Accession Number : AD0732132

Title :   Sensitivity Considerations in the Design of Linear Least Mean-Squared Error Filters,

Corporate Author : PRINCETON UNIV N J DEPT OF ELECTRICAL ENGINEERING

Personal Author(s) : Lugannani,R. ; Liopiros,K. J.

Report Date : OCT 1971

Pagination or Media Count : 7

Abstract : The usual approach to linear least mean-squared error filter design is concerned with minimizing a quadratic function of the error. Neglected in such an analysis is the effect that variations of the filter parameters may have upon the error criterion. In this paper the quadratic error term is combined with a suitable sensitivity measure to produce a new performance function, which is then minimized by selecting the appropriate linear filter. The resulting mean-squared error will be larger than the minimum error but less sensitive to parameter variations. To illustrate the advantages of this method, two specific examples are considered; both are concerned with estimating a random signal in while noise and exhibit a significant decrease in sensitivity with only a modest increase in mean-squared error. (Author)

Descriptors :   (*INFORMATION THEORY, ELECTRIC FILTERS), (*CONTROL SYSTEMS, MATHEMATICAL MODELS), LINEAR SYSTEMS, STATISTICAL ANALYSIS, POWER SPECTRA, INFORMATION THEORY, WHITE NOISE, TRANSFER FUNCTIONS

Subject Categories : Statistics and Probability
      Cybernetics

Distribution Statement : APPROVED FOR PUBLIC RELEASE