Joint Beamforming for RIS-Assisted Multicast System Based on Model-Driven Deep Learning
Chunxia Ding, Weijie Jin, Xiao Li, Jing Zhang · 2023
The reconfigurable intelligent surface (RIS) has recently been studied extensively due to its ability of intelligently changing the wireless channels to improve communication performance. In this paper, a RIS-assisted downlink multicast communication system is taken into consideration. Under the constraints of maximum transmit power, we aim to jointly design the base station active beamforming and RIS passive beamforming to maximize the achievable sum rate. A model-driven deep learning (DL) method is designed by unfolding the alternating projected gradient (APG) algorithm. We introduce trainable variables into the algorithm to accelerate its convergence. A delicate training method is proposed to ensure that the training results are rarely affected by the initial values, while achieving superior performance. The results of simulation indicate that the proposed DL method outperforms the benchmark algorithms in terms of complexity and achievable sum rate. Specifically, the model-driven DL method has a runtime that is approximately 10% of the APG algorithm and 0.05% of the majorization-minimization (MM) algorithm to achieve the same performance.