STUDY OF A 1-POINT ADAPTIVE FILTER ADVANCED ARRAY RESEARCH
Abstract
The design and evaluation of optimum, time-invariant filters (the classical approach to prediction problems) are subject to several limitations. Because of these limitations, an adaptive system was considered. This report describes a filtering system which can adjust to changes in either signal or noise, thereby overcoming many difficulties of time-invariant filtering. The small amount of required computational time to update the filter weights is another important feature of the scheme presented. In this report, an adaption algorithm applied to a simple time-series model was studied. For the case of stationary data, a tradeoff between the adaption rate and the mean-square-error performance of the filter exists. The adaption rate is inversely proportional to the convergence parameter lambda, while the mean-square-error is directly proportional to lambda. For the case of nonstationary data, a tradeoff between adapting too slowly and adapting too rapidly exists. The optimum rate of adaption appears to be approximately 10 times faster than the average time rate of change in the input data statistics.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 30, 1967
- Accession Number
- AD0826128
Entities
People
- Aaron H. Booker
- George D. Hair
- James E. Brown
- John P. Burg
Organizations
- Texas Instruments