Radiation Pattern Decoupling of 1×4 DRAs Using Machine Learning

This article presents a machine learning assisted decoupling method to decouple 1×4 dielectric resonator antennas with uniform radiation patterns, namely radiation pattern decoupling (RPD). Metallic vias and metal strips are used as decoupling structures. Gaussian process model and parallel Bayesian optimization method is used for optimization. A new cost function is designed to evaluate the effect of radiation pattern similarity. To verify the design idea, a prototype was fabricated and measured. The measured overlapping 10-dB impedance bandwidth is 6.4%. All radiation patterns can be found to radiate along the zenith directions, which verifies the RPD effects.