Design of wide operation range microwave transistors with a single multilayer perceptron network

In this work, Back Propagation Neural Network (BPNN) is aimed at building up a black box model of a microwave transistor for predicting the scattering parameter characteristics of the transistor having a wide range operation DC bias and frequency domain. For this purpose, a high technology transistor BFP181 is chosen with its wide range measured scattering parameters provided by the manufacturer's company. In order to increase the accuracy of the modelling in such a wide range of operation domain, this wide range of data is divided into sub -regions for each of which both individual training and testing processing is performed. Finally, accuracy and computational efficiency analysis is made for both different number of sub-regions and two different algorithms which are Levenberg Marquardt and Bayesian regression. It can be concluded that BPNN can also use for the fast and high accurate modelling scattering parameter characteristics of a microwave transistor having the wide range operation DC and frequency range.