Optimal control of Beneˇs optical networks assisted by machine learning Ihtesham Khana, Lorenzo Tunesia, Muhammad Umar Masooda, Enrico Ghillinob, Paolo Bardellaa, Andrea Carenaa, and Vittorio Curria aPolitecnico di Torino, Corso Duca degli Abruzzi 24, Torino, Italy bSynopsys Inc., Executive Blvd 101, Ossining, New York, USA ABSTRACT Beneˇs networks represent an excellent solution for the routing of optical telecom signals in integrated, fully reconfigurable networks because of their limited number of elementary 2x2 crossbar switches and their non- blocking properties. Various solutions have been proposed to determine a proper Control State (CS) providing the required permutation of the input channels; since for a particular permutation, the choice is not unique, the number of cross-points has often been used to estimate the cost of the routing operation. This work presents an advanced version of this approach: we deterministically estimate all (or a reasonably large number of) the CSs corresponding to the permutation requested by the user. After this, the retrieved CSs are exploited by a data- driven framework to predict the Optical Signal to Noise Ratio (OSNR) penalty for each CS at each output port, finally selecting the CS providing minimum OSNR penalty. Moreover, three different data-driven techniques are proposed, and their prediction performance is analyzed and compared. The proposed approach is demonstrated using 8x8 Beneˇs architecture with 20 ring resonator-based crossbar switches. The dataset of 1000 OSNRs realizations is generated synthetically for random combinations of the CSs using Synopsys® Optsim™ simulator. The computational cost of the proposed scheme enables its real-time operation in the field.

Optimal control of Beneš optical networks assisted by machine learning / Khan, Ihtesham; Tunesi, Lorenzo; Masood, Muhammad Umar; Ghillino, Enrico; Bardella, Paolo; Carena, Andrea; Curri, Vittorio. - ELETTRONICO. - (2022), p. 32. (Intervento presentato al convegno SPIE Photonics West tenutosi a San Francisco nel 2022) [10.1117/12.2608595].

Optimal control of Beneš optical networks assisted by machine learning

Khan, Ihtesham;Tunesi, Lorenzo;Masood, Muhammad Umar;Bardella, Paolo;Carena, Andrea;Curri, Vittorio
2022

Abstract

Optimal control of Beneˇs optical networks assisted by machine learning Ihtesham Khana, Lorenzo Tunesia, Muhammad Umar Masooda, Enrico Ghillinob, Paolo Bardellaa, Andrea Carenaa, and Vittorio Curria aPolitecnico di Torino, Corso Duca degli Abruzzi 24, Torino, Italy bSynopsys Inc., Executive Blvd 101, Ossining, New York, USA ABSTRACT Beneˇs networks represent an excellent solution for the routing of optical telecom signals in integrated, fully reconfigurable networks because of their limited number of elementary 2x2 crossbar switches and their non- blocking properties. Various solutions have been proposed to determine a proper Control State (CS) providing the required permutation of the input channels; since for a particular permutation, the choice is not unique, the number of cross-points has often been used to estimate the cost of the routing operation. This work presents an advanced version of this approach: we deterministically estimate all (or a reasonably large number of) the CSs corresponding to the permutation requested by the user. After this, the retrieved CSs are exploited by a data- driven framework to predict the Optical Signal to Noise Ratio (OSNR) penalty for each CS at each output port, finally selecting the CS providing minimum OSNR penalty. Moreover, three different data-driven techniques are proposed, and their prediction performance is analyzed and compared. The proposed approach is demonstrated using 8x8 Beneˇs architecture with 20 ring resonator-based crossbar switches. The dataset of 1000 OSNRs realizations is generated synthetically for random combinations of the CSs using Synopsys® Optsim™ simulator. The computational cost of the proposed scheme enables its real-time operation in the field.
2022
9781510649279
9781510649286
File in questo prodotto:
File Dimensione Formato  
2022_SPIE_Optimal_control_of_Bene__optical_networks_assisted_by_machine_learning (1).pdf

accesso aperto

Tipologia: 2. Post-print / Author's Accepted Manuscript
Licenza: PUBBLICO - Tutti i diritti riservati
Dimensione 1.17 MB
Formato Adobe PDF
1.17 MB Adobe PDF Visualizza/Apri
Khan-Optimal.pdf

accesso aperto

Tipologia: 2a Post-print versione editoriale / Version of Record
Licenza: PUBBLICO - Tutti i diritti riservati
Dimensione 1.26 MB
Formato Adobe PDF
1.26 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11583/2958548