Automated inverse design of bandpass filters with invariable layout through linear approximation of physical dimensions

In this paper, an efficient methodology for finding initial designs of microwave bandpass filters directly from specification (center frequency f and bandwidth BW) is presented. The proposed technique exploits an inverse linear surrogate model of the designable parameters with respect to the filter specifications, created using an available accurate design. Fast calculation of the model gradients is possible through efficient optimization in the feature space using a coarse discretization model. As a result, the proposed methodology allows obtaining an accurate approximation of the initial filter dimensions at the cost of a few fine model simulations. The approach is illustrated with a 4th-order substrate-integrated waveguide (SIW) filter design example. A study of the relative model error within the design space is included.