Neural network inverse model for multi-band unequal Wilkinson power divider

Abstract

Purpose The purpose of this paper is to build a neural network (NN) inverse model for the multi-band unequal-power Wilkinson power divider (WPD). Because closed-form expressions of the inverse input–output relationship do not exist, the NN becomes an appropriate choice, because it can be trained to learn from the data in inverse modeling. The design parameters of WPD are the characteristic impedances, lengths of the transmission line sections and the isolation resistors. The design equations used to train the NN inverse model are based on the even–odd mode analysis. Design/methodology/approach An inverse model of a multi-band unequal WPD using NNs is presented. In inverse modeling of a microwave component, the inputs to the model are the required electrical parameters such as reflection coefficients, and the outputs of the model are the geometrical or the physical parameters. Findings For verification purposes, a quad-band WPD and a penta-band WPD are designed. The results of the full-wave simulations verify the validity of the design procedure. The resulting NN model outperforms traditional time-consuming optimization procedures in terms of computation time with acceptable accuracy. The designed WPDs using NN are implemented by microstrip lines and verified by using full-wave analysis based on high-frequency structure simulator (HFSS). The results of the microstrip WPDs have good agreements with the corresponding results obtained by using ideal transmission line sections. Originality/value The associated time-consuming procedure and computational burden in realizing WPD through optimization are major disadvantages; needless to mention the substantial increase in optimization time because of the multi-band design. NNs are one of the best candidates in addressing the abovementioned challenges, owing to their ability to process the interrelation between electrical and geometrical/physical characteristics of the WPD in a superfast manner.