Multi-function Radar Cognitive Interference Decision-making Based on Improved DQN

The emergence of multi-function radar (MFR) has brought unprecedented challenges to the field of electronic reconnaissance and interference. In response to the problems of data redundancy and poor interpretability in traditional signal description methods for characterizing complex and variable MFR signals, we propose a joint representation method for radar state based on existing multi-level signal models. In response to the overestimation problem of the original Deep Q Network (DQN) algorithm in evaluating the value of interference actions, we apply three improved DQNs: DQN with target network, Double DQN, and Dueling DQN to the interference decision-making. Finally, we conduct simulation experiments using the operation modes of the Mercury Radar and common interference patterns as algorithm inputs to verify the performance of the improved algorithms, and optimize some parameters of the Dueling DQN. The simulation results show that the improved algorithms can effectively solve the overestimation problem of DQN, and significantly improve the convergence speed and decision performance of the algorithm.