Optimizing UAV Trajectory and Spectrum Access for Anti-Jamming Communications: A Heterogeneous MADRL Approach

Abstract

This letter addresses the challenging problem of joint UAV trajectory planning and spectrum access in adversarial environments, where malicious jammers pose a threat. We consider a multi-UAV-assisted communication system where UAVs and ground users (GUs) act as heterogeneous agents collaboratively combating jamming attacks. To maximize network throughput under energy, coverage, access, and signal-to-interference-plus-noise ratio (SINR) constraints, we formulate the problem as a Multi-agent partially observable markov decision process (MAPOMDP). Addressing the intricate coupling in anti-jamming decisions, we propose HAPPO, a novel heterogeneous multi-agent reinforcement learning approach featuring a specifically designed reward function. Simulation results demonstrate that HAPPO significantly outperforms benchmark solutions in terms of both system throughput and energy efficiency, achieving a throughput improvement of approximately 20% over the best baseline.