Summary

In this paper we present new multiuser MLSE algorithms for unknown, fast time-varying channels channels. We model the channels as Gauss-Markov, and marginalize over the channel model parameters to obtain the MLSE cost. For a large number of users and/or channels with deep memory, the optimum, exhaustive search, MLSE algorithm is impractical. We describe a PSP approach, based on list Viterbi, to prune the number sequences kept at each trellis stage. Additional computational savings is realized using a new iterative decision-feedback approach to RSSE.

In the simulation results presented, the proposed PSP/RSSE algorithm performs well compared to optimum and more computationally intensive approximate MLSE solutions. A more extensive performance analysis is in progress.