A two-stage adaptive multiuser detector in an additive white Gaussian noise code-division multiple-access channel is proposed and analyzed. Its first stage is an asynchronous one-shot decorrelator which in terms of computational complexity only requires inversion of K symmetric K/spl times/K matrices for all K users. In addition, the K inversions can be done in parallel, and the computed results for one user can be reused by all other users as well, resulting in a latency of only one bit, same as its synchronous counterpart. The decorrelated tentative decisions are utilized to estimate and subtract multiple-access interference in the second stage. Another novel feature of the detector is the adaptive manner in which the multiple-access interference estimates are formed, which renders prior estimation of the received signal amplitudes and the use of training sequences unnecessary. Adaptation algorithms considered include steepest descent (as well as its stochastic version), and a recursive least squares-type algorithm that offers a faster transient response and better error performance. Sufficient conditions for the receiver to achieve convergence are derived. The detector is near-far resistant, and is shown to provide substantial steady-state error performance improvement over the conventional and decorrelating detector, particularly in the presence of strong interfering signals.

A reduced-state sequence estimator (RSSE) for linear dispersive channels is described. The technique is based on partitioning the set of all channel states in a way that defines a trellis with fewer states. The new technique generalizes trellises that were previously obtained by Duel-Hallen and Heegard (1989) and Eyouboglu and Qureshi (1988). While the approach followed by Duel-Hallen results in powers of two-state trellises, the new technique generates trellises with an arbitrary number of states. Using the state partitioning approach one could also generalize the trellises obtained by Eyouboglu. Therefore, the approach described in this paper provides a better performance/complexity tradeoff.