A Learning Based Approach for Intelligent Reflecting Surface (IRS) Assisted Return Link Beamforming

Intelligent reflecting surface (IRS) is a promising technology to enhance wireless communication networks. The IRS is heralded as a cost-effective, energy-efficient solution for extending the range and enhancing capacity of the wireless systems. The main premise of the IRS is that it can impart controlled phase offsets to the multiple replicas of the transmitted signal such that they arrive phase coherently at the destination. In this paper, we propose a deep-learning (DL) based approach for an IRS-aided return link beamformer. Our proposal determines both the IRS phase shifts as well as the receiver-side beamforming coefficients. The Deep Neural Network (DNN) that we deploy directly generates the quantized form of the IRS phase coefficients that can be directly used at the IRS thereby jointly solving the problems of coefficient determination and their quantization. We conduct experimental evaluation of our proposal and and show that our scheme is effective in solving the dual problems of determining the IRS phase shifts and the beamforming coefficients.