Going Deeper into OSNR Estimation with CNN

As optical performance monitoring (OPM) requires accurate and robust solutions to tackle the increasing dynamic complicated network architectures, we experimentally demonstrate an end-to-end signal-to-noise (OSNR) estimation method based on convolutional neural (CNN), named OptInception. The design principles of proposed scheme are specified. idea behind combination Inception module finite impulse response (FIR) filter is elaborated as well. We evaluate mean absolute error (MAE) root-mean-squared (RMSE) OSNR monitored in PDM-QPSK PDM-16QAM signals under various symbol rates. results suggest that MAE reaches low 0.125 dB RMSE 0.246 general. OptInception also proved be insensitive rate, modulation format, chromatic dispersion. investigation kernels CNN indicates helps layers learn much more than a lowpass or bandpass filter. Finally, comparison complexity presents advantages