Overview of an environment for automated electrical behavioural modelling using data-based methods

This paper introduces an overview of an automated methodology for electrical black box behavioural modelling. The presented approach rests upon the transient simulation data of the in and output pins of an analogue or mixed signal circuit. This method extracts the magnitude and the phase information by measuring in- and output currents and voltages and creates a new electrical pin compatible black-box model which approximates the real electric behaviour of the original circuit. The behavioural model is made up of cross connected admittance models whose connections consist of complex functions which are approximated using Support Vector Machines. The electrical behavioural models have a speed-up factor up to 37x and an accuracy greater than 95% compared to the original circuit simulation. A special filter structure enables the modeling of frequency, amplitude and load dependent circuits. The electrical functioning of the methodology is shown by a passive second order Butterworth filter and a time variant and active charge pump circuit out of the automotive industry.