Vous êtes sur la page 1sur 4

SPACE-TIME SIGNAL PROCESSING FOR WIRELESS COMMUNICATIONS: A SURVEY

Constantinos B. Papadias Arogyaswami J. Paulraj Information Systems Laboratory, Stanford University, Stanford, CA 94305-9510, U.S.A.

This survey paper reviews space-time signal processing (STP) in mobile wireless communications. Space-time processing refers to manipulating signals that are received at or transmitted from an antenna array so as to improve performance. We focus on antenna arrays deployed at the base stations since such applications are of current practical interest. We rst introduce channel and signal models for base station antennas, and present the di erent structures present in the channel model after sampling and stacking the received signals. We then present single user (SU) and multiple user (MU) space-time processing techniques for the reverse link. Our focus is on blind signal separation, and intersymbol/co-channel interference cancellation. We consider a number of recent advances in batch and adaptive blind space-time processing that show promise in solving these problems. The rapid progress in radio technology has resulted in an increased interest for reliable and largely available wireless networks. However, a wireless designer is faced with a number of challenges. These are: i) a complex and rapidly varying propagation environment, ii) limited availability of radio spectrum, iii) limited energy storage capability of batteries in portable units, iv) user demands for higher bit rates, voice quality and longer talk times, v) operator demands for greater coverage of base stations, increased subscriber capacity and vi) lower infrastructure and operating costs. A number of di erent technologies have to be used to meet such diverse requirements. These include advanced multiple access techniques, e cient voice and channel coding and improved signal processing. Signal processing functions in wireless communication include modulation/ demodulation, channel coding/ decoding, channel equalization and estimation of transmitted signals, and reduction of co-channel interference. Since use of training bits for equalization is a waste of available bandwidth and also the training is not e cient in rapidly timevarying channels, blind channel equalization and blind estimation of multiple users' signals can have signi cant impact on overall network capacity and performance. Also, methods for co-channel interference reduction will have a direct bearing on network quality and cellular frequency re-use leading to increased capacity of wireless networks. A promising approach to attack the above problems is to use space-time (ST) signal processing. A space-time equalizer at the base station combines two signal processing aspects: equalization and co-channel interference reduction. Several techniques, based on higher- and second-order statistics, have been recently proposed for blind channel estimation and equalization. Similarly, many methods that exploit the structure and properties of the underlying base-

ABSTRACT

1. INTRODUCTION

band signals have been developed for blind separation of multiple users' signals. In space-time modems, the modem operates simultaneously on all the antennas. The key leverage that we get from this scheme is that the co-channel interference can be signi cantly reduced in a way which is not possible with single antenna modems. The reason why this is possible is that the co-channel interference and the desired signal almost always arrive at the antenna array (even in complex multipath environments) with distinct and well separated spatial signatures, thus allowing the modem to exploit this di erence to reduce the co-channel interference. Likewise, the space-time transmit modems can use spatial selectivity to deliver signals to the desired mobile while minimizing the interference for other mobiles. The spatial dimension can also be used to enhance other aspects of space-time modem performance. In receive, the antennas can be used to provide enhanced array gain, improve signal to thermal noise ratio and enhance diversity gain. In transmit, the spatial dimension can enhance array gain, improve transmit diversity and reduce delay spread at the subscriber's end. In this paper we review the key signal processing issues that underlie space-time modems. After presenting a mathematical signal model for receive (reverse-link) ST signal processing we discuss the structure included in the ST channel. We then focus on space-time blind channel and signal estimation and co-channel interference reduction and discuss brie y di erent approaches to the problem. Depending on whether the ST processing is performed at the base station and/or the subscriber end, at the forward or the reverse link, and with a single user or multiple user approach, several di erent transmit and receive con gurations can be obtained (see 1]). In this paper we will focus on reverse link STP, performed at the base station, and we assume a TDMA system. When a single user transmits the information signal and it is received at a base station with multiple antennas, the noiseless baseband signal xi (t) received by the base station at the i-th element of an m element antenna array is given by

2. SIGNAL MODEL

xi (t) =

L X l=1

ai ( l ) R l (t)u(t ? l ) + ni (t)

(1)

where L is the number of multipaths, ai ( l) is the response of the i-th element for an l-th path from direction l , R l (t) is the complex path fading, l is the path delay, u( ) is the transmitted signal that depends on the modulation waveform and the information data stream and ni (t) the additive noise. (1) can be written in a vector form as

x(t) =

L X l=1

Array Manifold

Time Manifold

Constant Modulus

a(

l) R l (t)u(t ? l )

+ n(t)

(2)
Finite Alphabet Block Toeplitz Structure H H X Low Rank X

where x(t)= x1 (t) xm (t)]T , a( )= a1 ( ) am ( )]T and n(t)= n1 (t) nm (t)]T . The fading j R (t)j is Rayleigh or Rician distributed depending on the propagation model. When we consider MU processing the signal model becomes

x(t) =

Lq Q X X q=1 l=1

=H

a( lq ) R lq (t)uq (t ? lq ) + n(t) (3)

..
H

where we have indexed each user signal, and corresponding path delay, angle and fading parameters by the user index q. The channel models described above use physical path parameters such as path gain, delay and angle of arrival. When the received signal is sampled at the receiver at symbol (or higher) rate, it is more convenient to use a discretetime sampled signal model. In the SU case, assuming x(t) is sampled at the symbol rate at instants t = t0 +kT , the antenna output can be written as x(k) = Hs(k) + n(k) (4) where H is the symbol response channel (a m N matrix) that captures the e ects of the array response, symbol waveform and path fading, m is the number of antennas, N is the channel length in symbol periods and n(k) the sampled vector of additive noise. Note that n(k) may be colored in space and time. s(k) is a vector of N consecutive elements of the data sequence and is de ned as s(k) = s(k) s(k?N +1)]T . Including, as before, the interference from Q?1 users, the sampled signal model becomes

Figure 1. Space-time structures indexed by : A = fa( )j 2 g, where is the set of all possible values of . A includes the e ect of array geometry, element patterns, inter-element coupling, scattering from support structures and objects near the base station, and when measured at the receiver baseband after digitization includes the e ects of cable and receiver gain/phase response, I-Q imbalance, A/D converter errors. Knowledge of A helps determine a( l ).

Temporal structure The temporal structure relates to

the properties of the signal u(t) and includes modulation format, pulse-shaping function and symbol constellation. Some typical temporal structures are: constant modulus (CM), nite alphabet (FA), non-Gaussianity and cyclostationarity. Also, the structure of the pulse shaping function u(t) can be captured with the de nition of a temporal manifold as the sampled response of a receiver to an incoming pulse with delay . We will now see how the above structures are exploited by di erent ST algorithms in order to recover blindly the transmitted signals. The term \blind" (other names are \self-recovering" or \unsupervised"), refers to methods that do not need training signals and exploit instead the structural properties mentioned above. We consider separately the SU and MU cases. direction-of-arrival (DOA) estimates as a basis for determining the optimum beamformer, and are typically applicable when the channel delay spread is negligible. In this case the sampled block signal takes the popular form (narrowband model) X(k) = A S(k) + N(k) (8) where A = a( 1 ) a( Q )]. Once A is estimated combining the knowledge of the array manifold with a DOA estimation technique (such as e.g. 3], 4], 5], 6], 7], 8], 9], 10], 11], 12], 13]), we have the array vector for the received signal. The minimum mean square error (MMSE) and maximum likelihood (ML) estimators for the desired user signal are then identical (under the white noise assumption) and given by 1 wq = R ? (9) xx a( q ) where wq is the beamformer for user q.

x(k) = Hs ss (k) +

Q ?1 X q=1

4. BLIND SPACE-TIME ALGORITHMS

Hq sq (k) + n(k)

(5)

If we collect M consecutive snapshots of x( ) corresponding to time instants k; : : : ; k + M ? 1, (and dropping subscripts for a moment), we get the block signal model of the following general form: X(k) = HS(k) + N(k) (6) where X(k) = x(k) x(k+M ?1)]T (m M ), and similarly for S(k) and N(k) (of dimensions N M and m M , respectively). Note that S(k) by de nition is constant along the diagonals and is therefore Toeplitz. Figure 1 shows the di erent types of structure that are involved in the general signal model of (6), which we discuss in the sequel:

4.1. Single-User methods Spatial structure methods These techniques use

3. SPACE-TIME CHANNEL STRUCTURE

Spatial structure It can be shown (see 1], 2]), that the j th column of H is given by H1:m;j =
L X l=1

a( l ) R l g ((Md + ?j )T ? l )

(7)

where Md is the maximum path delay and 2 T is the duration of the pulse shaping waveform g(t). a( l ) lies on the array manifold A, which is the set of array response vectors

Temporal structure methods These techniques span

a wide range from the well studied CM and higher-orderstatistics (HOS) methods to the more recent second order methods that exploit the cyclostationarity of the received signal. A widely known class of simple adaptive

blind algorithms that exploit implicitly the received signal's HOS is the so-called Bussgang class that contains among others the CM 1-2, CM 2-2, Sato, and Decision-Directed (DD) algorithms (see 14]). These algorithms, together with their more recent advanced versions (see 1]) attain at best MMSE-type solutions.

Polyphase methods Following the path-breaking paper by Tong, Xu and Kailath 15] that presented a blind channel identi cation method using oversampling that relied only on second order statistics, a number of techniques that exploit cyclostationarity have since dominated the blinddeconvolution literature. In the frequency domain, methods based on spectral factorization are studied in 16], 17] and 18], and those based on eigenstructure in 19] and 20]. In the temporal domain, subspace-based methods have been proposed in 21], 22] and 23]. Least-squares subspace approaches have been also proposed in 24], 25], 26], and a signal subspace method was developed in 27]. In the time domain, linear prediction methods have been proposed in 21], 28], 29] and 30]. Polyphase methods have also been proposed in order to improve performance of maximum likelihood techniques in 31], 21], and analyzed in 31] and 30]. Polyphase methods provide a blind solution by starting with the data X(k) = HS(k) + N(k) (10) or its second order statistics. They then extract H and S by exploiting the tallness structure (obtained via oversampling) of H. See 32] for a tutorial presentation of polyphase techniques.
The MU techniques are parallel to the single-user spatial and temporal blind methods. The spatial structure multiuser algorithms are again applicable. A number of temporal structure blind algorithms are known. For zero delay spread case these include FA-based techniques (e.g. 33], 34], 35]) and CM techniques ( 36], 37]). For nite delay spread, recent contributions include 38] and 39]. Again, CM-based techniques attend at their best MMSE-type solutions, whereas, ML-type techniques that are based on properties such as FA attain at their best ML-type solutions. At this point we describe three di erent MU approaches. 4.2.1. Finite Alphabet (FA) method This approach exploits the FA property of the digitally modulated signals. Assuming no delay spread and perfect multi-user symbol synchronization, the channel model is given by Eq. (8). The joint ML criterion for this reduces to the familiar minimization problem 2 min (11) A;S kX ? ASkF where subscript F denotes Frobenius norm. This is a joint ML problem where both the channel and data are unknown. The FA property allows us to solve Eq. (11) and estimate both A and S. Since the ML criterion is separable with respect to the unknowns, one approach to minimize the cost function in Eq. (11) is alternating projections. Starting with an initial estimate of A, we minimize Eq. (11) with respect to S, keeping A xed. This is a data detection problem. With an estimate of S, an improved estimate of A can be obtained by minimizing Eq. (11) with respect to A, keeping S xed. This is a standard least-squares problem. We continue this iterative process until a xed point is reached. The global solution is a xed point of the iteration. In order to avoid a computationally expensive search, two suboptimal iterative techniques, ILSP and ILSE 33, 40] can be used to make this minimization tractable.

4.2.2. Finite Alphabet - Oversampling (FA-OS) method In the presence of delay spread (and unsynchronized symbols), the FA algorithm has to be modi ed to estimate the space-time channel H as against the spatial channel A described earlier (Eq. (10). An attractive technique to estimate the temporal channel using a polyphase or oversampling method was recently proposed in 41]. Once again we can estimate H and S using a Joint-ML approach. This reduces to minimizing 2 min (12) H;S kX ? HSkF

4.2. Multi-User methods

A direct approach to Eq. (12) is computationally prohibitive. One approach 41] breaks up the joint problem into two smaller subproblems. First, the channels are equalized by enforcing the low rank and block-Toeplitz structure of H. This yields the row subspace of S. The FA property can now be enforced to determine the symbols in S. 4.2.3. MU Constant Modulus (MU-CM) algorithm The MU-CM approach is an adaptive alternative to the FA-OS method, which exploits the CM structure of the signals. We consider a linear spatio-temporal equalization structure WQ ] W = W1 (mM Q) where Wq , q = 1; : : : ; Q, denotes the spatio-temporal equalizer corresponding to the q-th signal. Then the Q 1 equalizer output at time instant k can be written as y(k) = WT (k)X (k), where X (k) is the regressor at time instant k, and y(k) = y1 (k) yQ (k)]T . The MU-CM algorithm determines W by minimizing a cost function that is the sum of a CM term and a cross-correlation term. The CM term penalizes the deviations of the equalized signals' magnitudes from a constant modulus, whereas the crosscorrelation term penalizes the correlations between them: min W J (W) = E
Q ? X q=1

jyq j2 ? 1 2 + 2

Q X l;n=1 ; l6=n

2 X

(13) where rln ( ) is the cross-correlation function between users l and n de ned as rln ( ) = E (yl (k)yn (k ? )). The resulting algorithm is described in 39]. The major advantage of this technique with respect to other approaches (e.g. 42]) is its ability to nd all the user signals. Global convergence can be guaranteed under realistic conditions. Space-time processing is a rapidly emerging eld that shows signi cant promise in improving the performance of wireless communication networks. Blind ST processing seems to be a powerful leverage for the improvement of performance. While these techniques are still at their infancy, more work will be needed in re ning and analyzing them before they become suitable for deployment in actual telecommunications systems.
1] A. J. Paulraj and C. B. Papadias. \Space-time processing for wireless communications". IEEE Signal Processing Magazine, to appear, 1997. 2] M. C. Vanderveen, C. Papadias, and A. Paulraj. \Joint angle and delay estimation (JADE) for multipath signals arriving at an antenna array". IEEE Communications Letters, 1(1):12{14, Jan. 1997. 3] R. O. Schmidt. \Multiple emitter location and signal parameter estimation". IEEE Trans. Antennas and Propagation, 34:276{280, March 1986.

= 1

jrln ( )j2

5. CONCLUSIONS

REFERENCES

4] A. Paulraj, R. Roy, and T. Kailath. \A subspace rotation approach to signal parameter estimation". Proceedings of the IEEE, 74(7):1044{1045, July 1986. 5] C. P. Mathews and M. D. Zoltowski. \Eigenstructure techniques for 2-D angle estimation with uniform circular arrays". IEEE Trans. Signal Processing, 42:2395{2407, Sept. 1994. 6] A. Nehorai and E. Paldi. \Vector-sensor array processing for electromagnetic source localization". IEEE Trans. Signal Processing, 42(2):376{398, Feb. 1994. 7] T. J. Shan, M. Wax, and T. Kailath. \On spatial smoothing for directions of arrival estimation of coherent signals". IEEE Trans. on Acoustics, Speech and Signal Processing, ASSP-33(4):806{811, April 1985. 8] S. Shamsunder and G. Giannakis. \Signal selective localization of non Gaussian cyclostationary signals". IEEE Trans. on Signal Processing, 42:2860{2864, Oct. 1994. 9] G. Xu and T. Kailath. \Direction-of-arrival estimation via exploitation of cyclostationarity: a combination of temporal and spatial processing". IEEE Trans. on Signal Processing, 40(7):1775{1786, July 1992. 10] S. Anderson, M. Millnert, M. Viberg, and B. Wahlberg. \An adaptive array for mobile communication systems". IEEE Transactions on Vehicular Technology, 40(1):230{236, Feb. 1991. 11] S. C. Swales, M. Beach, D. Edwards, and J. P. McGeehan. \The performance enhancement of multibeam adaptive base-station antennas for cellular land mobile radio systems". IEEE Transactions on Vehicular Technology, 39(1):56{67, Feb. 1990. 12] T. Trump and B. Ottersten. \Estimation of nominal direction of arrival and angular spread using an array of sensors". Signal Processing, 50:57{69, 1996. 13] P. Stoica and A. Nehorai. \MUSIC, maximum likelihood and the Cramer-Rao bound: further results and comparisons". IEEE Trans. ASSP, ASSP-38:2140{2150, Dec. 1990. 14] S. Haykin. Blind Deconvolution. Prentice Hall, Englewood Cli s, N.J., 1994. 15] L. Tong, G. Xu, and T. Kailath. \Blind identi cation and equalization of multipath channels: a time domain approach". IEEE Trans. on Information Theory, 40(2):340{ 349, March 1994. 16] Z. Ding. \Blind channel identi cation and equalization using spectral correlation measurements, part I: frequencydomain analysis". In Cyclostationarity in Communications and Signal Processing, editor W. A. Gardner, pages 417{ 436, New Jersey, USA, 1994. 17] W. A. Gardner, editor. Cyclostationarity in Communications and Signal Processing. IEEE press, New Jersey, USA, 1994. 18] Ph. Loubaton. \Egalisation autodidacte multi-capteurs et systemes multivariables". GDR 134 (Signal Processing) working document, February 1994, France, 1994. 19] L. Tong, G. Xu, and T. Kailath. \Blind identi cation and equalization based on second order statistics: A frequency domain approach". IEEE Trans. on Information Theory, 41(1):329{333, Jan. 1995. 20] G. Giannakis. \Linear cyclic correlation approach for blind identi cation of FIR channels". In Proc. 28th Asilomar Conf. on Signals, Systems and Computers, Paci c Grove, CA, Oct. 31 - Nov. 2 1994. 21] D. T. M. Slock. \Blind fractionally-spaced equalization, perfect-reconstruction lter banks and multichannel linear prediction". In Proc. ICASSP 94 Conference, pages IV{ 585{IV{588, Adelaide, Australia, April 1994. 22] E. Moulines, P. Duhamel, J. F. Cardoso, and S. Mayrargue. \Subspace methods for the blind identi cation of multichannel FIR lters". IEEE Trans. on Signal Processing, 1995.

23] K. Abed Meraim, P. Loubaton, and E. Moulines. \A subspace algorithm for certain blind identi cation problems". Submitted to IEEE Trans. on Information Theory, July 1994. 24] M. I. Gurelli and C. L. Nikias. \A new eigenvector-based algorithm for multichannel blind deconvolution of input colored signals". In Proc. ICASSP-93, volume IV, pages 448{ 451, Minneapolis, MN, April 1993. 25] L. A. Baccala and S. Roy. \A new blind time-domain channel identi cation method based on cyclostationarity". In Proc. 26th Conf. Informations Sciences and Systems, volume IV, pages 448{451, Princeton, NJ, March 1994. 26] G. Xu, H. Liu, L Tong, and T. Kailath. \A least-squares approach to blind equalization". IEEE Trans. Signal Processing, SP-43(12), Dec. 1995. 27] H. Liu and G. Xu. \Closed-form blind symbol estimation in digital communications". IEEE Transactions on Signal Processing, SP-43(11):2714{2723, Nov. 1995. 28] S. Halford and G. Giannakis. \Adaptive blind channel identi cation and equalization using cyclic correlations". In Proc. IEEE ICASSP, 1995. 29] K. Abed Meraim, P. Duhamel, D. Gesbert, P. Loubaton, S. Mayrargue, E. Moulines, and D. Slock. \Prediction error methods for time-domain blind identi cation of multichannel FIR lters". In Proc. IEEE ICASSP, 1995. 30] D. T. M. Slock and C. B. Papadias. \Further results on blind identi cation and equalization of multiple FIR channels". In Proc. ICASSP 95 Conference, volume 4, pages 1964{1967, Detroit, Michigan, May 1995. 31] Y. Hua. \Fast maximum likelihood for blind identi cation of multiple FIR channels". IEEE Transactions on Signal Processing, 44(3):661{672, March 1996. 32] H. Liu, G. Xu, L. Tong, and T. Kailath. \Recent developments in blind channel equalization: From cyclostationarity to subspaces". Signal Processing, 50:83{99, 1996. 33] S. Talwar, M. Viberg, and A. Paulraj. \Blind Estimation of Multiple Co-Channel Digital Signals Using an Antenna Array". IEEE Signal Processing Letters, SPL-1(2):29{31, Feb. 1994. 34] A. Swindlehurst and J. Yang. \Using least squares to improve blind signal copy performance". IEEE Signal Processing Letters, 1(5):80{82, May 1994. 35] L. K. Hansen and G. Xu. \Geometric properties of the blind digital co-channel communications problem". In Proc. ICASSP-96, Atlanta, May 1996. 36] A. van der Veen and A. Paulraj. \An analytical constant modulus algorithm". IEEE Trans. on Signal Processing, 44(5):1136{1155, May 1996. 37] A. Batra and J. R. Barry. \Blind cancellation of co-channel interference". In Proc. IEEE Globecom Conference, 1995. 38] A.J. van der Veen, S. Talwar, and A. Paulraj. \Blind identi cation of FIR channels carrying multiple nite alphabet signals". In Proc. ICASSP-95, volume 2, pages 1213{1216, 1995. 39] C. B. Papadias and A. Paulraj. \A constant modulus algorithm for multi-user signal separation in presence of delay spread using antenna arrays". IEEE Signal Processing Letters, to appear, 1997. 40] S. Talwar, M. Viberg, and A. Paulraj. \Blind separation of synchronous co-channel digital signals using an antenna array. Part I. Algorithms". IEEE Transactions on Signal Processing, 44(5):1184{1197, May 1996. 41] A.J. van der Veen, S. Talwar, and A. Paulraj. \Blind estimation of multiple digital signals transmitted over FIR channels". IEEE Signal Processing Letters, (2)-5:99{102, May 1995. 42] J. R. Treichler and M. G. Larimore. \New processing techniques based on the constant modulus adaptive algorithm". IEEE Trans. on Acoustics, Speech, and Signal Processing, ASSP-33(2):420{431, April 1985.

Vous aimerez peut-être aussi