Machine Learning-Driven Inspired MTM and Parasitic Ring Optimization for Enhanced Isolation and Gain in 26 GHz MIMO Antenna Arrays

Abstract

This paper presents an intelligent design framework for a high-performance 26 GHz MIMO antenna array tailored to 5G applications, built upon a compact single-element patch. The 11.5 mm × 11.5 mm × 1.6 mm microstrip patch on FR4 exhibits near-unity electrical length, an ultra-deep return loss (S11 < −40 dB at 26 GHz), and a wide operational bandwidth from 24.4 to 31.2 GHz (6.8 GHz, ~26.2%). A two-element array, spaced at λ/2, is first augmented with a inspired metamaterial (MTM) unit cell whose dimensions are optimized via a Multi-Layer Perceptron (MLP) model to maximize gain (+2 dB) while preserving S11. In the second phase, a closed-square parasitic ring is introduced between the elements; its side length, thickness, and position are predicted by a Random Forest (RF) model with Bayesian optimization to minimize mutual coupling (S12) from −25 dB to −58 dB at 26 GHz without significantly degrading S11 (remains below −25 dB). Full-wave simulations and anechoic chamber measurements confirm the ML predictions. The close agreement among predicted, simulated, and measured S-parameters validates the efficacy of the proposed AI-assisted optimization methodology, offering a rapid and reliable route to next-generation millimeter-wave MIMO antenna systems.