MAHTD-DDPG-Based Multi-Objective Resource Allocation for UAV-Assisted Wireless Network

As an aerial base station (BS), unmanned aerial vehicle (UAV) has been considered as a promising platform to provide wireless data service in future networks due to its flexible, swift and low cost features. However, since the suddenness and randomness of ground users’ (GUs’) data requirements, it is challenging for the UAV BSs to dynamically make decisions to provide real-time data services to GUs. In a multi-mode UAV-assisted wireless network, we formulate a multi-objective optimization problem to minimize the average peak age of information (APAoI) and energy consumption of UAVs, and to maximize the accumulated service data (ASD) for GUs. Therefore, this paper proposes the multi-agent hybrid twin delayed deep deterministic policy gradient (MAHTDDDPG) algorithm with hybrid action space design, which is empowered by centralized training and distributed execution (CTDE) framework. In the proposed algorithm, the UAVs can cooperatively make decisions by sharing the GU status information, in a result of jointly optimizing the UAV trajectory, mode selection and transmit power. Simulation results demonstrate that our proposed approach achieves 79.6% and 120.4% higher reward than MADDPG algorithm and HTD-DDPG algorithm respectively.