Convolutional Neural Network Based Blind Estimation of Generalized Mutual Information for Optical Communication

Johannes Ackermann, Maximilian Schädler, Christian Blümm

Abstract: We propose blind GMI estimation from received constellation diagrams, based on a convolutional neural network. For coherent optical transmissions between 600Gb/s and 960Gb/s with varying modulation schemes, mean errors of 1.4% without and 0.8% with channel specific fine-tuning are achieved, significantly outperforming previous methods.