Order Reduction Using Laguerre-FDTD with Embedded Neural Network

In this work, a neural network embedded Laguerre-FDTD method is proposed to greatly reduce the simulation order needed to recover the field waveform. By taking the lower order basis function coefficients as the training data, numerical results show up to half of the simulation order coefficients can be predicted by the neural network. Both time and frequency-domain (S-parameter) results converted from predicted basis function value show excellent accuracy compared with ground truth. As a result, the proposed method provides significant improvement on computational speed due to reduced number of sparse matrix solving operations.