Deep Decoder CsiNet for FDD Massive MIMO System

In order to achieve a higher gain in frequency division duplex (FDD) massive multiple-input multiple-output (MIMO) system, it is imperative to feedback downlink channel state information (CSI) from user equipment (UE) to base station (BS). However, excessive feedback overhead makes the task challenging. Hence, in recent years, various deep learning (DL)-based models have been introduced to effectively compress CSI to codeword at UE and then reconstruct back into CSI at BS in order to reduce feedback overhead. The authors of this letter have introduced a new network called Deep Decoder CsiNet (DDCsiNet), which utilizes a deep decoder approach for improving the quality of reconstructed CSI feedback. The proposed DDCsiNet leverages a residual on the residual structure to facilitate the low-frequency information flow. Additionally, it incorporates feature attention to exploit channel dependencies, resulting in a significant enhancement of reconstruction quality. Numerical results are presented to demonstrate that our introduced model can achieve significant performance and outperform other DL methods in terms of CSI feedback quality.