IRJ-MARL: Intelligent Radar and Jammer Zero- sum Waveform Game based on Multi-agent Reinforcement Learning

In modern electronic warfare, jammer adopts a more intelligent jamming strategy than before. Recent radar anti-jamming methods focus on using reinforcement learning (RL) to combat the fixed expert strategies which ignore the intelligence of the jammer. To address that issue, a method using multi-agent reinforcement learning (MARL) to design an intelligent radar and jammer zero-sum gaming scenario is proposed. Firstly, two sets of jamming and anti-jamming models are constructed, which incorporate changes in waveform duration and frequency. Secondly, a method which joint MARL and supervised learning (SL) is proposed to analyze the electronic environment and make decisions for radars and jammers. Finally, the simulation results confirm that the proposed method outperforms traditional RL algorithms in combat effectiveness, highlighting the impact of specifically tailored reward functions on the interplay between the radar and jammer.