An instrumental product moment test for model order estimation

The near singularity of the product moment matrix of observed input/output data provides a quick way of checking the order of a linear system. However, the technique becomes insensitive when significant amounts of extraneous noise are present, and while this difficulty can be alleviated it is only by additional computation and explicit use of a priori system knowledge. This short paper describes an instrumental variable modification to the product moment technique which overcomes these problems with a negligible amount of extra computation.