Deep Learning Based Passive Beamforming for IRS-Assisted Monostatic Backscatter Systems

Intelligent reflecting surfaces (IRS) can improve the performance of backscatter communication systems by employing reconfigurable phase shifts (or passive beamforming) to favorably configure the wireless propagation medium. However, the design of optimal IRS phase shifts requires channel state information (CSI), which is hard to acquire in a multi-reflection channel. In this paper, we propose a deep learning based framework that learns the desired IRS phase shifts without knowing the channels, to assist the communication of a passive backscatter tag. This is achieved by parameterizing the mapping from the received pilots to the desired configuration of IRS by training a deep neural network (DNN) BIRS-Net on a sufficiently large dataset covering a variety of channel realizations and possible power splitting ratios at the backscatter tag. Simulation results show that the proposed DNN based solution can efficiently learn to maximize the SNR of backscatter transmission and exhibits near optimal performance.