Reliability-based near maximum-likelihood multiuser detection

The prohibitive-exponential in the number of users-computational complexity of the maximum likelihood (ML) multiuser detector for direct-sequence code-division-multiple-access (DS/CDMA) communications has fueled an extensive research effort for the development of low complexity multiuser detection alternatives. Notable examples are the zero-forcing ("decorrelating") and minimum-mean-square-error (MMSE) linear filter receivers. We show that we can efficiently and. effectively approach the error rate performance of the ML optimum multiuser detector as follows. We utilize a multiuser zero-forcing or MMSE filter as a pre-processor and we establish that the output magnitudes, when properly scaled, provide a reliability measure for each user bit decision. An ordered reliability-based error search sequence of length linear in the number of users returns the most likely user bit vector among all visited options. Numerical and simulation studies for moderately loaded systems that permit exact implementation of the optimum detector indicate that the error rate performance of the optimum and the proposed detector are nearly indistinguishable over the whole pre-detection signal-to-noise ratio (SNR) range of practical interest. Similar studies for higher user loads (that prohibit comparisons with the optimum detector) demonstrate error rate performance gains of orders of magnitude in comparison with straight decorrelating or MMSE multiuser detection.