Scribd
Upload
32 views

Exercise

This document contains 12 exercises involving least squares problems and Kalman filtering. The exercises cover topics such as changing weights in the least squares solution, splitting normals into separate equations, using recursive least squares and Kalman filtering to solve least squares problems, and experimenting with MATLAB code for smoothing and clock filtering. The solutions to the exercises are plotted or involve computing matrix operations like QR decomposition, GSVD, or taking the square root of a matrix.

Uploaded by

Ahmet TOLUC
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
32 views

Exercise

This document contains 12 exercises involving least squares problems and Kalman filtering. The exercises cover topics such as changing weights in the least squares solution, splitting normals into separate equations, using recursive least squares and Kalman filtering to solve least squares problems, and experimenting with MATLAB code for smoothing and clock filtering. The solutions to the exercises are plotted or involve computing matrix operations like QR decomposition, GSVD, or taking the square root of a matrix.

Uploaded by

Ahmet TOLUC
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Exercises

1. (Influence on the solution of changing weights) Solve the least squares problem
given by
     
1 1 2 1 0 0
a) A =  1 2 , b =  1 , and C =  0 2 0  .
−1 1 0 0 0 1
 
1 0 0
b) Change C to C =  0 2 0  . This corresponds to not using the third
0 0 0
observation.
 
1 1
c) Show that the solution in b) could be obtained from A = , and
1 2
 
2
b= .
1
 
1 0 0
d) Find the solution when C =  0 2 0  .
0 0 ∞

Sketch the three lines given by Ax = b in a).


Next plot the solutions in a), b), and d). Show that they lie on a straight line.
In all the above exercises the third observation was given weights 1, 0, and ∞.
Any other weight for the third observation will produce a solution lying on this
line. Range of c3 = [−1, ∞).

1 21 0 0
     
1 0 0
1 1 8  1 2 0 0 2
2. Let A =  2
 1 3 , b =  8 , C =  0 0 1 0 , and σ0 = 1. Define W
   

1 4 20 0 0 0 2
as W = chol(C). Compute A1 = W A, b1 = W b and find x.
MATLAB has a function sqrtm that computes the square root of a matrix. Use
this function for an alternative definition of W and repeat the computations.
3. Let be given the normals
         
A B x1 b 4 8 x1 36
= 1 or = .
BT D x2 b2 8 26 x2 112

Use the formula

x2 = (D − B T A−1 B)−1 (b2 − B T A−1 b1 )

to estimate x2 and next by substitution find x1 .

7th December 1998 Kai Borre 1


4. We assume given the same the normals as in Exercise 3. We split them into

Ax = B y + Gz

or
      
4 8 y 4 8
= y+ z.
8 26 z 8 26

We compute the changed weight matrix (also a projector) P = I −B(B T B)−1 B T


and find the unknown ẑ from the reduced normals

GT P Gẑ = GT P b.

Finally the unknown ŷ can be computed from

ŷ = (B T B)−1 B T (b − Gẑ).

5. Let be given
   
1 0 0
A = 1 1 and b = 8 and C=I
1 3 8

and a constraint described by


   
B= 1 4 and d = 20 .

Find the least-squares solution via QR decomposition and GSVD. Furthermore


find the solution to the unconstrained system and compute the increase in the
sum of squared residuals when constraining the problem.
6. Solve the problem described in Exercise 5 by allocating a big weight to the con-
straint.
7. A least-squares problem is given by
   
1 0 0
1 1 8
A=   and b= 
 and C = I.
1 3 8
1 4 20

Fint the solution of this problem by means of recursive least squares.


8. A least squares problem is described by
     
1 1 2 1
A =  1 2 , b = 1 , and C= 2 .
−1 1 0 1

Solve this problem by means of a Kalman filter.

7th December 1998 Kai Borre 2


9. The M-file k_dd3 is described from the bottom of page 488 to the middle of
page 490. Try to change the diagonal entries of 6e,k and see if this affects the
result.
10. Run the M-file smoother with the calls smoother(1,1) and smoother(.1,1).
11. Run the M-file kalclock with the call kalclock(’pta.96o’,’pta.nav’,1)
12. Run the M-file rec_cloc.

7th December 1998 Kai Borre 3

You might also like