Robust Adversarial Attacks Against DNN-Based Wireless Communication Systems

Abstract

Deep Neural Networks (DNNs) have become prevalent in wireless communication systems due to their promising performance. However, similar to other DNN-based applications, they are vulnerable to adversarial examples. In this work, we propose an input-agnostic, undetectable, and robust adversarial attack against DNN-based wireless communication systems in both white-box and black-box scenarios. We design tailored Universal Adversarial Perturbations (UAPs) to perform the attack. We also use a Generative Adversarial Network (GAN) to enforce an undetectability constraint for our attack. Furthermore, we investigate the robustness of our attack against countermeasures. We show that in the presence of defense mechanisms deployed by the communicating parties, our attack performs significantly better compared to existing attacks against DNN-based wireless systems. In particular, the results demonstrate that even when employing well-considered defenses, DNN-based wireless communications are vulnerable to adversarial attacks.

AaN: Anti-adversarial Noise - A Novel Approach for Securing Machine Learning-based Wireless Communication Systems

“…Their results show that the performance of the modulation recognition system can be significantly improved by applying novel architectures and training methods. Adversarial Attacks Against DL-based WCS: Although DL has unique advantages in solving complex problems in radio communications, similar to other DNN-based applications, DNN-based WCS are susceptible to adversarial attacks [12], [38]- [52]. Small perturbations added to the input of the DNN-based WCS can fool the neural network classifier or regression model into making incorrect classifications or predictions.…”

“…Several works demonstrated the vulnerability of DNN-based WSC against these attacks. The susceptibility of DNN-based modulation recognition (MR) to adversarial attacks, in particular, was extensively investigated [12], [39], [44], [46], [51], [52]. [39] focused on targeted attacks, which aim to masquerade as a specific signal of interest wherein it was shown that a CNN-based MR breaks down intuitively When an adversary has direct access to the inputs of the system.…”

“…Authors of [52] explored the threats and impacts of adversarial attacks on the DNN-based MR in different environments and showed that, regardless of a white-box or a black-box model, the adversarial attack can reduce the accuracy of the target model. Authors of [12] proposed a novel input-agnostic adversarial attack against DNN-based WCS, which is more robust than previous attacks against different countermeasures.…”

Multiuser Adversarial Attack on Deep Learning for OFDM Detection

“…The attack leads to more misclassification than the conventional random noise. In [19], a perturbation generator against DNN-based wireless communication was tested. Thus, it is of great interest to examine different attack methods in OFDM.…”

“…In this letter, different adversarial attack methods will be evaluated against the DL based OFDM signal detector. Instead of adding the perturbation to the transmitted signal directly as in previous works [19], perturbation signal in this work will be regarded as an interfering user in the multiuser OFDM system, or as a multiuser adversarial attack. The DL model at the receiver recovers the transmitted signal against the attack disguised as a legitimate user.…”