Adaptive Channel-Robust Signal Fusion Method for Collaborative Device Fingerprint Identification

Radio frequency fingerprint identification (RFFI) has become a popular research topic due to its potential ability to address security issues. However, how to efficiently leverage the diversity gain of multi-source signals for enhancing the collaborative RFFI (Co-RFFI) performance is still a challenging and unsolved topic, especially under realistic conditions involving unknown channel variations and unbalanced signal-to-noise ratios (SNRs) across receivers. To this end, this letter proposes a high-performance Co-RFFI method, where an adaptive channel-robust signal fusion (ACRSF) method is designed. Specifically, we first generate the amplitude-limited denoised spectral quotient (ALDSQ) signal on each receiver in a channel-robust manner. Then, a novel signal-level fusion strategy utilizes SNR-based softmax weighting to adaptively combine these ALDSQ signals, thus prioritizing discriminative information from higher-quality sources. At last, a well-trained convolutional neural network (CNN) is employed to accomplish the final classification. Extensive simulation results demonstrate that the proposed Co-RFFI method, ACRSF-CNN, provides significant performance improvement compared to other existing fusion schemes under unknown channel-varying environments.