Artificial neural network‐based adaptive modulation for elastic optical networks. Issue 2 (23rd April 2021)

• Record Type: Journal Article
• Authors: Reihani, Amir Zarezadeh; Behdadfar, Mohammad; Sebghati, Mohammadali
• Is Part Of: Internet technology letters. Volume 5:Issue 2(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-04-23
• Subjects: artificial neural networks (ANN) -- elastic optical networks (EON) -- machine learning; Internet -- Periodicals; 004.67805
• DOI: 10.1002/itl2.291
• Languages: English
• ISSNs: 2476-1508

Abstract

Abstract : To cope with ever‐increasing complexity in call admission of next‐generation optical networks, deployment of machine learning stands as a promising solution. We investigate the feasibility of using machine learning to solve the problem of modulation type allocation for coherent optical orthogonal frequency‐division multiplexing (CO‐OFDM) elastic optical networks (EONs). We apply a multi‐layer perceptron (MLP) artificial neural network (ANN) to determine the suitable modulation type given the number of hops, path length, and bit error rate requirement of an incoming request. The proposed scheme is shown to outperform the distance adaptive and load‐aware methods in terms of spectral efficiency. Abstract : In this paper, we investigate the feasibility of deploying machine learning to allocate proper modulation types in the context of elastic optical networks (EONs). Using novel models of estimating quality‐of‐transmission (QoT), we exploit artificial neural networks to allocate the proper modulation type for a given source‐destination lightpath over a given EON architecture. Our proposed machine learning‐based call admission scheme deploys an multi‐layer perceptron ANN consisting of three inputs and seven outputs. Our simulation results demonstrate that the proposed scheme outperforms both distance adaptive and load‐aware methods in terms of average spectrum efficiency. In particular, the proposed scheme achieves up to a 34% increase of average spectrum efficiency compared to the conventional distance adaptive method.