Low-Complexity Beam-Domain Channel Estimation with Long-Term Statistics for Large MIMO Detection

This paper proposes low-complexity beam-domain channel estimation using long-term channel statistics in belief propagation (BP) based large multi-input multi-output (MIMO) detection. When the channel correlation matrix between the base station (BS) and each user equipment (UE) is available and used as prior information, maximum a-posteriori probability (MAP) estimation provides the optimal estimation performance. However, it requires undesirably complex large-scale matrix operations at any time the channel statistics is changed. By appropriately selecting beam-domain angular bins for each UE, the proposed method allows us to significantly reduce the computational cost while maintaining the near-optimal performance in terms of the mean square error (MSE) of estimated channel. The selection threshold is adaptively determined based on the prior information such as the channel correlation matrix, statistical beam, and receive SNR. For the subsequent BP-based signal detection, an appropriate covariance matrix is designed while considering the detrimental impact of channel estimation errors. Numerical results show that the proposed method can reduce the computational cost to less than 4% as compared to the MAP estimation, while providing similar MSE performance.