Towards Channel-Robust Radio Frequency Fingerprint Identification Using Contrastive Learning

Radio frequency fingerprint identification (RFFI) is an emerging device authentication technique that is based on intrinsic hardware impairments. Internet of things (IoT) devices can be identified and classified based on their wireless signals using RFFI. Developing a robust RFFI system that can maintain high classification accuracy across diverse communication scenarios is a critical challenge. In this paper, we proposed a contrastive learning-based RFFI approach to establish a channel-robust system using the spectrogram. Specifically, we leverage contrastive learning in the training stage, which has been implemented with data augmentation techniques to mitigate the influence of channels on RFFI. We carried out extensive experimental evaluations involving a public dataset and a self-collected dataset, both with ten LoRa devices. Utilizing these datasets, the performance of the system has been tested in various channel environments, including stationary, mobile, line-of-sight (LOS) and non-line-of-sight (NLOS) scenarios. The results demonstrated that our approach is effective and robust to channel variation, achieving 93% and 82% on static and dynamic channels. respectively.