Joint Beamforming Design for Distributed RIS-Aided Broadcast Communications with Deep Learning

In this paper, we investigate the feasibility of utilizing distributed reconfigurable intelligent surfaces (RISs) to assist the broadcast communication between a multiple-antenna base station (BS) and multiple users. In this context, JoBNet, a novel deep neural network (DNN) that aims at solving the non-convex optimization problem of maximizing the total spectral efficiency is proposed for the first time. In particular, unlike existing DNNs which are only capable of computing one configuration at a time, JoBNet is able to jointly output the beamforming at BS and phase shifts at multiple RISs simultaneously. Moreover, JoBNet operates based on unsupervised deep learning, hence, it requires no label for training. The numerical results show that JoBNet enjoys a satisfactory while providing a low-latency. In addition, the distributed RIS system provides a substantial enhancement in the spectral efficiency compared to the collocated RIS system where all the phase shift elements are utilized in a compact area.