Robust Radio Frequency Fingerprint Identification via Multi-Scale Complex-Valued CNN

Radio Frequency Fingerprint Identification (RFFI) exploits inherent hardware imperfections in RF signals at the physical layer to enable transmitter identification, playing a critical role in wireless security and Internet of Things (IoT) device authentication. However, the performance and generalization ability of existing deep learning-based models are often constrained by hardware heterogeneity and complex wireless environments. To address these challenges, this paper proposes a novel Multi-Scale Complex-Valued Convolutional Neural Network (MSCVCNN). Built upon the complex-valued convolutional neural network (CVCNN) foundation, MSCVCNN incorporates six multi-scale complex-valued convolutional blocks to capture features across diverse receptive fields. It further introduces a complex-valued simulation mechanism for joint amplitudephase modeling, enhancing robustness against signal distortion and incomplete feature representation. Residual connections are employed to improve information propagation across layers. Comprehensive evaluations are conducted on a real-world ADS-B dataset comprising signals from 30 emitters under varying signal-to-noise ratio (SNR) conditions and random feature masking scenarios. Experimental results show that MSCVCNN consistently outperforms benchmark models such as LSTM, ResNet, and DRCN, demonstrating superior robustness and generalization in complex wireless environments.