Deep Reinforcement Learning for Energy-Efficient Secure Dual UAV-Assisted Networks

Unmanned aerial vehicle (UAV) communications are expected to play a crucial role in next-generation wireless networks. However, in UAV-enabled communications, the dominant line-of-sight links can pose significant risks to the transmission of confidential data. To address this issue, this work aims to develop a secure dual UAV-assisted data transfer framework for ground users (GUs) while considering multiple eavesdroppers (Eves). Specifically, a rotary-winged UAV base station (BS) serves multiple GUs through user scheduling. Moreover, we employ a UAV jammer to provide data secrecy against multiple Eves that can wiretap the data by exploiting the UAV's aerial links. For this setup, we formulate an average secrecy rate (ASR) maximization problem by joint user scheduling and optimizing the transmit powers and trajectories of BS and jammer under the limited energy budget, maximum flight velocity, collision avoidance, and other mobility constraints. Due to the system's dynamic behavior, the optimization problem is modeled as a Markov decision process and tackled using a deep reinforcement learning approach, i.e., a deep Q-network algorithm. Lastly, the effectiveness of the proposed framework is evaluated by comparing it with several baseline schemes.