A Learning Approach Towards Secure Cognitive Networks with UAV Relaying and Active Jamming

Confidentiality of information flowing over the air is becoming a critical issue in cognitive radio (CR) towards future 5G networks. In this paper, we employ an unmanned aerial vehicle (UAV) relay to enhance information security in an underlay CR network without direct links. We propose a deep
$Q$
network (DQN)-based secure transmission scheme which jointly optimizes the jamming power of secondary receiver (SR) and the deployment of UAV to maximize secrecy rate. Due to the nonconvex nature of the formulated problem, we decompose it into two-layer sub-problems where the inner layer is to optimize the jamming power while the outer layer is to design the deployment of the UAV. A bi-section search algorithm is developed to solve the inner layer jamming sub-problem and a deep
$Q$
-learning algorithm is designed to approximate the optimal UAV deployment as the outer problem. Finally, numeric simulations are provided to demonstrate the performance superiorities of our proposal in terms of secrecy rate as compared with the baselines.