Hardware Fingerprint Authentication Based on Siamese Neural Networks in PON

This letter presents a novel approach to bolster the physical layer security of optical communication systems, specifically within Passive Optical Networks (PONs), through the utilization of device fingerprints. In this proposed scheme, we employ Optical On-Off Keying (OOK) modulation for signal transmission and subsequently extract distinct fingerprint features from the eye diagrams of these OOK signals. These fingerprint features are then subjected to dimensionality reduction via Siamese neural networks. Subsequently, a set of classifiers is utilized to discriminate among the downscaled feature data, thereby achieving robust authentication for up to 10 ONUs in a 20 km Single-Mode Fiber (SSMF) transmission. Remarkably, the recognition accuracy attained in our experiments reached 96.04%. Moreover, this system exhibits the capacity for transfer learning of fingerprint features when new devices are introduced into the network. This feature speeds up the authentication of new devices coming online.