230 likes | 378 Views
ASEN 5070: Statistical Orbit Determination I Fall 2013 Professor Brandon A. Jones Professor George H. Born Lecture 40: Information Filter. Announcements. No lecture quiz this week Final Exam Due December 16 B y noon for in-class students By 11:59pm for CAETE s tudents
E N D
ASEN 5070: Statistical Orbit Determination I Fall 2013 Professor Brandon A. Jones Professor George H. Born Lecture 40: Information Filter
Announcements • No lecture quiz this week • Final Exam Due December 16 • By noon for in-class students • By 11:59pm for CAETE students • Final Project Due December 16 • By noon for in-class students • By 11:59pm for CAETE Students
Project Kalman Filter Variance • Well, we know that the CKF has problems…
Project Kalman Filter Variance • How about the Joseph formulation of the measurement update?
Project Kalman Filter Variance • How about the EKF?
Project Kalman Filter Variance • How about the Potter square-root filter?
Project Kalman Filter Variance • What are we unable to do with Potter that we can do with the CKF?
Minimum Variance as a Sequential Processor • Time Update • Measurement Update:
Project Kalman Filter Variance • What if we go back to the minimum variance?
Minimum Variance as Sequential Processor • If I don’t want to invert the information matrix, do I have another option?
Information Filter Measurement Update • Well, that was easy. • What about the time update?
Information Filter Time Update Derivation • What can we do to simplify this? (Assume Qk non-singular)
Time Update of Information Matrix • Require that Qk be non-singular • Do not need to invert the information matrix Still need to maintain information matrix separate from D !
Time Update of State Estimate • From the time update of the information matrix:
Information Filter • Can I initialize the filter with an infinite a priori state covariance matrix? • What happens if we have very accurate measurements?
State/Covariance Generation • Once the information matrix is positive definite:
Why use the Information Filter? • Provides a more numerically stable solution • Stability equals that of the Batch, but in a sequential implementation • Don’t need to generate state/covariance until needed • Square-root information filter (SRIF) • Refined through extensive use in POD • Includes smoothing capabilities