Empowering Satellite-UAV MEC Networks via Matching-Aided Multi-Agent Deep Reinforcement Learning

In the sixth generation (6G) era, unmanned aerial vehicle (UAV) and satellite communications offer promising prospects in terms of the increasing demand for network coverage by the explosive growth of Internet of Things (IoT) devices. However, the lack of spectrum resources has become the bottleneck affecting network service performance. In this paper, we seek to use cognitive radio (CR) technology to assist the satellite-UAV networks. Specifically, we consider an integrated satellite-aerial network (SAN) and UAV-enabled MEC system in which CR is applied for the allocation and management of spectrum resources. Then, we formulate an optimization problem to maximize task execution and data transmission in the SAN system and minimize the energy consumption of UAVs by jointly optimizing the UAV trajectory and the strategy of task offloading. The problem is non-convex with high dynamic and hybrid action space, and thus we propose a matching-aided multi-agent deep reinforcement learning (MADRL)-based algorithm to solve the problem. Simulation results show that the proposed algorithm can improve the efficiency of task collection and reduce energy consumption while ensuring the anti-jamming ability.