Unsupervised Contrastive Learning for Robust RF Device Fingerprinting Under Time-Domain Shift

TL;DR

This paper demonstrates that the effectiveness of contrastive learning in improving device classification under domain shift in RF fingerprinting shows large and consistent improvements in accuracy, thus underscoring the effectiveness of contrastive learning in improving device classification under domain shift.

Abstract

Radio Frequency (RF) device fingerprinting has been recognized as a potential technology for enabling automated wireless device identification and classification. However, it faces a key challenge due to the domain shift that could arise from variations in the channel conditions and environmental settings, potentially degrading the accuracy of RF-based device classification when testing and training data is collected in different domains. This paper introduces a novel solution that leverages contrastive learning to mitigate this domain shift problem. Contrastive learning, a state-of-the-art self-supervised learning approach from deep learning, learns a distance metric such that positive pairs are closer (i.e. more similar) in the learned metric space than negative pairs. When applied to RF fingerprinting, our model treats RF signals from the same transmission as positive pairs and those from different transmissions as negative pairs. Through experiments on wireless and wired RF datasets collected over several days, we demonstrate that our contrastive learning approach captures domain-invariant features, diminishing the effects of domain-specific variations. Our results show large and consistent improvements in accuracy (10.8\% to 27.8\%) over baseline models, thus underscoring the effectiveness of contrastive learning in improving device classification under domain shift.

Figures (5)

• Figure 1: An overview of how training and test data are created.
• Figure 2: The proposed training procedure and contrastive-based framework for addressing the domain shift issue in RF fingerprinting. The overall domain adaptation process includes three stages: (a) pre-training, (b) training and (c) testing. The source inputs consist of IQ frames from one day, while the target inputs are from another day. The pre-training stage uses unlabeled data, meaning we do not know which device produced the data but we do know which captures come from the same transmission. Labeled data in the form of device labels is only used during the training stage to train the classifier.
• Figure 3: Illustrations of weak and strong data augmentation.
• Figure 4: Domain adaptation accuracy: from one day to another day on wired and wireless RF devices for CNN, AB and CTL.
• Figure 5: Confusion matrices between day 1 and day 2 on the wired setup for CNN, AB and CTL. Confusion matrices are normalized by row, enabling a clearer visualization of the predicted accuracy distribution across different classes.