Abstract
This study presents a high-gain, trident-shaped, band-notched, four-element ultra-wideband (UWB) multiple-input multiple-output (MIMO) antenna system, optimized using machine learning (ML) techniques. The single-element UWB antenna operates over a frequency range of 3.5 to 11.2 GHz, achieving a peak gain of 3.5 dBi. A band-notch characteristic is implemented in the 4.7–5.9 GHz range to reject the WLAN band and mitigate interference. The design is then extended into a four-element MIMO configuration, enhancing the peak gain to 5.1 dBi and achieving an impedance bandwidth of 114%, covering the 3.0–11.0 GHz range. The initial inter-element isolation exceeds 17 dB, which is further improved to more than 20 dB by introducing a decoupling structure.To further improve the gain and radiation characteristics, a metasurface comprising a 6 × 6 unit cell array is placed 23 mm above the MIMO structure, resulting in a peak gain of 9.5 dBi at 8.5 GHz. The proposed MIMO antenna system exhibits satisfactory diversity performance, with envelope correlation coefficient (ECC), diversity gain (DG), mean effective gain (MEG), and channel capacity loss (CCL) all falling within their respective acceptable limits. Additionally, various ML models are employed to optimize the reflection coefficient (∣ S 11 ∣) and gain. The predicted results align well with the CST Microwave Studio simulations. Among the models, Gaussian Process Regression (GPR) for ∣ S 11 ∣ and Fine Tree for gain yield the lowest root mean square error (RMSE) of 0.0082 and 0.3863, respectively, along with excellent R 2 scores of 0.9999 and 0.9847, confirming their superior prediction accuracy and generalization capability. Owing to these features, the proposed antenna is highly suitable for UWB band-notched MIMO applications, 5G sub-6 GHz communications, IoT networks, and radar systems.

This publication has 34 references indexed in Scilit: