Abstract-Design of low-complexity detection schemes that can approach near-optimum bit error rate (BER) performance in multiuser or other multiple-input multiple-output (MIMO) systems has always been highly challenging. In this paper we propose and investigate a so-called receiver multiuser diversity aided multi-stage minimum mean-square error multiuser detection (RMD/MS-MMSE MUD) scheme operated in the principles of successive interference cancellation (SIC). The BER performance of the RMD/MS-MMSE MUD is investigated in association with both the direct-sequence code-division multiple-access (DS-CDMA) communicating over both Gaussian and Rayleigh fading channels, and the spacedivision multiple-access (SDMA) communicating over Rayleigh fading channels. Furthermore, we assume that the DS-CDMA and SDMA systems are either full-load or overload. Our studies show that the RMD/MS-MMSE MUD in full-load cases is capable of converging to the optimum BER performance of the maximum likelihood (ML) multiuser detector (MUD). For the overload systems, the RMD/MS-MMSE MUD can make a DS-CDMA or SDMA system support K = 2N users, but still achieve much better BER performance than a corresponding DS-CDMA or SDMA system using the conventional MMSE-MUD to support K = N users, where N denotes the spreading factor of DS-CDMA or the number of receive antennas of SDMA.
I. INTRODUCTIONIn multiuser detection the optimum maximum a-posteriori (MAP) and ML MUDs [1][2][3] are capable of achieving the optimum BER performance that is close to the single-user BER bound. However, the complexity of these optimum MUDs is exponentially proportional to the number of users supported, which becomes extreme even when a moderate number of users are considered. Consequently, the application of the optimum MUDs in practice is limited. For this sake, research efforts have been put on finding the suboptimum solutions with practical complexity for the MAP or ML MUD problems, since the invention of the optimum MUDs [1,2]. In general, the suboptimum MUDs towards the optimum MUD problems can be divided into two classes. The first class aim at designing efficient search or non-search algorithms, such as those summarized in [4] and [5], in order to find approximate solutions for the optimum MUDs. Despite significant complexity reduction, however, these suboptimum algorithms either are still too complex to be implemented or achieve much worse BER performance than the optimum MUDs [5].The second class of suboptimum MUDs do not solve directly the optimum MUD problems. Instead, they approximate the optimum MUD problems by some suboptimum MUD problems that can be solved with lower complexity. This second class of suboptimum MUDs include typically the various types of decision feedback or interference cancellation (IC) assisted MUDs, such as those studied in [4,[6][7][8][9]. The BER performance of the conventional IC-aided MUDs is usually much worse than that of the optimum MUDs, as shown, e.g., in [3,4,6], due to the severe error propagation. In comparison with the conv...