Analyse de données de monitoring de performance dans les réseaux optiques cohérents

Abstract

The combined application of coherent detection, amplitude and phase modulation formats, as well as advanced Digital Signal Processing (DSP) and Forward Error Correction (FEC) techniques, has made it possible to increase the bit rate per channel, bringing the transmission capacity of optical transmission systems to more than 10 Tb/s per link. Such systems enable network operators to meet the growing and increasingly demanding needs of users. However, it is not enough to offer a high-speed connection to customers; operators must also ensure that service level agreements (SLAs) are met. In this context, Network Management Systems (NMS) continuously monitor the health of the network by collecting data from the devices deployed in the network. This data is used to analyze the causes of performance degradation in the network and equipment failures, to manage outages and maintenance interventions, and is used by network planners to plan future deployments based on traffic growth. On the other hand, the availability of data due to the development of the Internet of Things and the amount of connected data sources has given rise to data science which has already found applications in many areas. With the advent of coherent systems that collect data via DSP modules and massive data storage and processing capabilities, data science has recently emerged in the field of optical telecommunications. The central idea of the thesis is to extract the maximum of information and knowledge by using the techniques of performance data analysis of 155 lightpaths collected over a period of one year in an operator's production network with the objective to better understand the root cause of performance degradations as a first step towards proactive network management approaches. The first part of this thesis consists in characterizing the performance anomalies detected during the observation period with the objective of establishing correlations between the monitored parameters. A clustering of the 155 lightpaths based on their statistical characteristics is then done by applying an evolutive method of time series clustering in order to discover information underlying the data. Finally, a method for detecting optical loss degradation in an optical link, based on the use of three complementary statistical methods – Seasonal-Trend decomposition using LOESS, STL, Mann-Kendall test and Sen's slope – is proposed and validated with field data.

Date	23 Jul 2021
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Christine Tremblay (Supervisor) & Maurice O'Sullivan (Co-supervisor)