DFECsiNet: Exploiting Diverse Channel Features for Massive MIMO CSI Feedback

To enhance the performance of a massive multiple-input multiple-output (MIMO) system, downlink channel state information (CSI) should be fed back to the base station (BS) in frequency-division duplexing (FDD) mode. However, large numbers of antennas in massive MIMO systems will lead to an excessive feedback overhead, which poses a large challenge to practical systems. This paper proposes deep learning (DL) based CSI feedback method called DFECsiNet. It is designed based on a simple structure called DFEBlock, which uses two parallel feature extraction paths with different respective fields to extract diverse CSI matrices. Moreover, two specially designed refinement structures are used to progressively recover the initially recovered channel. Experiment results show that DFECsiNet outperforms CsiNet with similar complexities and achieves comparable performances to DS-NLCsiNet with much lower complexities.