On-line identification of linear systems

Abstract

This paper considers the on-line identification of a linear process in terms of an orthonormal expansion of its impulse response. The undesirable features of the earlier solutions such as flipping back tho input or using unstable networks are avoided by the use of correlation functions. Modified Laguorro functions have boen employed in this work for tho identification of tho impulso response. The scheme developed here is applicable for a wide class of input signals. Further, the use of correlation eliminates the effects of any additive noise uncorrected with the input. Digital computer results have boon included to domonstrate the practicability of the schome. The only significant limitation appears to be the requirement that the process should not change appreciably in one avoraging time.