Dictionary Learning Based Sparse Channel Representation and Estimation for FDD Massive MIMO Systems

Downlink beamforming in FDD Massive MIMO systems is challenging due to the large training and feedback overhead, which is proportional to the number of antennas deployed at the base station, incurred by traditional downlink channel estimation techniques. Leveraging the compressive sensing framework, compressed channel estimation algorithm has been applied to obtain accurate channel estimation with reduced training and feedback overhead, proportional to the sparsity level of the channel. The prerequisite for using compressed channel estimation is the existence of a sparse channel representation. This paper proposes a new sparse channel model based on dictionary learning which adapts to the cell characteristics and promotes a sparse representation. The learned dictionary is able to more robustly and efficiently represent the channel and improve downlink channel estimation accuracy. Furthermore, observing the identical AOA/AOD between the uplink and downlink transmission, a joint uplink and downlink dictionary learning and compressed channel estimation algorithm is proposed to perform downlink channel estimation utilizing information from the simpler uplink training, which further improves downlink channel estimation. Numerical results are presented to show the robustness and efficiency of the proposed dictionary learning based channel model and compressed channel estimation algorithm.