Regression machine learning-based highly efficient dual band MIMO antenna design for mm-Wave 5G application and gain prediction

Abstract

With the exponential growth of wireless communication systems, the need for compact, high-performance antennas operating at millimeter-wave (mm-Wave) frequencies has become increasingly critical. This paper presents a comprehensive design and performance analysis of a microstrip patch antenna system operating at dual frequencies of 28 GHz and 38 GHz, suitable for 5G and beyond applications. The antenna evolves from a single element to a 2-element array and a 4-port MIMO configuration, achieving high gains of 9 dB and 8.4 dB, respectively. It covers wide bandwidths of 2.55 GHz and 5.77 GHz within the operating ranges of 26.73–29.28 GHz and 34.96–40.73 GHz. Designed on a Rogers RT5880 substrate, the antenna measures 31.26 mm × 31.26 mm (2.92λ0 × 2.92λ0), offering a compact footprint with excellent performance. The system achieves isolation values greater than 35 dB and 29 dB, extremely low Envelope Correlation Coefficients (ECC) of < 0.0001 and Diversity Gain (DG) of > 0.999, and radiation efficiency exceeding 98% and 99%. A machine learning-based performance prediction framework was employed, where five regression models were evaluated using critical metrics, including variance score, R-squared, mean squared error (MSE), mean absolute error (MAE), and root mean square error (RMSE). Among them, the Extra Trees Regression model demonstrated the highest efficacy, achieving the lowest error rates of 14.04% for MAE, 4.42% for MSE, and 21.03% for RMSE, along with an isolation prediction accuracy of approximately 93%. With its outstanding performance, compact design, and intelligent prediction capabilities, the proposed antenna system is a strong contender for future high-capacity mm-Wave wireless communication networks.