Reinforcement Learning-Based WMMSE Precoding in UAV Networks

UAV communication has triggered extensive research due to its multifaceted advantages. In this paper, we focus on a scenario involving multiple clusters and the presence of multiple users within each cluster. In order to improve the communication performance, we aim to maximize the Weighted Sum Rate (WSR), and for this goal, we derive a Weighted Minimum Mean Square Error (WMMSE) precoding algorithm suitable for this scenario. However, the fairness of the data rates of different clusters of users within the system is of equal importance. To ensure this fairness, we adopt a strategy to minimize the rate differences among users of individual clusters by adjusting the weighting factor. To implement this strategy, we introduce the Deep Deterministic Policy Gradient (DDPG) reinforcement learning algorithm for training these weighting factors. After simulation experiments, it is demonstrated that the weighting factors obtained through the reinforcement learning algorithm possess effectiveness in improving the fairness of the system.