Polynomial modelling: Accuracy vs. shape

Behavioral modelling often aims in a simple black-box input-output response with minimal rms error with the used excitation. However, this may grossly overestimate the sensitivities to certain changes in the inputs, causing the model to be prone of big errors with a different excitation. This also makes it difficult to separate different nonlinear effects. In this paper we compare the effect of different model fitting strategies to the shape of the modeled I-V response and distortion contribution analysis, that very much relies on the fact, that besides accurate I-V match also the shape of the model and its higher derivatives are physically meaningful and correct.