Ceramic-Based Millimeter-Wave MIMO Antenna for 5G NR Band 261: A Machine Learning Optimization Approach

Abstract

This study introduces a novel four-port multiple input multiple output (MIMO) antenna system that employs cylindrical dielectric resonator antennas (CDRAs) for millimeter-wave (mm-wave) communication. The design addresses key challenges in mm-wave MIMO antenna development, including achieving high inter-port isolation, compact integration, and pattern diversity within limited bandwidth constraints. The proposed structure incorporates four CDRAs mounted on a single RT/Duroid 5880 substrate with a shared ground plane. Enhanced isolation is achieved through a plus-shaped aperture coupling, orthogonal feeding configuration, and strategic ground plane modifications. The antenna exhibits an impedance bandwidth (S₁₁ ≤ −10 dB) from 26.60 GHz to 28.73 GHz, with a high isolation of 34 dB at the resonant frequency of 27.54 GHz. Pattern diversity is realized by placing two CDRAs on the upper side and two on the lower side of the substrate, enhancing MIMO performance. To optimize the reflection coefficient (S₁₁), machine learning (ML) algorithms—k-nearest neighbors (KNN), random forest (RF), decision tree (DT), and extreme gradient boosting (XGBoost)—were applied to datasets obtained from parametric simulations in HFSS. The strong correlation between ML predictions and simulation results highlights the effectiveness of ML in guiding antenna design. This work offers advancements in isolation, pattern diversity, bandwidth, and fabrication simplicity, making it a strong candidate for next-generation 5G mm-wave communication systems in the NR 261 band. Experimental validation confirms its practical viability for advanced telecommunication applications aligned with current trends in mm-wave and MIMO technologies.