Analyzing Machine Learning Paradigms for the Linearization of 5G Radio over Fiber Systems

This work presents the use, analyses and proposals of improvement of Machine Learning (ML) techniques for compensating nonlinearities from the fifth generation of mobile communications (5G) Radio-over-Fiber (RoF) systems. RoF- based transmission networks enable us to take advantage of the high bandwidth of optical systems and the flexibility of wireless transmission. However, integrating these systems generates non-linear distortions in the modulated signal. We assess three machine-learning techniques to correct these nonlinearities, one based on Support Vector (SV), another based on regression trees, and a third based on Artificial Neural Networks (ANN). We apply them to an Orthogonal Frequency Division Multiplexing (OFDM) and investigate its performance based on two metrics, the Mean Squared Error (MSE) and the Power Spectral Density (PSD). We demonstrate ANN provides enhanced performance for this task, by means of providing 53% improvement in the MSE parameter and maintaining the low-complexity approach.