Fig. 5
From: Field and lab experimental demonstration of nonlinear impairment compensation using neural networks
The impact of activation function on the NN-NLC algorithm. a Plots of different activation functions. b The BER optimization trace of DNN over the CV dataset in the training with different activation functions. c The Q-factor of DNN with different activation functions as a function of channel power. d The Q-factor improvement of PPD and NN-NLC with Leaky ReLU w.r.t. the optimum CDC Q
