An Integrated Long and Short Term Fingerprinting Network for Wireless Transmitter Identification

Radio frequency fingerprint identification (RFFI) is an emerging authentication technique that takes advantage of intrinsic and transmitter-specific impairments for device identification. However, recent efforts on RFFI methods focus mainly on the long-term robust classification ability, while neglecting the impact of some discriminative features only with short-term invariant property, such as carrier frequency offset (CFO). To address this gap, we propose a novel hybrid RFFI approach, known as the Integrated Long- and Short-Term Fingerprinting Network (ILSTFN), which integrates both long-term and short-term features to improve classification accuracy. Specifically, we first employ the IQ samples to pre-train a long-term fingerprinting network (LTFN), incorporating CFO compensation in the process. Then, we conduct online semi-supervised training to adjust the Short-Term Fingerprinting Network (STFN) during the pre-testing phase. Finally, both LTFN and STFN are activated during the post-testing stage, where an integrated decision rule is designed to fuse their output. Experimental results demonstrate that ILSTFN outperforms both LTFN and STFN individually, suggesting a promising future of integrating long- and short-term fingerprints for enhanced device identification.