A Frequency Offset Decoupled Radio Frequency Fingerprint Identification Scheme for LoRa Devices

Radio frequency fingerprint identification (RFFI) is a crucial method for PHY-layer security authentication of LoRa devices. Due to its utilization of inherent hardware imperfections in devices, it possesses a tamper-resistant nature and is widely employed to enhance communication security. Nevertheless, a key feature used in RFFI—the Carrier Frequency Offset (CFO)—is subject to drift over time and with temperature variations, significantly degrading identification accuracy. This paper first conducts a series of experiments demonstrating the instability of CFO when used as the sole feature for RFFI. We then propose a frequency offset decoupling mechanism to mitigate this issue. Furthermore, a dedicated RFFI framework for LoRa is introduced, which includes: an efficient signal preprocessing algorithm called SPWT, a feature extractor based on a Deep Residual Shrinkage Network, and a robustness enhancement module incorporating triplet loss and voting mechanisms. To validate the framework, a CFO dataset is constructed using 20 Semtech SX1276 modules. Experimental results demonstrate that the proposed method achieves an accuracy of 90.5% in handling frequency offset drift, significantly outperforming conventional approaches. Furthermore, its deployment on the RK3588 platform validates the practical applicability of the method.