Generalizability of ML-Based Classification of State of Polarization Signatures Across Different Bands and Links
Paper in proceeding, 2025
We evaluate the Machine Learning (ML) model generalization for State of Polarization (SOP)-based event classification across spectral bands and links. Results show strong intra-system accuracy of up to 98.6% but limited cross-system generalizability, whereas multi-system training improves performance, highlighting the need for specific system-level knowledge.
Network sensing
Generalization
Optical networks
Author
Leyla Sadighi
Trinity College Dublin
Carlos Natalino Da Silva
Chalmers, Electrical Engineering, Communication, Antennas and Optical Networks
Stefan Karlsson
Micropol Fiberoptics AB
Lena Wosinska
Chalmers, Electrical Engineering, Communication, Antennas and Optical Networks
Eoin Kenny
HEAnet
Marco Ruffini
Trinity College Dublin
Marija Furdek Prekratic
Chalmers, Electrical Engineering, Communication, Antennas and Optical Networks
European Conference on Optical Communication, ECOC
Th.02.01.2
European Conference on Optical Communications (ECOC)
Copenhagen, Denmark,
Copenhagen, Denmark,
Efficient Confluent Edge Networks (ECO-eNET)
European Commission (EC) (EC/HE/101139133), 2024-01-01 -- 2028-12-31.
InfraTrust: Enabling trustworthy services over vulnerable physical network infrastructure
Swedish Research Council (VR) (2023-05249), 2024-01-01 -- 2027-12-31.
5G Trusted And seCure network servICes (5G-TACTIC)
European Commission (EC) (EC/H2020/101127973), 2023-12-01 -- 2026-11-30.
Areas of Advance
Information and Communication Technology
Subject Categories (SSIF 2025)
Telecommunications
DOI
10.1109/ECOC66593.2025.11263096