A Generalized Neural Network-based Optimization for Multiple IRSs-aided Communication System

Intelligent reflecting surfaces (IRSs) are considered a promising and revolutionary technology for promoting future six-generation networks with an ever-growing number of smart devices and applications. IRSs-aided wireless communication networks architecture maintains both spectrum and energy efficiency. In this paper, the joint design of the beamforming matrix of transmitting BS and reflect phase shifts of connected IRSs that minimize the total transmit power from the source BS is investigated. Using the conventional alternating approach to find converged optimal solutions is highly complex; hence, it is unsuitable for run time implementation with the dynamic environment. Motivated by this, we introduce a new generalized neural network (GRNN) - based optimization model that aims to optimize the joint design simultaneously as the GRNN output. Specifically, the proposed network is trained offline using a supervised learning approach with a wide range of dynamic channel instances, while real-time predictions are obtained at online deployment. Obtained simulation analysis shows that the proposed approach achieves robust training and validation performance while significantly reduces the total computation complexity compared with the alternating-based algorithms.