Flex-Form Generative Neural Network for Multifrequency Modes Antennas

This communication introduces a flex-form generative neural network (FGNN) that is capable of inverse designing the dimensions of a multifrequency modes antenna, which is implemented with a metallic monopole patch containing variable numbers of slots, based on a target reflection coefficient across 0.3-3 GHz. The FGNN integrates three models: the forward prediction model (FPM) for predicting the antenna’s reflection coefficient $S_{11}$ based on geometric parameters, i.e., patch and slot dimensions, the auxiliary classification model (ACM) for determining slots count, and finally, the inverse design model (IDM) for generating geometric parameters from the desired reflection coefficient. The proposed method enables flexible adjustment of antenna structure by allowing parameters output of indefinite length. Antenna supporting up to three bands is achieved in this demonstration, with minimal amount of dataset required. The network potentially provides an efficient framework for various electromagnetic (EM) structures design with flexible forms of geometry parameters or targets.