Neural Fictitious Self-Play for Radar Antijamming Dynamic Game With Imperfect Information

Abstract

One emerging issue in modern electronic warfare is the competition between the radar and jammer, which in principle can be viewed as a noncooperative game with two players. In practice, the interaction between the radar and jammer involves multiple rounds as well as partial observation. This makes the competition become a dynamic game with imperfect information. Antijamming strategy design for such kind of a game is still unclear. In this work, the competition between a frequency agile radar and a transmit/receive time-sharing jammer is considered. We utilize an extensive-form game (EFG) with imperfect information to model the multiple rounds interaction between the radar and jammer. For the established EFG, finding Nash equilibrium (NE) strategies is a computationally-intractable task since the number of information states grows exponentially with respect to game stages. Instead, a sampled-based learning method called neural fictitious self play algorithm is used to find approximate NE strategies (ANES). Simulation results show that ANES can be obtained and outperform the common elementary and advanced strategies from the perspective of detection performance.