Beamforming Design in Vehicular Communication Systems With Multiple Reconfigurable Intelligent Surfaces: A Deep Learning Approach

In this paper, we propose an innovative framework for vehicular communication utilizing reconfigurable intelligent surfaces (RIS) in the millimeter-wave (mmWave) spectrum. Specifically, we consider a scenario where a base station (BS) employs multiple RISs to serve a vehicular user, considering both perfect and imperfect channel state information (ICSI) to account for real-world conditions. The vehicular user sends an uplink training pilot sequence, which is received by the BS through multiple RISs. We formulate an optimization problem to simultaneously optimize the precoder at the BS and the passive beamforming/passive phase shift matrix at each RIS in order to maximize the achievable rate for the vehicle. By leveraging the received signals with varying phase shifts, we provide a solution based on deep learning (DL) that effectively learns to utilize these signals for predicting the optimal phase shift matrix. Through extensive numerical analysis, we validate the effectiveness of our proposed solution by comparing it to the successive refinement (SR) benchmark scheme. Furthermore, it demonstrates that the proposed DL-based beamforming solution attains a performance level near to the maximum achievable rate with the increase in dataset size, eliminating the need for additional training overhead. The incorporation of RIS amplifies the achievable rate, especially in high-mobility scenarios, without necessitating additional complex beamforming solutions. In addition, we also evaluate the performance of the proposed solution in the presence of ICSI. We analyze the impact of the key system parameters, including the number of elements in the RIS, the speed of the vehicle, the transmit power, the distance between devices, and other relevant factors, on the performance of the considered system.