Forward and inverse modeling of multiple input multiple output microstrip antennas for satellite communication using machine learning algorithms

Abstract

This article presents a supervised machine learning (SML) structure as a forward and inverse model for electromagnetic problems. For this purpose, we proposed ridge, lasso, elastic net, and K-nearest neighbor’s regression for both forward and inverse modeling tasks of duel port and four-port multiple-input multiple-output microstrip antennas for satellite communication. The duel-port and four-port multi input multi outputs microstrip antennas consist of three and two notched bands, respectively. In addition, The dual-port and four-port antennas achieved maximum notched frequencies at 6 and 8.2 GHz, respectively. We compare the results predicted by various ML techniques with those obtained from a high-frequency structure simulator(HFSS) to validate their accuracy. SML approaches are utilized to predict the mean square error (R_MSE) and error of coefficient (R_R) for both the physical and frequency parameters of dual-port and four-port microstrip antennas. However, the results obtained by K-nearest neighbor’s regression algorithms are significantly better than those from ridge regression, lasso regression, and elastic net regression.