Comparative analysis of ML algorithms for the design of intelligent EDFA

Abstract Design and optimization of erbium-doped fiber amplifiers (EDFAs) is one of the key factors in the improvement of the efficient working of modern optical communication systems. The less adaptable operational parameters assessed by traditional modeling of EDFA characteristics usually utilize analytic or simulations-based methods, which are computationally intensive and do not compute as well as adaptive to a wide range of operational parameters. In this paper, a comparative analysis of several machine learning (ML) regression algorithms is provided to create an intelligent EDFA model to predict the main parameters related to performance of the amplifier, such as signal gain, noise figure, and output signal power with high accuracy. The dataset that is considered as one of the important tasks of this project is produced by controlling the input parameters in the OptiSystem environment, taking into consideration the changes in the input signal power, wavelength, pump power, fiber length, and the pump wavelength. Different regression models like Linear Regression (LR), Random Forest (RF), Support Vector Regression (SVR), Gradient Boosting (GB), and Artificial Neural Networks (ANNs) are trained and tested on the basis of statistical performance measures, including coefficient of determination (R2), root mean square error (RMSE), and mean absolute error (MAE). The findings suggest that ensemble and neural network–based models are superior to linear models to model nonlinear EDFA behavior, which is an effective starting point to intelligent amplifier design and real-time optimization of optical networks.