Data-Selective Learning Algorithm Using Resonance Parameters Based on Stacked Data Augmentation for Wideband Impedance Prediction of Printed Spiral Coils
In recent years, various fields have conducted extensive research on neural network learning to address the growing demand for miniaturization and multi-functionalization of wireless devices. In this paper, we propose a data-selective learning algorithm that uses resonance parameters based on stacke...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
The Korean Institute of Electromagnetic Engineering and Science
2025-03-01
|
| Series: | Journal of Electromagnetic Engineering and Science |
| Subjects: | |
| Online Access: | https://www.jees.kr/upload/pdf/jees-2025-3-r-261.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In recent years, various fields have conducted extensive research on neural network learning to address the growing demand for miniaturization and multi-functionalization of wireless devices. In this paper, we propose a data-selective learning algorithm that uses resonance parameters based on stacked data augmentation to predict the wideband impedance characteristics of printed spiral coil (PSC) structures, which are widely used as radio-frequency interference measurement probes. The proposed model utilizes a multilayer perceptron (MLP) neural network to predict the impedance of PSCs. The training data used in this study comprised 604 PSC design structures, with the self-impedance of the PSC corresponding to 600 frequencies. To achieve efficient data learning for wideband impedance prediction, a data selection algorithm that uses the difference between the resonance parameters of the predicted and target impedances in the high frequency range is proposed. To further enhance learning efficiency and improve model stability, we introduced a novel method that combines data selection and stacked data augmentation. The model with the proposed data selection and augmentation algorithm demonstrated efficient learning and accurate impedance prediction using approximately 54.4% less training data than a conventional MLP neural network model. Furthermore, the proposed model was validated through electromagnetic field simulation, showing an accuracy of up to 6 GHz. |
|---|---|
| ISSN: | 2671-7255 2671-7263 |