Parametric Modeling of a 5-order Waveguide Bandpass Filter Based on Neuro-TF Approach

This paper presents an advanced parametric mod-eling technique based on the Neuro-Tf approach and establishes a knowledge-based neural network (KBNN) model for a 5-order waveguide filter. The model takes the S-parameter of the electromagnetic (EM) simulation as the output and the geometric parameters of the filter as the input. By using the poles and residues of the transfer function as a knowledge rule, we accelerate the solving speed of nonlinear problems. This method performs better generalization than the traditional multi-layer perceptron (MLP) network model.