AUT Journal of Electrical Engineering

AUT Journal of Electrical Engineering

Machine-Learning-Assisted Inverse Design of an Artificial-Magnetic-Conductor-Backed Antenna for Wearable and Internet-of-Things Applications

Document Type : Research Article

Authors
Department of Electronics and Communication Engineering, NIT Jamshedpur, Jamshedpur 831014, Jharkhand, India
10.22060/eej.2026.26568.6160
Abstract
The relationship between the geometric parameters and electromagnetic responses of artificial magnetic conductor (AMC) backed wearable antennas is very nonlinear in nature. However, the conventional full-wave optimization requires a long computation time particularly for the simultaneous optimization of the antenna parameters and AMC parameters. A data-driven machine-learning approach for an AMC-backed wearable antenna is proposed along with graphical user interfaces (GUIs) for an automatic processing of the electromagnetic data and very fast geometry prediction.First, a small size meander-line monopole antenna and a periodic AMC unit cell are designed and analyzed based on the electromagnetic theory and CST Microwave Studio simulation. All 654 antenna configurations are generated by the parametric sweeps in a frequency band of 2.0-6.0 GHz. The eight supervised-learning models are comparatively evaluated by using mean absolute error (MAE), root mean square error (RMSE) and coefficient of determination (R²); where the support vector regressor (SVR) achieved the best results with a MAE of 0.00519 GHz and R² of 0.9998 in the prediction of resonance frequency and minimum S₁₁ of 0.248 dB and 0.9441, respectively.The proposed framework performs an inverse design by finding the geometry of the antenna that can achieve the desired resonance frequency and matching condition, and an inverse multilayer perceptron is used to predict the AMC-unit-cell geometries with desired 0° reflection-phase crossing, which is then verified in CST. The coupled antenna–AMC configuration, a gain improvement from 1.16 to 4.46 dBi and a decrease in peak 1-g SAR from 34.3 to 1.23 W/kg are obtained.
Keywords
Subjects


Articles in Press, Accepted Manuscript
Available Online from 21 November 2026