Modeling the Optical Gain of Erbium-Doped Fiber Amplifiers in Strong Cross-Gain Modulation Regime Employing Artificial Neural Networks

Erbium-doped fiber amplifiers (EDFAs) represent a key enabling component in many modern optical communication systems. Their accurate modeling is, therefore, essential not only to aid in their design but also to appropriately dimension the optical system. The modeling of EDFAs is particularly challenging in wavelength division multiplexing (WDM) systems operating in the high cross-gain modulation regime since the gain experienced by a certain channel depends on the power of the other channels. This effect, denominated cross-gain modulation, is usually simulated using the Giles Desurvire model, which requires the integration of a system of coupled partial equations and the characterization of multiple physical parameters. In order to reduce the computational cost and enable fast computing, in this paper, we propose, optimize, and analyze an alternative model based on a simple artificial neural network (ANN). Simulation results considering a 4-channel WDM system reveal that a single-layer multi-layer perceptron with a hyperbolic tangent activation function and 80 neurons can predict the output power with an error between 0.08 and 0.11 dB.