Channel Estimation for FDD Multi-User Massive MIMO: A Variational Bayesian Inference-Based Approach

Abstract

This paper addresses downlink channel estimation for frequency division duplex multi-user massive multiple-input multiple-output systems. Suppose that a base station communicates with $K$ mobile users, then the task is to estimate $K$ channel matrices, each corresponding to one user. Due to the limited scattering in physical propagation, each channel matrix is sparse in the virtual angular domain. Besides, different user links tend to share some common scatterers. As such, different channel matrices may have a partially common sparsity pattern. These observations motivate us to take a variational Bayesian inference based approach for channel estimation. Specifically, we design a Gaussian mixture prior model, which can efficiently capture the individual sparsity in each channel matrix and the partially joint sparsity shared by different channel matrices. Furthermore, we develop a variational expectation maximization strategy to estimate the hyperparameters associated with the prior model and the channel matrices. Compared with the existing counterparts, the proposed approach achieves much better performance in terms of the channel estimation accuracy, while maintaining a low computational complexity.