Adaptive 3D Deep Learning for Multi-carrier Coherent Optical Communications

摘要

Fiber-induced nonlinearities significantly limit the transmission performance of coherent optical signals. Here, a novel adaptive 3D deep learning nonlinear equalizer based on an artificial neural network is experimentally demonstrated for multi-channel coherent optical orthogonal frequency division multiplexing. It is shown that adaptive 3D deep learning outperforms 2D machine learning and the deterministic gold-standard digital back-propagation at 3200 km of single-mode fibre transmission. This occurs since our technique can tackle both deterministic and stochastic nonlinear distortions.