Robust Radio Frequency Fingerprint Identification Under Modulation Variations Via Invariant Multi-Scale Feature Learning

Authors:

Abstract
Radio Frequency Fingerprint Identification (RFFI) enables the identification of individual wireless devices by extracting inherent radio frequency ($\mathbf{R F}$) features, offering substantial potential for enhancing wireless communication security. However, conventional RFFI methods often suffer from significant performance degradation under varying modulation schemes, limiting their applicability in dynamic wireless environments. To address this issue, this paper proposes a robust RFFI approach based on invariant feature learning. A simulated RF signal dataset with diverse modulation types is constructed using a transmitter impairment model to comprehensively evaluate cross-modulation performance. The proposed method adopts a multi-scale feature extraction architecture, which enhances the network's capacity to learn modulation-invariant representations, thereby improving generalization across different modulation types. Experimental results demonstrate that the proposed method achieves an average identification accuracy of $\mathbf{8 5. 7 5 \%}$ in cross-modulation scenarios, substantially outperforming conventional baselines. Notably, the method exhibits strong transferability between high-order and multi-carrier modulation schemes, while maintaining high robustness in complex and variable environments.

This publication has 18 references indexed in Scilit: