Automated Model Generation Approach Using MATLAB

High level modelling (HLM) for operational amplifiers (opamps) has been previously carried out successfully using models generated by published automated model generation (AMG) approaches. Unfortunately, there are not any publications describing the use of AMG approaches for HLFM at a system level. For straightforward system simulation relatively simple models may be adequate, but they can prove inadequate during high level fault modelling (HLFM). The accuracy and speedup of existing models may be doubted when fault simulation is implemented because faulty behaviour may force (non-faulty) subsystems into highly nonlinear regions of operation, which may not be covered by their models.

There are several broad methodologies for AMG; a fundamental decision is the model structure, which in general terms divides into linear time-invariant (LTI), linear time-variant (LTV), nonlinear time-invariant and nonlinear time-variant types. An estimation algorithm is then required in order to obtain parameters for these models. These algorithms may be lookup tables, radial basis functions (RBF), artificial neural networks (ANN) and its derivations such as fuzzy logic (FL) and neural-fuzzy network (NF), and regression. Model generators can also be categorized into the black, grey or white box approaches, depending on the level of existing knowledge of the system’s structure and parameters.

The aim of the chapter is to introduce a self-tuning algorithm (i.e., AMG) using MATLAB for automated analogue circuit modelling suitable for HLM and HLFM applications. It can generate multiple models to cover a wide range of input conditions by detecting nonlinearity through variations in output error, and can achieve bumpless transfer between models and handle nonlinearity.

The properties of the tuning algorithm were investigated by modelling a two-stage CMOS opamp, and comparing simulations of the macromodel against those of the original SPICE circuit utilizing transient analysis. HLFM results show that the models generated can handle both linear and nonlinear situations with good accuracy in a low-pass filter.