Variational autoencoder-assisted unsupervised hardware fingerprint authentication in a fiber network.

Area of Science:

• Optical communication security
• Machine learning for network security

Background:

• Traditional physical-layer authentication (PLA) requires extensive labeled data, limiting its scalability.
• Optical networks face threats like masquerade and active injection attacks.

Purpose of the Study:

• To develop an unsupervised hardware fingerprint authentication method for optical networks.
• To enable effective identification of rogue optical transmitters without relying on pre-labeled datasets.

Main Methods:

• Utilized a variational autoencoder (VAE) for unsupervised learning.
• Generated training data through variational inference on unlabeled optical spectra.
• Employed a feature extractor trained on generated triplets for generalization.

Main Results:

• Successfully classified eight optical transmitters after 20 km of standard single-mode fiber transmission.
• Achieved 99% recognition accuracy and a 0% miss alarm rate, even with multiple rogue devices.
• Demonstrated comparable performance to supervised learning methods.

Conclusions:

• The proposed unsupervised VAE-based PLA offers a scalable and effective solution for securing optical networks.
• The method exhibits strong generalization capabilities, allowing authentication of previously unknown transmitters.
• This approach enhances optical network security against sophisticated attacks.