Reconfigurable Intelligent Surface for Joint Sensing and Communication in Vehicular Networks Based on Deep Reinforcement Learning
TL;DR
The results demonstrate that integrating time–space RIS partitioning with constraint- aware deep reinforcement learning can effectively balance communication utility, information timeliness, and sensing reliability in dynamic vehicular environments.
Abstract
Reconfigurable Intelligent Surface (RIS)-assisted Integrated Sensing and Communication (ISAC) has the potential to improve coverage and sensing capability in vehicular networks under blockage and high-mobility conditions. However, the rapid variation of vehicular channels makes joint communication and sensing resource allocation highly challenging. This study proposes RIS-JSC, a time–space dual-domain RIS-assisted framework that combines A/B-segment time division with RIS aperture partitioning for joint communication and sensing. A sensing reliability threshold is introduced through CFAR calibration. To solve the resulting coupled non-convex optimization problem, a Constraint-Aware Soft Actor– Critic (CA-SAC) algorithm is developed to jointly optimize time splitting, RB reuse, transmit power, RIS phase shifts, and communication/sensing sub-array control. Simulation results show that the proposed method reduces the average V2I AoI, improves the V2V within-deadline success probability, and increases system throughput compared with SAC, PPO, TD3, DDPG, and fixed/random RIS baselines, while satisfying the sensing reliability constraint. The results demonstrate that integrating time–space RIS partitioning with constraint- aware deep reinforcement learning can effectively balance communication utility, information timeliness, and sensing reliability in dynamic vehicular environments. The proposed framework provides an effective deep reinforcement learning solution for joint sensing and communication optimization in RIS-assisted vehicular networks.