Resource Allocation in Vehicular Communications using Graph and Deep Reinforcement Learning

Cellular based vehicle-to-everything (V2X) communications have recently gained more interest from both academia and industry. However, there exist many challenges in cellular-based V2X communications in which resource allocation is one of the main challenges. In this paper, we propose a graph and deep reinforcement learning-based resource allocations in which channels for vehicular communications are assigned in a centralized manner by the base station whereas vehicular user equipment uses deep reinforcement learning for distributed power control. Graph-based channel allocation includes a weighted bipartite matching and clustering scheme and relies on strictly limited channel state information (CSI). Whereas, power selection is performed using deep reinforcement learning where each agent selects the transmission power to maximize the aggregated V2V data rate. Our proposed scheme relies on realistic channel assumption with minimum transmission overhead. In addition, we have also performed simulations and have shown that our scheme is better compared to previous schemes in terms of sum V2V and sum V2I capacity.