Hybrid Nearest-Neighbor Approximation for Message-Passing MIMO Detectors

Recently, message passing algorithms (MPAs) have been widely investigated in multiple-input multiple-output (MIMO) detection for achieving a good balance between performance and complexity. Some MPAs approximate the messages as Gaussian distribution using moment matching (MM), which involves hardware-unfriendly operations. In this paper, we demonstrate a general hybrid nearest-neighbor approximation (HNNA) optimized by multi-objective evolutionary algorithm (MOEA) to reduce the complexity of MM. The grouping strategy is then proposed to reduce the search space with improved optimization efficiency. Applied to approximate message passing (AMP) detector, numerical results indicate that with the proposed HNNA-based MM, AMP detector can achieve up to 17.6% complexity reduction with negligible performance loss. The proposed HNNA-based MM can also be applied to other MPAs, being a promising approach in simplifying the complexities.