Fiber-Agnostic Machine Learning-Based Raman Amplifier Models

Uiara C. de Moura, Darko Zibar, A. Margareth Rosa Brusin, Andrea Carena, Francesco Da Ros

Research output: Contribution to journalJournal articleResearchpeer-review

81 Downloads (Pure)

Abstract

Machine learning techniques have been applied to solve many open and highly complex problems in optical communications. In particular, neural networks (NN) have proved to be effective in learning the complex mapping between pump powers and gain profiles in Raman amplifiers. Since the Raman scattering process is highly dependent on the optical fiber, these NN-based Raman amplifier (RA) models are specific for a single optical fiber type. Therefore, countless NN models are required, one for each optical fiber type. In this work, we first show that by combining experimental data from different optical fiber sources, we can build a fiber-agnostic (or general) NN-based RA model. This single NN model can predict the gain profile of a new fiber type (unseen by the model during training) with a maximum absolute error (averaged over 1500 cases) as low as 0.22 dB. However, this generalization is only possible when the unseen fiber is similar to the fibers used to build the model. Therefore, a training dataset with a wide range of optical fibers is needed to enhance the chance of accurately predicting the gain of a new fiber. The first contribution of this work aims at avoiding the time-consuming experimental measurements of countless fibers. For that, here we extend and improve our general model by numerically generating the dataset. By doing so, it is possible to generate uniformly distributed data that covers a wide range of optical fiber types. To guarantee that the numerical simulation is able to reproduce the experimental results, we also propose a fitting procedure to extract the Raman gain coefficient from a few experimental measurements. The proposed numerical data-based general model is used to predict the gain of real fibers considering their corresponding Raman gain coefficient (extracted by the fitting procedure). The results show that the averaged maximum prediction error is reduced when compared to the limited experimental data-based general models. However, even after the fitting procedure, this fiber-agnostic model purely based on numerical data is still limited by the discrepancies between numerical and experimental measurements. Therefore, as the second and final contribution of this work, we propose the use of transfer learning (TL) to re-train the numerical data-based general model using just a few experimental measurements. Compared with the fiber-specific models, this TL-upgraded general model reaches very similar accuracy, with just 3.6% of the experimental data. These results demonstrate that the already fast and accurate NN-based RA models can be upgraded to have strong fiber generalization capabilities. They can, therefore, significantly reduce the total number of RA models that would be needed in multi-span optical networks.
Original languageEnglish
JournalJournal of Lightwave Technology
Volume41
Issue number1
Pages (from-to)83 - 95
ISSN0733-8724
DOIs
Publication statusPublished - 2022

Keywords

  • Optical communications
  • Optical amplifiers
  • Machine learning
  • Neural networks

Fingerprint

Dive into the research topics of 'Fiber-Agnostic Machine Learning-Based Raman Amplifier Models'. Together they form a unique fingerprint.

Cite this