关键词: 深度神经网络, 调制识别, 对抗样本, 特征梯度, 均值化

Abstract: Concern the issue that modulation aiming jamming could reduce the communication performance by identifying the modulation mode of signal through Deep Neural Network (DNN), a adversarial attack algorithm of modulation signal based on equalization of characteristic gradient was proposed. The algorithm was different from the traditional method of label back propagation to obtain the gradient, but instead used the rich space-time characteristic of the modulation signal in the DNN high-dimensional space to calculate the gradient, and used the local average characteristic gradient to replace the single point characteristic gradient for the algorithm iteration, which solved the problem of unreliable gradient caused by the local oscillation of the loss function surface. Based on the processed gradient and existing momentum attack methods, it could generate more subtle adversarial disturbance and added to the normal communication signal to structure the adversarial sample, so as to reduce the recognition rate of DNN to the communication signal and weaken the effect of modulation aiming jamming. The experimental results on the RADIOML 2016.10A dataset showed that, compared to FGSM (Fast Gradient Sign Method) and MI-FGSM (Momentum Iterative Fast Gradient Sign Method), although the running time of the proposed algorithm on VTCNN2 (Visual Transformer Convolutional Neural Network) model respectively increased by 1.36h and 0.58h, the attack effect of the no-target adversarial samples generated by the proposed algorithm was significant. At a signal-to-noise ratio of 10dB, the success rates of white box attack respectively increased by 36 and 26 percentage points. These were directly transferred to the CLDNN (Convolutional Long Short-Term Memory-Deep Neural Network) model, the success rates of black box attack increased by 19 and 14 percentage points respectively. The proposed algorithm improves the attack success rate of adversarial samples and indicate good transferability.

Key words: Deep Neural Network(DNN), modulation recognition, adversarial sample, characteristic gradient, equalization