Machine Learning for Future Fiber-Optic Communication Systems

Chapter Two: Machine learning for long-haul optical systems

Shaoliang Zhang^a; Christian Häger^b    aAcacia Communication Inc., Maynard, MA, United States^bChalmers University of Technology, Gothenburg, Sweden

Abstract

In this chapter, machine learning (ML) algorithm is introduced in single-step perturbation and multi-step digital back-propagation methods to compensate for intra-channel fiber nonlinearity. A wide & deep neural network is first introduced to use intra-channel cross-phase modulation and intra-channel four-wave mixing triplets as input features to predict the fiber nonlinearity at symbol rate. In the second part, a parameterized physics-based ML model is reviewed by emulating linear dispersion step as the weight matrices in a neural network ...