Systematic pole-zero sorting method for neuro-TF modeling of electromagnetic response.

Area of Science:

• Electromagnetics
• Computational Methods
• Signal Processing

Background:

• Neuro-transfer functions (neuro-TF) are popular for parametric modeling of electromagnetic (EM) filter responses.
• Vector fitting methods for extracting pole-zero data can lead to discontinuities with geometric parameter changes, hindering neuro-TF accuracy.
• Existing neuro-TF methods struggle with the disorder of positive and negative values due to small data points.

Purpose of the Study:

• To propose a novel systematic pole-zero sorting method for neuro-TF parametric modeling.
• To address the discontinuity issue in pole-zero data extraction for improved neuro-TF accuracy.
• To enhance the modeling accuracy and stability of neuro-TF models in electromagnetic filter design.

Main Methods:

• Developed a systematic pole-zero sorting algorithm specifically for neuro-TF parametric modeling.
• Ensured continuous and smoother pole-zero data transitions with respect to geometrical parameter variations.
• Implemented the sorting method to manage and correct disorders in positive and negative values within the data.

Main Results:

• The proposed method yields continuous pole-zero data that exhibits smoother changes relative to geometrical parameters.
• Significantly improved modeling accuracy compared to existing neuro-TF methods lacking systematic sorting.
• Successfully resolved issues related to the disorder of positive and negative values in extracted data.

Conclusions:

• The novel systematic pole-zero sorting method substantially enhances the accuracy of neuro-TF parametric modeling for EM filters.
• This approach overcomes limitations of existing methods by providing continuous and stable pole-zero data.
• The proposed technique offers a more robust and reliable solution for establishing and training neuro-TF models.