A neural network for forward and inverse nonlinear Fourier transforms for fiber optic communication

TL;DR

This paper introduces a neural network capable of performing both forward and inverse nonlinear Fourier transforms, demonstrating accurate and generalizable performance for fiber optic communication signals.

Contribution

The paper presents a novel neural network architecture that efficiently performs both NFT and INFT, capturing their true characteristics and generalizing beyond training data.

Findings

Achieves RMSE of 5e-3 for forward and 3e-2 for inverse transforms.

Successfully handles a mix of NFDM-QAM signals.

Generalizes to arbitrary pulses beyond training data.

Abstract

We propose a neural network for both forward and inverse continuous nonlinear Fourier transforms, NFT and INFT respectively. We demonstrate the network's capability to perform NFT and INFT for a random mix of NFDM-QAM signals. The network transformations (NFT and INFT) exhibit true characteristics of these transformations; they are significantly different for low and high-power input pulses. The network shows adequate accuracy with an RMSE of 5e-3 for forward and 3e-2 for inverse transforms. We further show that the trained network can be used to perform general nonlinear Fourier transforms on arbitrary pulses beyond the training pulse types.