Learning-Assisted User Scheduling and Beamforming for mmWave Vehicular Networks

Millimeter-wave (mmWave) communication is a promising wireless technology for supporting various intelligent vehicle applications. In mmWave vehicular network systems, acquiring accurate and timely channel state information (CSI) is challenging due to the high mobility of vehicles, making user scheduling and beamforming more difficult. This work aims to enhance both communication throughput and reliability for mmWave vehicular networks without the help of explicit CSI. A closed-form optimal scheduling policy is proposed for the single road side unit (RSU) case based on the Lyapunov optimization framework. For the multiple-RSU case, a multi-agent deep reinforcement learning (DRL) framework is proposed to jointly optimize user scheduling, beamforming, power allocation, and handover decisions. Simulation results demonstrate that the proposed DRL framework significantly enhances communication throughput under reliability constraints compared to baseline algorithms.