Machine Learning in Optical Fibre Networking Under Uncertainty

  • Fanchao Meng

Student thesis: Doctoral ThesisDoctor of Philosophy (PhD)


During the past three decades, there have been major achievements on accurate modelling of behaviour and operation of telecommunication networks utilising classical methods with analytical or heuristic models. Currently, there is a big hype on the application of Artificial Intelligence (AI) and Machine Learning (ML) in the telecommunication space. However, there is a gap on scientific scrutiny of advantages of AI and ML compared to existing methods. Analogue telecommunication networks, i.e., optical and wireless networks seem to be the most suitable problem space for AI and ML. They are complicated network systems in nature that are highly dependent on ubiquitous physical layer
uncertainties induced by subsystems and transmission mediums such as amplifiers, fibres, switches and transceivers. The problem space becomes even more complicated with recent advances in optical and wireless technologies that allow the development and operation of a fully programmable and dynamic network. This work focuses on benefits and applications of learning agents built on top of cognitive optical networks. It discusses the appropriateness of employing AI methods for specific problems in optical networks and address the importance of online learning with restricted monitoring data. It proposes and experimentally demonstrates a brand new way of carrying out optical
network analytics utilising hybrid probabilistic and generative learning model which differs from traditional deterministic models. The result of this investigation, for the first time, can shed light into future AI and ML research in the optical network planning under uncertainty
Date of Award7 May 2019
Original languageEnglish
Awarding Institution
  • The University of Bristol
SupervisorDimitra Simeonidou (Supervisor)

Cite this