Expectation Propagation-Based Signal Detection for Highly Correlated MIMO Systems

Abstract

Large-scale multiple-input-multiple-output (MIMO) systems typically operate in dense array deployments with limited scattering environments, leading to highly correlated and ill-conditioned channel matrices that severely degrade the performance of message-passing-based detectors. To tackle this issue, this paper proposes an expectation propagation (EP)-based detector, termed overlapping block partitioning EP (OvEP). In OvEP, the large-scale measurement vector is partitioned into partially overlapping blocks. For each block and its overlapping part, a low-complexity linear minimum mean square error (LMMSE)-based filter is designed according to the partitioned structure. The resulting LMMSE outputs are then combined to generate the input to the denoiser. In this combining process, subtracting the overlapping-part outputs from the block outputs effectively mitigates the adverse effects of inter-block correlation induced by high spatial correlation. The proposed algorithm is consistently derived within the EP framework, and its fixed point is theoretically proven to coincide with the stationary point of a relaxed Kullback- Leibler (KL) minimization problem. The mechanisms underlying the theoretically predicted performance improvement are further clarified through numerical simulations. The proposed algorithm achieves performance close to conventional LMMSE-EP with lower computational complexity.