Operator-Ready Fiber Disturbance Classification, Localization, and LLM Interpretation in Operational Metro Networks

Deployed optical fibers can act as distributed sensors because external mechanical disturbances modify the received state of polarization (SOP). For a network operator, however, an SOP-based alarm is useful only if it identifies the disturbance, indicates where it occurred, and makes clear how far these conclusions are supported by the measurements. This paper addresses these three requirements through a field study on the operational 31 km GARR-T metropolitan fiber network in Turin, Italy. Controlled disturbances are applied through a robotic arm to an accessible fiber segment and grouped into eight events comprising vibrations at six frequencies, mechanical tapping, and fiber bending. Six supervised classifiers are compared using features extracted from the received Stokes parameters, together with a feature-reduction study and a processing-cost analysis. The multilayer perceptron (MLP) achieves a macro-F1 score of 99.82%, matching LightGBM and SVM, and retains this performance with only five of its 171 input features, reducing the combined feature-extraction and inference time to 0.21 ms per window. To locate the disturbance, cross- correlation and a cepstrum-based delay estimator are applied to counter-propagating Stokes recordings. Across 50 acquisitions, the mean cross-correlation estimate lies within 16 m of the true disturbance position and the mean of eight cepstral candidates lies within 29.3 m, with the spread of individual estimates quantified to define the search window reported to the operator. Finally, an evidence-grounded interface supplies the predicted condition, the localization status, and a predefined operator policy to a locally hosted large language model. Mistral 7B achieves the highest factual fidelity (97%) among three candidate models and, on the testbed, delivers operator reports within about 5 s of an alert, answering follow-up questions on the event, its location, and the required action consistently with the sensing results. The results show how SOP sensing can move beyond event detection towards transparent, operator-ready alarms for metropolitan optical networks.

Authors

Publication Details

Published
2026-11-17
DOI
https://doi.org/10.1364/opticaopen.35024884
Primary Topic
Advanced Fiber Optic Sensors
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Operator-Ready Fiber Disturbance Classification, Localization, and LLM Interpretation in Operational Metro Networks

Stefano Straullu, Gabriele Maria Galimberti, Antonio Napoli, Walid Wakim et al.
2 links
Advanced Fiber Optic Sensors
preprint

Operator-Ready Fiber Disturbance Classification, Localization, and LLM Interpretation in Operational Metro Networks

Stefano Straullu, Gabriele Maria Galimberti, Antonio Napoli, Walid Wakim, Mashboob Cheruvakkadu Mohamed, Gulmina Malik, Sai Kishore Bhyri, Vittorio Curri, Muhammad Umar Masood, Joao Pedro, ahtisham ali
preprint en

Abstract

Deployed optical fibers can act as distributed sensors because external mechanical disturbances modify the received state of polarization (SOP). For a network operator, however, an SOP-based alarm is useful only if it identifies the disturbance, indicates where it occurred, and makes clear how far these conclusions are supported by the measurements. This paper addresses these three requirements through a field study on the operational 31 km GARR-T metropolitan fiber network in Turin, Italy. Controlled disturbances are applied through a robotic arm to an accessible fiber segment and grouped into eight events comprising vibrations at six frequencies, mechanical tapping, and fiber bending. Six supervised classifiers are compared using features extracted from the received Stokes parameters, together with a feature-reduction study and a processing-cost analysis. The multilayer perceptron (MLP) achieves a macro-F1 score of 99.82%, matching LightGBM and SVM, and retains this performance with only five of its 171 input features, reducing the combined feature-extraction and inference time to 0.21 ms per window. To locate the disturbance, cross- correlation and a cepstrum-based delay estimator are applied to counter-propagating Stokes recordings. Across 50 acquisitions, the mean cross-correlation estimate lies within 16 m of the true disturbance position and the mean of eight cepstral candidates lies within 29.3 m, with the spread of individual estimates quantified to define the search window reported to the operator. Finally, an evidence-grounded interface supplies the predicted condition, the localization status, and a predefined operator policy to a locally hosted large language model. Mistral 7B achieves the highest factual fidelity (97%) among three candidate models and, on the testbed, delivers operator reports within about 5 s of an alert, answering follow-up questions on the event, its location, and the required action consistently with the sensing results. The results show how SOP sensing can move beyond event detection towards transparent, operator-ready alarms for metropolitan optical networks.

Advanced Fiber Optic Sensors
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon