Chromatic Dispersion Monitoring for Long-haul 28 GBaud PDM-QPSK Optical Signal based on CNN

A CNN-based neural network for chromatic dispersion monitoring method is proposed in this paper, named OptInception. It adopts the parallel structure for convolutional layers with different sizes. Experimentally, the mean absolute error and root-mean-squared error of estimation on 28 GBaud PDM-QPSK optical signals are below 0.1% with less than 32,000 ps/nm of residual chromatic dispersion. The effects of OSNR and signal power on estimation performance are investigated in experiment as well. In general, the proposed method presents high accuracy, robustness and adaptive characteristic in practice.