Limited Quantized RIS-Aided Communication Systems Based on End-to-End Online Learning

This paper investigates an end-to-end learning-based finite quantized reconfigurable intelligent surfaces (RIS)-assisted communication system, which jointly optimizes the passive beamforming at the RIS, along with the active beamforming at both base stations and users. The system design emphasizes the solution against the impact of limited quantization of RIS phase shifts on communication performance. The experimental results indicate that 3-bit is a crucial quantization threshold. If the RIS quantization is below this threshold, end-to-end optimization cannot achieve the forward error correction limit. However, when the RIS quantization is no less than this threshold, our method enables the RIS-assisted communication system to approach the bit error rate performance of the full-precision system.