Flying Ad-Hoc Network Covert Communications with Deep Reinforcement Learning

Flying ad-hoc networks (FANETs) enable an unmanned aerial vehicle (UAV) to work as both the task operator and the relay node, providing adaptive communication coverage for remote areas. However, the FANET's open communication lines pose significant security risks. In particular, when the eavesdropper detects communication activities between FANET and ground base stations, there is a possibility of utilizing artificial intelligence (AI) technologies to decipher traditional encryption, thereby posing a risk of data leakage. To address the above issue, we propose a FANET covert communications architecture, in which every UAV uses deep reinforcement learning (DRL) to optimize FANET's covertness. Specifically, the UAV applies DRL to help FANET prevent eavesdropping by automatically adjusting the hovering position transmit power, and optimizing the artificial noise power of some UAVs to act as jammers to meet both the high-quality communication requirement and the covert constraints. Moreover, an improved multi-agent deep deterministic policy gradient (MADDPG) algorithm is adjusted to adaptive requirements of the FANET network, and has passed simulation verification. Simulation results show that our scheme can maximize the throughput of FANET under the constraints of concealment and network adaptability, and reduce energy consumption by about 10 percent.