Wideband dielectric resonator antenna for K-Band/n258 applications with intelligent response filtering

Abstract

This work presents a wideband rectangular dielectric resonator antenna (RDRA) integrated with a machine learning (ML) response filtering framework for K-band and 5 G NR n258 applications. The antenna utilizes an RT/Duroid 5880 substrate, a partial ground plane, a heart-shaped structure integrated with an L-shaped feed, and a top parasitic patch. The fabricated RDRA achieves a wide impedance bandwidth of 29.78% (20–27 GHz) and a peak measured gain of 6.53 dBi. To accelerate antenna design-space screening, a lightweight cascaded ML framework is developed as a predictive digital twin. The first stage acts as a gatekeeper, filtering out non-compliant designs based on a strict S11<−20 dB threshold to provide a manufacturing safety margin. The second stage automatically classifies qualified responses into discrete 1 GHz resonant-frequency intervals. Among the evaluated classifiers (RFC, ANN, and kNN), the Random Forest Classifier achieves mean accuracies of 96% for impedance validation and 93.86% for frequency-band classification. The proposed screening pipeline enables rapid evaluation of candidate antenna geometries, significantly reducing the computational overhead associated with high-volume mmWave antenna manufacturing and automated quality testing.