Fiber nonlinearity mitigation with a perturbation based Siamese neural network receiver

We simulate nonlinear distortion compensation of single frequency dual polarization signals propagating in single mode optical fibers with a combination of Siamese Neural Networks (SNNs) and a perturbative nonlinearity compensation technique. We find that a 2-branched SNN can enhance the quality factor (Q-factor) of a 3200 km 16-QAM optical system from the 8 dB that is associated with only applying chromatic dispersion compensation (CDC) to 8.9 dB. We then investigate the Q-factor improvement associated with different numbers and widths of SNN branches. We finally demonstrate that the number of inputs to the SNN and hence the computational complexity can be reduced by employing Principal Component Analysis (PCA). This results in a 0.75 dB Q-factor enhancement with 50% fewer inputs than previous designs.

Automated summary

This study demonstrates the improvement of quality factor (Q-factor) in single frequency dual polarization signals propagating in optical fibers by combining Siamese Neural Networks (SNNs) and perturbative nonlinearity compensation technique. The research shows that optimizing the number and width of SNN branches, along with implementing Principal Component Analysis (PCA), can significantly enhance the Q-factor while reducing computational complexity.