This repository contains the MATLAB code for training neural networks (NN) to equalize nonlinear frequency division multiplexing (NFDM) systems. The NFT-domain NN and time-domain NN are trained to regress towards the transmitted QPSK-modulated b-coefficients.
The provided data files are obtained from 2800 km and 1400 km transmission simulations with 40 km span lengths. The 2800 km transmission data is used to train the NFT-domain NN, while the 1400 km data is used to train the time-domain NN.