Joint Active and Passive Beamforming for RIS-Aided Semantic Communication

Reconfigurableintelligent surface (RIS) is a promising technology to enhance the performance of the semantic communication system. Despite its great potential, the performance gain of integrating RIS into semantic communications is not fully exploited. This is mainly because the design objective of the physical layer does not match with that of the learning-based codecs. In this paper, we introduce a unified design framework that aligns the learning and communication objectives based on the Infomax principle. In this framework, the learning (feature encoding and classification) and communication (joint active and passive beamforming) modules are separately designed to reduce complexity. We further develop a block coordinate descent (BCD) algorithm to solve the joint beamforming problem. We conducted the classification task on the CIFAR-10 dataset. Simulation results demonstrate that the proposed method achieves superior classification accuracy compared to the benchmarks.