Deep Learning-Based Intelligent Reflecting Surface Phase Shift Control

Hyunsoo Kim, Jiao Wu, Yosub Park, Seungnyun Kim, Byonghyo Shim

Abstract: Intelligent refiecting surface (IRS), a planar meta- surface consisting of a large number of reflecting elements, is a promising solution for improving the spectral efficiency of future wireless systems. In order to maximize the throughput gain of IRS, the base station (BS) needs to acquire not only the conventional direct channel between the BS and user equipment (UE) but also the IRS reflected channel. Since the dimension of the IRS reflected channel is proportional to the number of reflecting elements, the pilot overhead as well as the channel estimation error are tremendous, resulting in a significant data rate degradation. In this paper, we propose a deep learning (DL)-based approach to find out the IRS phase shift maximizing the data rate of IRS-aided communication systems. To achieve this goal, we express the relationship between the noisy estimated channel and the IRS phase shifts using the deep neural network. We then train the network parameters in the direction of maximizing the data rate formulated with the ideal channel. From the simulation results, we demonstrate that the proposed scheme outperforms the benchmark schemes by a large margin.