Surrogate Model for Adaptive Control of Optical Amplifier Operating Point based on Machine Learning

Freitas, Leandro Martins de; Barboza, Erick de A.; Bastos-Filho, Carmelo J. A.; Martins-Filho, Joaquim F.

doi:10.1109/latincom53176.2021.9647798

Cited by 1 publication

(3 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Thus, it is necessary to evaluate the performance of the proposed model in links with different lengths to perform a better analysis of the surrogate model proposal. Therefore, in this work, we extend the results presented in [11] by considering links with 2, 4, 5, 6, 7, and 8 amplifiers.…”

Section: Problem Contextmentioning

confidence: 75%

“…In [11], we investigated the creation of a surrogate model to obtain a solution for the ACOP as good as the multi-objective algorithm but obtained in less time. The objective was to evaluate how to create a model that can replace the MOO evolutionary algorithm with a good level of precision in the task of defining amplifier gains and VOAs losses in an optical link.…”

Section: Problem Contextmentioning

confidence: 99%

“…In [11], a new strategy for building the database is proposed, as well as new approaches for allocating and training the models. The articles [1] and [4] were the basis of this study, in the sense that the MOO algorithm was used to create the database and surrogate models to solve the problem.…”

Section: Autonomous Optical Amplifiersmentioning

confidence: 99%

See 2 more Smart Citations

Applying Machine Learning to create a Surrogate Model for Adaptive Control of Optical Amplifier

Freitas

Silva

Barboza

et al. 2022

Preprint

View full text Add to dashboard Cite

The Adaptive Control of the optical amplifier Operating Point (ACOP) is one of the problems presented in the challenge of Dynamic operation in optical communication networks. The ACOP approaches aim to define the gains of the optical amplifiers dynamically to increase the transmission quality after a cascade of amplifiers. The most recent ACOP approach uses a multi-objective evolutionary optimization algorithm to define the gains of the amplifiers to maximize the Optical Signal to Noise Ratio (OSNR) and minimize OSNR ripple. Despite the promising results regarding Quality of Transmission, relying on an evolutionary algorithm to make decisions in real-time is not desirable because its iterative method usually implies a high execution time. In previous work, we proposed a surrogate model that can obtain solutions as good as the multi-objective evolutionary algorithm but in less time. We considered five Machine Learning (ML) regression techniques, trained with the optimization algorithm solutions. This article extends the previous work by experimenting with different link lengths, analyzing how the solution is impacted given different numbers of amplifiers on the link. Results showed that the regression median error is less than 1 dB for all cases. One regressor can be used to define amplifiers' gains and variable optical attenuators' losses of an entire optical link. It also showed that the most straightforward regressor is 28,000 times faster than the evolutionary optimization approach.

show abstract

Section: Problem Contextmentioning

confidence: 75%

Section: Problem Contextmentioning

confidence: 99%