Downlink beamforming in FDD Massive MIMO systems is challenging due to thelarge training and feedback overhead, which is proportional to the number ofantennas deployed at the base station, incurred by traditional downlink channelestimation techniques. Leveraging the compressive sensing framework, compressedchannel estimation algorithm has been applied to obtain accurate channelestimation with reduced training and feedback overhead, proportional to thesparsity level of the channel. The prerequisite for using compressed channelestimation is the existence of a sparse channel representation. This paperproposes a new sparse channel model based on dictionary learning which adaptsto the cell characteristics and promotes a sparse representation. The learneddictionary is able to more robustly and efficiently represent the channel andimprove downlink channel estimation accuracy. Furthermore, observing theidentical AOA/AOD between the uplink and downlink transmission, a joint uplinkand downlink dictionary learning and compressed channel estimation algorithm isproposed to perform downlink channel estimation utilizing information from thesimpler uplink training, which further improves downlink channel estimation.Numerical results are presented to show the robustness and efficiency of theproposed dictionary learning based channel model and compressed channelestimation algorithm.
展开▼
