An overview of some recent advancements in fast RLS algorithms
(1991) Algorithms and Parallel VLSI Architectures. Lectures and Tutorials Presented at the International Workshop — Location: Pont-a-Mousson, France (10.June.1990)
Files
No attached file found for this publication.
Details
- Authors
- Abstract
- Discusses a number of recent developments in the area of fast recursive least squares (RLS) adaptive filtering algorithms. The driving force behind the developments considered is numerical robustness, which is a quality that is of paramount importance when considering an actual implementation of these (and most other) algorithms. Two groups of fast RLS algorithms have coexisted for well over a decade now. These are the fast lattice (FLA) and the fast transversal filter (FTF) algorithms, whose names reflect the filtering structure which the algorithms adapt. Both groups have recently been augmented with new variants. Fast RLS algorithms have been derived based on the QR decomposition of the data matrix. This development was motivated by the good numerical properties of orthogonal transformations, which are ubiquitous in sound algorithms in numerical analysis. The resulting FQR algorithms turn out to resort under the FLA category, but nevertheless constitute a new variant based exclusively on orthogonal rotations. In the FTF algorithms on the other hand, the introduction of computational redundancy has been considered, allowing the dynamics of the propagation of numerical errors to be altered and stabilized via output feedback.
- Affiliations
Citations
Slock, D. T. M. (1991). An overview of some recent advancements in fast RLS algorithms. In Deprettere, F.; Van Der Veen, A.-J.; (ed.), Algorithms and Parallel VLSI Architectures. Lectures and TutorialsPresented at the International Workshop (p. Vol. A, p. 241-65). Elsevier. https://hdl.handle.net/2078.5/227255