Self-adaptive erbium-doped fiber amplifiers using machine learning

Abstract

This paper presents a method to autonomously adjust the operating point of amplifiers in a cascade using an approach based on machine learning. The goal is to smoothly adjust the gain of each amplifier in the cascade in order to reach predefined input and output power levels for the entire link, aiming to minimize both the noise figure and the gain flatness of the transmission system. The proposal uses an iterative method and performs feedforward and backward error adjustments based on local information. The experimental results indicate that our proposal can optimize the performance of the link ensuring predefined input and output power levels, which is important in a network scenario. As an example, our proposal was capable to define the gain of 6 amplifiers returning a link with a noise figure equal to 30.06 dB and a gain flatness equal to 5.26 dB, while maintaing the input and output powers around 3 dBm with an error lower than 0.1 dB.