Set membership identification in digital signal processing

Set membership (SM) identification refers to a class of techniques for estimating parameters of linear systems or signal models under a priori information that constrains the solutions to certain sets. When data do not help refine these membership sets, the effort of updating the parameter estimates at those points can be avoided. An intuitive development is given, first in one dimension and then in the general case, of an SM algorithm based on least-squares estimation. Two useful versions of the method are described, one of which can be implemented on a systolic array processor. The relationship of the featured SM method to both historical and current developments is discussed. Application to real speech data illustrates the developments.<>