Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Recursive least squares filtering algorithms with systolic array architectures: a geometrical approach

Thesis/Dissertation ·
OSTI ID:5868829

This dissertation presents a geometrical derivation of recursive least squares algorithms for general space-time (multi-channel) filtering. These algorithms are recursive in both the time index and filter order, as well as having a simple systolic array architecture. Thus, by taking advantage of concurrent processing, these algorithms obtain a high throughput and are therefore suitable for real-time least squares filtering. The derivation is based upon developing fundamental order and time update theorems for various vectors in a pseudonorm space, where the pseudonorm determines the least squares criterion. The use of a pseudonorm instead of a true norm leads to a simplified derivation of a time-update theorem. This time update theorem is derived for a sliding memory, exponentially weighted pseudonorm, and is therefore a generalization of a previously derived theorem used in the derivation of the least squares lattice algorithm. However, unlike the least squares lattice algorithm, the order update theorem makes no assumption concerning the time shift properties of the input data, and therefore the general space-time filtering problem is solved without the need of matrix manipulation.

Research Organization:
California Univ., Los Angeles (USA)
OSTI ID:
5868829
Country of Publication:
United States
Language:
English

Similar Records

Recursive least-squares learning algorithms for neural networks
Conference · Sun Dec 31 23:00:00 EST 1989 · OSTI ID:6698032

Realizations and performances of least-squares estimation and Kalman filtering by systolic arrays
Thesis/Dissertation · Wed Dec 31 23:00:00 EST 1986 · OSTI ID:7245060

Recursive least squares algorithm for estimation of heat source
Journal Article · Thu Nov 07 23:00:00 EST 1996 · Numerical Heat Transfer. Part A, Applications · OSTI ID:418044