Downlink Channel Estimation in FDD Massive MIMO Systems Based on Multi-Resolution Discrimination Dictionary Learning

Compressive sensing (CS) techniques can be used to reduce the pilot overhead, and to improve the performance of channel estimation in massive multiple-input multiple-output (MIMO) systems. Most existing methods adopt the DFT matrix as a basis, which leads to direction mismatch and energy leakage problem in practice. However, the properties of geometry-based stochastic channel model (GSCM) are usually overlooked, but can be exploited to improve the performance of channel estimation. In this paper, a multi-resolution discrimination dictionary learning (MRDDL) method is proposed for downlink sparse channel estimation in frequency-division duplexing (FDD) MIMO systems. By taking into consideration that far scatterers in a specific cell are fixed at a certain position in the space and multipath angle of arrival (AOA) from far scatterers is concentrated in a fixed range, we design a specific dictionary for each far scatterer to reduce the redundant atoms. Simulations are conducted to validate the robustness and effectiveness of the MRDDL method over existing channel estimation methods.