Bounded error parameter estimation: noise models, recursive algorithms and H/sub /spl infin// optimality

The first part of the paper deals with the relationship between various noise models and the "size" of the resulting membership set. Next, we present algorithms for various commonly encountered noise models that have the following properties: 1) they are recursive and easy to implement, and 2) after a finite "learning period" yield an estimate that is guaranteed to be in the membership set. Finally, we propose algorithms that not only have nice worst-case performance characteristics similar to those of LMS and LS, but also yield estimates that are in the membership set or "close" to it.