The Off-Line Use of a Sequential Estimator.
Abstract
The straightforward application of a sequential estimator to nonlinear regression (curve fitting) problems is generally not possible when good a priori parameter estimates are not available and also when minimizing the error over a local portion of the data does not insure that it is minimized globally. However, a sequential estimator may be easily utilized to perform off-line processing in such a situation. The key is to process the measurements in a random order rather than the causal order in which they occur. The off-line use of an extended Kalman filter is illustrated in terms of a particular application. This technique is essentially a sequential version of the Gauss-Newton minimization procedure with relinearization being performed after each measurement is processed. Fictitious measurement noise is necessary to prevent filter divergence and is included in a very simple manner. Computational savings over more conventional iterative minimization techniques are possible if the functions and partial derivatives involved are sufficiently complex to evaluate. But there is a real question regarding the extent to which convergence can be assured. The results of simulations are presented. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 1980
- Accession Number
- ADA088178
Entities
People
- Stuart C. Schwartz
- Thomas G. Robertazzi
Organizations
- Princeton University