Machine Learning Optimized Slotted Patch Antenna With Modified Aperture for 5G mm-Wave Applications

Abstract

A compact slotted patch antenna intended for 5G mm-wave applications is presented in this article. The dimension of this antenna is 7 × 7 × 0.8 mm³ and operates at a frequency of 28.5 GHz, and it is made of FR4 epoxy material. The measured and simulated bandwidths are 16.1% and 16.12%, respectively, with corresponding gains of 5.3 dBi. Compactness is achieved by introducing the slots in the radiating patch. The slots effectively modify current distribution and enable the antenna to resonate at the desired frequency while reducing its overall physical size. Incorporating a half-sinusoidal arc with varying slot amplitudes significantly enhances bandwidth, radiation efficiency, impedance matching, and polarization purity, making the antenna highly suitable for 5G mm-wave applications. Five machine learning regression algorithms, such as Gradient Boosting Regressor (GBR), Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosting (XGB), and K-Nearest Neighbor (KNN), are employed to predict and optimize the return-loss (S11) signature of the antenna. A purpose-built dataset of 3864 samples, generated in HFSS by varying the slot length and width, is split into 80% training and 20% testing data. Model accuracy is reported as the coefficient of determination (R²) on the held-out test set, together with MSE, MAE, and MAPE. The Extreme Gradient Boosting Regressor (XGB) provides the best performance, achieving the lowest MSE of 6.2288 and the highest R² of 0.8077 (≈81%) among the five models. This work demonstrates the use of ML to accurately and efficiently optimize a 5G antenna for high-speed connectivity. The proposed antenna is suitable for V2X communication systems, automotive radar, IoT, and 5G mm-wave applications in the n257 and n261 bands.