW§

conventional ones. The orthogonal transform can be any that is easily realizable, such as the discrete Fourier transform (DFT, for which fast Fourier transform or FFT may be used in its realization) or the Fast Hadamard transform (FHT). The SFI is a pseudo-random permutation in space (antenna) and in frequency. For ease of realization (in both the transmitter and the receiver), we propose a separable design which does permutations in the spatial and the frequency domains sequentially separately. Mathematically, this can be achieved by having a permutation matrix that can be written as a product of a spatial permutation matrix and a frequency permutation matrix , such as . Fig. 2 illustrates the concept graphically. Data are first arranged according to antennas and subcarriers. In frequency interleaving, data associated with the same subcarrier at different antennas are moved as a group to another subcarrier in a pseudo-random manner. Then in spatial interleaving, the permuted data that are associated with the same subcarrier at different antennas are permuted again pseudo-randomly in the spatial domain. In the receiver, after OFDM demodulation, block-based TDFE is used to detect the data. In the initial stage when no prior decision output are available, linear minimum meansquare error (MMSE) equalization may be used to obtain the initial decision output. The FEC decoder should have soft output capability, such as the soft-output Vitribi algorithm (SOVA) for convolutional codes, if we want to obtain better convergence behavior for the TDFE. The MIMO TDFE,

where is the number of paths, is the channel response matrix of path , is the relative delay of path , and is the additive noise vector at time . Note that are integers in the equivalent lowpass discrete-time representation. (Some leakage may result due to truncation [16], which is assumed negligible in the present work.) The channel estimator may proceed with a knowledge of the longest possible path delay, but it may perform better if it knows the delays of the significant paths. Our method may work under either condition. The delays of the significant paths may be estimated, for example, via a subspace acquisition algorithm such as ESPRIT [17] or MUSIC [18]. For quasi-static correlated multipath Rayleigh fading chancan be modeled as [19], [20] nels, (2) 

© 97 % 3 C ) 3 C ¦4AB§ @ % 0(& )' © 97856 % 342¨ #!£ ©  1 © "   % %$ 

§

%

§ 

©  

¦ ¦ 

© ¨

(1)

¢ ¡ ¥¡ £¤   

¢

《Taiwan-Japan Joint Conference on Communications Technology 2007》

, is the vector of additive where noise, and with entry being the circulant data matrix transmitted from antenna . Furthermore, we can write
  § ¤   £  © ¨ §   ¡ ¢ ¢ § § ¢ ¢ ¥ ¢

and 

B. Frame Structure and Initial Channel Estimation Note that in performing TDFE, the channel responses need be estimated first. For this, consider the frame structure illustrated in Fig. 3 which we call a pilot-data separated frame structure. A frame consists of a number of bursts where each burst starts with an all-pilot symbol as the preamble. It is then followed by a number of OFDM symbols. The pilot symbol is used to obtain an initial estimate of the channel response. During subsequent data symbols, adaptive channel prediction is used to track channel variation. For each symbol, iterative joint CE-DD is carried out, as described in more detail below. For the pilot symbol, an LS channel estimation can be obtained by minimizing the cost function
P

III. J OINT C HANNEL E STIMATION A. Turbo DFE

AND

DATA D ETECTION

When the channel responses are available, the maximumlikelihood (ML) data detection can be formulated as a constrained LS (CLS) problem which minimizes the cost function

(8)

The solution is
¢

For channel prediction in subsequent symbols (which yields the initial channel estimate for each data symbol), we can employ block-based linear filtering whose basic form is

and is the ratio of noise variance to the variance of decision error. The shaped version of the block gradient descent algorithm is given by
P Q1 # 0 ) D # I # & 5 H E

38

h i

H

e

(12)

" "#  H 5 © D £

p qf 

F

f g

9

&

E

)

9

E

(11)
C

where represents one tap of the estimated channel response at time and is one tap of the predicted channel response. If the mobile speed is known and the fading follows Jakes’ model, then the optimal prediction coefficient can be shown to be (17)

# 

©

9

) 

E 7

3

where 

©

9

E 6d e c

P 

© V1 9 

H £

V

F ©

9

E

#

¥

b

) 

© "&

a Y W U`X

¥

$q £  ©

9

&

E

where with and being the FFT size. If is error-free, then the TDFE can cancel the intersymbol interference (ISI) perfectly. In the initial iterations, however, may contain many decision errors. For this, we have proposed using a multi-stage shaping filter to reduce error propagation [7]. The shaping filter is given by (10)

where . To achieve the minimum mean-square error in the estimate, the pilot symbol should be such that [10] (15) 

6¦ ¥

T

(9)
3 C

b` # " £` H ` 1 b` £  ` ©

!

)

¢

T

¥

1

T

! R

)

T U0

¢

subject to the constraint that is a valid codeword. As such minimization may not be easy, one way to obtain a potentially suboptimal solution is via the block gradient descent algorithm [11], which leads to a receiver based on block TDFE as shown in Fig. 1. In block TDFE, let denote the SFT output in the st iteration. The TDFE output of iteration can be obtained as
$ %

$

£

E

is a vector of fading coefficients of the channel from where antenna to antenna whose entires are zero-mean complex Gaussian random variables, , is a diagonal matrix whose diagonal elements are unity only where there exists a propagation path at the corresponding delay position and otherwise zero, and with being the Kronecker matrix product. 
¨ ¨           £ ¤

S

£

  ¡ ¤

1 (

¢

£ r e Xq e` f(ihphihf H e` "f e` c £ ` a ` ` £` f ` £` f #phihphf f H f " ` r u q re q e` #phihph Hf e` c "f e` £ c £ ` ` v5 ` £ b` u " ( ) )b " # Y £)
) 1 2 A B # 9 @

" ( Y ) 1b ) 5( Y £ Y " ) )# ) ) H H £ "  ¦ 5 © b £ ) )

" ( 

Y H Y ) 1 b £  Y©

1

!

# 

Y " ( Y H iƒ ) " H ƒ ¦ £ 1b 5# Y £ Y " ( Y

8

E 7

! 7 !

0 )

5 

 

¢

! "

)

¥ ¦

' (

0

¢

)

#

'

8

#

! 7 !

A B

£

#

9

D

5 6

#

'

&

F

3 4

F G

a

  ¡

with
  £

being circulant matrices and the first column giving the channel impulse response from antenna antenna . Moreover, we also have

`

`ƒ `ƒ
'

of to

(5)

Fig. 3.

The pilot-data separated frame structure considered.

(6) (7) Since, in actual realization, it may be hard to estimate the variance of the decision errors, we thus proposed a threestage approach: 1) Initial stage: Perform MMSE block linear equalization (since the decision output is initially unavailable). 2) Middle stage: Use a fixed shaping filter with which can tolerate a large range of decision errors. 3) Final stage: Use the unshaped gradient descent algorithm.

(13)

(14)

(16)

《Taiwan-Japan Joint Conference on Communications Technology 2007》

IV. S IMULATION S TUDY In this section, we illustrate the performance of the proposed SFT MIMO OFDM and compare it with that of the conventional MIMO OFDM [12] and the MIMO CPBSC [12], [14], all with FEC coding, by way of simulation. In fact, the MIMO CPBSC can be considered a particular kind of SFT MIMO OFDM. For the SFT, we use FFT for the orthogonal transform. The simulated systems employ a rate-1/2 convolutional code with generators . Bit interleaving over one MIMO OFDM symbol period is performed to effect BICM. The data are QPSK-modulated. A frame contains 5 bursts, with each burst made of 5 MIMO OFDM symbols following 1 short pilot symbol. The FFT size is 1024 for data symbols but only 256 for the pilot symbol. Therefore, the pilot-data ratio is 1/20 in terms of bandwidth consumption. The pilot symbol contains BPSK-modulated sequences that are mapped to the subcarriers of each antenna in an interleaved manner [10]. In the receiver, the adaptive channel predictor has five predictor taps. The maximum number of iterations of the joint CE-DD is 10 and the iteration may stop early if the decoder output contains no error. Shaping filtering is used in the earlier iterations of the TDFE and the switching from shaped to unshaped operation depends on the convergence of the output log-likelihood ratio (LLR) of the FEC decoder [7]. MIMO block time-varying channel with Consider a the ETSI Vehicular A power-delay profile (where the channel varies between OFDM symbols). Let each propagation path have the same transmit and the same receive correlation matrix given by
P

(18)

For each data symbol, following the initial channel estimation, we perform jointly iterative CE-DD. In each iteration, the TDFE output can be used to re-estimate the channel responses in the LS manner as in (14). However, the pseudo-inverse operation costs a very large computational effort. Hence, herein we propose performing the channel re-estimation using the projected gradient descent algorithm and combining it with the TDFE to yield the desired jointly iterative CE-DD. For this, note first that the projected gradient descent algorithm is an iterative approach to solving the convex CLS problem such as that given in (13). Let be the estimate of in iteration . By applying a similar recursion as in TDFE, we can obtain an iteration for channel estimation as 
#

(19) (20)

(22) (23)

, , and with denoting the combined actions of the inverse SFT, the soft-output FEC decoding, and the SFT. In the initial pass through the iterative loop, the channel estimate used is the output of the channel predictor and the TDFE output is that from linear MMSE equalization. If shaping filtering is considered in the earlier iterations as discussed previously, (23) is replaced by
# # ¢ P # )

To reduce the complexity, one may use a constant shaping filter calculated using the predicted channel responses.
D 

#

" ( )

¢

# 

#

D

" 5( Y £ Y

#

#

#

&

#

"

#

where
# &

(24)

Let the carrier frequency be 2.5 GHz and the channel bandwidth be 10 MHz. Consider three mobile speeds: 50, 100, and 200 km/h, which correspond roughly to maximum normalized Doppler shifts of 0.0125, 0.025, and 0.05, respectively. Figs. 4 and 5 illustrate the block error rate (BLER) and the bit error rate (BER) performance of the three transmission schemes, where a block means the period of one OFDM symbol. At the two slower mobile speeds, the channel estimation mechanism proposed in this work can track the channel variation well. Its tracking ability is reduced at the highest mobile speed, as indicated by the error floors in the plots. In all cases, the SFT MIMO OFDM performs better than the conventional MIMO OFDM and the MIMO CPBSC. The performance gain is roughly 4 dB over conventional MIMO OFDM and 2 dB over MIMO CPBSC at the lower mobile speeds.

39

5 6$

P

$

P

4

h

(21)

$

4

$

2 3

£ £ % %

1

Combining the above with the TDFE, we obtain the jointly iterative CE-DD procedure as

£ Y

` ` ter e q f £ # ) " f ` "( f(` ( " "

&

#

# 

)

) 

P 

#

&

¢

#

# 

¦

¥

# 

¢ 

! ' ¦(

#

' (

&

and

0 ( & $ ©)' %#

¡

# ¡ ¤¢

C. Jointly Iterative Channel Estimation and Data Detection

)

where is the autocorrelation matrix of the channel response and is the crosscorrelation of the channel response and its delayed version. Under Jakes’ model, the th element in is given by and the th element in by , where is zeroth-order modified Bessel function and is the maximum Doppler shift normalized to sample spacing. In practice, the mobile speed may not be known and the fading may not follow Jake’s model. In this case, we may use an adaptive prediction filter in place of the filter above. By minimizing the cost function , the simplest LMS adaptive algorithm yields the following adaptive prediction filter for :
C

Note also that, since the super-matrices and are composed of circulant matrices whose FFTs are diagonal matrices, many computations can be simplified by using this property. To limit the length of this paper, we omit the details.

¥ 

f ©

A

¡

G © g 9 ©  © 6 © £  5 © 1 © 5 H 9 2© © 1 

§ ¥ 

GR6u u y1 u 2 2  5 ©©
P $ ¦ E " 9

 

&

E

` £` £` " ( ` 1 b` " 1 (` £ `

" #`

 

§ ¥ £ ¡ ¨¦¨©

#

¢

§ ¥ £ ¡ ¨¦¤¢

9

&

¢

E

¥ 

A

" " ©'  f © ' £ 6£ ©u 6u © D
E R 

) 

)

9

E

#

¢

' (

¥ 

 

h f #ihphihf e H f ©e " c £ " 5# Y £ Y 5 "# ` £ ` 1 b` £ `

p

¥

'

f

h

9

 

#

#

¢

f 

#

$

e

#

¢

e

1

¢

p qf

$ 

h

#

#

#

#

&

&

&

¢

¢

9

e

f  

Y© ' u " D 

f g h

p qf

D

`

Y 

¢

(25)

"

《Taiwan-Japan Joint Conference on Communications Technology 2007》
0

10

ACKNOWLEDGMENT This work was supported by the National Science Council of R.O.C. under grant no. 95-2219-E-009-003. R EFERENCES
SFT,0.0125 F T
d s

10

−1

10

−2

10

−3

BSC,0.0125 FdTs OFDM,0.0125 FdTs SFT,0.025 F T
d s

10

−4

BSC,0.025 FdTs OFDM,0.025 F T SFT,0.05 F T
d s d s

10

−5

BSC,0.05 FdTs OFDM,0.05 F T
d s

4

6

8

10

12 E N
b o

14

16

18

20

22

Fig. 4. Block (symbol) error rates of different transmission schemes over block time-varying channel.
−2

10

10

−3

10

−4

BER

SFT,0.0125 FdTs

10

−5

BSC,0.0125 FdTs OFDM,0.0125 FdTs SFT,0.025 FdTs

10

−6

BSC,0.025 FdTs OFDM,0.025 FdTs

10

−7

SFT,0.05 FdTs BSC,0.05 FdTs OFDM,0.05 FdTs

10

−8

4

6

8

10

12 E N
b o

14

16

18

20

22

Fig. 5. Bit error rates of different transmission schemes over block timevarying channel.

V. C ONCLUSION The recently proposed MIMO OFDM transmission with space-frequency transform (SFT) can achieve near-minimum pairwise codeword error probability in coded transmission. In this work, we considered jointly iterative channel estimation and data detection for this transmission technique. The proposed receiver technique can be applied not only to SFT MIMO OFDM but also to other typically considered transmission techniques based on OFDM or block single carrier (BSC) modulation. In addition, it can reduce the need of pilot subcarriers in OFDM and BSC-based transmission, resulting in better spectrum efficiency. Simulation results demonstrated that the proposed receiver technique could track the channel variation well and that the SFT MIMO OFDM, together with the proposed receiver technique, could outperform typically considered transmission techniques based on OFDM or BSC modulation.

[1] V. Tarokh, N. Seshadri, and A. R. Calderbank, “Space-time codes for high data rate wireless communication: Performance criterion and code construction,” IEEE Trans. Info. Theory, vol. 44, no. 2, pp. 744–765, Mar. 1998. [2] S. M. Alamouti, “A simple transmit diversity technique for wireless communications,” IEEE J. Sel. Areas Commun., vol. 16, no. 8, pp. 1451– 1458, Oct. 1998. [3] Y. Xin, Z. Wang, and G. B. Giannakis, “Space-time diversity systems based on linear constellation precoding,” IEEE Trans. Wireless Commun., vol. 2, no. 2, pp. 294–309, Mar. 2003. [4] J. Boutros and E. Viterbo, “Signal space diversity: A power- and bandwidth-efficient diversity technique for the Rayleigh fading channel,” IEEE Trans. Info. Theory, vol. 44, no. 4, pp. 1453–1467, July 1998. [5] S. Yiu, R. Schober, and L. Lampe, “Distributed space-time block coding,” IEEE Trans. Commun., vol. 54, no. 7, pp. 1195–1206, July 2006. [6] R. A. Horn and C. R. Johnson, Matrix Analysis. Cambridge, U.K.: Cambridge Univ. Press, 1999. [7] K.-C. Hung and D. W Lin, “Theory and design of near-optimal MIMO OFDM transmission system for correlated multipath Rayleigh fading channels,” J. Commun. Networks, vol. 9, no. 2, pp. 150–158, June 2007. [8] C. Laot, A. Glavieux, and J. Labat, “Turbo equalization: Adaptive equalization and channel decoding jointly optimized,” IEEE J. Sel. Areas Commun., vol. 19, pp. 1744–1752, Sep. 2001. [9] A. M. Chan and G. W. Wornell, “A class of block-iterative equalizers for intersymbol interference channels: Fixed channel results,” IEEE Trans. Commun., vol. 49, no. 11, pp. 1966–1976, Nov. 2001. [10] I. Barhumi, G. Leus, and M. Moonen, “Optimal training design for MIMO OFDM systems in mobile wireless channels,” IEEE Trans Signal Process., vol. 51, pp. 1615–1624, June 2003. [11] R. Freund, G. Golub, and N. Nachtigal, “Iterative solution of linear systems,” Acta Numer., vol. 1, pp. 57–100, 1992. [12] J. Coon et al., “A comparison of MIMO-OFDM and MIMO-SCFDE in WLAN environments,” in IEEE Global Telecommun. Conf., vol. 6, Dec. 2003, pp. 3296–3301. [13] S. Y. Park and C. G. Kang, “Complexity-reduced iterative MAP receiver for interference suppression in OFDM-based spatial multiplexing systems,” IEEE Trans. Vehic. Tech., vol. 53, no. 5, pp. 1316–1326, Sep. 2004. [14] J. Siew et al., “A channel estimation algorithm for MIMO-SCFDE ,” IEEE Commun. Lett., vol. 8, no. 9, pp. 555–557, Sep. 2004. [15] C. Xiao et al., “A discrete-time model for triply selective MIMO Rayleigh fading channels,” IEEE Trans. Wireless Commun., vol. 3, no. 5, pp. 1678–1688, Sep. 2004. [16] C. Pirak et al., “Adaptive channel estimation using pilot-embedded data-bearing approach for MIMO-OFDM systems,” IEEE Trans. Signal Process., vol. 54, no. 12, pp. 4706–4716, Dec. 2006. [17] B. Yang, K. B. Letaief, R. S. Cheng, and Z. Cao, “Channel estimation for OFDM transmission in multipath fading channels based on parametric channel modeling,” IEEE Trans. Commun., vol. 49, pp. 467–478, Mar. 2001. [18] M. Oziewicz, “On application of MUSIC algorithm to time delay estimation in OFDM channels,” IEEE Trans. Broadcasting, vol. 51, no. 2, pp. 249–255, June 2005. [19] O. Oyman, R. U. Nabar, H. Bolcskei, and A. J. Paulraj, “Characterizing the statistical properties of mutual information in MIMO channels,” IEEE Trans. Signal Process., vol. 51, no. 11, pp. 2784–2795, Nov. 2003. [20] A. K. Sadek, W. Su, and K. J. R. Liu, “Diversity analysis for frequencyselective MIMO-OFDM systems with general spatial and temporal correlation model,” IEEE Trans. Commun., vol. 54, no. 5, pp. 878–888, May 2006. [21] T. Hwang and Y. Li, “Novel iterative equalization based on energy spreading transform,” IEEE Trans. Signal Processing, vol. 54, pp. 190– 203, 2006.

BLER

40