Resonance frequency prediction of dielectric antennas for liquid sensing via support vector regression

Abstract

This paper presents a slot antenna integrated with a split ring resonator (SRR) and feed line, designed to achieve a high Q-factor while maximizing channel capacity utilization. By incorporating a lens into the dielectric resonator antenna (DRA), we enhance both bandwidth and directivity, with the dielectric material’s permittivity serving as a key control parameter for radiation characteristics. We explore water and ethanol as controllable dielectrics within the terahertz (THz) frequency range (0.5-1 THz), implementing these liquids through microfluidic techniques. This novel design serves two purposes, functioning as both an antenna system and a highly sensitive material sensing device. The antenna’s performance is evaluated using the Debye model for pure water and ethanol, with electromagnetic full-wave simulations employing the Finite Integration Technique (FIT) to model both the antenna and microchannel structures. For predicting resonant frequencies based on antenna dimensions, we implement a Support Vector Regression (SVR) algorithm, comparing its performance against various models including Linear Regression, Regression Trees, Ensemble Bagged Trees, Ensemble Boosted Trees, and Three-layered Neural Network Models. The SVR demonstrates superior prediction accuracy by effectively capturing non-linear relationships between antenna dimensions and resonant frequencies.