Circular MIMO antenna with ML-based bandwidth and isolation prediction for 6G communications

The demanding requirements of next-generation 6G wireless systems necessitate the development of compact, wideband, and high-isolation terahertz (THz) MIMO antennas, while conventional full-wave electromagnetic optimization remains computationally expensive for complex multi-parameter designs. To overcome these challenges, this work introduces a compact circular MIMO antenna integrated with a machine learning (ML)-based framework for efficient performance prediction and design optimization. The proposed antenna consists of two co-oriented circular radiating elements, enhanced with a concentric ring and side stubs to improve impedance matching and broaden bandwidth. High inter-element isolation is achieved through the incorporation of isolation walls and an optimized partial ground structure. The polyimide-based antenna, with compact dimensions of 130 × 70 μm², operates at 5.55 THz and provides a wide bandwidth of 4.56-5.86 THz, a peak gain of 8.04 dB, a radiation efficiency of 85.64%, a diversity gain (DG) of 9.985 dB, and a total active reflection coefficient (TARC) below - 35 dB. To enable rapid performance estimation, an ML-based predictive model employing five supervised regression algorithms is trained using crucial geometrical parameters, including inner ring radius, feedline width, stub width, element spacing, and substrate height. Among the evaluated models, Cat Boost regression achieves the highest prediction accuracy, with R² scores of 97.05% for bandwidth and 92.56% for isolation. These results demonstrate that the proposed circular MIMO antenna, supported by ML-based predictive modeling, offers a promising solution for compact, high-performance antennas in 6G THz communication systems.