The estimation of spectral spacing (guard band) among optical channels in gridless WDM systems would be decisive for making swift decisions in reconfigurable optical add-drop multiplexers (ROADMs) to avoid linear interchannel interference (ICI) effects during the channel aggregation process in transit nodes. In this work, we propose a method based on the construction of heat scatter images from constellation diagrams along with convolutional neuronal networks (CNN) to identify when optical channels are spectrally overlapped as well as the value in GHz of the channel separation in a specific optical channel without adjacent channels information. We validate our method in a gridless 16-QAM Nyquist-WDM system with different channel spacing and optical signal-to-noise ratio (OSNR). Experimental results demonstrated that overlap detection can be achieved with 3×16GBd accuracy. Additionally, the estimation of spectral spacing achieved an error (RMSE) of less than 0.6 GHz. The penalty in accuracy and RMSE is only ∼98% and , respectively, when there is no knowledge of the OSNR value. Thus, this method has the potential to be integrated into monitoring tools designed for future dynamic gridless optical transmission systems.
Read full abstract