Improvement of an autoencoder network for jointly information transfer by RIS and transmitter

This paper proposes an autoencoder (AE)-based deep learning (DL) framework for reconfigurable intelligent surface (RIS)-assisted wireless communication systems. In this framework, RIS transmits information through its passive reflecting elements. In wireless networks, RIS reflects incident signals toward the receiver by adjusting the phase shifts of their passive elements. They also need to support synchronization and power allocation by transmitting status updates and channel state information to a central controller. Accordingly, this work develops a DL-based framework for joint RIS information transmission and signal reflection to mitigate channel impairments and improve communication reliability. The proposed framework jointly performs transmitter beamforming and reliable detection of the information transmitted by both the transmitter and the RIS. However, the joint design of the transmitter, RIS, and receiver results in a non-convex optimization problem. To address this issue, three end-to-end (E2E) DL models based on AE architectures are proposed. These networks model the transmitter, RIS, and receiver using custom-designed input features. Simulation results demonstrate improved system performance and a significant reduction in symbol error rate compared with conventional RIS-assisted communication schemes.