Fiber-Agnostic Machine Learning-Based Raman Amplifier Models

Abstract

General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. Users may download and print one copy of any publication from the public portal for the purpose of private study or research.  You may not further distribute the material or use it for any profit-making activity or commercial gain  You may freely distribute the URL identifying the publication in the public portal If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

Machine learning meets photonics

“…Additionally, solving the governing equation might require solving a system of partial differential equations through numerical methods, for example in the case of Raman amplification. 19 To increase accuracy and inference speed, black-box neural-network models of Raman and EDFA have been proposed throughout the years [20][21][22][23][24][25][26][27][28][29][30][31] progressively providing higher generalizability, e.g. to multiple physical units of similar make (ED-FAs 23 ), to multiple input channel loads 24 and chosen fiber type.…”

“…to multiple physical units of similar make (ED-FAs 23 ), to multiple input channel loads 24 and chosen fiber type. 25 As the model is required to generalize to a larger dimensional space of parameters, model complexity and availability of training data become challenging. Therefore data-augmentation and transfer learning approaches have been proposed.…”

“…Therefore data-augmentation and transfer learning approaches have been proposed. 25 Additionally, new directions are being explored to focus on grey-box models where physical knowledge is embedded in the network 26,27 or even nearly-white-box models where ML is used only to represent the parts of the physical model that are challenging to compute and/or optimize. As an example, in order to optimize a Raman amplifier, the Raman gain spectrum is normally not described in a form that is differentiable with respect to the pump frequency, but it can be made differentiable through fitting.…”

Modeling optical amplifiers: from inverse design to full system optimization

“…However, due to the inherent nonlinear fitting ability of a NN, a model overfitted to a single physical device would not be of great interest. The ability of black-box models to generalize to multiple physical devices has been investigated for EDFAs [13] as well as Raman amplifiers [21], showing promising prospects for moving beyond a unit-specific model. In the latter work, generalization to amplifiers relying on different fiber types has been achieved by proposing the use of a NN model pre-trained on synthetic data generated through a loosely fit numerical model, followed by a quick re-training stage (following the paradigm of transfer learning) using experimental measurements [14].…”

ML-Based Spectral Power Profiles Prediction in Presence of ISRS for Ultra-Wideband Transmission

“…One model directly predicts the spectral and spatial evolution of power profiles, while the other predicts the parameters required in the closed form-model formula demonstrated in [20] for NLI estimation, where instead these parameters are determined through fitting. This machine learning (ML) based solution finds support in the promising results obtained when ML and ANNs were applied to other optical communication systems problems, such as in the analysis and design of Raman amplifiers [21]- [24].…”