Previous work has identified iterative multiuser decoding techniques that are often only limited by theoretical channel capacity. These investigations usually dealt with the case where user waveforms are decorrelated to some degree. Investigations with identical waveforms indicate that the theoretical capacity limits are not always achieved. Performance has a threshold-like behaviour, as one might expect with a capacity limit, but the threshold is, in some cases, above the theoretical capacity for the channel under investigation. In this paper, an explanation for this behaviour is proposed.

Two multi-user soft-interference cancellation algorithms are presented for synchronous narrow-hand BPSK users. The two techniques are designed for multiple user detection where the users use highly correlated waveforms for modulation. The techniques perform close to the optimal cross-entropy multi-user decoder on the K-symmetric channel for cross-correlations up to 0.5. For higher cross-correlations on this channel, we introduce an initial decorrelation step prior to the first iteration, which greatly improves the performance for the high values of cross-correlation. The complexity per bit per iteration increases linearly for the two interference cancellation algorithms rather than exponentially as for the cross-entropy decoder.

The combination of forward error correction (FEC) coding and random interleaving is shown to overcome the limitations of multiuser detectors/decoders when the user cross correlations are high. In particular, one can asymptotically achieve single-user performance in a highly correlated multiuser system. In addition, an optimal iterative multiuser detector is derived from iterative techniques for cross-entropy minimization. A practical suboptimal implementation of this algorithm is presented, and simulations demonstrate that, even with highly correlated users, it achieves optimal asymptotic efficiency. The effects of the theoretical limits on channel capacity are evident in many of the simulation results. The complexity of the suboptimal algorithm is approximately (O(2/sup K/)+O(2/sup /spl kappa//)) per bit per iteration where K is the number of users and /spl kappa/ is the code constraint length.

In this correspondence, the relationship between iterative decoding and techniques for minimizing cross-entropy is explained. It is shown that minimum cross-entropy (MCE) decoding is an optimal lossless decoding algorithm but its complexity limits its practical implementation. Use of a maximum a posteriori (MAP) symbol estimation algorithm instead of the true MCE algorithm provides practical algorithms that are identical to those proposed in the literature. In particular, turbo decoding is shown to be equivalent to an optimal algorithm for iteratively minimizing cross-entropy under an implicit independence assumption.