Novel Decomposition Technique on Rational-Based Neuro-Transfer Function for Modeling of Microwave Components

Area of Science:

• Electromagnetics
• Microwave Engineering
• Computational Modeling

Background:

• Parametric modeling of electromagnetic (EM) behavior in microwave components is crucial.
• Rational-based neuro-transfer function (neuro-TF) methods are popular but suffer from high sensitivity issues with increasing model order.
• High sensitivity leads to training difficulties and reduced accuracy due to small coefficient errors causing large output errors.

Purpose of the Study:

• To address the high-sensitivity issue in high-order neuro-TF models.
• To propose a novel decomposition technique for improving the accuracy and trainability of neuro-TF models.
• To develop new formulations for determining sub-model parameters.

Main Methods:

• Decomposition of a high-order neuro-TF model into multiple lower-order sub-neuro-TF models.
• Reformulation of the overall model as a combination of these sub-models.
• Derivation of new formulations to determine the number of sub-models and their respective transfer function orders.

Main Results:

• The proposed decomposition technique significantly decreases the sensitivity of the model response to transfer function coefficients within each sub-model.
• This reduction in sensitivity leads to improved overall modeling accuracy.
• The method was demonstrated effectively on two electromagnetic (EM) parametric modeling examples.

Conclusions:

• The decomposition technique offers a viable solution to the high-sensitivity problem in neuro-TF modeling.
• This approach enhances the accuracy and robustness of parametric modeling for microwave components.
• The method provides a pathway for more reliable and precise EM behavior modeling.