Noise-resilient Hardware Fingerprinting for Physical-layer Authentication in Passive Optical Networks

In this letter, we propose a noise-resilient hardware fingerprinting method for device authentication in passive optical networks (PON). By converting time-domain signals into time-frequency representations via short-time fourier transform (STFT) and introducing a double-stream convolutional neural network (DSCNN) that synergistically fuses time-domain and frequency-domain features using an attention mechanism, our approach effectively separates noise components and preserves intrinsic hardware features. Experimental results on a practical PON testbed demonstrate the superiority of the proposed scheme. The method achieves a recognition accuracy of 99.5% at a 20dB signal-to-noise ratio (SNR), outperforming comparative approaches and demonstrating exceptional resistance to noise.