A Reinforcement Learning-based Intelligent Learning Method for Anti-active Jamming in Frequency Agility Radar

Active jamming's flexibility and variability pose significant challenges for frequency-agility radar (FAR) detection, as it can continuously intercept and retransmit radar signals to suppress or deceive the radar. To tackle this, we propose an intelligent learning method for FAR based on reinforcement learning (RL), integrating signal processing with compressed sensing (CS). We introduce an inter-pulse carrier-frequency hopping combined with intra-pulse sub-frequency coding (IPCFH-IPSFC) signal model to address time-domain discontinuities caused by active jamming, enabling effective mutual masking of pulses through agile waveform parameters. We develop jamming signal models and design four jamming strategies based on two common types of active jamming, providing essential data for the FAR intelligent learning method. To enhance FAR's adaptive anti-jamming and target detection performance, we propose an RL-based intelligent learningmodel.This model includes five submodules: signal processing, anti-jamming evaluation, target detection, optimization constraint design, and optimization algorithm design.We apply a proximal policy optimization combined with a generative pre-trained transformer (PPO-GPT) to solve this model, allowing FAR to adaptively learn jamming strategies and optimize IPCFH-IPSFC waveform parameters for effective anti-jamming.Simulation results confirm that our method achieves robust performance and rapid convergence, finding optimal anti-jamming strategies in just 215 training iterations.The FAR effectively counteracts jamming while accurately estimating target range and velocity.