Investigo. Repositorio Institucional de la Universidade de Vigo
Publicación
Contribución de congreso (ConferenceOutput)
Extracción de modelos comportamentales para transistores de alta frecuencia
This communication describes the software tools
developed in MATLAB environment to extract behavioral
models for microwave transistors from RF calibrated
experimental data (Load-Pull measurements). These tools are
designed to extract both linear and nonlinear frequency domain
models. Linear models are limited to a single impedance and
small signal operation. In the linear case, small-signal S, Y and
Z parameters are computed. Extracted nonlinear models are
based on Cardiff model formulations: Cardiff conventional A-B
model, Cardiff Admittance I-V model and Cardiff advanced
formulations. Auxiliary apps have been designed to import
calibrated measurement data and also for drawing test graphs
to evaluate models extraction. Finally, another type of model
has been explored: artificial neural networks based models.
Hence, another MATLAB tool has been designed for training
shallow neural networks (a Multi-Layer Perceptron, MLP; with
one hidden layer trained for approximate fitting). A quality
FoM (SNR) is computed comparing measurements with model
predictions for all extracted models obtaining good results for
all methods and allowing a comparison between them., Agencia Estatal de Investigación | Ref. PID2020- 116569RB-C33