Machine Learning based Reliability Assurance in Optical Networks

Efficiently handling failures and ensuring transmission quality are crucial for high-speed optical networks. As data demands rise, disruptions can lead to significant financial losses and data loss. Automated, dynamic methods are needed to manage failures in the complex and evolving optical networks. While traditional approaches rely on conservative designs, machine learning algorithms offer proactive solutions for early failure detection and degradation. Accurate transmission quality estimation is essential for maximizing network capacity and enabling self-management in future networks. Although acquiring monitoring data from wavelength division multiplexed signals is challenging, optical spectrum analyzers can transparently obtain signal information without demultiplexing.

This work presents novel approaches for quality of transmission estimation and soft-failure management based on the optical spectrum obtained through optical spectrum analyzers. Multiple machine learning algorithms were evaluated for quality of transmission estimation under uncertain parameters using optical spectral data. The spectral data utilization increased the quality of transmission estimation accuracy. Furthermore, this work investigates optical spectra driven soft-failure management using experimentally emulated component failures. For fault detection, a variational autoencoder was used, yielding high accuracy. Integrating a generative adversarial network into the framework, enabled by the variational autoencoder's generative capabilities, achieved exceptional performance in identifying soft-failures with very low amounts of available training data. In combination with advanced training methods for generative adversarial networks, unknown spectra were reliably identified enabling unknown failure detection and classification.

Rights

Use and reproduction:


CC BY-SA 4.0

Please note that individual components of the publication may be subject to other licensing or copyright conditions.

Cite

Citation style:
Could not load citation form.