Transformer-Based RIS Phase Shift Control for Ultra-Low Latency V2X Systems
Hyunsoo Kim, Seungnyun Kim, Jiao Wu, Byonghyo Shim
Abstract
To meet the stringent latency and reliability requirements of vehicular applications (e.g., vehicle platooning and autonomous driving), reconfigurable intelligent surface (RIS)-aided vehicle-to-everything (V2X) communications have received great attention recently. A major challenge in the RIS-aided V2X systems is channel aging effect, the mismatch between the channels at the channel estimation stage and the data transmission stage caused by the short coherence time of V2X channel. In this paper, we propose a deep learning (DL)-based phase shift control technique for RIS-aided V2X systems to overcome the channel aging effect. Key idea of the proposed Transformer-based RIS phase shift control (T-RPSC) is to predict the proper phase shifts from the past RIS-aided channels using Transformer. The attention mechanism in Transformer quantifies the long and short-term correlations between the input (i.e., past RIS-aided channels) and output (i.e., future RIS phase shifts) to assign proper attention weights to input data based on the measured correlations. In doing so, we can effectively extract temporally correlated features, thereby improving the transmission latency reduction performance of RIS phase shifts. From the simulation results, we demonstrate that T-RPSC achieves more than 23% latency reduction over the alternating direction method of multiplier (ADMM)-based phase shift control scheme.