Modulation Classification with Convolutional Neural Network Based Deep Learning in Elastic Optical Network

Abstract

In this paper, the modulations are classified by six types of deep learning (DL) methods in Elastic Optical Network (EON). In EON, flexible coherent transceivers are applied that can demodulate received signals in different modulations. Here, DL-based methods are proposed for modulation classification in flexible receivers. Three Transfer Learning (TL) methods as AlexNet, GoogleNet, and InceptionV3, are applied, and three convolution neural networks based on different structures (3, 4, and 5 layers) are proposed to show that DL-based methods are practical to identify the usual modulation formats. The performance is evaluated in EON, which includes five different modulation types in four different symbol rates. The total number of studied scenarios are 248, with various link in terms of fiber, power, and span number. In practical scenarios, the numerical results show that, Modulation classification can be done with 99% accuracy.