Design of an energy efficient resource allocation techniques for D2D based V2X communication

Vehicle-to-Everything (V2X) communication technology is expected to become an integral part of daily life in the near future. To enhance its efficiency and reliability while supporting autonomous driving, it must incorporate advanced features. This study utilizes both cellular links (CUE) and device-to-device (D2D) communication for transmitting safety-critical information. To optimize spectrum utilization, resources are dynamically shared between cellular and D2D users. To achieve efficient power and resource allocation, a multi-agent deep Q-learning approach is proposed, operating in a distributed manner. Given the rapidly changing environmental conditions and the absence of complete Channel State Information (CSI), deep Q-learning offers significant advantages. Specifically, it enhances energy efficiency, improves the successful transmission of safety-critical DUE information, and increases the sum capacity of CUE links for media content transmission. The multi-agent deep reinforcement learning (MADRL) model is initially trained in a centralized framework and subsequently deployed in a distributed manner. The proposed algorithm is benchmarked against Single-Agent Reinforcement Learning (SARL) and a Random Baseline approach. Simulation results demonstrate that MARL outperforms both SARL and the Random Baseline across all evaluated parameters, highlighting its effectiveness in V2X communication.