Q-learning and Policy Gradient-Based Reinforcement Learning Method to Decision Making of Phased Array Radar Jamming

Abstract

The rapid advancement of phased array radar has greatly improved the guiding and anti-jamming capacities of radar seekers. In contemporary electronic warfare, traditional radar-jamming decision-making methods have proven ineffective, prompting the use of reinforcement learning methods to tackle performance and efficiency shortcomings. This research employs Q-learning and Policy Gradient techniques for radar jamming decision-making, with signal-level simulation rather than functional-level simulation as used in previous research. Signal-level simulation offers a more realistic and understandable representation of the interference mechanism that affects missile terminal guidance. The proposed reinforcement learning methods discover optimal actions for interference equipment, thereby increasing jamming efficiency. Simulation findings show an 18% increase in interference efficiency over conventional techniques, with models attaining 92% accuracy in optimal decision-making. The precision of the signal-level simulation model and the effectiveness of reinforcement learning in improving interference performance are confirmed.