RRF: A Robust Radiometric Fingerprint System that Embraces Wireless Channel Diversity

Abstract

Radiometric fingerprint schemes have been shown effective in identifying wireless devices based on imperfections in their hardware electronics. The robustness of fingerprint systems under complex channel conditions, however, is a critical challenge that makes their application in real-world scenarios difficult. We systematically evaluate the wireless channel's impact on radiometric fingerprints and find that the channel impacts fingerprint features in a very particular way that depends on the channel's properties. Based on the insights, we present RRF, a system that provides a robust identification/authentication service even under complex channel fading disturbance. Our design deploys a hybrid architecture that combines wireless channel simulation, signal processing and machine learning. In this pipeline, RRF first utilizes a series of structured channel simulations to strategically improve system tolerance towards multipath channel interference. On top of that, in the identification phase, RRF relies on noise compensation and a feature denoising filter to augment the system's stability in noisy conditions with weak signals. Our experimental results show that RRF achieves an average accuracy consistently above 99% in empirical scenarios with complex channels, where the baseline approach from previous work rarely exceeds 50%.