CAP-Net: A Deep Learning-Based Angle Prediction Approach for ISAC-Enabled RIS-Assisted V2I Communications

In the vehicle-to-infrastructure (V2I) scenarios, it is challenging to acquire accurate channel state information (CSI) due to the high mobility of the communication channels, which leads to considerable pilot overhead. To tackle this issue, we employ both integrated sensing and communication (ISAC) technique and reconfigurable intelligent surface (RIS) technique to reduce the pilot overhead significantly while maximizing the achievable rate. In particular, we consider a RIS-assisted ISAC system serving the vehicle and design a transmission protocol based on communication-sensing-computing integration architecture for the proposed system, where the ISAC base station (BS) and the dedicated sensors deployed at the RIS receive the reflected echo signals from the user equipment (UE) via the BS-UE-BS link and the BS-UE-sensors link, respectively. Then the CSI of the UE can be acquired from the received signals. Furthermore, to provide both high-quality and low-latency communication services, we propose a covariance-based angle prediction neural network (CAP-Net) to predict the angle parameters facilitating the joint transmit and reflective beamforming design for the next time slot. Simulation results show that the proposed RIS-assisted ISAC system with the CAP-Net achieves better communication performance compared with other baseline schemes and can approach the upper bound in terms of achievable rate.