Enhancing optical fiber transmission performance through advanced link and system design

Distributed Raman amplification (DRA) is a key technology that can improve the performance of fiber optic communication systems. This amplification scheme provides several advantages over the Erbium-Doped Fiber Amplifiers (EDFAs), in terms of Noise Figure (NF), broadband gain, and flexibility in design by means of multi-pumping schemes. Due to its distributed amplification, DRA enables to control of the shape of signal power evolution in both frequency and fiber distance. This is crucial for attaining some of the long-term objectives in fiber optic communications, including optimization of Signal-to-Noise Ratio (SNR) and compensating for nonlinear impairments. However, the optimization of the pump power and wavelength values poses a challenge to DRA configurations. In this thesis, we utilize Machine Learning (ML) and optimization techniques to design signal power evolution in two dimensions (2D), i.e. frequency and fiber distance, using Raman amplifiers. First, an inverse system model based on a Convolutional Neural Network (CNN) is used to map the 2D signal power profiles to their corresponding Raman pump power and wavelength values. The CNN model has shown a statistically low error in learning the inverse mapping. However, its performance is not accurate for designing 2D profiles of practical interest, such as a 2D flat or a 2D symmetric (with respect to the midpoint in the distance). To accurately design the practical 2D profiles, we use an online optimization framework based on Differential Evolution (DE). In this framework, the DE adjusts the pump power values online on the setup aiming to reduce the cost value between the desired and the designed 2D profiles. The DE framework is also combined with the CNN inverse model to achieve better accuracy, more reliable optimum values, and faster convergence. Finally, we experimentally validate the performance of the CNN model, the DE, and the CNN-assisted DE framework using an amplifier setup employing four counter-propagating Raman pumps. Different target power profiles defined jointly in the entire C-band and in fiber distance are aimed to be designed. Moreover, the DE framework is tested and showed promising performance in an experimental multi-objective design scenario to achieve 2D profiles with flat gain levels at the end of the span, jointly with minimum spectral excursion over the entire fiber length.