MACHINE LEARNING BASED DESIGN AND INVERSE OPTIMIZATION OF PATCH ANTENNAS FOR 5G APPLICATIONS

Abstract

In this work, a machine learning (ML) assisted approach is proposed for the design and optimization of compact inset-fed microstrip patch antenna (MPA) intended for 28 GHz 5G applications. First, the antenna design and simulation are carried out using HFSS for symmetrical semicircular notches to obtain better impedance matching over 24-31GHz. Then, HFSS generated dataset are obtained for varying dimensions of patch width, substrate height, semicircle radius and inset feed depth. These datasets are used on four different machine learning (ML) models for prediction of antenna performance indices such as (a) return loss, (b) resonant frequency and (c) gain. However, the best estimations with regard to accuracy has been obtained from the (i) Extra Trees Regressor model for (a) and (ii) Random Forest Regressor model for both (b) and (c). The optimized design has resulted        -49.08 dB return loss at 27.97 GHz, 7.16 dBi gain and 4.7 GHz bandwidth which implies a good agreement with the HFSS simulations. Moreover, it is seen that the ML integrated electromagnetic simulation significantly accelerates the antenna design by preserving the predictive accuracy.