Scribd
Upload
95 views5 pages

Exam SC4040

This document provides information about an exam on filtering and identification taken on April 7, 2014 at Delft University of Technology. The exam consists of 4 exercises worth a total of 100 points. Exercise 1 (30 points) involves fusing prior information about an unknown random variable with new information from a colleague to update an estimate and its covariance matrix. Exercise 2 (25 points) uses Kalman filtering to estimate an unknown constant vector from a given signal generation model. Exercise 3 (25 points) applies subspace identification techniques to determine the order of an unknown minimal state space model using input-output data. Exercise 4 (20 points) involves using an unbiased estimate of a cross-correlation vector to check the statistical independence of noise sequences.

Uploaded by

kricks92
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
95 views5 pages

Exam SC4040

This document provides information about an exam on filtering and identification taken on April 7, 2014 at Delft University of Technology. The exam consists of 4 exercises worth a total of 100 points. Exercise 1 (30 points) involves fusing prior information about an unknown random variable with new information from a colleague to update an estimate and its covariance matrix. Exercise 2 (25 points) uses Kalman filtering to estimate an unknown constant vector from a given signal generation model. Exercise 3 (25 points) applies subspace identification techniques to determine the order of an unknown minimal state space model using input-output data. Exercise 4 (20 points) involves using an unbiased estimate of a cross-correlation vector to check the statistical independence of noise sequences.

Uploaded by

kricks92
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

Delft Center for Systems and Control

Delft University of Technology


Mekelweg 2
2628 CD Delft

Exam SC4040
FILTERING A N D IDENTIFICATION
Monday A p r i l 7, 2014, 18.30 - 21.30

This exam consists of 5 pages and 4 exercises.


The maximum number of points you can earn is indicated for each exercise. The total number of
points you can earn w i t h this written exam is 100.
You are allowed to use t h e b o o k [1] Filtering and System Identification:
A Least Squares
Approach,
Cambridge University Press, 2007 by the authors M . Verhaegen and V . Verdult during this exam.

G o o d luck!!!
1

E X E R C I S E 1: (30 p o i n t s )
[Least Squares]
The following prior information is given about an unknown random variable x G M " , x,^
{x,P),
such that F > 0, n > 1 and
= /3. A colleague from another department has convmcmg
arguments that the first component of x has to be equal to a, i.e.
x{l)

^ a ^ f 3

Fuse this information of your colleague w i t h the prior information to provide a new estimate of x
and its covariance matrix.
To answer this problem address the following questions:
( a ) [5 p o i n t s ] Express the prior information into a data equation format. [Hint: use a lower
triangular square root of P for this purpose.]
( b ) [5 p o i n t s ] Update the data equation of part (a) w i t h the info from your colleague and the
fact that

= (3 ^ a. [Hint: Partition the data equation!]

( c ) [20 p o i n t s ] Use the result of part (b) to update the estimate and (the square root) of its
covariance matrix.

(EXERCISE 2 ON T H E N E X T PAGE)

E X E R C I S E 2: (25 p o i n t s )
[Kalman Filtering]
Use the theory of Kalman filtering to estimate the unknown constant vector q of the followine
signal generation model (SGM):

:{k+l)
y{k)

Ax{k)+Bq

Cx{k)+v{k)

w i t h x{k) e K " , y{k) G

+ w{k)
E[

E[w{k)]^G
vik)
w{k)

(1)
R

A{k

and q M'" w i t h m < n.

( a ) [10 points] Propose a new SGM w i t h an augmented state containing both x{k) and q{k),
I.e. the value of q at time instance k. This new SGM is then used i n the design of a Kalman
filter to reconstruct both the system state x{k) and q{k).
( b ) [15 points] Show that the augmented SGM is observable provided the pair {A, C) is observable and the matrix C{A - I)-'^B having f u l l column rank.

(EXERCISE 3 O N T H E NEXT PAGE)

E X E R C I S E 3: ( 2 5 p o i n t s )
[Subspace Identification]
Consider an unknown though nrinimal L T I system belonging to the class of systems described by
finite dimensional state equations w i t h stochastic zero-iriean white noise input,

3;(fc + l )

Ax{k)+Ke{k)

e{k)GR

y{k)

Cx{k) + e{k)

x{k)

eR^

y{k)R

w i t h x{0) = 0 and e{k) a zero-mean white noise sequence w i t h covariance matrix

R>OL

For the subspace identification problem, let us consider the following data equation (for . and N
chosen such that s > n ,

> n).

( a ) [10 points] Prove that for the given state space model,
E[x{k)e{f]

= 0

Vfc, and ^ > / c > 0

( b ) [15 points] Prove that


N

A n d draw the conclusion from this result about finding the order of the state space model
f r o m the data.

(EXERCISE 4 ON T H E NEXT PAGE)

E X E R C I S E 4: (20 p o i n t s )
[The Identification Cycle]
Explain how the unbiased estimate of the cross-correlation vector i?e on the top of p. 389 and its
covariance matrix can be used w i t h some probability to check whether e{k) and u{i) are statistically
independent.
( E N D OF T H E E X A M )

You might also like