Sparse Bayesian Learning Using Complex t-Prior for Massive Multi-User MIMO Channel Estimation

This paper proposes a novel beam-domain channel estimation (CE) algorithm based on sparse Bayesian learning (SBL) using complex t-prior for massive multi-user multiple-input multiple-output (MU-MIMO) systems. Due to the sidelobe leakage and insufficient observation resolution, the equivalent channel after digital beamforming at the receiver does not have a sparse structure strictly consisting of zero/non-zero elements, but has a structure characterized by differences in signal intensity consisting of a large number of small non-zero elements and a few large elements. To fully capture this pseudo-sparse structure, a complex t-distribution with appropriate degrees of freedom (DoF) is incorporated into the SBL algorithm as a hierarchical Bayesian model. This heavy-tailed prior allows for efficient beam-domain CE accounting for small but non-negligible elements, which is verified by the analysis of regularization based on an equivalent optimization problem. Simulation results show that the proposed method significantly outperforms the state-of-the-art (SotA) sparse signal recovery (SSR)-based alternatives in sub-6 GHz wireless communication scenarios.