Joint Trajectory and Resource Optimization for Secure UAV Communications Based on Graph Attention Reinforcement Learning

UAV communications have been a vital technology for 5G and the forthcoming 6G wireless communications. However, the realization of secure communication remains a critical challenge due to the broadcast nature of UAV links. In this paper, we investigate a UAV-assisted framework, where the UAV is deployed as a flying base station (BS) to serve multiple user equipments (UEs) against eavesdropping threats. We propose a Graph Attention Reinforcement Learning for Secure Communications (GARL-SC) algorithm, where the complex relational features among UAV and UEs are extracted by Graph Attention Network (GAT) for jointly optimizing the UAV trajectory. We further develop Secure Hierarchical Bandwidth Allocation (SHBA) algorithm and Adaptive Hybrid Power Allocation (AHPA) algorithm for optimizing transmission power control and bandwidth allocation, aiming at maximizing the overall throughput and secrecy rate. From the experimental results, the proposed algorithm has considerable performance gains over existing traditional and Deep Reinforcement Learning (DRL)-based baselines in terms of the optimization objective.