Evaluating Adversarial Evasion Attacks in the Context of Wireless Communications

Flowers, Bryse; Buehrer, R. Michael; Headley, William C.

doi:10.48550/arxiv.1903.01563

Cited by 9 publications

(15 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the case of modulation recognition, adversarial perturbations have also been shown to be effective and to require much less power than additive white gaussian noise (AWGN) to fool the network [8], [10], [11]. Adversarial perturbations in this case are constrained relatively to the signal power, using the signal-to-perturbation ratio (SPR) metric [8].…”

Section: Related Workmentioning

confidence: 99%

“…We split the RML2016.10a IQ signals into 70% of training and 30% of test signals, using 5% of the training as validation for hyperparameter tuning, as proposed by [10]. For the RML2018.01a dataset, we use 1 million signals as training set as proposed in [5].…”

Section: A Datasetsmentioning

confidence: 99%

“…First, we do a small experiment on the RML2016.10a dataset, to see how adversarial training changes the model predictions on adversarial examples. We use the robustness framework described before with the VT CNN2 BF architecture [10] as our AMC model. For both the training and testing adversarial examples, we choose 20 dB for the SPR since it decreases significantly the performance of the model without compromising in perceptibility [11].…”

Section: B Performance Analysismentioning

confidence: 99%

See 2 more Smart Citations

SafeAMC: Adversarial training for robust modulation recognition models

Maroto,

Bovet,

Frossard

2021

Preprint

View full text Add to dashboard Cite

In communication systems, there are many tasks, like modulation recognition, which rely on Deep Neural Networks (DNNs) models. However, these models have been shown to be susceptible to adversarial perturbations, namely imperceptible additive noise crafted to induce misclassification. This raises questions about the security but also the general trust in model predictions. We propose to use adversarial training, which consists of fine-tuning the model with adversarial perturbations, to increase the robustness of automatic modulation recognition (AMC) models. We show that current state-of-the-art models benefit from adversarial training, which mitigates the robustness issues for some families of modulations. We use adversarial perturbations to visualize the features learned, and we found that in robust models the signal symbols are shifted towards the nearest classes in constellation space, like maximum likelihood methods. This confirms that robust models not only are more secure, but also more interpretable, building their decisions on signal statistics that are relevant to modulation recognition.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: A Datasetsmentioning

confidence: 99%

Section: B Performance Analysismentioning

confidence: 99%

See 1 more Smart Citation

SafeAMC: Adversarial training for robust modulation recognition models

Maroto,

Bovet,

Frossard

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…For our first experiments, we use the robustness framework described before to measure how robust the model is to adversarial perturbations. To choose an effective value of SPR, we test the VT CNN2 BF model, 9 which has good performance and it is fast to train, on the RML2016.10a dataset under different values of SPR. To minimize the impact of AWGN in the decision we use the cleaner signals from the testing set, which have 18dB SNR.…”

Section: Performance Analysismentioning

confidence: 99%

“…Understanding the reasons of such vulnerabilities, how we can robustify DNNs against adversarial perturbations, and, more generally, how we can promote desired properties on the networks, currently forms an active line of research. Several authors 7,8,9 have studied the impact of adversarial perturbations against modulation recognition DNN models and have found that, like the computer vision models, they are vulnerable to attacks. This poses a real threat since DNNs are very popular due to their performance, even though there is a general lack of awareness about their vulnerability in communication systems.…”

Section: Introductionmentioning

confidence: 99%

On the benefits of robust models in modulation recognition

Maroto¹,

Bovet²,

Frossard³

2021

Preprint

View full text Add to dashboard Cite

Given the rapid changes in telecommunication systems and their higher dependence on artificial intelligence, it is increasingly important to have models that can perform well under different, possibly adverse, conditions. Deep Neural Networks (DNNs) using convolutional layers are state-of-the-art in many tasks in communications. However, in other domains, like image classification, DNNs have been shown to be vulnerable to adversarial perturbations, which consist of imperceptible crafted noise that when added to the data fools the model into misclassification. This puts into question the security of DNNs in communication tasks, and in particular in modulation recognition. We propose a novel framework to test the robustness of current state-of-the-art models where the adversarial perturbation strength is dependent on the signal strength and measured with the "signal to perturbation ratio" (SPR). We show that current state-of-the-art models are susceptible to these perturbations. In contrast to current research on the topic of image classification, modulation recognition allows us to have easily accessible insights on the usefulness of the features learned by DNNs by looking at the constellation space. When analyzing these vulnerable models we found that adversarial perturbations do not shift the symbols towards the nearest classes in constellation space. This shows that DNNs do not base their decisions on signal statistics that are important for the Bayes-optimal modulation recognition model, but spurious correlations in the training data. Our feature analysis and proposed framework can help in the task of finding better models for communication systems.

show abstract

Deep Learning for Wireless Communications

Erpek

O’Shea

Sagduyu

et al. 2019

Studies in Computational Intelligence

130

View full text Add to dashboard Cite

Existing communication systems exhibit inherent limitations in translating theory to practice when handling the complexity of optimization for emerging wireless applications with high degrees of freedom. Deep learning has a strong potential to overcome this challenge via data-driven solutions and improve the performance of wireless systems in utilizing limited spectrum resources. In this chapter, we first describe how deep learning is used to design an end-to-end communication system using autoencoders. This flexible design effectively captures channel impairments and optimizes transmitter and receiver operations jointly in single-antenna, multiple-antenna, and multiuser communications. Next, we present the benefits of deep learning in spectrum situation awareness ranging from channel modeling and estimation to signal detection and classification tasks. Deep learning improves the performance when the model-based methods fail. Finally, we discuss how deep learning applies to wireless communication security. In this context, adversarial machine learning provides novel means to launch and defend against wireless attacks. These applications demonstrate the power of deep learning in providing novel means to design, optimize, adapt, and secure wireless communications.

show abstract

Evaluating Adversarial Evasion Attacks in the Context of Wireless Communications

Cited by 9 publications

References 0 publications

SafeAMC: Adversarial training for robust modulation recognition models

SafeAMC: Adversarial training for robust modulation recognition models

On the benefits of robust models in modulation recognition

Deep Learning for Wireless Communications

Contact Info

Product

Resources

About