Dynamic Channel Matching based on Deep Reinforcement Learning for D2D Communications

Abstract

This paper studied the problem of autonomous channel matching for device-to-device (D2D) pairs in a multiuser cellular network. The goal of each D2D pair is to match an optimal wireless channel to maximize its reward. The reward is defined as the rate of the D2D pairs and limited by the SINR (Signal to Interference plus Noise Ratio) of the cellular user on the current channel. This strategy maximizes a certain network D2D throughput in a distributed manner without requiring online coordination or message exchange among users. We describe this problem as a random non-cooperative game with multiple players (D2D pairs), where each player becomes a learning agent, whose task is to learn its best strategy (based on locally observed information). Then, we designed a multi-user learning algorithm based on double deep Q-network (DDQN), which converged to Nash equilibrium (NE) of mixed strategy. After simulation verification, the algorithm can enable each user to obtain a best strategy to obtain a high communication rate through online or offline learning. And the algorithm has a faster convergence speed compared to the similar method.