TR2018-047

Fiber Nonlinearity Equalization with Multi-Label Deep Learning Scalable to High-Order DP-QAM



We use deep neural network (DNN) to compensate for Kerr-induced nonlinearity in fiber-optic communications. The proposed DNN is scalable to high-order modulations by employing multi-label classification, achieving greater than 1.2 dB gain in nonlinear regimes