DeepAntiJam: Stackelberg Game-Oriented Secure Transmission via Deep Reinforcement Learning

In this letter, we present a novel deep reinforcement learning-assisted anti-jamming transmission scheme (DeepAntiJam) to guarantee reliable and creditable transmission in the presence of one smart jammer and multiple eavesdroppers. Specifically, we formulate secure transmission as a Stackelberg game in which the jammer, as the leader, adaptively adjusts its jamming power while the transmitter, as the follower, selects its transmit power and secrecy rate accordingly. Furthermore, the existence and uniqueness of Stackelberg equilibrium are proved. To achieve the Stackelberg equilibrium when the prior knowledge of jammer is unknown, DeepAntiJam is proposed to improve the secrecy throughput, where a solver neural network is utilized to determine the optimal transmission parameter according to the transmission strategy obtained by deep reinforcement learning. Moreover, transfer learning is introduced into the initialization of DeepAntiJam to avoid unnecessary initial random exploration. Simulation results validate that DeepAntiJam can enhance secrecy throughput significantly under the coexistence of smart jammer.