Time-Domain Scattering Parameters-Based Neural Network Inverse Model for Antenna Designs

A multilayer perceptron neural network (MLP-NN) inverse model based on the time-domain scattering (TDS) parameters (TDS-MLP-NN) is proposed for antenna design. Inputting desired antenna electromagnetic responses, the MLP-NN inverse model can directly output the antenna geometric parameters. Integration of TDS into MLP-NN allows for reducing the complexity of the neural network topology and the dimension of the input data, thus contributing to an improved generalization ability of the MLP-NN. The feasibility and effectiveness of the TDS-MLP-NN model are investigated using a resonant and traveling-wave hybrid antenna of Fabry–Perot unidirectional dielectric radiator (UDR) antenna, and it is found that the proposed model can output desired geometric parameters even when antenna performances as input to the neural network are better than those in the training dataset. The proposed model yields a better antenna performance than the UDR antenna, with an impedance bandwidth of 30.07% and a maximum gain of 9.59 dBi. Moreover, the TDS-MLP-NN is shown to be able to significantly improve the generalization ability of the MLP-NN inverse model and the design efficiency of antennas.