An Energy Effective RIS-Assisted Multi-UAV Coverage Scheme for Fairness-Aware Ground Terminals

Unmanned aerial vehicle (UAV)-assisted communications are critical in regional wireless networks. Using reconfigurable intelligent surfaces (RISs) can significantly improve UAVs' throughput and energy efficiency. Due to limited communications resources, the data transfer rate of ground terminals (GTs) could be slower, and the throughput may be low. Using RIS-assisted UAVs can effectively address these limitations. This paper focuses on optimizing the three-dimensional (3D) trajectory of the UAV and the scheduling order of the GTs and designing the phase shift of the RIS to maximize energy efficiency while meeting the limited energy and fair service constraints in the case of fair service GTs. To address the non-convexity of this problem, we propose a triple deep q-network (TDQN) algorithm, which better avoids the overestimation problem during the optimization process. We propose an improved k-density-based spatial clustering of applications with noise (K-DBSCAN) clustering algorithm, which is characterized by the ability to output the initial movement range of the UAV and prune the deep reinforcement learning (DRL) state space by the initial movement range to speed up DRL training based on the completion of the partitioning deployment work. A fair screening mechanism is proposed to satisfy the fairness constraint. The results show that the TDQN algorithm is 2.9% more energy efficient than the baseline. The K-DBSCAN algorithm speeds up the training of the TDQN algorithm by 59.4%. The fair screening mechanism reduces the throughput variance from an average of 114099.9 to an average of 46.9.