{"id":{"repo_id":"cambridge","oai_identifier":"oai:www.repository.cam.ac.uk:1810/347239"},"canonical_url":"https://search.dev.ndltd.org/etd/cambridge/oai:www.repository.cam.ac.uk:1810/347239","repository":{"repo_id":"cambridge","name":"Cambridge University","base_url":"https://api.repository.cam.ac.uk/server/oai/request"},"display":{"title":"Machine Learning for Optical Fibre Communication Systems","abstract":"Global demand for internet traffic is growing at a rapid rate, driven by the adoption of new technologies and increased demand from consumers. This continued growth is exerting pressure on optical fibre communication systems and networks, which carry the bulk of modern internet traffic, leading to a risk of a capacity crunch. The overarching goal of this thesis is to alleviate this pressure by increasing the capacity of optical fibre communication systems. Machine learning is an attractive technology to help achieve this goal, due to the vast quantities of data generated by these systems, the complexity of their operation in the face of nonlinearities and the demonstrated success of machine learning in a vast array of applied fields, including optical networks. We highlight a number of desirable properties for machine learning methods applied within the optical fibre communications domain, namely the effective use of a priori knowledge, interpretable model outputs, well-quantified predictive uncertainty and transparent model design, and discuss to what extent these properties are satisfied by the work in this thesis. First, we focus on estimation of the physical layer parameters at the receiver, to increase the capacity by reducing the uncertainty associated with the physical layer in a non-disruptive way. Gaussian process regression is leveraged to learn a probabilistic, computationally cheap mapping from the physical layer parameters to SNR. This is applied within both a novel history matching-based parameter estimation technique and a novel approach for optimisation of the physical layer parameters in terms of the gain afforded by digital backpropagation. We then consider how to embed our a priori knowledge within the machine learning model itself, for both link and network scale problems. A novel technique is presented based on multi-task learning for combining a priori information from physical models of transmission with measured data from and experimental optical fibre communication link within the framework of a Gaussian process. Then, we show how a priori knowledge can be used to increase the efficacy of machine learning for network scale problems, embedding information describing the current network state into the action space of a reinforcement learning solution for routing and wavelength assignment in a simulated optical network. Planned future work will focus on extending the presented techniques to better incorporate the desirable machine learning properties outlined and increase the scope of applicability to more complex systems.","abstract_html":"Global demand for internet traffic is growing at a rapid rate, driven by the adoption of new technologies and increased demand from consumers. This continued growth is exerting pressure on optical fibre communication systems and networks, which carry the bulk of modern internet traffic, leading to a risk of a capacity crunch. The overarching goal of this thesis is to alleviate this pressure by increasing the capacity of optical fibre communication systems. Machine learning is an attractive technology to help achieve this goal, due to the vast quantities of data generated by these systems, the complexity of their operation in the face of nonlinearities and the demonstrated success of machine learning in a vast array of applied fields, including optical networks. We highlight a number of desirable properties for machine learning methods applied within the optical fibre communications domain, namely the effective use of a priori knowledge, interpretable model outputs, well-quantified predictive uncertainty and transparent model design, and discuss to what extent these properties are satisfied by the work in this thesis. First, we focus on estimation of the physical layer parameters at the receiver, to increase the capacity by reducing the uncertainty associated with the physical layer in a non-disruptive way. Gaussian process regression is leveraged to learn a probabilistic, computationally cheap mapping from the physical layer parameters to SNR. This is applied within both a novel history matching-based parameter estimation technique and a novel approach for optimisation of the physical layer parameters in terms of the gain afforded by digital backpropagation. We then consider how to embed our a priori knowledge within the machine learning model itself, for both link and network scale problems. A novel technique is presented based on multi-task learning for combining a priori information from physical models of transmission with measured data from and experimental optical fibre communication link within the framework of a Gaussian process. Then, we show how a priori knowledge can be used to increase the efficacy of machine learning for network scale problems, embedding information describing the current network state into the action space of a reinforcement learning solution for routing and wavelength assignment in a simulated optical network. Planned future work will focus on extending the presented techniques to better incorporate the desirable machine learning properties outlined and increase the scope of applicability to more complex systems.","abstract_has_math":false,"creators":["Nevin, Joshua"],"institution":"University of Cambridge","degree_name":"Doctor of Philosophy (PhD)","degree_level":"Doctoral","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Savory, Seb"],"committee_chairs":[],"committee_members":[],"year":2022,"date_issued":"2022-09-30","date_published":"2022-09-30","updated_at":"2026-07-22T22:24:21Z","subjects":["Machine learning","Optical fibre communications"],"languages":["eng"],"rights":[],"rights_urls":["https://www.rioxx.net/licenses/all-rights-reserved/"],"identifier_entries":[{"key":"dc:creator.authoridentifier","label":"Author Identifier","values":["0000000260676892"],"render_values":[{"text":"0000-0002-6067-6892","href":"https://orcid.org/0000-0002-6067-6892","code":true}]}]},"links":{"outbound_url":"https://doi.org/10.17863/CAM.94656","outbound_label":"DOI","outbound_source":"dc:identifier.doi"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Savory, Seb"]},{"key":"dc:creator","label":"Author","values":["Nevin, Joshua"]},{"key":"dc:creator.authoridentifier","label":"Author Identifier","values":["0000000260676892"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2022-09-30"]},{"key":"dc:publisher.institution","label":"Dc Publisher Institution","values":["University of Cambridge"]},{"key":"dc:relation.isreferencedby.uri","label":"Dc Relation Isreferencedby URI","values":["https://www.repository.cam.ac.uk/handle/1810/347239"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"dc:type.qualificationlevel","label":"Dc Type Qualificationlevel","values":["Doctoral"]},{"key":"dc:type.qualificationname","label":"Dc Type Qualificationname","values":["Doctor of Philosophy (PhD)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Machine learning","Optical fibre communications"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["https://www.rioxx.net/licenses/all-rights-reserved/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.doi","label":"DOI","values":["10.17863/CAM.94656"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/fdd6a2f5-bcbf-4571-87a2-409a4f31db1e/download"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Global demand for internet traffic is growing at a rapid rate, driven by the adoption of new technologies and increased demand from consumers. This continued growth is exerting pressure on optical fibre communication systems and networks, which carry the bulk of modern internet traffic, leading to a risk of a capacity crunch. The overarching goal of this thesis is to alleviate this pressure by increasing the capacity of optical fibre communication systems. Machine learning is an attractive technology to help achieve this goal, due to the vast quantities of data generated by these systems, the complexity of their operation in the face of nonlinearities and the demonstrated success of machine learning in a vast array of applied fields, including optical networks. We highlight a number of desirable properties for machine learning methods applied within the optical fibre communications domain, namely the effective use of a priori knowledge, interpretable model outputs, well-quantified predictive uncertainty and transparent model design, and discuss to what extent these properties are satisfied by the work in this thesis. First, we focus on estimation of the physical layer parameters at the receiver, to increase the capacity by reducing the uncertainty associated with the physical layer in a non-disruptive way. Gaussian process regression is leveraged to learn a probabilistic, computationally cheap mapping from the physical layer parameters to SNR. This is applied within both a novel history matching-based parameter estimation technique and a novel approach for optimisation of the physical layer parameters in terms of the gain afforded by digital backpropagation. We then consider how to embed our a priori knowledge within the machine learning model itself, for both link and network scale problems. A novel technique is presented based on multi-task learning for combining a priori information from physical models of transmission with measured data from and experimental optical fibre communication link within the framework of a Gaussian process. Then, we show how a priori knowledge can be used to increase the efficacy of machine learning for network scale problems, embedding information describing the current network state into the action space of a reinforcement learning solution for routing and wavelength assignment in a simulated optical network. Planned future work will focus on extending the presented techniques to better incorporate the desirable machine learning properties outlined and increase the scope of applicability to more complex systems."]},{"key":"dc:format.checksum.md5","label":"Dc Format Checksum Md5","values":["1dd8f4c9291941335295192936534def"]},{"key":"dc:title","label":"Title","values":["Machine Learning for Optical Fibre Communication Systems"]}]}],"canonical_facts":{"dc:contributor.advisor":["Savory, Seb"],"dc:creator":["Nevin, Joshua"],"dc:creator.authoridentifier":["0000000260676892"],"dc:date.issued":["2022-09-30"],"dc:description.abstract":["Global demand for internet traffic is growing at a rapid rate, driven by the adoption of new technologies and increased demand from consumers. This continued growth is exerting pressure on optical fibre communication systems and networks, which carry the bulk of modern internet traffic, leading to a risk of a capacity crunch. The overarching goal of this thesis is to alleviate this pressure by increasing the capacity of optical fibre communication systems. Machine learning is an attractive technology to help achieve this goal, due to the vast quantities of data generated by these systems, the complexity of their operation in the face of nonlinearities and the demonstrated success of machine learning in a vast array of applied fields, including optical networks. We highlight a number of desirable properties for machine learning methods applied within the optical fibre communications domain, namely the effective use of a priori knowledge, interpretable model outputs, well-quantified predictive uncertainty and transparent model design, and discuss to what extent these properties are satisfied by the work in this thesis. First, we focus on estimation of the physical layer parameters at the receiver, to increase the capacity by reducing the uncertainty associated with the physical layer in a non-disruptive way. Gaussian process regression is leveraged to learn a probabilistic, computationally cheap mapping from the physical layer parameters to SNR. This is applied within both a novel history matching-based parameter estimation technique and a novel approach for optimisation of the physical layer parameters in terms of the gain afforded by digital backpropagation. We then consider how to embed our a priori knowledge within the machine learning model itself, for both link and network scale problems. A novel technique is presented based on multi-task learning for combining a priori information from physical models of transmission with measured data from and experimental optical fibre communication link within the framework of a Gaussian process. Then, we show how a priori knowledge can be used to increase the efficacy of machine learning for network scale problems, embedding information describing the current network state into the action space of a reinforcement learning solution for routing and wavelength assignment in a simulated optical network. Planned future work will focus on extending the presented techniques to better incorporate the desirable machine learning properties outlined and increase the scope of applicability to more complex systems."],"dc:format.checksum.md5":["1dd8f4c9291941335295192936534def"],"dc:identifier.doi":["10.17863/CAM.94656"],"dc:identifier.uri":["https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/fdd6a2f5-bcbf-4571-87a2-409a4f31db1e/download"],"dc:language":["eng"],"dc:publisher.institution":["University of Cambridge"],"dc:relation.isreferencedby.uri":["https://www.repository.cam.ac.uk/handle/1810/347239"],"dc:rights":["https://www.rioxx.net/licenses/all-rights-reserved/"],"dc:subject":["Machine learning","Optical fibre communications"],"dc:title":["Machine Learning for Optical Fibre Communication Systems"],"dc:type":["Thesis"],"dc:type.qualificationlevel":["Doctoral"],"dc:type.qualificationname":["Doctor of Philosophy (PhD)"]},"updated_at":"2026-07-22T22:24:21Z"}