Multi-User Collaborative Jamming Deception for UAV Communications: A Multi-Agent Reinforcement Learning Based Method

Abstract

Due to the flexible mobility and high-quality of line-of-sight (LoS) channels, unmanned aerial vehicle (UAV) has begun to play an important role in wireless communications. However, the broadcasting nature of wireless communications and the limited payload of UAVs render the spectrum vulnerable to malicious jamming attacks. To guarantee the performance of UAV communications, this paper focuses on reactive jamming, and sets up an actively exposed deception band to attract partial power of jamming. Specifically, we first model the anti-jamming process with a Stackelberg game model, under the assumption that the rational behavior of jamming is known. Then, we analyze the theoretical optimal strategies of the jamming as well as the users in UAV communication to reach the equilibrium of the above game model. Finally, we design a collaborative multiagent jamming deception method to achieve anti-jamming in the absence of environmental and jamming information. This method is based on the centralized evaluation network at the UAV and decentralized policy network at each user. Simulation results show that the anti-jamming performance of the proposed method can approach the theoretical upper bound and significantly outperform other benchmark methods.