“I feel truly honored and delighted. It is one of the most important conferences in our field and being recognized among so many high-quality contributions was both motivating and humbling. It also felt very rewarding to see our research efforts acknowledged by the broader optical communications community.” says Leyla Sadighi, PhD student in Optical Networks at the Department of Electrical Engineering, who was selected as the runner-up for the Best Student Paper Award at the ECOC Conference 2025.
This year’s European Conference on Optical Communication (ECOC) took place in Copenhagen, Denmark, from September 28 to October 2, 2025. During the event, the Best Student Paper Award was presented, and Leyla Sadighi’s paper, “Generalizability of ML-Based Classification of State of Polarization Signatures Across Different Bands and Links" was selected from among many strong contributions.
In her research, Leyla Sadighi investigates how well machine-learning models can generalize when analyzing State of Polarization (SOP) signatures collected from different fiber links and spectral bands for fiber sensing.
“We conduct experiments using real-world data from two optical systems and evaluate how ML models trained on one system perform when tested on another. Our study shows that while models achieve very high accuracy within the same system, their performance drops significantly when applied to a different band or link. We also show that combining data from multiple systems improves the results. This work highlights important considerations for deploying ML-based sensing solutions in heterogeneous optical networks.” explains Leyla.
Co-authors of the study are Carlos Natalino Da Silva, Chalmers, Stefan Karlsson, Micropol Fiberoptics AB, Marco Ruffini, Trinity College Dublin, Eoin Kenny, HEAnet, Lena Wosinska, Chalmers and Marija Furdek Prekratic, Chalmers.
Leyla is currently finalizing her doctoral thesis and preparing for her PhD defence in February.
“In parallel to this, I will continue extending the results from this conference paper by developing more generalizable and adaptive ML techniques for polarization-based sensing to submit the outcomes to the invited journal version of this work. These methods are intended to enhance model robustness across different fiber types, operating conditions, and network environments.” Leyla concludes.
- Visiting Researcher, Communication, Antennas and Optical Networks, Electrical Engineering