Adversarial Attack and Defense on Deep Learning for Air Transportation Communication Jamming

Abstract

Air transportation communication jamming recognition \nmodel based on deep learning (DL) can quickly and \naccurately identify and classify communication jamming, to \nimprove the safety and reliability of air traffic. However, due \nto the vulnerability of deep learning, the jamming recognition \nmodel can be easily attacked by the attacker’s carefully designed \nadversarial examples. Although some defense methods have been \nproposed, they have strong pertinence to attacks. Thus, new \nattack methods are needed to improve the defense performance of \nthe model. In this work, we improve the existing attack methods \nand propose a double level attack method. By constructing the \ndynamic iterative step size and analyzing the class characteristics \nof the signals, this method can use the adversarial losses of feature \nlayer and decision layer to generate adversarial examples with \nstronger attack performance. In order to improve the robustness \nof the recognition model, we use adversarial examples to train \nthe model, and transfer the knowledge learned from the model to \nthe jamming recognition models in other wireless communication \nenvironments by transfer learning. Simulation results show that \nthe proposed attack and defense methods have good performance.