Radar Jamming Waveform Optimization Method Based on Self-Adaption DeepFool Adversarial Attacks

Jamming pattern recognition is an important aspect of radar anti-jamming technology. As jammers, we can launch corresponding attacks to cause the radar to misidentify jam-ming patterns and react incorrectly. Therefore, we propose a jamming pattern optimization method based on the adversarial attacks method. On the basis of the jamming waveform pattern transmitted by the jammer, this method iteratively calculates the perturbation. So that the minimum perturbation that can make the radar's deep network recognition error is generated. By adding this perturbation to the original jamming waveform, the radar will recognize the wrong jamming type. Meanwhile, the original jamming waveform's effect has not been influenced. This paper tests the deep neural networks that are widely used in jamming pattern recognition, including ResNet, VGG, and AlexNet. We horizontally compared four adversarial attack methods. Simulation results indicate that our method significantly reduces radar recognition accuracy and outperforms other methods.