Abstract
The exponential growth of global data traffic, driven by the proliferation of 5Gnetworks, AI-driven services, real-time applications, and cloud computing, is
placing increasing demands on existing optical networks. These networks are
now required to deliver significantly higher capacity, greater flexibility, and enhanced
scalability to support the dynamic and heterogeneous nature of modern applications. In
parallel, the rapid advancement of distributed computing, high-performance computing
devices, and quantum computing introduces further challenges for optical network security, as emerging threats increasingly target the physical layer of communication systems.
More recently, AI-assisted dynamic optical network control, multi-core fibre (MCF)-based
space-division multiplexing (SDM), and quantum-secured optical transmission have
emerged as promising research directions in response to the increasing demands of
modern network applications. However, several critical challenges remain. Specifically,
existing AI implementations predominantly focus on single-link QoT prediction and lack
effective cross-domain intelligent control mechanisms. Furthermore, the switching and
topology design for MCF-based SDM networks are not yet sufficiently hardware-efficient
or scalable. In addition, the secure integration of quantum and classical channels over
shared fibre continues to be limited by nonlinear impairments.
This thesis addresses these challenges through four key contributions. First, it develops AI-enhanced machine learning (ML) models for network-level and spectrum-level
quality of transmission (QoT) prediction, validated through extensive field trials. These
i
models enable high-precision QoT prediction for both fixed-grid and flex-grid dynamic
optical networks, significantly improving spectrum efficiency and service reliability.
An AI-engine-driven cross-domain orchestrator is further deployed in a multi-domain
field-trial testbed to demonstrate intelligent service delivery.
Second, a programmable reconfigurable optical add-drop multiplexer (ROADM) with
bypass capability is designed for MCF-based SDM networks, achieving 10–20% hardware
cost reduction compared with fully flexible SDM/WDM ROADM architectures.
Third, a performance-aware and hardware-efficient topology optimisation framework
is proposed for MCF-based SDM networks. Validated across a comprehensive set of
diverse traffic scenarios, this approach achieves hardware savings of up to 51% while
maintaining service performance and improving spectrum utilisation.
Finally, the thesis presents a comprehensive study on quantum-classical channel
coexistence in C-band over shared fibre. An ANN-based model is developed and enhanced
with transfer learning to accurately predict nonlinear impairments and to identify
feasible and infeasible coexistence regions across varying transmission scenarios.
Collectively, these contributions establish a robust framework for the future of optical
networking, encompassing AI-driven dynamic configuration, SDM-based scalability, and
QKD-secured resilience, to meet the demands of next-generation ICT ecosystems.
| Date of Award | 30 Sept 2025 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Shuangyi Yan (Supervisor) & Dimitra Simeonidou (Supervisor) |