Bayesian learning based multiuser detection for M2M communications with time-varying user activities

Abstract

Machine-to-Machine (M2M) communication plays a significant role in supporting Internet of Thing (IoT). This paper is concerned about multiuser detection (MUD) for massive M2M supported by Low-Activity Code Division Multiple Access (LA-CDMA). In previous work, maximum likelihood (ML) and maximum a posterior probability (MAP) detectors have been developed for such system. The ML detector has exponential complexity, while the MAP detector requires perfect knowledge of user activity factor. In practice, the user activity factor may not be known and could change from time to time. To design MUD detectors addressing these problems, in this paper, we formulate multiple measurement vector (MMV) model for uplink LA-CDMA system with time-varying user activities. Since the transmitted signals have block sparse structure, we introduce the pattern coupled spare Bayesian learning (PCSBL) by using the neighbour coherence of each transmitted signal, which effectively solves the user activity factor unknown problem. Furthermore, we embed the generalized approximate message passing (GAMP) to PCSBL and develop a novel algorithm, called generalized approximate message passing pattern coupled sparse Bayesian learning (GAMP-PCSBL). The GAMP-PCSBL does not require activity factor either, and greatly reduces the computational complexity. Simulation results have shown that the proposed algorithms have superior recovery performance than the conventional algorithms.