Generative Forgery Attack for Radio Frequency Fingerprints Spoofing

Radio frequency (RF) fingerprints are unique signal distortions resulting from hardware imperfections during manufacturing. These unique characteristics help identify radio devices and mitigate network attacks such as identity spoofing and impersonation. However, advancements in artificial intelligence (AI) have significantly enhanced generative models, enabling RF fingerprint forgery. This paper identifies a security threat enabled by surrogate models, where conditional generative networks produce forgery samples with specified labels, deceiving classifiers into misidentifying the intended identity. Specifically, the generative model crafts perturbations, allowing a transmitter to mimic another device’s fingerprint and mislead receiver identification. Extensive experiments validate the effectiveness of forged fingerprints, demonstrating high spoofing success rates and robustness under diverse SNR conditions and channel fading scenarios. Furthermore, the proposed approach achieves notable performance even in challenging black-box attack settings, leveraging surrogate models to successfully deceive the target classifiers. To minimize the impact on communication quality, a power constraint mechanism is implemented, ensuring that perturbations remain imperceptible and practical for real-world deployment.