Prediction of axial ratio using machine learning (ML) for a dual-band circularly polarized dielectric resonator antenna (DRA)

Abstract

In this paper, an asymmetrical dielectric resonator antenna (DRA) is designed to achieve circular polarization at dual frequencies within the C-band (4–8 GHz) and X-band (8–12 GHz) ranges. A regression-based machine learning (ML) technique is employed to predict the antenna’s axial ratio. The DRA structure consists of three rectangular ceramic blocks with a uniform permittivity of 9.8 and is excited by a microstrip feedline. Two dielectric resonators (DRs) of different heights are placed on the same plane, while the third DR is positioned atop the first two with a 60° rotation, enabling a wide 3 dB axial ratio (AR) bandwidth across both bands. The antenna operates from 5.2 GHz to 12.0 GHz, achieving a 10 dB impedance bandwidth of 79%. The AR bandwidth (≤ 3 dB) spans two bands: 6.0–8.7 GHz (lower band) and 9.9–11.8 GHz (upper band). The antenna maintains a gain of over 4 dBi throughout the bandwidth. Additionally, the antenna is analyzed in an array configuration, and beam steering at 30° is demonstrated through simulation. A dataset comprising various antenna dimensions and their corresponding axial ratio values is generated using parametric sweep analysis. A supervised regression-based ML approach is then employed to predict the axial ratio at two circularly polarized frequencies: 7.2 GHz and 10.5 GHz. Several regression algorithms are tested, and the Extra Trees Regression model achieves the lowest prediction error and highest accuracy.