Domain Adaptation: the Key Enabler of Neural Network Equalizers in Coherent Optical Systems

TL;DR

This paper presents a domain adaptation method that significantly reduces training time for neural network equalizers in coherent optical systems by effectively using synthetic data, demonstrated across multiple experimental setups.

Contribution

The paper introduces a domain adaptation and randomization approach that enables neural network equalizers to be trained efficiently with synthetic data, reducing training time by up to 99%.

Findings

Achieved up to 99% reduction in training process.

Validated approach across three experimental setups.

Effective calibration of neural network equalizers using synthetic data.

Abstract

We introduce the domain adaptation and randomization approach for calibrating neural network-based equalizers for real transmissions, using synthetic data. The approach renders up to 99\% training process reduction, which we demonstrate in three experimental setups.