Tracking characteristics of an OBE parameter-estimation algorithm

Abstmct-Recently there seems to have been a resurgence of interest in recursive parameter-bounding algorithms. These algorithms are applicable when the noise is bounded and the bound is known to the user. One of the advantages of such algorithms is that 100% confidence regions (which are optimal in some sense) for the parameter estimates can be obtained at every time instant, rather than asymptotically as in the case of the least squares type algorithms. Another advantage is that these recursive algorithms have the inherent capability of implementing discerning updates, particularly that of allowing no updates of parameter estimates in the recursion. This paper investigates tracking properties of one such algorithm, referred to as the Dasgupta-Huang optimal bounding ellipsoid (DHOBE) algorithm. Conditions that ensure the existence of these 100% confidence regions in the face of small-model parameter variations are derived. For larger parameter variations, it is shown that the existence of the 100% confidence regions is guaranteed asymptotically. A modification is also proposed here to enable the algorithm to track large variations in model parameters. Simulation results show that in general, the modified algorithm has tracking performance comparable, and in some cases superior, to the exponentially weighted recursive least squares algorithm.