Optical Nonlinearity Monitoring and Launch Power Optimization by Artificial Neural Networks

Abstract

We present a linear-to-nonlinear power ratio monitor based on a shallow artificial neural network and the optical power spectrum. The neural network is trained with experimental pairs of input single-channel optical power spectra and output optimal power corrections, i.e., power amendments that lead to the power level maximizing the performance in terms of the signal-to-noise ratio. The technique is tested and shows the capability of providing up to 1 dB of signal-to-noise ratio gain in the ±3 dB region around the actual optimal power. Furthermore, the neural network does not recommend power variations resulting in a severe signal-to-noise ratio penalty (max −0.12 dB). Here, we extend our previous conference contribution by providing further insight into the theoretical background and some additional technical results in the direction of proving the connection between the optical power spectrum and the optimal power correction, i.e., the linear-to-nonlinear power ratio.