Abstract
Elastic optical network (EON)s offer flexible spectrum allocation, allowing for the accommodation of varying bandwidth traffic. Multiband and multicore fiber (MCF) EONs that extend the capacity in spectral and spatial dimensions have been proposed to meet future communication demands. Planning the resource usage of these networks has been the subject of extensive research. Careful design of routing, spectrum, core (if MCF), and modulation assignment algorithms enable efficient use of network resources in the wake of an increasing number of network users. In addition, consideration of physical layer impairments in EONs is important in the network planning stage for long-haul systems, especially when dealing with large dynamic traffic.
This dissertation investigates the resource provisioning problem for multiband and multicore fiber EONs using heuristic and machine learning approaches. The contributions of the dissertation work are mainly divided into four parts: (1) a Q-learning (reinforcement learning) based routing selection for C+L band multi-band network is proposed considering dynamic network status; (2) the impact of different impairments in a dynamic multicore fiber provisioning scheme is analyzed that highlights the importance of inclusion of non-XT impairments; (3) a multi-attribute decision-making based route and core selection method and a new spectrum assignment scheme is proposed for multicore networks while considering network fragmentation and energy efficiency; and (4) machine learning (ML) is introduced to capture the hidden relationships in the network to judge the suitability of candidate resources and to estimate the links noises, thereby speeding the quality of transmission constraint verification in comparison to the analytic Gaussian Noise model.
For each part of the dissertation, the performance of the proposed algorithms is evaluated and compared with standard and published algorithms as benchmarks. Results show that both multiband and multicore are potential candidates for handling increasing network traffic. Although each of them comes with new associated impairments, it is found that multiband can handle the demand increase only to a certain level, and multicore can considerably multiply the capacity, also providing greater flexibility in the impairment-aware design. Heuristic solutions enable the design of resource allocation so that the individual network aspects can be tackled by designing the rules for the selection and utilization of network resources in a unique manner while also being straightforward and easy to design and implement. On the other hand, the ML based solutions, especially the supervised approaches, enable learning unknown relationships of the network, allowing for different insights to help design the allocation solutions. The incorporation of ML can also help reduce the computation time for tasks that can become time-consuming when considering dynamic resource allocation for large loads and traffic. However, ML usage also comes with the need for a careful design and expertise with the problem domain. Hence, depending on the problem’s goal and the long-term network requirements, either heuristic, ML based, or a combination of their approaches can be chosen. The proposed algorithms are simple, practical, and can be used easily by the network operators.
