Evaluation of Frequency-Domain Learned Digital Back-Propagation Nonlinear Compensation for Unrepeatered Optical Links

We have implemented a technique for nonlinear compensation in optical transmission based on neural network optimization applied to digital back-propagation and evaluated its performance with experimental data from an unrepeatered link, sweeping the parameters most relevant to computational complexity. This technique enabled mutual information gains over 0.1 bit/symbol in all tested scenarios when compared with the non-optimized counterpart, or 0.15 bit/symbol when compared with similar complexity linear compensation.