Make your plans to attend
Vaccination required
>
Conference 12028 > Paper 12028-32
Paper 12028-32

Optimal control of Beneš optical networks assisted by machine learning

Abstract

We propose a Machine Learning based solution for the calculation of optimal the control state in Beneš networks: for a given topology we deterministically calculate all the control states (CSs) allowing for the requested permutation and, using a pre-trained neural network, we estimate for each CS the associated OSNR at each output port, finally selecting the CS allowing for the optimal overall OSNR. We applied this approach to an 8x8 network, with 20 Mach-Zender based crossbar switches, and trained the network with a data set of 3000 OSNRs estimated for random combinations of the CSs using Synopsys Optsim.

Presenter

Paolo Bardella
Politecnico di Torino (Italy)
Author
Politecnico di Torino (Italy)
Author
Politecnico di Torino (Italy)
Author
Muhammad Umar Masood
Politecnico di Torino (Italy)
Author
Synopsys, Inc. (United States)
Presenter/Author
Paolo Bardella
Politecnico di Torino (Italy)
Author
Andrea Carena
Politecnico di Torino (Italy)
Author
Vittorio Curri
Politecnico di Torino (Italy)