Blind multiuser detection in multirate CDMA based on cyclic LMS adaptation

The problem of blind adaptive multiuser detection in multirateCDMA systems is considered. Indeed, since symboldetection in multirate CDMA systems requires periodicallytime-varying processing of the observables, classical LMS and RLSadaptive algorithms, which assume that the solution to be trackedis time-invariant or slowly time-varying, are not suited for blindadaptive multiuser detection in a multirate system. While a cyclicRLS algorithm has recently appeared in the literature, thispaper focuses on the development of LMS-based cyclic filteringalgorithms. In particular, cyclic versions of the standard LMSalgorithm, of the LMS algorithm with iterate averaging and of theLMS algorithm with adaptive step-size are derived. Interestingly,the last two algorithms are shown to exhibit a convergence speed close to thatof the cyclicRLS procedure, but with an order of magnitude lower computationalcomplexity.An adaptive procedure for the automatic selection ofthe algorithm periodicity is also presented, which is based on aminimum mean-output-energy criterion, and that obviates theneed for knowledge of the transmitted data-rates from theinterfering signals.Moreover, the case of known multipathfading channels is also examined. In particular, it is shown that theproposed cyclic LMS algorithms can be used to achieve RLS-likeperformance also in the presence of multipath distortion.Extensive computer simulation results, along with some analytical convergence results, confirmthat the proposed algorithms are effective and achieve very satisfactoryperformance.