An Inverse Reinforcement Learning Method to Infer Reward Function of Intelligent Jammer

This paper takes into account a frequency-agile radar system designated as “us” and an adversarial spot jammer designated as “enemy”, where the enemy jammer is modeled as an agent, and the radar’s strategy is characterized by its state transition probability in a Markov decision process (MDP). To comprehend the underlying attack motives of an intelligent jammer, a radar system must infer the utility function of the jammer. To accomplish this, the radar system must gather a dataset of actions taken by the jammer through observation of their transmitted signal. Next, a maximum entropy inverse reinforcement learning approach is utilized to recover the jammer’s reward function. After recovering the jammer’s reward function, the radar can infer the jammer’s attack intent, and develop appropriate anti-jamming strategies to effectively mitigate against the intelligent jammer. Simulation results demonstrate the efficacy of the proposed method.