Estimation of Resonant Frequency, Bandwidth, Input Impedance and Gain using Deep Learning Model for Wearable Patch Antennas

Wearable microstrip patch antennas play a vital role in non-invasive monitoring of medical vitals and also in body-worn wireless applications. However, designing and predicting performance parameters of body-worn antennas pose significant challenges, especially when dealing with lossy materials such as fabrics and denim. To address these challenges, this study proposes a deep learning model to simplify the estimation of the post-fabrication antenna performance parameters. Initial design of the microstrip patch antenna is based on analysis and electromagnetic (EM) simulation, followed by data training for the deep learning model, which is structured with three input parameters: substrate height (h), relative permittivity (εr) and loss tangent (tan δ) of the wearable dielectric substrate. The model outputs four parameters: resonant frequency (fr), impedance (real, imaginary, and total impedance), bandwidth (BW) and gain (G). Training of the artificial neural network based deep learning model has been conducted using the Levenberg-Marquardt back propagation algorithm, which can accurately predict the wearable antenna performance parameters. In this proposed work, a rectangular inset fed microstrip patch antenna is studied-but, the employed ANN model can be trained for any arbitrary shaped antenna structure like square, triangular, circular etc.