Deep Learning-Based Joint Channel Estimation and CSI Feedback for RIS-Assisted Communications

Abstract

In reconfigurable intelligent surface (RIS)-assisted communication systems, downlink channel state information (CSI) is essential for designing beamforming at both the base station and the RIS. Hence, accurate downlink channel estimation and CSI feedback are crucial for ensuring optimal system performance. Current methods focus on independently enhancing the processes of channel estimation and CSI feedback. However, this separate optimization strategy is inefficient and introduces errors from the previous process into the subsequent one, resulting in degraded system performance. To address this issue, this letter presents a novel deep learning-based joint channel estimation and CSI feedback scheme named JDCNet for RIS-assisted communications. This scheme directly incorporates preliminary channel estimation with errors into the compression and feedback process, enabling the joint design and optimization of these two critical tasks. Simulation results demonstrate that the proposed scheme outperforms benchmark methods regarding channel reconstruction accuracy while significantly reducing time and space complexity.