Gaussian-categorical Bayesian inference for massive MIMO downlink channel estimation

Abstract

In this paper, we investigate wideband channel estimation in downlink massive multiple-input multiple-output (MIMO) systems, where the channel sparsity patterns may vary significantly along the large base station (BS) array. We tackle the problem of channel estimation under the framework of Bayesian inference. To characterize the spatial non-stationarity in massive MIMO, a hierarchical Gaussian-categorical (GC) prior model is designed for the channel vectors to be estimated. The hyper-parameters associated with the GC prior are artfully coupled such that the GC model has the potential to characterize the varying channel sparsity patterns along the BS array. By resorting to the variational Bayesian inference methodology, we develop an iterative algorithm to adaptively infer the hyper-parameters in the prior model from pilot observations, thereby obtaining the estimates of the channel vectors. Simulations demonstrate that the proposed method remarkably outperforms the state-of-the-art counterparts, and can approach the performance bound realized by the genie-aided least square (LS) method.