Dispersion-Oriented Inverse Design of Photonic-Crystal Fiber for Four-Wave Mixing Application

1、The calculation of effective refractive index data for photonic-crystal-fiber with COMSOL using parametric scanning (14,105 data points) (d-diameter of air hole      Λ- distance between the centers of the two air holes)Λ : ranging 2-5μm steps 0.1μmd /Λ: ranging 0.25-0.45 steps 0.05operating wavelength: 0.5-5μm  2、The calculation of effective refractive area data for photonic-crystal-fiber with COMSOL using parametric scanning (2635 data points)Λ : ranging 2-5μm steps 0.1μmd /Λ: ranging 0.25-0.45 steps 0.05operating wavelength: 0.8-1.6μm  3、Table of fitting parameters for the empirical refractive index formula. The expression of empirical formulas itself. (You can easily approximate the refractive index of PCF with given d, d/Λ, λ) This empirical formula is obtained by fitting 14105 simulated point from Section 1.  4、All values of Aeff in the 5 training dataset are generated from 2652 simulated results in Section 2. Data processing methods instruction documents.  5、Training data for DNN (5*801*12621)The five groups of labeled data, corresponding to five pump power configurations (i.e., 100W,1kW,10kW, 100kW, and 1000kW), are prepared for the DNN models. In detail, each group data of power configurations contains 12621 phase-matching curves linked with PCF structural parameters of Λ and d/Λ ratio (define as d/Λ). The parameter Λ varies from 2 to 5μm (a step of 5nm) and the ratio changes from 0.25 to 0.45 (a step of 0.01). Effective index in is represented by empirical formula in section3 .The phase-matched curves are computed via scanning pump wavelength from 0.8 to 1.6μm by an interval of 1nm. The labeled 12621 data are divided into the training, validation, and testing sets in the amounts of 10000, 2000, and 621, respectively. Λ : ranging 2-5μm steps 0.1μmd /Λ: ranging 0.25-0.45 steps 0.05operating wavelength: 0.8-1.6μmTotal: 5*801*12621（21*601）Every groups of training data is a Three-dimensional arrays .mat file. "Reshape" function is helpful for open it.  6、The training process of DNN, expressed by the decline of loss and RMSE7、PCF structure predicted by deep-learning-neural net work. Ptest – untrained test data Pdnn-DNN predicted structure