Performance evaluation of a machine learning-driven dual broadband antenna with beam steering for 5G sub-6 GHz applications

Abstract

This research evaluates the performance of a compact, lightweight, dual-band switched-beam antenna designed for 5G sub-6 GHz communications, operating at 0.7 GHz and 2.6 GHz. The antenna structure integrates grooves and parasitic features to improve bandwidth performance. Three machine learning models, decision tree, random forest, and XGBoost, are utilized to optimize seven critical dimensional parameters for the achievement of broadband operation. The model is trained using 244 samples collected from 305 simulated datasets designed by CST Microwave Studio to replicate an extensive selection of conditions, targeting resonance frequencies and corresponding bandwidths. Beam steering is achieved with the incorporation of a short-circuit mechanism, allowing the antenna to guide its main beam in two directions across both frequency bands. The proposed design attains bandwidths of 21.65 MHz at 0.63 GHz and 192.89 MHz at 2.48 GHz, representing enhancements of 11.52 MHz (113.72%) and 128.72 MHz (200.59%), respectively, in comparison with the conventional reference antenna. The design was evaluated using simulation and empirical measurement, demonstrating significant concordance and affirming the antenna’s efficacy in beam switching and wideband dual-band functionality. Its simple structure, compact dimensions, and machine learning performance provide a promising alternative for next-generation wireless systems.