Radio Frequency Fingerprinting for WiFi Devices Using Oscillator Drifts

Abstract

Radio frequency fingerprint (RFF) identification is a promising technique that exploits hardware impairment-induced features to achieve specific device identification. Among RFF features, carrier frequency offset (CFO) as a hotspot feature has received widespread attention. Since CFO is time-variant, existing research suggests compensating for its drift; however, this article emphasizes using the drift of CFO. Correspondingly, a novel RFF feature, named cyclic similarity (cyc-similarity), is proposed to depict the oscillator drift. Simply combining the cyc-similarity feature with a K-nearest neighbor (KNN) classifier, the system can achieve superior temporal and receiver generalization performance. On a public dataset of WiFi devices, the proposed method outperforms the existing methods.