Mobile Edge Intelligence and Computing With Star-RIS Assisted Intelligent Autonomous Transport System

When communication signals are weak, the advantages of on-board edge intelligence cannot be fully utilized. To tackle this challenge, the introduction of a key technology in the sixth-generation mobile network (6G)—reconfigurable intelligent surface that can simultaneously transmit and reflect signals (star-RIS)—is proposed. In star-RIS-enhanced intelligent transportation system (ITS), intelligent vehicles use star-RIS to upload local training models and perform global model training on the roadside unit (RSU) side. In this paper, the goal is to minimize system delay and loss function of the learning model, and comprehensively considering constraints such as system bandwidth, star-RIS phase shift, vehicle transmission power and beamforming, and vehicle selection. Firstly, for the optimization of phase shift, transmission power and beamforming caused by the introduction of star-RIS, the block coordinate descent method and Lagrange dual algorithm are used to simplify the solution. Secondly, for system delay and global model training, federated learning (FL) algorithm based on double deep Q-network (DDQN) is utilized. This approach leverages policy optimization and data privacy protection to provide an intelligent resource optimization scheme for ITS. In addition, the effectiveness of the proposed algorithm is validated through extensive simulations and numerical analyses. The results show that the algorithm enhances the adaptability and service quality of the system significantly.