Reverse-engineering nonlinear analog circuits with evolutionary computation

Abstract

The design of analog circuits by hand is a difficult task, and many successful approaches to automating this design process based on evolutionary computation have been proposed. The fitness evaluations necessary to evolve linear analog circuits are relatively straightforward. However, this is not the case for nonlinear analog circuits, especially for the most general class of design tasks: reverse-engineering an arbitrary nonlinear 'black box' circuit. Here, we investigate different approaches to fitness evaluations in this setting. Results show that an incremental algorithm outperforms naive approaches, and that it is possible to evolve robust nonlinear analog circuits with time-domain output behavior that closely matches that of black box circuits for any time-domain input. © 2014 Springer International Publishing Switzerland.