Enhancing the Accuracy of Microwave Element Models by Artificial Neural Networks

Abstract

In the recent PSpice programs, five types of the GaAs FET model havebeen implemented. However, some of them are too sophisticated andtherefore very difficult to measure and identify afterwards, especiallythe realistic model of Parker and Skellern. In the paper, simpleenhancements of one of the classical models are proposed first. Theresulting modification is usable for the accurate modeling of both GaAsFETs and pHEMTs. Moreover, its updated capacitance function can serveas an accurate representation of microwave varactors, which is alsoimportant. The precision of the updated models can be strongly enhancedusing the artificial neural networks. In the paper, both using anexclusive neural network without an analytic model and cooperating acorrective neural network with the updated analytic model will bediscussed. The accuracy of the analytic models, the models based on theexclusive neural network, and the models created as a combination ofthe updated analytic model and the corrective neural network will becompared.