Exploiting Dynamic Sparsity for Downlink FDD-Massive MIMO Channel Tracking

Abstract

Accurate channel tracking with a small pilot overhead is vital for real-time massive multiple-input and multiple-output (MIMO) communication over a dynamic channel. Recently, compressive sensing has been applied to reduce the pilot overheads by exploiting the spatial and/or temporal correlation of massive MIMO channels. However, most existing channel estimation and tracking algorithms are based on oversimplified channel models with restrictive assumptions, and thus perform poorly under realistic channels. In this paper, we consider downlink frequency-division duplexing-massive MIMO system operating with limited scattering around the base station and flat fading channel is considered. We propose a two-dimensional Markov Model (2D-MM) to capture the 2-D dynamic sparsity (i.e., structured sparsity in the spatial domain and probabilistic temporal dependency of channel in the temporal domain) of massive MIMO channels. The 2D-MM has the flexibility to model different propagation environments in practice. We derive an effective message passing algorithm called dynamic turbo orthogonal approximate message passing (D-TOAMP) to recursively track a dynamic massive MIMO channel with a 2D-MM prior. The proposed D-TOAMP algorithm does not require knowledge of the 2D-MM channel parameters, which could be automatically learned through the expectation maximization framework. Extensive simulations show that the proposed D-TOAMP can achieve significant gains over the existing algorithms under realistic channels.