Design a Slotted Patch Antenna by Using Deep Learning Algorithms for Mobile Communication

Abstract

—A structured machine-learning-assisted methodology is proposed for designing and predicting slotted microstrip patch antenna performance. The antenna targets mobile communication applications in the S-and lower C-bands. A rectangular FR-4 patch is loaded with a single parametric rectangular slot. The slot is defined by length L g , width W g , and position X 0 . These parameters perturb the surface current distribution and effective electrical length in a controlled manner. A comprehensive dataset of 250 full-wave configurations was generated in HFSS using waveguide excitation and open boundary conditions, with resonant frequency f r extracted at S 11 < − 10 dB , impedance bandwidth BW defined at the same threshold, and peak realized gain G obtained from far-field patterns at f r . The nonlinear mapping between aperture geometry [ L g , W g , X 0 ] and antenna responses [ f r , BW, G ] was modeled using four regression algorithms. Fully connected neural network (FCNN), linear regression, decision tree regressor, and support vector machine (SVM)—have been used to train and predict the antenna parameters. The models show a substantial agreement between the results obtained by HFSS software and ML. The FCNN demonstrated superior predictive accuracy with a mean absolute error below 1% across all targets. The proposed technique achieves a frequency tuning range of 1.52–4.18 GHz, a bandwidth enhancement of up to 47%, and gain control from -3 dBi to 3 dBi.