Practical design and analysis of a simple 'neural' optimization circuit

Abstract

It is shown that simple neural networks can have potential problems of stability depending on their circuit implementation. A technique for transforming neural networks to equivalent circuits containing Schmitt triggers with varying thresholds is demonstrated. Using the example of an A/D converter, it is shown that the hysteresis of the Schmitt triggers play a large part in determining stability. Techniques of analysis first used by K.W. Cattermole (1969) to describe a class of circuits known as equilibrium converters are developed, which suggest methods to improve the performance of the A/D neural networks.