Machine Learning-Driven Design and Experimental Validation of a Highly Miniaturized Dual-Band MIMO Antenna for Sub-6...

Area of Science:

• Electrical Engineering
• Antenna Theory and Design
• Computational Electromagnetics

Background:

• The proliferation of 5G and Internet of Things (IoT) networks necessitates compact Multiple-Input Multiple-Output (MIMO) antennas.
• Achieving high inter-port isolation in miniaturized MIMO antennas presents significant computational challenges for traditional optimization methods.

Purpose of the Study:

• To propose and validate a novel Deep Surrogate Active Learning framework for the autonomous design of ultra-compact dual-band MIMO antennas.
• To overcome computational bottlenecks in designing compact MIMO antennas with stringent isolation requirements.

Main Methods:

• A Deep Surrogate Active Learning framework combining a custom-penalized Deep Neural Network with dynamic boundary reduction was developed.
• A surrogate-assisted closed-loop strategy was employed to minimize reliance on computationally expensive full-wave simulations.
• An initial surrogate model was trained using 440 full-wave responses from a design-of-experiments (DOE) stage.

Main Results:

• An optimized nested-loop geometry with a partial defected ground structure (DGS) was identified, occupying a footprint of 1634 mm².
• The antenna achieved simulated dual-band operation (3.35-3.88 GHz and 4.34-5.05 GHz) with inter-port isolation exceeding 13.8 dB and 14.9 dB.
• Measurements confirmed dual-band behavior, a maximum gain of 4.54 dBi, and radiation efficiencies of 51-63%.

Conclusions:

• The proposed Deep Surrogate Active Learning framework effectively enables the autonomous design of ultra-compact dual-band MIMO antennas.
• The optimized antenna geometry demonstrates suitability for sub-6 GHz 5G and IoT applications requiring small form factors and high isolation.
• The methodology significantly reduces computational burden compared to conventional optimization techniques.