ADAPTIVE SPACE-TIME PROCESSING TECHNIQUES FOR W-CDMA SYSTEMS

Abstract

Narrowband direct sequence code-division multiple-access (DS-CDMA) has been proposed and implemented recently as an attractive alternative to traditional frequencydivision multiple-access (FDMA) and time-division multiple-access (TDMA) techniques for use in mobile communications. Wideband DS-CDMA (W-CDMA) based third generation systems such as universal mobile telecommunications system (UMTS) offer higher capacity, together with other features such as multipath resistance, inherent frequency diversity, and potential use of advanced receiver structures coupled with multiple antennas for achieving even higher capacity. Proposed data rate varies from 8 Kbps to 2 Mbps depending upon the application and the mobility of the user. Previously, narrowband DS-CDMA based systems, like 1S-95, offer data rate of up to 9.6 Kbps only. The W-CDMA/DS-CDMA detector receives a signal consisting of contributions from all users' signals, which overlap in time and frequency. In conventional DS-CDMA detector (matched filter receiver), desired user's signal is detected by correlating the entire received signal with that of desired user's code waveform. However, this approach ignores the presence of multiple-access interference (MAI), which limits the capacity and performance of the DS-CDMA systems. MAI results because of random asynchronicity of transmission amongst different users, which makes it impossible to design the code waveforms to be completely orthogonal. As the number of interferers increases, MAI becomes substantial, causing degradation in system performance. It also introduces near-far problem, resulting from a weak signal received by the base station from a distant user being overwhelmed by a strong signal coming from an interferer situated nearby. One common strategy to deal with the near-far problem is to use power control. In order to overcome fading, even for single path, transmitter would have to adjust their power at least a few hundred times per second, causing wastage to precious bandwidth. Another problem introduced at high data rates is intersymbol interference (ISI) caused by multipath fading, which severely limits the performance of the conventional receiver, because the conventional matched filter (MF) receiver is unable to exploit the multipath diversity. An effective strategy to overcome MAI, ISI and near-far problem is to use multiuser detection (MUD). Here, information about multiple users is used jointly to better detect each individual user. MUD techniques provide significant performance improvement over the conventional detector. In 1986, Verdu proposed an optimum receiver for joint detection of users based on maximum likelihood sequence estimator (MLSE). In this receiver aViterbi decoder is placed immediately after the bank of matched filters. However, the drawback of this receiver is it's computational complexity, which grows exponentially (o(lK)) with the number of users K. With multipath fading, the computational complexity increases even further. In view of the high computational complexity of the optimum MLSE receiver, several suboptimum receivers are proposed in the literature with complexity linear to the number of users. Decorrelator detector and minimum mean-square error (MMSE) detector, belongs to this category. The minimum mean-square error (MMSE) detector tries to minimize mean-square error (MSE) between the transmitted and the estimated bits. The estimated bits are obtained by using linear transformation. The optimum solution for the filtering problems is provided by the Wiener-Hopf equations using the stochastic gradient approach. For obtaining the optimum filter tap weights, the correlation matrix needs to be inverted which is computationally expensive involving the complexity 0(n].), where Nr is the number of filter taps. Amajor limitation of non-adaptive detectors is that they require the knowledge of the spreading codes, timings and amplitudes of all the users, which may not be available to single user receiver. Adaptive detectors, in which minimum side information is needed, may be used in their place with less number of computations. In training based adaptive detector implementation, atraining sequence is transmitted so that detector self-tunes itself. In this work, we study the adaptive space-time processing techniques for W-CDMA systems. Our motivation is to develop low computational complexity adaptive linear multiuser MMSE receiver structure employing multiple receive antennas with reasonable . performance in terms of convergence speed, residual mean-square error (MSE) or probability of error, near-far resistance and number of users supported. We highlight the advantage of adding spatial dimension (processing) to the temporal signal processing. We have considered 'adaptive iterative (turbo) equalization' and its advantage in ISI and MAI mitigation, so that near optimum performance can be obtained. We have first considered three different adaptive linear receiver architectures based on MMSE criterion for additive white Gaussian noise (AWGN) channel. All the three ii architectures employ chip-matched filter (CMF) at their front ends and the samples collected are used to adapt the filter weights. In the two receiver architectures, namely non-connected (NC) and fully-connected (FC) architectures, we only need time delay ofeach user other than the training sequence. The difference between the NC and FC architecture is that in the FC receiver architecture sampled output of CMF corresponding to each user is shared amongst the users, whereas in the NC receiver architecture samples corresponding to each user's CMF output is not shared amongst users. The FC architecture jointly detects all the users and exploits the diversity present by sharing the sampled CMF outputs. In the third architecture, which we call as chip fractional sampling (FS) receiver architecture, there is no need to estimate the time delays of the users. In its place, we use fractional sampling and increase the filter length to twice of the processing gain. The NC architecture corresponds to single user adaptive receiver as proposed earlier in the literature, whereas the other two architectures correspond to multiuser adaptive receivers. Simulation is carried out for both synchronous and asynchronous CDMA systems. Performance of least-mean-square (LMS), normalized LMS (NLMS), and recursive-least-square (RLS) is compared for the three receiver architectures. Computational complexity involved in the LMS, NLMS, and RLS algorithms is compared for the three receiver architectures. It is shown that all the three receiver architectures offer good near-far resistance. Number of users accommodated for the different receiver architectures using LMS, NLMS, and RLS adaptive algorithms are also reported in terms of output signal to interference and noise ratio (S1NR). Probability of error calculation is carried out assuming the output S1NR as Gaussian random variable. It is shown that the FC receiver architecture outperforms both the NC and the FS receiver architectures. The key issues involved in adaptive implementation are rate of convergence, computational complexity, and misadjustment error. The LMS algorithm requires 0(NT) computations but it converges slowly, while, the RLS algorithm, on the other hand, converges much faster but requires large number of computations 0(/V72). Next, we propose to use regularized affine projection algorithm (APA) for the adaptive implementation of multiuser detector (based on MMSE criteria), whose performance is found to be superior to NLMS algorithm in terms of convergence characteristics, with significantly less computational complexity as compared to the RLS algorithm. The computational complexity offered by APA is linear in the length of the transversal filter with an additional term ofthe order of 0{L2) and a matrix inversion of dimension LxL, where Lis the order of filter, 11! which is considerably smaller than the length ofthe filter. In our simulations, we use order four APA. The APA may be viewed as an intermediate adaptive algorithm between the NLMS algorithm and RLS algorithm in terms of both computational complexity and performance. Its performance is found to be superior to block or partial rank algorithm (PRA). In the adaptive filtering algorithms increasing the step-size results in increasing convergence rate, but italso results in increasing the misadjustment error. To overcome, these contradictory requirements of faster convergence and smaller misadjustment error, various time-varying step-size algorithms are proposed in the literature. These algorithms work on the principle ofemploying large step-size in the initial stages ofconvergence to achieve fast convergence and smaller value of step-size, when the algorithm reaches its optimal solution to provide small misadjustment error. We next propose a novel variable step-size APA (VSSAPA), whose performance is found to be superior to that of fixed step-size APA (FSS-APA). In VSS-APA, step-size is varied in proportion to the square of the autocorrelation between adjacent error blocks. VSS-APA marginally increases the complexity ofFSS-APA by L+5 multiplication per iteration. We have calculated the number ofusers supported by both FSSAPA and VSS-APA using simulation in terms ofthe output SINR. With the ever-increasing demand on the mobile (wireless) communication systems, new methods are used for increasing the capacity ofthe systems. Advanced spatial diversity technique is one ofthe methods that can significantly increase the capacity ofthe wireless communication systems. For proper system design and analysis of these systems, we need realistic channels models, which includes both spatial and temporal characteristics of the channel. We propose a new space-time fading channel model, which includes both temporal and spatial correlation of the fading channel. This model can be used for the analysis of wireless communication systems employing antenna array, which are being used in third generation mobile communication systems and are also proposed for the fourth generation mobile communication systems. We have extended the multipath Rayleigh fading wide-sense stationary uncorrelated scattering (WSSUS) channel model using an efficient Monte Carlo method, to include spatial (antenna) correlation. In general, antenna correlation is a function of carrier wavelength, antenna separation, angular spread, and mean direction of arrival (DOA). Our model includes all these effects with small computational complexity. We have taken Laplacian distributed angular spread, which approximates well in realistic outdoor IV environment, though any distribution can be accommodated. We have obtained both Doppler power spectrum and power delay profile of the radio channel using simulations that give expected results. We have also computed antenna correlation of the channel as a function of antenna separation, angle spread, and mean DOA. Results obtained are in agreement with the theoretical results. The time-domain signal processing techniques can be improved upon by using multiple receive antennas and antenna array processing, which enhances the output SINR and offers diversity to mitigate the impairments caused by fading. Next, we have considered the performance of adaptive space-time multiuser detector based on MMSE criteria using simulations in multipath Rayleigh fading channels. We show that by adding antennas at the receiver, SINR gain can be increased even if the correlation between antenna elements is high. We highlight the additional diversity gain obtained if either angle spread is increased or independent fading is assumed across antenna elements (this situation occurs when antennas are kept widely apart). Tracking performance of FC system using VSS-APA is also obtained. Performance of the uncoded systems can be improved by using forward error control (FEC) codes. We have considered the performance of convolutionally coded CDMA systems employing adaptive space-time MMSE receiver for interference suppression in both AWGN and frequency-selective fading channel. The concept of iterative (turbo) equalization was extended to interference suppression in CDMA systems, in which soft information is exchanged between multiuser detector and channel decoder. We next propose a novel adaptive space-time iterative (turbo) equalizer receiver for coded CDMA systems, where the conventional matched filter front end is replaced by adaptive MMSE filter, for the generation of sufficient statistics. The receiver employs an adaptive soft-input soft-output (SISO) multiuser detector and a bank of single user SISO channel decoders. In the first iteration, multiuser detector processes received samples from CMF and provides soft bit estimates. After deinterleaving, these bit estimates are decoded by SISO channel decoders. We have used optimum maximum a posteriori (MAP) decoders. For comparison sake, we have also implemented max-log-map and SOVA (soft output Viterbi algorithm) algorithms. The soft output decoders provide soft outputs for both decoded and coded bits. These soft estimates of coded bits are fed to a symbol spaced adaptive multiuser MMSE equalizer in second or higher iterations, which further provides soft interference cancellation. It is shown that through an iterative process the effects of MAI and ISI can almost be completely overcome and near optimum performance can be obtained.