Abstract
This paper investigates several novel machine learning procedures that employ two machine learning stages to mitigate nonlinearity in dual polarized optical fiber systems. These employ a neural network pre-compensator at the transmitter and a classifier at the receiver. Different types of classifiers such as neural network and decision tree classifiers as well as a number of ensemble methods including boosting, random forest, and extra trees are investigated at the receiver. Here the extra trees classifier is found to yield the greatest Q-factor with ∼1.3 dB enhancement and lowest training computational time.
Disclosure statement
No potential conflict of interest was reported by the author(s).