Neural network based multiuser detectors for DS-CDMA

Abstract

DS-CDMA is a popular multiple access technique for wireless communications. However its performance is limited by multiple access interference. The work presented here focuses on Multiuser detection that deals with the demodulation of mutually interfering signals to improve the performance of DS-CDMA. The Optimal Multiuser Detector (OMD) proposed by Verdu is known to possess a prohibitive complexity that makes it unsuitable for practical implementation. This has resulted in formulation of several sub-optimal multiuser detectors with reduced complexity and near optimum performance.Two novel sub-optimum multiuser detectors are presented in this report: Hybrid Matrix Graduated Non-Convexity Annealed Neural Network (Hybrid-MGNANN) and Hybrid Radial Basis Function Network (Hybrid-RBFN). We carried out extensive simulations in order to compare the error probability performance of our proposed detectors with other competing sub-optimum multiuser detectors. We show that the error probability performances of our proposed detectors are significantly better than other suboptimum MUDa??s and approaches the performance of OMD.