SignCRF: Scalable Channel-agnostic Data-driven Radio Authentication System

Abstract

Radio Frequency Fingerprinting through Deep Learning (RFFDL) is a data-driven IoT authentication technique that leverages the unique hardware-level manufacturing imperfections associated with a particular device to recognize ("fingerprint") the device itself based on variations introduced in the transmitted waveform. Key impediments in developing robust and scalable RFFDL techniques that are practical in dynamic and mobile environments are the non-stationary behavior of the wireless channel and other impairments introduced by the propagation conditions. To date, the existing RFFDL-based techniques have only been able to demonstrate a desirable performance when the training and testing environment remains the same, which makes the solutions impractical. SignCRF brings to the RFFDL landscape what it has been missing so far: a scalable, channel-agnostic data-driven radio authentication platform with unmatched precision in fingerprinting wireless devices based on their unique manufacturing impairments that is independent of the dynamic nature of the environment or channel irregularities caused by mobility. SignCRF consists of: (i) a classifier developed in a base-environment with minimum channel dynamics, and finely trained to authenticate devices with high accuracy and at scale; (ii) an environment translator that is carefully designed and trained to remove the dynamic channel impact from RF signals while maintaining the radio's specific "signature"; and (iii) a Max Rule module that selects the highest precision authentication technique between the baseline classifier and the environment translator per radio. We design, train, and validate the performance of SignCRF for multiple technologies in dynamic environments and at scale (100 LoRa and 20 WiFi devices, the largest datasets available in the literature). We demonstrate that SignCRF can significantly improve the RFFDL performance by achieving as high as 100% correct authentication for WiFi devices and 80% correctly authenticated LoRa devices, a 5x and 8x improvement when compared to the state-of-the-art respectively. Furthermore, we show that SignCRF is resilient to adversarial actions by reducing the device recognition accuracy from 73% to 6%, which translates into zero mis-authentication of adversary radios that try to impersonate legitimate devices, which has not been achieved by any prior RFFDL techniques.

Comprehensive RF Dataset Collection and Release: A Deep Learning-Based Device Fingerprinting Use Case

“…During the inference stage, only the device-specific features, which encompass characteristics that remain invariant across the source and target domains, are utilized. Similarly, SignCRF [19] leverages a cycle-consistent generative adversarial network to construct an environment translator that effectively decouples hardware impairments from channel and environmental conditions. Adversarial domain adaptation techniques have also been employed in this context.…”

“…Impairment compensation techniques, on the other hand, target specific channel impairments, limiting their generalizability across different environments and wireless channels. Recently, domain adaptation techniques have emerged as a means to reduce the presence of domain information in the feature vector through adversarial learning [17], or to decouple device-specific and domain-specific features through adversarial disentanglement learning [18] and generative adversarial networks [19]. However, these techniques do not scale well and fail to maintain a practically acceptable performance for medium to large numbers of devices.…”