Research Article
A Robust Signal Modulation Recognition Method Against Black-Box Detection Attack
@INPROCEEDINGS{10.1007/978-3-031-23902-1_25, author={Zhihui An and Peihan Qi and Xiaoyu Zhou and Yongchao Meng}, title={A Robust Signal Modulation Recognition Method Against Black-Box Detection Attack}, proceedings={Mobile Multimedia Communications. 15th EAI International Conference, MobiMedia 2022, Virtual Event, July 22-24, 2022, Proceedings}, proceedings_a={MOBIMEDIA}, year={2023}, month={2}, keywords={Modulation recognition Adversarial attack Knowledge distillation Robust network}, doi={10.1007/978-3-031-23902-1_25} }- Zhihui An
Peihan Qi
Xiaoyu Zhou
Yongchao Meng
Year: 2023
A Robust Signal Modulation Recognition Method Against Black-Box Detection Attack
MOBIMEDIA
Springer
DOI: 10.1007/978-3-031-23902-1_25
Abstract
Deep learning (DL) models have been widely used in the recognition of modulation types with outstanding recognition effects. With the improvement of modulation recognition, the perturbation from the attacker has also changed from adding physical interference to the original signal to an adversarial attack based on the neural network. The adversarial attack adding subtle perturbation which is imperceptible to the human eye, makes the neural network produce false recognition results with high confidence. This kind of perturbation is hard to be reflected in spectrogram or constellation diagram, so it is seriously destructive to the modulation recognition algorithm based on neural networks. In response to adversarial attacks, we propose a modulation recognition method against black-box detection attacks. In this paper knowledge distillation is used to defend against the attack that comes from the attacker’s black-box detection. The experimental results demonstrated that the defense method constructed in this paper can improve the ability to defend adversarial samples and keep the recognition accuracy of the recognition network. This article aims at improving the robustness of the network and constructing a robust modulation recognition network.
