Nonlinear estimation with sparse temporal measurements

Download
Author
Frontera, Paul J.
Date
2016-09Advisor
Ross, I. Michael
Metadata
Show full item recordAbstract
Nonlinear estimators based on the Kalman filter, the extended Kalman filter (EKF) and unscented Kalman filter (UKF) are commonly used in practical application. The Kalman filter is an optimal estimator for linear systems; the EKF and UKF are sub-optimal approximations of the Kalman filter. The EKF uses a first-order Taylor series approximation to linearize nonlinear models; the UKF uses an approximation of the states' joint probability distribution. Long measurement intervals exacerbate approximation error in each approach, particularly in covariance estimation. EKF and UKF performance under varied measurement frequency is studied through two problems, a single dimension falling body and simple pendulum. The EKF is shown more sensitive to measurement frequency than the UKF in the falling body problem. However, both estimators display insensitivity to measurement frequency in the simple pendulum problem. The literature's lack of consensus as to whether the EKF or UKF is the superior nonlinear estimator may be explained through covariance approximation error. Tools are presented to analyze EKF and UKF measurement frequency sensitivity. Covariance is propagated forward using the approximations of the EKF and UKF. Each propagated covariance is compared for similarity with a Monte Carlo propagation. The similarity of the covariance matrices is shown to predict filter performance. Portions of the state trajectory susceptible to EKF divergence are found using the Frobenius norm of the Jacobian matrix, limiting the need to consider covariance propagation along the entire state trajectory. Long measurement intervals also reveal a commonly overlooked challenge in UKF application: sigma point selection methods may produce sigma point vec-tors that violate physical state constraints. Although the UKF can function under this condition over short measurement intervals, unexpected failure may occur without consideration of physical constraints. A novel constrained UKF, using the scaled unscented transform, is proposed to address this issue.
Rights
This publication is a work of the U.S. Government as defined in Title 17, United States Code, Section 101. Copyright protection is not available for this work in the United States.Collections
Related items
Showing items related by title, author, creator and subject.
-
A frequency domain analysis of the linear discrete Kalman filter
Costello, Walter Jeremiah (Monterey, California. Naval Postgraduate School, 1980-03);The linear discrete Kalman filter was analyzed using a frequency-domain approach. Process and measurement noise covariances are shown to be critical design parameters which, together with the assumed prior state and ... -
Sub-optimal gain schedules for the discrete Kalman filter.
Crotteau, Robert Allen (Naval Postgraduate School, 1969);The object of this study is to find an approximation to the discrete optimal Kalman filter gain schedule by closed-form analytic expressions. In doing so, required table storage and/or on-line computation time can be ... -
Application of the Kalman filter to spherical discrete track smoothing and prediction
Demetry, James S.; Hudson, R.E. (Monterey, California. Naval Postgraduate School, 1966); Technical Report/ Reserach Paper No. 70This paper investigates the effectiveness of a method for circumventing the problem of nonlinear observation of a linear system's state variables in the Kalman filter discrete estimation scheme. The method employs a ...