Power Control for D2D Communication Using Multi-Agent Reinforcement Learning

Device-to-device (D2D) communication is a promising and rapidly evolving technology and it plays a significant role in reducing pressure on the base station. In this paper, we focus on how to achieve the goal of maximizing system throughput by adjusting the transmitted power of each D2D user. Due to the uncertainty of channel states and state transition probabilities, this problem can be modeled as the reinforcement learning (RL) algorithm. We first use the fuzzy clustering algorithm to group D2D users of which the attribute values are in a large dissimilarity so as to reduce interference, then each group is treated as an agent in the RL algorithm. Therefore, a multi-agent RL based on fuzzy clustering algorithm is established. Finally, we verified the superiority of the proposed algorithm through simulations.