Skip to main navigation Skip to search Skip to main content

Développement d'un modèle cognitif de calcul de la qualité de transmission dans les réseaux optiques

Translated title of the thesis: Development of a cognitive model for the quality of transmission computation in optical networks
  • Sandra Aladin

Student thesis: Master's thesisMaster in Engineering: Engineering

Abstract

The strong growth of IP traffic, powered by the Internet on smartphones and tablets, cloud services, games and video on the Internet, requires an increase in bandwidth as well as the use of new transmission technologies. This increase in transmission capacity and the diversity of services sharing the same link lead to transmission quality parameters that are difficult to optimize. The concept of cognitive optical networks has been proposed to solve this problem. Solutions have been proposed for optical connections at Gb/s bit rate and an On-Off Keying OOK modulation format. Case-Based Reasoning CBR technique has been proposed for the classification of optical connections prior to their establishment. In recent works, the technique has been applied to an optical network configuration with higher bit rates and a more advanced modulation format. A more recent solution takes into account different bit rates and modulation formats as well as nonlinear effects through margins. Our research on a model of estimation of the quality of the transmission takes into account the nonlinear effects by means of analytical formulas described in Gaussian noise model described. We apply three learning techniques K-NN, RF and SVM to these generated synthetic data to predict the transmission quality of optical connections as a function of link and signal parameters. Three classifiers are proposed and evaluated according to the following performance metrics: the classification accuracy, data processing time and scalability. An analysis of the results makes it possible to determine the best method to adopt for the development of the proposed quality of transmission estimation tool. The Support Vector Machine (SVM) method performs better than the K-Nearest Neighbors (K-NN) method or the Random Forests (RF). The conclusion is that cognition based on machine learning techniques can be successfully implemented in heterogeneous optical networks supporting different bit rates and modulation formats. In addition, this strategy can improve the processing time for estimating QoT optical connections. Nevertheless, studies based on field data as well as the application of optimization techniques to algorithms and to data remain aspects to be explored.
Date29 Jun 2018
Original languageFrench
Awarding Institution
  • École de technologie supérieure
SupervisorChristine Tremblay (Supervisor)

Cite this

'