Regression machine learning methods for isolation prediction and massive gain broadband MIMO antenna design for 28 GHz applications

This research paper investigates the design and analysis of a miniaturized mm-Wave MIMO antenna array tailored for fifth-generation applications. The antenna demonstrates a calculated 10-dB impedance bandwidth of 16.61 % (27.137–31.788 GHz). To enhance performance, a combination of pi and L-shaped slots is employed. Constructed from low-loss dielectric material, specifically Rogers RT Druid 5880, the antenna features a dielectric constant of 2.2 and a tangent loss of 0.0009, with an ultrathin height of just 0.8 mm. The dimensions of both the substrate and ground for a single element are 0.653λ₀ × 0.653λ₀ mm, while the overall MIMO antenna design measures 2.8λ₀ × 2.8λ₀, targeting the lowest frequency. In addition to its compact dimensions, the proposed design achieves a maximum gain of 9.129 dB, isolation greater than 26 dB, and an efficiency rating of 82.95 % at its optimal configuration. The Envelope Correlation Coefficient (ECC) is below 0.0012, and the Diversity Gain (DG) exceeds 9.99 dB. Various metrics are available to evaluate the performance of Machine Learning (ML) models, including variance score, R-squared, Mean Square Error (MSE), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). Among the five ML models assessed, Gaussian Process Regression (GPR) showcases the highest accuracy, exhibiting the lowest prediction error in isolation assessments. The results obtained from CST and ADS modeling, alongside actual and expected outcomes from machine learning, indicate that the proposed antenna is a strong candidate for 5G applications.