Flexible Raman Amplifier Optimization Based on Machine Learning-Aided Physical Stimulated Raman Scattering Model

The problem of Raman amplifier optimization is studied. A differentiable interpolation function obtained for the gain coefficient using machine learning (ML), which allows gradient descent forward-propagating pumps. Both frequency and power an arbitrary number pumps in a forward pumping configuration are then optimized data channel load span length. propagation model combined with experimentally-trained ML backward-pumping to jointly optimize amplifier's powers backward joint demonstrated unrepeatered transmission 250 km. flatness $< $ 1 dB over 4 THz achieved. amplifiers validated numerical simulator.