Kalman Filter and Simple Limited-Memory Estimation Algorithms,
Abstract
The purpose of the paper is to develop some simple estimation algorithms intended to increase measurement accuracy by taking into account more samples. These algorithms reduce to weighted means of few samples of noisy observed output. The weights or coefficients of these means are analyzed studying initial behaviour of the Kalman filter for some particular cases, especially when a priori information is incomplete. It is shown that when input noise is not present, the estimator coefficients form a sequence of increasing values for a stable process, of equal values for constant parameter estimation, and of decreasing values for unstable process. The increase in the input noise co-variance result in the increases of values of the coefficients related to most recent measurements. Output noise has an inverse effect on the coefficients. Some numerical examples are given as well as some suboptimal algorithms which can be constructed in an easy manner, without computing the whole Kalman filter. In the appendix, some general properties of the matrices involved in the estimator equations are proved. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- May 07, 1971
- Accession Number
- AD0886456
Entities
People
- Andrzej Manitius
- Feliks Gadzinski
Organizations
- National Air and Space Intelligence Center