End-to-end Learning of a Constellation Shape Robust to Variations in SNR and Laser Linewidth

TL;DR

This paper introduces an autoencoder-based geometric shaping method that creates a constellation resilient to SNR and laser linewidth variations, maintaining significant mutual information gains over traditional QAM schemes.

Contribution

The novel autoencoder-based constellation design demonstrates robustness to SNR and laser linewidth estimation errors, preserving shaping gains across diverse conditions.

Findings

Maintains up to 0.3 bits/symbol mutual information gain over QAM

Robust to variations in SNR and laser linewidth

Effective geometric shaping for optical communication

Abstract

We propose an autoencoder-based geometric shaping that learns a constellation robust to SNR and laser linewidth estimation errors. This constellation maintains shaping gain in mutual information (up to 0.3 bits/symbol) with respect to QAM over various SNR and laser linewidth values.