Robust Fast Electromagnetic Optimization of SIW Filters Using Model-Based Deviation Estimation and Jacobian Matrix Update

In this paper, a simple fast electromagnetic (EM) optimization method for substrate integrated waveguide (SIW) filters is presented. The feature parameter deviations are estimated by direct response approximation based on the feature parameter model with few times quasi-Newton method. The parameter extraction process is more concise than the parameter identification techniques based on the rational polynomial model and can be used for lossless networks as well as lossy networks. The dimension correction is based on the Jacobian matrix which relates the physical dimensions and the feature parameters. An analytical general mapping model for SIW filters is extracted and applied to generate and update the Jacobian matrix for the first time. The extracted model can be directly used for SIW filter design without repeated extraction. This helps to minimize the number of EM simulations and avoid extra work for model extraction. Using general mapping model can overcome the robustness issue introduced by Broyden update methods. The number of EM simulations required to meet the specifications can be significantly reduced. The effectivity and robustness of the proposed method is validated by two design examples: a 10-order SIW Chebyshev filter and a 5-order SIW cross-coupled filter. The proposed method exhibits low complexity, good stability, good convergence speed and the ability to avoid being trapped in local minimum.