Design optimization of wearable antennas using an adaptive network-based fuzzy inference system

This paper presents a data-driven optimization framework for wearable microstrip patch antennas operating near the 2.4 GHz ISM band using an Adaptive Network-Based Fuzzy Inference System (ANFIS). The proposed approach predicts key antenna geometry parameters-patch width, patch length, and feed point coordinates-from target electromagnetic performance metrics, including resonant frequency, reflection coefficient, bandwidth, and substrate thickness. A dataset of 500 samples was extracted from parametric HFSS simulations, and the ANFIS model was trained using a 70/30 training-testing split. Four membership function types "triangular, trapezoidal, Gaussian, and generalized bell" were systematically evaluated to analyze their influence on prediction accuracy and convergence behavior. Results demonstrate strong agreement between ANFIS predictions and HFSS simulations, with sub-millimeter mean absolute errors for patch dimensions and high regression coefficients across parameters. Among the tested membership functions, the generalized bell function showed the most stable convergence and the lowest overall prediction error. The proposed approach reduces reliance on repeated full-wave simulations by providing rapid estimation of antenna geometry from performance specifications, enabling faster prototyping of wearable antennas. The results highlight the importance of membership function selection in neuro-fuzzy modeling for antenna design and demonstrate the feasibility of ANFIS-based surrogate modeling for wearable communication devices.