Compressed Machine Learning-Based Inverse Model for the Design of Microwave Filters

This paper presents an inverse model for the optimization of the geometrical parameters of a parallel coupled-line pass-band filter. Given the overall structure of the filter, the least square support vector machine is combined with the principal component analysis with the aim of constructing an inverse model able to estimate the geometrical parameters of the filter starting from a frequency-domain mask. Such model is trained via a set of scattering parameters computed via a 2D solver for few configurations of the filter geometrical parameters. The feasibility and the accuracy of the proposed optimization scheme is investigated by comparing its predictions with the corresponding optimal configuration estimated via a commercial tool.