Ralf Müller

Power and Bandwidth Efficiency of Multiuser Systems with Random Spreading

Band 10, Kommunikations- und Informationstechnik, Herausgeber: B. Girod und J. Huber, Shaker Verlag, Aachen, 1999.


This work introduces a simplified model for uplink (reverse link) communication applying the code-division multiple-access (CDMA) protocol with powerful forward error-correction coding on the mobile radio channel. This model comprises synchronous transmission, channel coding at capacity limit, spherical random spreading sequences, perfect channel state information about slow fading available at both transmitter and receiver site, and negligence of multipath propagation and fast fading. The position in the power-bandwidth plane is used as fundamental means of performance evaluation.

As optimum multiuser coding implies prohibitive complexity, several suboptimum methods are analyzed which allow to break up multiuser coding into multiuser detection and single-user coding. They include linear detection, decision feedback detection, maximum likelihood, reduced state, and iterative detection with hard decisions. The latter ones turn out not to offer adequate performance related to their complexity. Therefore, only linear and decision feedback detectors are analyzed more detailed.

These detection methods simplify quantitative comparisons between their various particular realizations, as their respective performances hardly depend on both number of users and spreading factor, but only on the ratio of the latter parameters which is termed load. This means that for fixed load neither the spreading factor nor the number of users significantly affects performance. More precisely, the supportable information rate slightly decreases for increasing spreading factor if the load is fixed. The number of users is assumed to be infinite for sake of simplicity, analytical tractability, and, as it turns out, to serve as a lower bound on possible performance. This allows the application of theorems on the asymptotic eigenvalue distributions of random covariance matrices.

The decision feedback detectors are shown to significantly outperform the linear detection methods for all signal-to-noise ratios at the expense of only little increase in complexity. In particular, minimum mean-squared error (MMSE) decision feedback detector achieves the same performance as optimum joint multiuser coding if the users rates are appropriate. If they are not, however, the drawback is still very small.

In the presence of large differences among the signal attenuation factors of the users which can be tracked by both transmitter and receiver, e.g. caused by shadowing and/or path loss, the advantages of decision feedback detectors further grow, as they provide capabilities to make sophisticated use of these differences in signal attenuations, while linear detectors cannot do better than combating them by power control.

If perfect single-user coding, that is coding at channel capacity, and perfect interference cancellation, i.e. no residual interference remains after subtraction of the interfering signals, are available, randomly spread CDMA with MMSE or conventional decision feedback detection can even outperform any orthogonal multiple-access method with respect to averaged transmitted power. However, even small imperfections of multiuser systems based on decision feedback structures lead to significant degradations in many cases. Thus, in practice it is a difficult challenge to actually beat the performance of orthogonal systems, especially if the time slot durations or the widths of the subbands of the latter ones are able to be adapted according to the instantaneous attenuation factors.