Evaluation of Autoencoder-based Communications with Reconfigurable Intelligent Surfaces

Abstract

Reconfigurable intelligent surfaces (RIS) have emerged as a promising technique for wireless communication in 5G and beyond. In an environment with lots of physical obstacles, RIS provides an alternative solution to the coverage problem by beam-forming signal towards the desired direction of the receiver. Due to the fixed constellation diagram, traditional modulation schemes cannot be used for the RIS-aided communication system with the receiver movement. In this work, we aim to design and evaluate an autoencoder-based communication model with RIS support, which can adapt to changes in the environment and the receiver’s position. Specifically, we use neural networks to present the encoder and decoder of the system and the parameters of these networks are trained to minimize the reconstruction error at the receiver. The simulation setup is based on characteristics of the environment and real RIS. Performance results show the benefits of RIS when a physical obstruction is present between the transmitter and receiver. Moreover, we prove that the selection of RIS codeword plays an important role in system performance