Predication of negative dispersion for photonic crystal fiber using extreme learning machine

The photonic crystal fiber (PCF) is a resourceful optical device that can be used in various applications. The dispersion is a major impediment for such optical waveguides. We propose a modified PCF that evaluates the negative dispersion coefficient−3126 ps/(nm–km) at 1.55 μm wavelength. The precise value calculation of the design parameters is helpful to improve the desired output. The machine learning approaches are now more in fashion to predicate such parameters. The dispersion parameters are obtained for three different PCF models as conventional PCF with fixed radius air holes and type 1 and type 2 models with dual radius air holes. Further, the negative dispersion of a type-I PCF is modeled using an extreme learning machine (ELM) as a regression task and its performance is tested against benchmark models such as support vector machine with linear and radial basis function kernel function, Gaussian process regression, and artificial neural network. Results indicate superior performance of ELM as a regressor both, in terms of prediction and computation time.