Machine learning‐based approach for bandwidth and frequency prediction of circular SIW antenna
Machine Learning (ML) has significantly transformed antenna design by enabling efficient optimization of geometrical parameters, modeling complex electromagnetic behavior, and accelerating performance prediction with reduced compu tational cost. This study presents an ML-based approach to accurately...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | en |
| Published: |
King Saud University
2025
|
| Subjects: | |
| Online Access: | https://umpir.ump.edu.my/id/eprint/46318/1/Machine%20learning%E2%80%90based%20approach%20for%20bandwidth%20and%20frequency.pdf https://doi.org/10.1007/s44444-025-00010-0 https://umpir.ump.edu.my/id/eprint/46318/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Machine Learning (ML) has significantly transformed antenna design by enabling efficient optimization of geometrical parameters, modeling complex electromagnetic behavior, and accelerating performance prediction with reduced compu tational cost. This study presents an ML-based approach to accurately predict the resonance frequencies and bandwidth of a novel triple-band circular Substrate Integrated Waveguide (SIW) antenna intended for K- and Ka-band satellite com munication. The antenna features four symmetrically etched ring slots on the radiating patch and circularly arranged vias within a Rogers RT/Duroid 5880 substrate (20 × 15 × 1.6 mm3). The interaction between the TE₁₁ cavity mode and the ring slots facilitates controlled electromagnetic field leakage, enhancing radiation performance. A predictive ML framework was developed using six regression algorithms trained on significant geometrical parameters, such as ring slot radius, via diameter, and feedline width. Among them, the Extra Trees Regression model achieved over 98% accuracy, with errors below 0.1% for both resonance frequency and bandwidth predictions. The approach was validated through Computer Simulation Technology (CST) simulations and ML-based predictions, both demonstrating strong agreement. Although experimental fabrication was not included in this phase, the model offers a reliable foundation for future physical validation and prototyping. The results confirm that the proposed antenna structure, combined with the predictive power of ML, presents a promising solution for compact, high-performance antennas in satellite communication systems. |
|---|