Soft-output multiuser estimation for asynchronous CDMA channels

We consider several multiuser demodulation schemes for a direct-sequence code-division multiple access (DS-CDMA) system that provide soft-outputs to enhance single-user soft-decision decoders. First, we derive an optimum soft-output multiuser estimation (OSOME) algorithm for asynchronous CDMA channels with white Gaussian noise (WGN). The algorithm generates a posteriori probabilities (APP) for coded bits given the received signal. This forward-recursive algorithm requires complexity and storage exponential in the number of users K and linear in the decision lag L. Second, we outline a reduced-computation suboptimal version of this algorithm (SSOME) which compares favorably in complexity and storage with multiuser soft-output Viterbi algorithm (SOVA). Furthermore, we note that the Bayesian conditional decision feedback estimation (BCDFE) algorithm provides symbol APP estimates that can be considerably improved upon by a slight modification. We derive the modified BCDFE algorithm for whitened multiuser channels. Finally, we present simulation results of a 4-user, convolutionally coded, asynchronous CDMA system. The results indicate that the SSOME algorithm with L=K-1 achieves a 0.4-0.5 dB gain over SOVA with much greater lag (and complexity) and keeps within 0.25 dB of the OSOME algorithm. Moreover, the modified BCDFE algorithm provides a good performance/complexity tradeoff.