Abstract
In this paper, we investigate a Deep Reinforcement Learning based delay-constrained relay selection for secure buffer-aided Cognitive Relay Networks (CRNs). We model the relay selection problem in secure buffer-aided CRNs as a Markov Decision Process (MDP) problem, and introduce Deep Q-Learning to solve this MDP problem. In the proposed scheme, delay constraint is considered when the packets arriving at the receiver in CRNs. Moreover, we consider the security of data transmissions in buffer-aided CRNs with an eavesdropper which can intercept the signals from the source and relays. Furthermore, we introduce r-greedy strategy to balance the exploitation and exploration. The result shows compared with Max-Ratio scheme, the proposed scheme enhances the throughput with both delay and security constrained significantly in secure CRNs.