Fast and accurate identification of electronic circuit parameters using regularised feedforward neural networks

This paper postulates that feedforward artificial neural networks (FANN) can be used to identify parameters of electronic circuits. The problem of identification accuracy is discussed to show that errors can be made small by learning a multi-variable function and its partial derivatives with respect to the unknown parameters. Results of computer simulation and measurements are analysed using examples of a semiconductor diode and two bandpass filters. They show that high accuracy can be obtained with small-size, single-hidden-layer FANNs. The identification process can be made very fast, limited only by the signal propagation through the actual FANN structure, with no need for the iterative calculations that are normally required using traditional model fitting techniques.