A Novel Neuro-TF Modeling Technique Incorporating Parametric Sanathanan–Koerner Iteration of Continuous Pole/Residue Extraction

This article proposes a novel neurotransfer function (neuro-TF) modeling technique incorporating parametric Sanathanan–Koerner iteration (S–K) of continuous pole/residue extraction for electromagnetic (EM) parametric modeling of microwave components. For the first time, parametric S–K iteration is applied to extract the coefficients of the transfer function instead of vector fitting for neuro-TF modeling. In the process, S–K iteration is introduced to construct and derive the novel and systematic formulation for continuous pole/residue extractions with respect to geometrical parameter variations. Since the coefficients follow the geometrical basic function within our limits, they can remain smooth when being extracted. Due to the robustness issue for the neuro-TF modeling, those extracted coefficients need to be converted to poles and residues. The continuity of poles/residues in the conversion process is rigorously proven. Compared with the existing quadratic approximation vector fitting method of poles/residues extraction and standard vector fitting method, the parametric S–K iteration can provide better coefficient continuity and smoothness with strong robustness. The developed models using the proposed method can achieve better modeling accuracy in a larger range of geometrical parameters. Three microwave filters are employed as examples to verify the effectiveness of the proposed technique.