An Analysis of Gauss Elimination for Adaptive Beamforming.
Abstract
There are many formulations of the Adaptive Beamforming (ABF) problem. We are motivated to use an algorithm that uses the least number of divisions or square root operations, since these operations are very expensive in time or chip area. One method solves the solution of a Hermitian matrix. The Gauss Elimination (GE) algorithm has no square roots. In this paper, we analyze dynamic range requirements and precision analysis for an integer processor implementation of GE (as opposed to fixed point or floating point), to solve the ABF problem. This analysis differs from the standard Wilkinson analysis which is based on fixed point arithmetic, where the multiplier is formed once per row, whereas the integer formulation requires that we perform explicit divisions on each element of the reduced matrix. We present formulas to determine input quantization, storage wordlengths and accumulator wordlengths for a desired weight precision.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 09, 1994
- Accession Number
- ADA298992
Entities
People
- Barry J. Kirsch
- Peter R. Turner
Organizations
- Naval Air Warfare Center Aircraft Divison