Reinforcement Learning based Anti-jamming Frequency Hopping Strategies Design for Cognitive Radar

Frequency agile (FA) radar is capable of altering carrier frequency randomly, which is especially useful for radar anti-jamming designs. Obviously, random frequency hopping is not the best choice if the radar can learn the jammer' strategy. In this paper, a novel frequency hopping strategy design method is proposed for cognitive radar to defeat the smart jammer, in which the radar does not know the exact jamming model. Q-learning and deep Q-network (DQN) is utilized to solve this problem. By applying the reinforcement learning algorithm, the radar is able to learn the jammer's strategies through the interaction with environment and adopt the best action to obtain high reward. The learning performance of DQN is much better than that of Q-learning especially when the available frequencies are large. The proposed method can improve the signal-to-interference-plus-noise ratio (SINR) for the radar when the jamming model is not available. Numerical results are given to illustrate the effectiveness of the proposed method.