Datasets for ML-Assisted Optimal Power and GSNR Estimation in Multi-band Elastic Optical Networks

Here’s a draft description you can use for your Zenodo dataset upload, tailored to your paper and emphasizing the datasets (C+L, C+L+S, C+L+S+E):

This dataset accompanies the paper “ML-Assisted Optimal Power and GSNR Estimation in Multi-band Elastic Optical Networks” by K. Ghodsifar, F. Arpanaei, H. Beyranvand, M. Ranjbar Zefreh, C. Natalino, P. Monti, S. Yan, Ó. González de Dios, J. M. Rivas-Moscoso, J. P. Fernández-Palacios, A. Sánchez-Macián, D. Larrabeiti, and J. A. Hernández.

The work addresses one of the main challenges in next-generation intelligent and autonomous optical networks: fast and accurate estimation of power and generalized signal-to-noise ratio (GSNR) in multi-band elastic optical networks (MB-EONs). Traditional analytical approaches (e.g., GN/EGN semi-closed form models) provide accurate results but are too computationally intensive for online network planning and optimization.

To overcome this limitation, the study:

• Employs a semi-closed form GN/EGN model to generate large-scale synthetic datasets.

• Trains machine learning (ML) models (Gradient Boosting and Neural Networks) to predict per-span power and GSNR profiles.

• Achieves prediction errors below 0.04 dB for power and 0.1 dB for GSNR in 99% of cases.

• Demonstrates that ML-assisted power optimization is 25–50× faster than analytical approaches, with negligible accuracy loss (≤0.1 dBm).

About the Datasets

The datasets are designed to train and validate ML-assisted models for power and GSNR estimation across different multi-band transmission scenarios. They are topology-independent and include diverse conditions (span length, channel loading, modulation format, and launch power).

We provide datasets for three key optical spectrum configurations:

• C+L band

• C+L+S band

• C+L+S+E band

Each dataset contains multiple scenarios, covering a wide range of system parameters:

• Span length: 40–100 km

• Launch power per channel: -5 to +5 dBm

• Loading factor: 50%–100%

• Modulation formats: Cardinality 1–6

• Channel configurations: Up to 160×75 GHz channels with inter-band guard bands

These datasets allow researchers and practitioners to:

• Reproduce the ML-assisted models for QoT estimation.

• Benchmark alternative ML or analytical approaches.

• Explore optimization strategies for multi-band elastic optical networks.