OSNR prediction for optical links via learned noise figures

Sarah Kamel, Hartmut Hafermann, Dylan Le Gac, Ludovic Dos Santos, Balázs Kégl, Yann Frignac, Gabriel Charlet

Abstract: We predict the per-channel OSNR of optical links with up to 23 EDFAs via a machine learning model based on learned noise figures from experimental data. For a 20 span link, the error margin to cover 99% of cases is less than 0.35 dB.