Abstract

A modular neural network architecture for accurate small/large signal microwave MESFET/HEMT modeling is presented. This is achieved by means of an original neural architecture having two main modules. A network captures the nonlinear dynamic Pulsed I/V characteristic of the device, which is mainly responsible of the large signal behavior, while the second network estimates the high order derivatives at the operation point, which are responsible of the IMD behaviour, by means of a neural network and then it locally reconstructs the current function by means of a third order Taylor series around that point. Finally, in order to have a maximum of coherence, the two networks are combined into a global model by means of a simple fuzzy controller. Computer simulations and experimental measurements validate this flexible modeling technique.