Proposal of a Fiber/Wireless System Assisted by Machine Learning Towards 6G Communications

This paper reports the performance evaluation of a fiber/wireless (FiWi) system assisted by a machine learning (ML) algorithm envisioned for the sixth generation of mobile networks (6G). An augmented real-valued time delay neural network (ARVTDNN) is trained to learn the inverse response of the non-linear distortions introduced by the communication chain. The trained ML algorithm operates as a digital pre-distortion (DPD) scheme, which allows an analog radio-over-fiber (A-RoF) system to be used to connect the central office (CO) with a low-cost remote radio head (RRH) installed in remote areas. A Rayleigh channel was employed to model the wireless radio-frequency (RF) signal transmission. The RF signal is linearized by the DPD scheme before being radiated, aiming to provide broadband communications for remote and rural areas. Numerical results demonstrate that the ML-based DPD scheme enables the seamless integration of A-RoF into the wireless system.