RF Fingerprint Identification Based on Multi-Channel Time-Frequency Attention

The rapid development of the Internet of Things (IoT) presents new security challenges. Unlike conventional authentication schemes, radio frequency fingerprint identification (RFFI) provides an effective approach to enhance security at the physical layer. However, existing channel attention mechanisms in RFFI cannot generate adaptive channel attention weights as the signal-to-noise ratio (SNR) changes, and there is limited research on attention mechanisms that fuse time-domain and frequency-domain features in the spectrogram of received signals to enhance RFFI accuracy in low SNR environments. To solve these two problems, we propose the multi-channel time-frequency attention network (MCTFN), which incorporates core components such as the dynamic channel attention Squeeze-and-Excitation module (dynamic SE module) and the time-frequency attention (TFA) module. The dynamic SE module generates suitable channel attention weights that vary with changes in SNR for multi-channel tensors, while the TFA module extracts distant relationships within both the time and frequency domains and fuses them to obtain enhanced time-frequency features. Numerical results demonstrate that our proposed MCTFN achieves relative identification accuracy improvements of 7.49% and 3.94% compared to other state-of-the-art networks at SNR levels of -5 dB and 0 dB, respectively.