Adaptive filtering and change detection 2005

Prerequisites

Undergraduate courses in mathematical statistics, linear algebra, calculus, filter theory, automatic control, stochastic processes and signal processing.

Schedule

The course starts in mid January 2005 and continues for approximately 14 weeks. The course will be coordinated with a similar course at KTH/UU. The periodicity is every two years.

Credits

To be decided, but probably 6 credits. 3 credits (orientation course) or 6 credits (advanced course). 2 more credits to the advanced course for mini-projects.

Organization

The goal is to get an understanding for the theory of model based filtering and change detection, with particular attention to applications. The course is divided into three parts: signal, parameter and state estimation. The tools and algorithms are very similar for these parts, but the applications and engineering disciplines in which this theory is commonly described in are quite different. The course consists of 10 lectures.

There is one possibilty to follow the course as an orientation. This shortcut covers adaptive filters and Kalman filters, and it includes all the lectures, Chapters 1,3,5,8 (also 2 and 4, but not in detail) and the introductory basics sections in chapters 6,7,9,10,11 of the compendium.

To the advanced course, there is a possiblity to do application oriented mini-projects, which gives another 2 credit points.

Besides the lectures, the participants organize classes themselves for the exercises. Solutions from earlier years students are available.

Content

Detection theory.
To separate a signal from noise.
State estimation in dynamical systems using Kalman filters.
Adaptive filtering using parametric models.
Change detection in state space models and parametric models.
Applications in digital communication: equalizers, speech coders, echo detection.
Applications in automatic control: fault detection in dynamic systems, parameter estimation for adaptive control, fault detection and diagnosis in state space models.
Applications in avionic systems: target tracking, navigation and sensor fusion.
Applications in signal processing: signal conditioning, surveillance, analysis and segmentation of signals in medicine, physics, biology and seismology.

Literature

Book 1: Selected parts of
Fundamentals of statistical signal processing: Volume II, detection theory.
Steven M. Kay
Prentice Hall PTR, New Jersey, 1998
ISBN: 0-13-504135-X Hdbk.
Book 2,
Adaptive filtering and change detection.
Fredrik Gustafsson,
John Wiley & Sons, Ltd, 2001.
ISBN: 0471 49287 6, Hdbk.
This is available at the best possible price at Linus and Linnea, tel 013-282850, for approximately 750SEK.
An errata list to the first printing of the book (2000): pdf. These are fixed in the second (2001) printing. There is yet no errata list to second printing.
Exercises pdf. Selected exercises to Kay's book: pdf

A Matlab Toolbox. From inside the department, do addpath('/home/rt/fredrik/adfilt') in Matlab (or add this to your startup.m file). The manual is available as the file manual.pdf (with browsing facilities) in the toolbox directory. The toolbox is for personal use and none of the parts may be distributed.

Course leader

Fredrik Gustafsson , tel 282706 email: fredrik@isy.liu.se.

Examination

There will be a three day take home examination available. The examination can be checked out during April.

Lecture plan

Standard lecture time is Wednesday 10-12 in Algoritmen . The schedule is still preliminary, and may be adjusted for conferences etc.

Nr.	Time	Content	Chapter
1. pdf	v4, 26/1, 10-12	Detection theory from Kay's book. Single and multiple hypothesis tests.	Selected chapters.
2. pdf	v5, 2/2, 10-12	Applications. Detection of mean changes: likelihood principles	1, 2, 3
3. pdf	v6, 9/2, 10-12	Detection of mean changes: The parallel filter concept.	3,4, Appendix
4. pdf	v9, 2/3, 10-12	Parameter estimation: recursive least squares. Residual based detection.	5
5. pdf	v10, 9/3, 10-12	Parameter estimation: parallel filters.	6
6. pdf	v11, 16/3, 10-12	Parameter estimation: filter bank change detection.	7
7. pdf	v13, 30/3, 10-12	State estimation: Kalman, EKF, residual based detection.	8, 9
8. pdf	v14, 6/4, 10-12	State estimation: parallel filters and filter bank change detection	9,10
9. pdf	v15, 13/4, 10-12	State estimation: parity space diagnosability and PCA.	11, Artikel

Slides will be available to download, when ready.

Projects

The projects will be performed individually or in group. The subject is either on particular implementations or applications, as defined in discussion with the examiner.