Multidomain Algorithm Evaluation. Volume I.
Abstract
The purpose of this study is to evaluate different algorithms for solving for up to 200 adaptive weights in an adaptive array radar, using the sample covariance matrix inversion technique. The sample covariance matrix inversion technique was studied because of its ability to handle adaptation in many domains, i.e., spatial, temporal, and polarization. The algorithms and their implementations on different computer architectures, such as associative, parallel, vector pipeline, and sequential, are considered. Both theoretical timings and actual timings on currently available machines are obtained. Major conclusions reached are that because of the strong dependence of an algorithm's implementation on the computer architecture, it is not possible to choose the best algorithm by operation counts; it is possible to greatly improve system performance by using separate processors to perform the covariance matrix computation and weight calculations; parallel complex arithmetic implemented in hardware would greatly improve a system's performance; and associativity is not useful for this problem. Finally, an architecture designed specifically for solving for adaptive weights is outlined. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1978
- Accession Number
- ADA054357
Entities
People
- Irving S. Reed
- James C. Demmel
- John D. Mallett
- Lawrence E. Brennan
- William C. Liles
Organizations
- Technology Service Corporation