Vous êtes sur la page 1sur 5

1246

An Efficient Channel Estimation Technique for OFDM Systems with Transmitter Diversity
Won Gi Jeon, Kyung Hyun Paik, and Yong So0 Cho School of Electronic and Electrical Engineering, Chung- Ang University yscho@cau.ac.kr

ABSTRACT
In this paper, we propose an efficient channel estimation technique for orthogonal frequency-division multiplexing (OFDM) systems with transmitter diversity. The proposed technique estimates uniquely all channel frequency responses needed in space-time coded OFDM systems using comb-type training symbols. The computational complexity of .the proposed technique is reduced dramatically, compared with the previous minimum mean-squared error (MMSE) technique, due to the processing made all in the frequency-domain. Also, several other techniques for mitigating random noise effect and tracking channel variation are discussed to further improve the performance of the proposed approach. The performances of the proposed approach are demonstrated by computer simulation for mobile wireless channels.

receiver antenna pair, required for space-time coded OFDM systems [8]. The MMSE technique, derived by utilizing the information on channel delay profile, is shown to have a good performance with a reduced computational complexity. However, it still requires a high computational complexity, especially for channels with a large delay spread, in addition to the necessary transformations between frequency-domain and time-domain in the estimation process of channel frequency responses. In this paper, we propose an efficient least square (LS) estimation technique for channel frequency responses of the OFDM systems with transmitter diversity. The proposed technique estimates the channel frequency responses directly in the frequency-domain using comb-type training symbols. Since all the processing in the proposed technique is made in the frequency-domain, the computational complexity reduces dramatically, compared with the previous MMSE technique. Also, several techniques are discussed to further improve its performance with a slightly increased complexity. This paper is organized as follows: In Section 11, we briefly overview the basic structure of a space-time coded OFDM system. In Section 111, we propose an efficient LS estimation technique for OFDM systems with transmitter diversity using training symbols. Also, several other techniques for improving the performance of the proposed approach to some extent are discussed. A modified decision-directed channel estimation (DDCE) technique for OFDM systems with transmitter diversity is discussed to track the channel variation caused by Doppler frequency. In Section IV, the performances of the proposed techniques are demonstrated by computer simulation for mobile wireless channels. Section V concludes this paper.

I. INTRODUCTION
Due to many advantages over single-carrier schemes, especially for high data-rate transmission in time-dispersive channels, orthogonal frequency-division multiplexing (OFDM) has been proposed and standardized for a variety of applications such as digital audio broadcasting (DAB), digital TV broadcasting, wireless LAN/ATM, and asymmetric digital subscriber lines (ADSL). Recently, Advanced Cellular Internet Service (ACIS) based on OFDM was developed to achieve reliable high-speed transmission to wide-area mobile and portable cellular subscribers with very high spectrum efficiency [l]. The link budget and dispersive fading limitations of the cellular mobile radio environment were overcome by combining the OFDM with transmitter antenna diversity and Reed-Solomon coding (clustered OFDM) [2]. More recently, a space-time coded OFDM was proposed for OFDM systems with transmitter diversity, and shown to have high code efficiency, good performance, and no delay time usually caused by interleaving [3]. However, it requires channel state information to decode the coded OFDM symbols correctly. In the previous literatures, many techniques have been proposed to estimate channel characteristics of OFDM systems with no diversity [4]-[6]. Also, a robust channel estimation technique is proposed for an OFDM system with receiver diversity [7].However, these techniques are inadequate for channel estimation of OFDM systems with transmitter diversity since the signal received at the receiver antenna is the superposition of the signals transmitted from multiple transmitter antennas, implying that the signals from other transmitter antenna act as interferences. A new channel estimation technique based on the minimum mean-squared error (MMSE) criterion is proposed to estimate the channel frequency responses, corresponding to each transmitter and
0-7803-6465-5/00 $10.00 0 2000 IEEE

11. A SPACE-TIME CODED OFDM SYSTEM


The simplified block diagram of a space-time coded OFDM system is shown in Fig. 1 [3]. Here, the estimation techniques are derived on the assumption that the number of antennas is two at both transmitter and receiver for notational convenience. However, the proposed techniques can be easily extended to

Fig. 1. A block diagram for a space-time coded OFDM system.

1247
the case with multiple antennas more than two. Input source bit stream is first encoded into symbols by a space-time encoder and then modulated by inverse fast Fourier transform (IFFT). The modulated OFDM signal transmitted from the i-th antenna at the I-th symbol period can be expressed as follows:
(.)H denotes Hermitian transpose. &I and H2 can be estimated by minimizing the cost function, given in (4),as follows: v s = - X F Y + Xl"XlHI + xl"x2H20 = (5)
~

H:

Vi:S

= -XFY

+ XFXIHl + X2X2H2= 0

(6)

where V denotes the complex gradient. From ( 5 ) and (6). one can see that the channel frequency responses cannot be where N denotes the number of subchannels. Here, estimated for a general input case since the received signal in X,[I,k] and z,[I,n] represent the coded symbol at the k-th ( 3 ) is the superposition of the signals transmitted from two transmitter antennas, implying that the signals from other subcarrier and the corresponding time-domain sample at the transmitter antennas will become interference. That is, in order n-th moment, respectively. The process of insertion and to estimate H I from ( 5 ) , XI and X2 must satisfy the removal of a cyclic prefix are omitted in this figure. The following two conditions: XyX2 = 0 and XI# 0 . On the received signal at the j-th receiver antenna is converted into a discrete-time signal by analog-to-digital converter (ADC) and other hand, to estimate H2 from (6), the conditions, then demodulated by FFI'. By assuming the cyclic prefix X t X l = 0 and X2 # 0 , must be satisfied. Any longer than the largest delay spread of a multipath channel, the combination of XI and X2 cannot meet these conditions FFT output can be expressed as for the general input case. However, for the case where training symbols are available, the channel frequency responses can be 2 estimated by elaborate YJII,m]= ~X,[l,m]H,~[I,m]+WJ[l,m],m=0,1,~~~,N-1(2) selection of training symbols. Note that, for wireless burst modems employing coherent OFDM z=1 transmission scheme such as an IEEE 802.11a modem, the first where HaJII, represents the channel frequency response of m] block of a packet is the training block (preamble) that is used the multipath channel, at the m-th subchannel, between the for time and frequency synchronizations and for channel i-th transmitter and j-th receiver antennas. Also, W,[I,n] estimation. Two OFDM symbols are allocated for channel represents the discrete Fourier transform (DFT) of additive estimation in the preamble of the IEEE 802.11a modem [9]. white Gaussian noise (AWGN) with a zero mean and variance of a$. In order to apply the LS estimation technique for the case where one training symbol is available, we first partition the training symbol, X,, into several subsets. The number of subsets must be the same as the total number of transmitter antennas. When the transmitter antennas are two, the subsets of

X, for z =1,2 are defined as X ~ C N / 2 x N / 2 for E


J = 1,2. Here, the subscript 3 is used for subset index. Note that the subscriptj denotes thej-th receiver antenna as given in (2). The subsets are partitioned to satisfy the following condition: Xt,E X, , X;nXf=0 , XtuXf = * .

x,

Also, we can partition Y in (3) into two subsets, ,.VJ for J = 1,2 . Then, the received symbol vector can be expressed as

y ' = X ; H ;tXpl$+W'
p=X@f:+x$!;+W2

(7)
(8)

-x,[I,o] 0

e..

x,=

X,[I,l]

... ...

... ...

0
2

where and W J are subsets of H, and W , respectively. By applying LS criterion to (7) and (8). we can obtain the following four equations:

OX,[I,N-l]

(X;)H iu;+ ( x ; ) * Xiv; = (X:)Hy' x; ;


(Xh)Hx;iv; +(x;)"X; = (X;)Hy' ivi

(9)

(10)
(11)

( X y iuf+ ( X ? ) H Xf
111. CHANNEL ESTIMATION
The LS criterion is used to identify the channel frequency responses from demodulated signals. The cost function is defined as

iu; = ( X f ) H Y 2 (X;)Hxf +(Xi)H iv;= ( X ; ) H y 2 . ivf 4

(12)

By selecting the subsets as X$=Xf= , the previous 0

(Xi)"X;= 0, conditions for subsets, ( X { ) " X ; =


and (A;?)"X;= ($)HA';= 0, $#

xi#0

0 , can be satisfied. By

s = &HE = ( Y -Y)H(Y

-Y)

(4)

substituting the training symbol satisfying these conditions, the channel estimates, can be obtained by

where P is given by Xlfk1 + X z H ; . Here, and H 2 represent the estimates of HI and H2, respectively. Also,

e, e= x y y " , (
2

=42.

(13)

1248
Although many subsets are available satisfying the conditions of Xj+ 0 for i = j and Xi=O for it j , a simple training symbol to estimate the rest part of subchannels, X!, by using the pre-determined channel parameters is a comb-type symbol. For the case of comb-type training symbol, the subsets can be expressed as due to Doppler effect is negligible during two consecutive symbol periods, the following equation can be obtained from ( 5 ) for the l-th symbol period:

X i p l J f i l J + X p 2 J f i 2 , l = x63

(18)

where Xl,l and X2,1 are the training symbols transmitted at the first training symbol period.
fil,l

Xi=diag[X,[Z,O]X,[1,2]... X,[I,N -211

(14) (15)

, f i 2 , 1 , and

A:= d i u g [ X ~ [ l , lX&3] ... X,[I,N -111 ]

where diug[ 1 represents the diagonal matrix. One example 1 1 would be Xl[l,2m]= X2[I,2m+ 1 = D and X1[l,2m+ 1 =

represent the channel frequency responses and received symbol, respectively. Also, another equation can be obtained from (6) for the (l+l)-th symbol period as follows:

X~+1~1,I+1fi1,1+1+ X5+lX2,1+1~2,1+1 X$+lY+l =

(19)

x 2 [ l , 2 m ] 0 for nz = O,l;..,N/2 -1. Here, D represents =


an arbitrary complex symbol, e.g., QPSK. In this case, for i,j = 1,2 are given by

Xi

Xi= [Hl[l,O]Hl[l,2]...H1[I,N-2]IT
Xf=[H1[1,1] H1[1,3]...H1[I,N -1]IT

where Xl,l+l and X2,1+1 are the training symbols transmitted at the second training symbol period. fil,l+l and fi2,1+1, q+lrepresent the channel frequency responses and = and received symbol, respectively. Assuming fil,l fil,l+l and fi2,1 = fi2,1+1 , training symbols are constructed as follows: (20) x , += X2,l = - 11 1 x , . 2 1 1 x , += 1 1 By substituting (20) into (18) and (19), the estimates of channel frequency responses are obtained as

Xi= [H2[l, H2[I,21.. .H2[l,N - 2]$ 01 Xi= [H2[1,1] H2[1,3]...H,[l,N -1]IT


The rest parts of subchannels,

k!:

and

%a,

can be obtained

by interpolating the channel estimates, ki and , respectively. When the simplest linear interpolation technique is applied, the final channel estimates are given by

ki

1 + 1- i]+ 4 [ 1 , 2 m+ 3 - 2 1 ) 2 fi1[1,1]=*1[l,O], 1 [ l , N - l ] = k l [ l , N - 2 ] k (16)


k,[l,2m+ 2 - i]= -(H,[l,2m

Since Xl,l is diagonal matrix, (21) can be expressed as

H,[I+l,m]=[Y[I,m] (-l),Y[l+ l , m ] ] . (22) + 2x1 [l,ml


This estimator can provide a better performance than the one with one training symbol due to noise averaging effect. When Doppler frequency is high, the channel variation between two successive OFDM symbol periods becomes non-negligible. To reduce the performance degradation caused by Doppler effect, the correlation characteristic of channel between two successive symbol periods can be incorporated. One possible method for determining current channel estimate is to assign different weights to the pre-estimated channel frequency responses. If k,[l,m] and H,[I+ l,m] are the channel frequency responses estimated by using (17) at the first and second symbol period, respectively, the final estimate are obtained by averaging these two pre-estimated channels in time-domain (ACE-T) as follows:

k2[I,0] H2[[I,1], = H2[l,N-21 = k2[1,N-11


where m varies over the range { - l + z , i , l + i , ~ ~ ~ , N / 2 - 3 + z. } More sophisticated interpolation techniques such as higher-order polynomial interpolation can be used for better estimates of channel frequency responses at the expense of computational complexity. The channel parameters estimated in the symbol period are used for the space-time decoding in the data transmission period. The performance of the simple estimation technique in (13) and (16) can be degraded by additive random noise, especially at low signal-to-noise ratio (SNR). In order to reduce the random noise effect at subchannels, the initial channel estimates are weighted and averaged over three neighboring subchannels under the assumption that the channel frequency response varies slowly over adjacent subchannels. Averaged channel estimation in the frequency-domain (ACE-F) is defined as

G i [ l + l , m ]= r k , [ I + l , m ] + ( l - y ) k , [ I , m ]

(23)

H,[l,m] O f i , [ l , m - l ] + a H , [ I , m ] =
L

where y represents a weight factor, determined by Doppler frequency and SNR. If Doppler frequency is zero, the major effect becomes noise. To reduce the noise effect at each subchannel, y is set to 0.5. Then, (23) can be rewritten by

(17)

+(l-a)k,[l,m+l], 2

m = 1,2,...,N-2 .

The weighting factor, a , affecting the performance of the channel estimator is determined by the following basic rules: (1) the value of 0 needs to be increased when the channel frequency response varies fast over adjacent subchannels (2) the value of a needs to be decreased as SNR decreases. When two OFDM symbols are available for training symbols, ( 5 ) and (6) can be modified as follows: If the channel variation

1 k , [ I + l , m ]= -(H,[1+1,m]+ H , [ l , m ] + ~ [ l + l , m ] + W , [ l , m ] ) 2 = H,[I + l,m]+ V,[I+ 1,m]


where

Vi[l 1,m] denotes (W$ +

+ 1,m]+ W,[l,m])/2 and ,

(24)

has a variance of 03 = / 2 . About 3 dB SNR gain can be obtained in this situation. If SNR is infinite and Doppler frequency is high, the degree of channel variation becomes dominant effect where the value of y should be set close to 1.0

05,

1249
to track the channel variation. In a practical situation where both noise and Doppler frequency exist, y can be chosen in the range between 0.5 and 1.O. Once the initial estimates of channel frequency responses are obtained using one or two training symbols, the channel variation must be tracked during data transmission period. Two techniques are available for channel tracking; pilot symbol aided estimation techinque [5] and DDCE,technique [6]. Here, we extend the DDCE technique for a space-time coded OFDM system. The conventional DDCE technique is modified to track the variation of channel frequency responses uniquely from the superposition of signals transmitted from multiple transmitter antennas. The modified DDCE technique for two transmitter antennas (DDCE-M) is given by

-24;

"

"

10

"

12

14

"

16

18

"
20

SNR (dB)

(a) z = 1 . 2 5 , ~ s ,

+ (1- p)Iit[em],
where &[l,m] for

2 = 1,2

(25) i = 4 2 represents the estimate of X,[l,m] obtained by the space-time decoder in the receiver. Also, p denotes the weighting factor for channel updating.

2 symbols (ACE-F)

IV. SIMULATION
The performances of the proposed LS estimator are evaluated by computer simulation for mobile wireless channels. The mobile radio channels used for simulation are two-path channels, each experiencing independent Rayleigh-fading, under two different Doppler frequencies, f d = 40 Hz and f d = 100 Hz. The parameters of the simulated OFDM system are set as follows. The entire bandwidth of 800 KHz is divided into 128 subchannels where 4 virtual subcarriers are left unused at both edge of the frequency spectrum. A 16-state space-time code with QPSK modulation is applied. One OFDM symbol in this scheme carries 236 bits of information. A gross bit rate becomes 1.18 Mbps since the period of one OFDM symbol = including effective symbol period, Tsub 160 ps, and guard , interval, 40 ~ s becomes 200 ps. It is assumed that high-power amplifier is ideal, and symbol and frequency are perfectly synchronized. Fig. 2 shows the MSE performances of the proposed channel estimator for different rms delay spreads defined as z = (q - zo)/2 where ~0 and z represent the delay time for , , the first and second path, respectively. Fig. 2(a) shows the MSE performance of the proposed estimator for z, = 1.25 ps. From this figure, one can see that the performance with two training symbols outperforms the case with one training symbol due to averaging effect for the channel with a relatively low frequency, fd = 40 Hz. One can also see that the channel estimator combined with ACE-F ( a = 0.4) improves the performance over the case with no averaging ( a = l.O), especially for f d = 40 Hz. However, as both Doppler frequency and SNR increase, the effect of averaging in the estimator becomes insignificant due to the dominant effect of channel variation. The same results can be found from Fig. 2(b) for z, = 5.0 ps. The parameter y in (23), used for two training symbols, is set to 0.5 for the simulation. Fig. 3 shows the BER performance of the proposed estimator
for the same environment as in Fig. 2. When one training symbol is used, 4 dB loss in SNR for f d = 40 Hz and 5 dB loss

10

12

14

16

18

SNR (dB)

(b) z = 5.0,~s , Fig. 2. MSE performance of the proposed channel estimator (combined with ACE-F and ACE-T) forfd = 40 and f d = 100 Hz. for f d = 100 Hz occur at 0.1% BER, compared to the ideal case. When two symbols are used for channel estimation, only 1 dB loss occurs for f d = 40 Hz. For f d = 100 Hz, the performance degradation for both cases becomes larger, especially at high SNR, due to rapid channel variation. To further improve the performance of the channel estimator for a fast-varying channel, the parameter y in (23) should be properly selected. From Fig. 4, one can see that about 2 dB SNR gain at BER of lo4 can be obtained at high SNR when y = 0.9. At low SNR ranging from 0 dB to 12 dB, the performance of the estimator with equal weights, y = 0.5, slightly outperforms the case with 0.9 due to the averaging effect of random noise. Fig. 5 shows tracking performances of the DDCE-M technique in (25). In the training mode, the initial estimates obtained by (17) is used while. in the tracking mode, the equation (25) with the parameter, p , set to 0.4 for fd = 40 Hz and 0.8 forfd = 100 Hz is used. Also, ACE-F with a = 0.4 is applied to the channel estimates obtained by (13) and (16). When f d is 40 Hz and training data is 10 %, only 1 dB SNR loss at BER of 104is occurred, but for f d = 100 Hz and 20 % training data, 3 dB SNR loss is occurred. Thus, DDCE-M can be seen as an effective technique for tracking a slowly-varying channel. For robust tracking over a fast fading channel, the channel estimation technique based on pilot symbols can be used. In this case, careful selection schemes of pilot tones and processing techniques of the received signals for OFDM system with transmitter diversity need to be developed.

1250
1oo
8 initial estimate - DDCE-M. 20 % training data 4 4 DDCE-M. 10 % training data

10-

10-

10-

10-

t
2
4

x
l8

8 SNR10(dB) 12

14

16

.0 2

10-1

1 ; 12 SNR (dB)

14

16

1s

do

(a) z = 1.25p.s ,

Fig. 5. BER performance of the proposed channel tracking technique (DDCE-M) for z = 5 . 0 , ~ ~ = 0.4). , (a of the space-time codes, the computational complexity for channel estimation can be reduced significantly. Also, several frequency-domain techniques such as ACE-F for reducing random noise effect, ACE-T for mitigating channel variation effect, and DDCE-M for channel tracking, are discussed. It is shown by computer simulation that the proposed channel estimation technique can be effectively used for a slow-varying fading channel, regardless of the amount of channel delay spread. Currently, we are investigating enhanced channel estimation techniques exploiting symbol correlation between successive OFDM symbols in the frequency-domain, and using pilot tones for tracking variation of a fast fading channel.

10 12 SNR (dB)

14

18

18

20

REFERENCES
[l] L. J. Cimini, Jr., J. C.-I. Chuang, and N. R. Sollenberger, Advanced cellular internet service (ACIS), IEEE Commun. Mag., vol. 36, no. 10, pp. 150-159, Oct. 1998. [2] L. J. Cimini, Jr., B. Daneshrad, and N. R. Sollenberger, Clustered OFDM with transmitter diversity and coding, in Proc. IEEE GLOBECOM, pp. 703-707.1996. [3] D. Agrawal, V. Tarokh, A. Naguib, and N. Seshadri, Space-time coded OFDM for high data-rate wireless communication over wideband channels, in Proc. IEEE VTC, pp. 2232-2236, 1998. [4] 0. Edfors, M. Sandell, J. J. van de Beek, S. K. Wilson, and P. 0. Borjesson, OFDM channel estimation by singular value decomposition, IEEE Trans. Commun., vol. 46, no. 7, pp. 931-939, July 1998. [5] F. Tufvesson and T. Maseng, Pilot assisted channel estimation for OFDM in mobile cellular systems, in Proc. IEEE VTC, pp. 1639-1643, May 1997. [6] V. Mignone and A. Morello, CD3-OFDM: a novel demodulation scheme for fixed and mobile receivers, IEEE Trans. Commun., vol. 44, pp. 1144-1151, Sep. 1996. [7] Y. Li, L. J. Cimini, Jr., and N. R. Sollenberger, Robust channel estimation for OFDM systems with rapid dispersive fading channels, IEEE Trans. Commun., vol. 46, no. 7, pp. 902-915, July 1998. [8] 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, no. 3, pp. 461-471, Mar. 1999. [9] IEEE, Wireless MAC and PHY Specifications: High Speed Physical Layer in the 5 GHz Band, P802.11alD7.0, July

(b) Z = 5.0p.s , Fig. 3. BER performance of the proposed channel estimator (combined with ACE-F and ACE-T) forfd = 40 andfd = 100 Hz (a= 0.4).

lo,

L 6
2 4

Y
8 IO 12 SNR (dB)
14

16

1s

Fig. 4. BER comparison of the proposed channel estimators (combined with ACE-T) forfd = 100 Hz (a= 0.4).

V. CONCLUSION
In this paper, an efficient channel estimation technique for a
space-time coded OFDM system is proposed. Since the proposed technique estimates the channel frequency responses directly in the frequency-domain, which is crucial for decoding

1999.

Vous aimerez peut-être aussi