Machine learning-based EDFA Gain Model Generalizable to Multiple Physical Devices

TL;DR

This paper presents a neural network model for EDFA gain prediction that accurately generalizes across different physical devices, reducing errors significantly compared to previous models.

Contribution

The authors develop a neural network-based EDFA gain model that generalizes well to multiple physical devices, demonstrating low prediction errors on unseen units.

Findings

Low prediction error for trained device (MSE ≤ 0.04 dB²)

Effective generalization to different physical units (MSE ≤ 0.06 dB²)

Neural network models can reliably predict EDFA gain across devices.

Abstract

We report a neural-network based erbium-doped fiber amplifier (EDFA) gain model built from experimental measurements. The model shows low gain-prediction error for both the same device used for training (MSE $\leq$ 0.04 dB$^2$) and different physical units of the same make (generalization MSE $\leq$ 0.06 dB$^2$).