Cognitive radio technology is an important branch in the field of wireless communication, and automatic modulation classification (AMC), which plays critical roles in both civilian and military applications, is a major part of cognitive radio technology. Adversarial examples can reduce the performance of a deep learning-based AMC system with a hardly perceptible perturbation, which can prevent the signal being intercepted by non-cooperative deep learning-based signal receiver. In this paper, we test the attack efficiency of two kinds of adversarial examples in dataset RML2016.04C, and discuss the reason why adversarial examples can attack the deep learning-based AMC successfully, verifying the underfitting of networks of AMC is ubiquitous. Modulations which are similar in time domain and frequency domain, such as {8PSK, QPSK}, {QAM16, QAM64}, {AM-DSB, WBFM}, are easily vulnerable to adversarial examples.
Paper
Full text
Application of Adversarial Examples in Communication Modulation Classification
Semantic Scholar · Computer Science · 2019
Abstract
Cognitive radio technology is an important branch in the field of wireless communication, and automatic modulation classification (AMC), which plays critical roles in both civilian and military applications, is a major part of cognitive radio technology. Adversarial examples can reduce the performance of a deep learning-based AMC system with a hardly perceptible perturbation, which can prevent the signal being intercepted by non-cooperative deep learning-based signal receiver. In this paper, we test the attack efficiency of two kinds of adversarial examples in dataset RML2016.04C, and discuss the reason why adversarial examples can attack the deep learning-based AMC successfully, verifying the underfitting of networks of AMC is ubiquitous. Modulations which are similar in time domain and frequency domain, such as {8PSK, QPSK}, {QAM16, QAM64}, {AM-DSB, WBFM}, are easily vulnerable to adversarial examples.