Transcript Lecture 21
ASEN 5070: Statistical Orbit Determination I
Fall 2015
Professor Brandon A. Jones
Lecture 21: A Bayesian Approach to the Kalman
Filter Derivation
University of Colorado
Boulder
Homework 6 Due Friday
No lecture quiz this week
University of Colorado
Boulder
2
Homework 6 – Common Question
University of Colorado
Boulder
3
What are the dimensions of the Htilde matrix?
Since the observations are generated via a single
ground station, what is the partial w.r.t. to the
other stations?
Need to add logic to your code to properly select
the non-zero columns for the ground station
partials!
University of Colorado
Boulder
4
The Kalman Filter – A Bayesian Approach
Ho and Lee, “A Bayesian Approach to Problems in
Stochastic Estimation and Control”, IEEE
Transactions on Automatic Control,
DOI: 10.1109/TAC.1964.1105763
University of Colorado
Boulder
5
University of Colorado
Boulder
6
University of Colorado
Boulder
7
We start with a previous state PDF at some
time tk-1:
Assume a linear description of the dynamics:
University of Colorado
Boulder
8
If we map the (Gaussian) previous-state PDF
through a set of linear equations, what is the
output?
University of Colorado
Boulder
9
University of Colorado
Boulder
10
A linear relationship between the state and
the observations, i.e.,:
All input PDFs are independent and Gaussian:
University of Colorado
Boulder
11
As you will show in HW7:
University of Colorado
Boulder
12
University of Colorado
Boulder
13
University of Colorado
Boulder
14
Do we know anything about the PDF of ε ?
Do we know if ε is independent of x ?
University of Colorado
Boulder
15
University of Colorado
Boulder
16
University of Colorado
Boulder
17
University of Colorado
Boulder
18
We have a solution, but it is not “elegant”
Can we manipulate the terms in the exponent
to look like something a little more familiar?
(Perhaps a Gaussian…)
We can, but we need a couple of tricks…
University of Colorado
Boulder
19
Schur Identity (Appendix B, Theorem 4):
University of Colorado
Boulder
20
We need to “complete the square”:
After applying those tricks and about 1-2
pages of linear algebra…
University of Colorado
Boulder
21
We have the Kalman filter as derived using
Bayes theorem!
University of Colorado
Boulder
22
In this derivation, what did we assume?
University of Colorado
Boulder
23
Since the Kalman and the Batch processor are
mathematically equivalent, then the batch can
also be derived via Bayes theorem, right?
◦ Yes! (See book section 4.5)
Both proofs/arguments work, but this
important derivation of the Kalman filter was
not included in the book
University of Colorado
Boulder
24