Intelligent Radar Anti-Jamming Strategy with High Sample Efficiency Based on MBRL

As jammer intelligence advances, some reinforcement learning methods are used to develop radar anti-jamming strategies. However, some of the methods used in the training process require a substantial number of interactive samples of radar and jammer. This paper employs a model-based reinforcement learning algorithm to generate radar anti-jamming strategies to address this issue. It constructs a virtual environment to produce virtual samples from actual anti-jamming interaction samples. Experiments show that achieving the same anti-jamming performance with this method requires fewer samples than model-free reinforcement learning algorithms,