Parameter-Efficient Transformer Network for Radar Anti-Jamming Strategy Design

In modern electronic warfare (EW), the competition between radar and jammer systems is presenting a critical challenge. Recently, active countermeasures employ frequency agile technique and conceptualize anti-jamming as an online decision-making process. However, most traditional online methods suffer from poor sample efficiency and struggle with the real-world, large-scale scenarios. This paper addresses the limitations by introducing a transformer-based neural network tailored for radar anti-jamming strategy design. Our approach utilizes the transformer architecture's capabilities in representing large-scale scenarios, and incorporates prior knowledge of the jamming systems to significantly enhance sample efficiency. A parameter-efficient fine-tuning approach is further employed to optimize computational demands. Moreover, a signal-level simulation platform is developed to facilitate an in-depth evaluation of our approach against a range of established online algorithms and various jamming strategies. Rigorous experiments on this platform confirms our method's superiority in both sample efficiency and computational performance.