Abstract
Modern societies increasingly rely on fast, reliable, and energy-efficient communication networks to support education, healthcare, commerce, artificial intelligence (AI), and critical public services. The rapid growth of cloud computing, AI-driven workloads, and next-generation mobile systems is pushing optical fiber infrastructures toward their fundamental capacity limits. A primary barrier to further scaling is nonlinear interference caused by the Kerr effect, which constrains transmission performance and spectral efficiency. Accurate modeling of these nonlinear effects and their incorporation into network-level optimization are therefore essential to sustain future high-capacity communication systems
This dissertation is organized into two main parts addressing nonlinear interference modeling and its role in next-generation optical communication systems. The first part focuses on analytical estimation of nonlinear interference in point-to-point coherent optical fiber links. A comprehensive nonlinear interference model is derived for four-dimensional modulation formats that jointly encode information across both polarizations, overcoming limitations of existing models restricted to polarization-multiplexed signaling or symmetric constellations. The proposed framework accurately captures nonlinear interference for both symmetric and asymmetric four-dimensional constellations. In addition, a generalized nonlinear model is developed to account for practical pulse shaping, enabling accurate estimation of cross-phase modulation for raised-cosine spectra with arbitrary roll-off factors. These models provide a unified and accurate foundation for quality-of-transmission estimation and advanced modulation design.
The second part of the dissertation investigates the impact of nonlinear interference at the network level, with emphasis on band- and space-division elastic optical networks that form a critical component of modern cloud, data-center, and artificial-intelligence infrastructures. Physical-layer-aware routing and resource allocation frameworks are proposed that explicitly incorporate nonlinear interference, inter-channel stimulated Raman scattering, and inter-core crosstalk. Both multi-core-fiber and bundled multi-fiber-pair architectures are analyzed to identify their performance trade-offs and operational regimes. Channel-centric and impairment-aware optimization strategies are shown to significantly improve network efficiency and robustness by dynamically adapting to frequency-dependent impairments and network conditions. Large-scale network studies demonstrate that the proposed approaches substantially reduce blocking probability and enable more effective utilization of spectral and spatial resources in high-capacity optical backbone networks.
