A double network structure anti-jamming algorithm based on deep reinforcement learning

At this stage, when dealing with intelligent jammer with environmental cognition, most anti-jamming methods only consider how to maximize avoidance of it. However, the user’s previous waveform and frequency actions have been leaked. As the intelligent jammer continues to learn, the performance of anti-jamming will decline. Existing literature has proposed the idea of hidden anti-jamming. The user uses rational communication action-making to prevent the jammer from obtaining the information of the user so that the jammer cannot target the user for jamming, but its model is relatively simple and cannot cope with complex jamming types. This paper designs a kind of decision related judgment module between jammer and user based on Generative Adversarial Networks (GAN), which uses the environment state received by the user to reversely infer the user’s decision whether to avoid jammer’s sense. Specifically, the neural network is used to fit the environment state under known user information, and calculate the loss between it and the real environment state to evaluate whether the user’s decision is to avoid jammer’s sense. Then, this paper proposes a deep reinforcement learning algorithm with double-network structure, which can deal with various types of complex jamming while ensuring anti-jamming performance. The final simulation results show that the improved deep reinforcement learning algorithm proposed in this paper can improve the anti-jamming performance, achieve hidden anti-jamming, and have a wider range of applications.