Related Experiment Videos
Physics-informed-GPR data augmentation framework for flower-shaped antenna performance optimization and prediction
Lijaddis Getnet Ayalew1, Tsion Yigzaw Kumlachew2, Demessu Kebede Chaka2
1Faculty of Electrical and Computer Engineering, Bahir Dar Institute of Technology, Bahir Dar University, P.O. Box 26, Bahir Dar, Ethiopia. lijaddisg@gmail.com.
Scientific Reports
|July 12, 2026
Summary
This study introduces a physics-informed Gaussian process regression (PI-GPR) framework to enhance millimeter-wave (mmWave) antenna design. The method efficiently generates data for designing compact multiple-input multiple-output (MIMO) antennas for Internet of Things (IoT) applications.
Area of Science:
- Electrical Engineering
- Electromagnetics
- Antenna Theory
Background:
- Designing wideband millimeter-wave (mmWave) antennas and compact multiple-input multiple-output (MIMO) configurations for Internet of Things (IoT) applications is challenging due to the need for extensive simulations.
- Existing methods often require computationally expensive full-wave electromagnetic simulations, limiting design efficiency and scalability.
Purpose of the Study:
- To develop a simulation-efficient data augmentation framework for designing and optimizing mmWave antennas and MIMO configurations.
- To integrate physics-informed Gaussian process regression (PI-GPR) with machine learning models for accurate performance prediction.
- To enable the design of compact, high-performance MIMO antennas for interference-prone IoT environments.
Main Methods:
- A physics-informed Gaussian process regression (PI-GPR) framework was employed for data augmentation, expanding an initial dataset to 2,014 physics-consistent samples.
- The framework integrated closed-form analytical modeling, electromagnetic constraints, and data-driven GPR to avoid exhaustive full-wave simulations.
- Five predictive machine learning models (RSM, ANN, RR, RF, GB) were developed and cross-validated, with Random Forest (RF) showing superior performance.
Main Results:
- The Random Forest model achieved high accuracy in predicting resonant frequency (R-squared: [Formula: see text], RMSE: [Formula: see text]) and impedance bandwidth (R-squared: 0.028, MSE: 0.001).
- An optimized single-element antenna achieved a gain of 3.0 dBi at 28 GHz with a wide impedance bandwidth (22-41.378 GHz).
- A compact quad-port MIMO antenna exhibited a bandwidth of 19.833 GHz, peak gain of 6.22 dBi, low envelope correlation coefficient (ECC < 0.0035), and high diversity gain (DG > 9.983 dB).
Conclusions:
- The proposed PI-GPR data augmentation framework offers a scalable and simulation-efficient approach for mmWave MIMO antenna design.
- The methodology effectively reduces reliance on computationally intensive simulations, accelerating the design cycle.
- The developed framework and optimized antenna designs are suitable for advanced IoT applications requiring robust wireless communication.