Censored Approximate Message Passing Based Multiuser Detection in Massive MIMO

Abstract

Recently, a low-complexity multiuser detector has been constructed for massive multiple-input and multiple-output (MIMO) based on generalized approximate message passing (GAMP). It is featured on avoidance of matrix inversion and parallelization on hardware implementation. In order to further decrease the complexity of the GAMP detector, this paper proposes a censored GAMP (CEN-GAMP) detector, which iterates between a one-step GAMP iteration and message censoring operation. Different to GAMP, CEN-GAMP will judge the signal elements to be converged-or-not based on their values on current and previous iterations. The converged ones will be directly censored and not updated anymore. Consequently, CEN-GAMP enjoys much lower computational complexity in comparison to GAMP. Simulation results indicate that, similar as GAMP, CEN-GAMP still apparently outperforms the classical minimum mean square error (MMSE) detector. Moreover, its complexity is almost one order of magnitude lower than that of GAMP, and its detection performance shows only 0.3-0.5 dB signal-to-noise ratio (SNR) losses.