Machine learning assisted frequency reconfigurable concentric split ring resonator inspired antenna for sub-6 GHz 5G and IEEE 802.11ba/Be applications

This work proposed a wideband concentric split ring resonator frequency reconfigurable antenna using the machine learning approach for 5G (sub-6 GHz) and IEEE 802.11ba/Be applications. The concentric split ring resonator and defected ground structures are employed to achieve wide bandwidth. The proposed antenna covers 1–1.2 GHz, 1.2–1.5 GHz, 1.5–1.7 GHz, 3.8–5.2 GHz and 4–5.3 GHz frequency bands. The proposed reconfigurable antenna offers excellent tuning range (99.18%) and total spectrum (99.80%) by using two Positive-Intrinsic-Negative diode (PIN) diodes (with a maximum gain of 6.2 dBi and maximum radiation efficiency of 96%). Six regression ML algorithms such as K-nearest Neighbour (KNN), Decision Tree (DT), Random Forest (RF), Extreme Gradient (XG) Boost, Support Vector Machine (SVM) and Artificial Neural Network (ANN) are employed to optimize the antenna design. Among all ML algorithms, Random Forest (RF) provides the highest accuracy i.e. 99.30% with maximum $R^2$ score and minimum mean square error (MSE) and execution time for all switches configurations. Additionally, 10-fold cross-validation, paired t-test techniques and SHAP analysis are employed to ensure the accuracy of RF model. The antenna characteristics have been investigated using the Ansys HFSS simulator, and it is compared with the experimental results, which found to be in good agreement. These results indicates that the proposed antenna is best suitable for single-band and double-band sub-6 GHz applications such as Global Positioning System (GPS), mobile phones, Wi-Fi, WLAN, WiMAX, 5G and IEEE 802.11 ba/Be communication applications.