Bilinear Gaussian Belief Propagation for Massive MIMO Detection With Non-Orthogonal Pilots

We propose a novel joint channel and data estimation (JCDE) algorithm via bilinear Gaussian belief propagation (BiGaBP) for massive multi-user MIMO (MU-MIMO) systems with non-orthogonal pilot sequences. The contribution aims to reduce significantly the communication overhead required for channel acquisition by enabling the use of short non-orthogonal pilots, while maintaining multi-user detection (MUD) capability. Bilinear generalized approximate message passing (BiGAMP), which is systematically derived by extending approximate message passing (AMP) to the bilinear inference problem (BIP), provides computationally efficient approximate implementations of large-scale JCDE via sum-product algorithm (SPA); however, as the pilot length decreases, the estimation accuracy is severely degraded. To tackle this issue, the proposed BiGaBP algorithm generalizes BiGAMP by relaxing its dependence on the large-system limit approximation and leveraging the belief propagation (BP) concept. In addition, a novel belief scaling method complying with the data detection accuracy for each iteration step is designed to avoid the divergence behavior of iterative estimation in the early iterations due to the use of non-orthogonal pilots, especially in insufficient large-system conditions. Simulation results show that the proposed method outperforms the state-of-the-art schemes and approaches the performance of idealized (genie-aided) scheme in terms of mean square error (MSE) and bit error rate (BER) performances.