Design optimization of erbium-doped fiber amplifiers using artificial neural networks

Abstract

An artificial neural network (ANN)-based model for erbium-doped fiber amplifiers (EDFAs) is developed. For a network with two input neurons and a single hidden layer, the number of neurons in the hidden layers is optimized by monitoring the mean square error (MSE) at the outputs. The model is used to optimize the design of EDFAs for central signal gain, 3-dB bandwidth, and noise factors, taking into account the influence of amplifier constructional parameters in the amplifier performance. For a given pump power and pulse width that correspond to the two input neurons, the model computes the constructional parameters to obtain values of central signal gain, 3-dB bandwidth, and noise figures in optimal range. The effects due to intrapulse stimulated Raman scattering (ISRS) are also considered by varying pulse widths from subpicosecond regime to hundreds of picoseconds. Self frequency shift due to ISRS is found to be prominent in the subpicosecond regime. The proposed model generates results with a MSE of 1×10 -4 , and needs relatively very low computational time compared to other analytical models.