Capturing the Sparsity and Tracking the Channels for Massive MIMO Networks

Abstract

Due to its enhanced capacity and energy efficiency, massive MIMO has become a critical technology for the 5 G and beyond mobile networks. However, fully achieving its benefits requires the accurate estimation of channel. In this article, we propose a low-overhead characteristic learning, tracking, and monitoring mechanism for the time-varying massive MIMO channels. Specially, we exploit the common spatial sparsity and temporal correlation of the channels. Firstly, using the virtual channel representation and modeling the temporal correlation as an autoregressive process, we formulate the time-varying massive MIMO channel as a sparse signal model. Then, a sparse Bayesian learning (SBL) scheme based on the expectation maximization (EM) is proposed to determine the model parameters of the channel. To achieve the posteriors of different model parameters, the approximate message passing (AMP) is used in the expectation step. Furthermore, the Kalman filtering (KF) with a reduced dimension is used to track the downlink (DL) channel. To observe the change of model parameters and start the relearning process, a monitoring scheme based on the Bayesian Cramer-Rao bound (BCRB) is also developed. Finally, numerical results are provided to demonstrate the performance of our proposed scheme.