Pattern-Shared Sparse Bayesian Learning for Channel Estimation in FDD Massive MIMO Systems

Channel state information (CSI) is indispensable to fully unleash the great potentials of massive multiple-input multiple-output (MIMO) systems. In this paper, we propose a pattern-shared sparse Bayesian learning (PS-SBL) method for downlink CSI estimation in frequency division duplex (FDD) massive MIMO. To characterize the common sparsity shared by different antennas, we design a pattern-coupled Gaussian prior model. The coefficients in channel vector are divided into several groups of equal length. Each group is associated with one common hyperparameter such that the coefficients in the group bear the same sparsity. Furthermore, by exploiting the expectation maximization (EM) procedure, we develop an iterative algorithm for Bayesian inference, where we treat the channel coefficients as hidden variables and the hyperparameters as unknown parameters. As a result, the obtained posterior mean of channel vector is used as the channel estimate. Simulations show that the proposed PS-SBL method remarkably outperforms the counterparts, and can approach the performance bound enabled by the genie-aided least square (LS) method.