Low-cost feature-based modeling of microwave structures

A simple technique for accurate and low-cost modeling of microwave structures is presented. Our approach utilizes response surface approximation at the level of suitably selected response features rather than direct approximation of frequency responses (e.g., S-parameters) of the structure of interest. By exploiting the fact that dependence of the features on designable parameters is much less nonlinear than similar dependence for original responses, accurate quasi-global surrogates can be constructed at a fraction of computational cost necessary by conventional modeling techniques. Numerical verification involving two microstrip filters is provided.