Using AI to Improve MIMO Antennas with SRR for 26 GHz by Analyzing Data

Abstract

This paper introduces a database-based design methodology aimed at optimizing a 26 GHz MIMO antenna system through machine learning (ML) techniques. The procedure is divided into two primary phases. Initially, a rectangular microstrip patch antenna is designed and enhanced using analytical models alongside ML algorithms that are trained on a detailed dataset of geometric parameters. This yields effective impedance matching (S11 < −45 dB) and a high gain (~6.64 dBi), which serve as the foundation for the MIMO structure. In the second phase, split ring resonator (SRR) configurations are integrated between the antenna elements to reduce mutual coupling. A specialized dataset, featuring varied dimensions of SRR, quantities of unit cells, and spatial placements, is utilized to train Random Forest models that forecast arrangements achieving optimal isolation (S21 < −40 dB) while maintaining low reflection losses. Additionally, a secondary dataset is constructed to investigate the best strategies for SRR placement, ensuring an optimal balance between isolation and return loss. The ultimate MIMO design is validated via comprehensive full-wave electromagnetic simulations and experimental measurements. The proposed system exhibits noteworthy performance enhancements, including an envelope correlation coefficient (ECC) < 0.005, diversity gain (DG) ≈ 9.99 dB, channel capacity loss (CCL) < 0.3 bits/s/Hz, total active reflection coefficient (TARC) < −30 dB, radiation efficiency exceeding 80%, and a maximum gain increase up to 10.22 dB. The close correlation between predicted and measured outcomes validates the effectiveness of the ML-driven approach in expediting antenna optimization for 5G and future applications.