Filtering and Identification

Lecture 1:
Introduction and refreshment LA

Michel Verhaegen and Jan-Willem van Wingerden

1/39

Delft Center for Systems and Control

Delft University of Technology
Introduction to the course

• A Challenging/Motivating Application
Adaptive Optics

2/39

Delft Center for Systems and Control

Introduction to the course

• A Challenging/Motivating Application
Adaptive Optics
• Key problems

2/39

Delft Center for Systems and Control

Introduction to the course

• A Challenging/Motivating Application
Adaptive Optics
• Key problems
• Goal of the Course:
To DERIVE Numerical solutions to the
problems to meet the challenges!

2/39

Delft Center for Systems and Control

Introduction to the course

• A Challenging/Motivating Application
Adaptive Optics
• Key problems
• Goal of the Course:
To DERIVE Numerical solutions to the
problems to meet the challenges!
• Outline and Organization of the Course

2/39

Delft Center for Systems and Control

Introduction to the course

• A Challenging/Motivating Application
Adaptive Optics
• Key problems
• Goal of the Course:
To DERIVE Numerical solutions to the
problems to meet the challenges!
• Outline and Organization of the Course
• Refreshement

2/39

Delft Center for Systems and Control

Smart Optics Systems
Star

Telescope / Plane wavefront

Collimator Turbulent Atmosphere
Disturbed wavefront
Deformable
mirror

Controller

Tip−tilt mirror
Beam splitter

Wavefront
sensor

Camera

Adaptive Optics Active correction of wavefront aberrations by a

deformable mirror. What is needed from a control engineer?
3/39

Delft Center for Systems and Control

Smart Optics Systems
Star

Telescope / Plane wavefront

Collimator Turbulent Atmosphere
Disturbed wavefront
Deformable
mirror

Controller

Tip−tilt mirror
Beam splitter

Wavefront
sensor

Camera

Adaptive Optics Active correction of wavefront aberrations by a

deformable mirror. What is needed from a control engineer?
3/39

Delft Center for Systems and Control

Smart Optics Systems
Star

Telescope / Plane wavefront

Collimator Turbulent Atmosphere
Disturbed wavefront
Deformable
mirror

Controller

Tip−tilt mirror
Beam splitter

Wavefront
sensor

Camera

Adaptive Optics Active correction of wavefront aberrations by a

deformable mirror. What is needed from a control engineer?
3/39

Delft Center for Systems and Control

I. Domain Knowledge
Optical Field (spatial do-
main):

Ψ(r) = a(r)ejφ(r)
 
→ W (r, t) = Re Ψ(r)ejωt
wavenumber
(
ω
c for c (speed of light)
k = 2π
λ for λ (wavelength)

4/39

Delft Center for Systems and Control

I. Domain Knowledge
Optical Field (spatial do- φ(x, y, t)
main):

Ψ(r) = a(r)ejφ(r)
 
→ W (r, t) = Re Ψ(r)ejωt
y x
wavenumber Video Demo: Gemini Telescope (1:25)
( P

ω
for c (speed of light)
y x

c
k = 2π
λ for λ (wavelength)

4/39

Delft Center for Systems and Control

The Essential Role of the Wavefront (Phase) in Imaging

5/39

Delft Center for Systems and Control

II. Sensor
Schematic Shack-Hartmann Keck Measurements
Sensor
0.5

s(1)
φ(x, y) 0
-0.5
0 50 100 150

s(xi , yi )

s(N )
0.5
0
[From M. Konnik, 2010]
-0.5
0 50 100 150
with s(xi , yi ) an approximation
 
∂φ
of the slope  ∂x  → φ̂? • — current time index t = 150.
∂φ
∂y
6/39

Delft Center for Systems and Control

III. Actuator
 
Let Φa = vec φa (xi , yi ) ,
then in the low-bandwith
regime, the relation between
the actuator inputs and the
wavefront induced by the
mirror is:

Φa = M u

Taking time into consideration

[Ph.D thesis Roger Hamerlinck
we denote this as:
and Rogier Ellenbroek, (TuE - TUD)]

Φdm (t) = M udm (t)

7/39

Delft Center for Systems and Control

 udm
b
Φ(t) η(t)
VI. Control
Φdm (t)
udm (t) DM − WFS + s(t)
M (z −1 ) SH-sensor

Controller

With t = 150(e.g.) the current time instance, we want to influence

Φ(t + 1). How to determine udm (t + 1)?

8/39

Delft Center for Systems and Control

 udm
b
Φ(t) η(t)
VI. Control
Φdm (t)
udm (t) DM − WFS + s(t)
M (z −1 ) SH-sensor

Controller

With t = 150(e.g.) the current time instance, we want to influence

Φ(t + 1). How to determine udm (t + 1)?
b̂
• Criterium: Cov(Φdm (t + 1) − Φ(t + 1)). With the DM model
this is equal to:

b̂
Cov(M udm (t + 1) − Φ(t + 1))

8/39

Delft Center for Systems and Control

 udm
b
Φ(t) η(t)
VI. Control
Φdm (t)
udm (t) DM − WFS + s(t)
M (z −1 ) SH-sensor

Controller

With t = 150(e.g.) the current time instance, we want to influence

Φ(t + 1). How to determine udm (t + 1)?
b̂
• Criterium: Cov(Φdm (t + 1) − Φ(t + 1)). With the DM model
this is equal to:

b̂
Cov(M udm (t + 1) − Φ(t + 1))

• What is in the end needed to minimize this cost function?

8/39

Delft Center for Systems and Control

Wide scale Applicability?
Photonics: Driving Economic Growth in Europe

Microscopy

Lithography

Astronomy

Opthalmology

9/39

Delft Center for Systems and Control

Introduction to the course

• A Challenging/Motivating Application
Adaptive Optics
• Key problems
• Goal of the Course:
To DERIVE Numerical solutions to the
problems to meet the challenges!
• Outline and Organization of the Course
• Refreshement

10/39

Delft Center for Systems and Control

Teaching Staff (DCSC)

Lecturers
Prof.dr.ir. Michel Verhaegen | [email protected]
Dr.ir. Jan-Willem van Wingerden | [email protected]
Teaching Assistants (TA’s):
Baptiste Sinquin (MSc) | [email protected]
Pieter Piscaer | [email protected]
Student Assistant (SA):
Peter Várnai | [email protected]

11/39

Delft Center for Systems and Control

Objectives of the course
After studying this course you should

be able to derive estimation, filtering and identifica-

a
tion algorithms based on the the linear least
squares method
a
And control (H2 , etc.)

12/39

Delft Center for Systems and Control

Objectives of the course
After studying this course you should

be able to derive estimation, filtering and identifica-

tion algorithms based on the the linear least
squares method
And control (H2 , etc.)

1. Deriving a Mathematical Model (linear State Space Model)

from Data = Identification.
2. Determining the optimal (state) estimate given the (State
Space) Model = Filtering.
3. The Fundamental Basics is Linear Least squares =
Estimation.
12/39

Delft Center for Systems and Control

Course material
• Book:
Filtering and System Identification: An
Introduction, by Michel Verhaegen and
Vincent Verdult, Cambridge University Press,
2007.

13/39

Delft Center for Systems and Control

Course material
• Book:
Filtering and System Identification: An
Introduction, by Michel Verhaegen and
Vincent Verdult, Cambridge University Press,
2007.
• Student Guide: Rules, Grading, Schedule,
office-hours, · · · , and hand-outs, home works,
matlab exercises, Practical Exercise

13/39

Delft Center for Systems and Control

Course material
• Book:
Filtering and System Identification: An
Introduction, by Michel Verhaegen and
Vincent Verdult, Cambridge University Press,
2007.
• Student Guide: Rules, Grading, Schedule,
office-hours, · · · , and hand-outs, home works,
matlab exercises, Practical Exercise
• Register on Piazza - the tool we will use for
communication between students and
teaching staff! All on blackboard!
13/39

Delft Center for Systems and Control

Outline of the course

• Nine lectures
See Course schedule on the website

14/39

Delft Center for Systems and Control

Outline of the course

• Nine lectures
See Course schedule on the website
• Four instruction lectures/Four home-works
See Course schedule on the website

14/39

Delft Center for Systems and Control

Outline of the course

• Nine lectures
See Course schedule on the website
• Four instruction lectures/Four home-works
See Course schedule on the website
• Two matlab hands-on sessions
See Course schedule on the website

14/39

Delft Center for Systems and Control

Outline of the course

• Nine lectures
See Course schedule on the website
• Four instruction lectures/Four home-works
See Course schedule on the website
• Two matlab hands-on sessions
See Course schedule on the website
• Practical exercise
Handed-out (see course schedule) —
Register in groups of 2 (TA)!

14/39

Delft Center for Systems and Control

Exam
• Four sets of homework exercises: 6%
Deadline Hand-in sets see
“Guidelines/Rules/Schedule sc4040:
2016-2017” [1]

15/39

Delft Center for Systems and Control

Exam
• Four sets of homework exercises: 6%
Deadline Hand-in sets see
“Guidelines/Rules/Schedule sc4040:
2016-2017” [1]
• Two sets of matlab exercises: 4%
Hand-in sets see [1]

15/39

Delft Center for Systems and Control

Exam
• Four sets of homework exercises: 6%
Deadline Hand-in sets see
“Guidelines/Rules/Schedule sc4040:
2016-2017” [1]
• Two sets of matlab exercises: 4%
Hand-in sets see [1]
• ···

15/39

Delft Center for Systems and Control

Recommendation and Rules [1]

• Lectures: Highly recommended!

a
[1] “Guidelines/Rules/Schedule sc4040: 2016-2017”
16/39

Delft Center for Systems and Control

Recommendation and Rules [1]

• Lectures: Highly recommended!

• Homework/matlab/Practical exercises:
Exercises available and handed-in as outlined
on blackboard [1]!

[1] “Guidelines/Rules/Schedule sc4040: 2016-2017”

16/39

Delft Center for Systems and Control

Recommendation and Rules [1]

• Lectures: Highly recommended!

• Homework/matlab/Practical exercises:
Exercises available and handed-in as outlined
on blackboard [1]!
• During the help-session for finalizing the
homework - sit on oneven rows only.
Advise: Come prepared!

[1] “Guidelines/Rules/Schedule sc4040: 2016-2017”

16/39

Delft Center for Systems and Control

Recommendation and Rules [1]

• Lectures: Highly recommended!

• Homework/matlab/Practical exercises:
Exercises available and handed-in as outlined
on blackboard [1]!
• During the help-session for finalizing the
homework - sit on oneven rows only.
Advise: Come prepared!
• ...

[1] “Guidelines/Rules/Schedule sc4040: 2016-2017”

16/39

Delft Center for Systems and Control

 Filtering and identification
Let’s start!

17/39

Delft Center for Systems and Control

Mathematical ingredients?

(Linear) Least Squares

min ǫT ǫ y = Fx + ǫ
x
y
f2
• Matrix Theory
F f1
• Probability Theory

• Signal/System Theory

• Domain Knowledge

18/39

Delft Center for Systems and Control

Overview Linear Algebra (LA)

• The matrix concept!

• The Usefull matrix factorization: The SVD
• A Quick view on its potential!
• “Matrix-crimes”
• The Useful matrix Lemma

19/39

Delft Center for Systems and Control

Definition of a matrix

A matrix A ∈ Rm×n is a
two-dimensional table of numbers:
 
a11 a12 · · · a1n
a a a  h i
 21 22 2n 
A= . ..  = a1 a2 · · · an
 
am1 am2 · · · amn

with aij ∈ R, ai ∈ Rm .

20/39

Delft Center for Systems and Control

A matrix represents a (linear) mapping

A matrix is (also) a mapping between two

Euclidean vector spaces:
A : Rn → Rm : ∀x ∈ Rn , ∃y ∈ Rm : Ax = y
n
R m
R
"A"

0 0

21/39

Delft Center for Systems and Control

The “Four” key spaces of a linear mapping

R
n
R
m The linear mapping: A : Rn
(“domain”) → Rm (“Image or
"A"

0 0
Range space”) is character-
ized by four subspaces:
• range(A) = {y ∈ Rm : y = Ax for some x ∈ Rn }

• range(AT ) = {x ∈ Rn : x = AT y for some y ∈ Rm }

• ker(A) = {x ∈ Rn : Ax = 0}

• ker(AT ) = {y ∈ Rm : AT y = 0}

The rank of A equals the dimension of range(A).

22/39

Delft Center for Systems and Control

Special class of matrices
A “square” matrix Q ∈ Rn×n is
Definition:
orthogonal if

QT Q = QQT = In
This means:
1. Each column vector of an orthogonal matrix has length · · · ?
2. Two different column (row) vectors of an orthogonal matrix
satisfy?
3. What is the inverse of an orthogonal matrix?
4. And many more useful (numerical) advantages ...

23/39

Delft Center for Systems and Control

Overview Linear Algebra (LA)

• The matrix concept!

• The Usefull matrix factorization: The SVD
• A Quick view on its potential!
• “Matrix-crimes”
• The Useful matrix Lemma

24/39

Delft Center for Systems and Control

The Singular value decomposition (SVD)

The SVD-Theorem: Let A ∈ Rm×n , then there exists

a pair of orthogonal matrices:
h i
U = u1 · · · um ∈ Rm×m : U U T = U T U = Im
h i
V = v1 · · · vn ∈ Rn×n : V V T = V T V = In

such that,
" #
Σ 0
A=U V T ∈ Rm×n , Σ = diag(σ1 , · · · , σp )
0 0

with σ1 ≥ σ2 ≥ · · · ≥ σp ≥ 0 and p = min(m, n).

25/39

Delft Center for Systems and Control

  
Example SVD 1 0 1
(CondA.m)
 
1 1 
A = 2 2 1
⇒

0 1 1
 √ √ √  √  √ √ √ T
3 2 6 3 2 6 2 3
− 0 0 − −
 √3 2
√ 
6  2   √6 2
√ 
3
 6   √ 
A= − 33 0  
3  0 1 0  − 66 − 2
2
− 33 
 √ √ √  √ √ 
− 33 − 22 6
6
0 0 0 − 36 0 3
3
| {z }| {z }| {z }
U Σ VT

[U,Sigma,V]=svd(A);
• Column vectors of the matrix U : left singular vectors

• Column vectors of the matrix V : right singular vectors

• Diagonal elements of Σ: the singular values

26/39

Delft Center for Systems and Control

 RangeDemo.m

27/39

Delft Center for Systems and Control

Observations from RangeDemo.m
 
• columns of A lie in a plane ⊂ R3 ⇔ dim spancol (A) = 2 ⇔
# non-zero singular values (sv’s) = 2
• the left singular vectors u1 , u2 corresponding to the non-zero
singular values:
2
X
A= σi ui viT
i=1

form an orthogonal basis for spancol (A).

• the left singular vector u3 corresponding to the zero singular
value (i = 3) is a basis for ker(AT ).
• the left (and right) singular vectors are orthogonal and are of
unit length.
28/39

Delft Center for Systems and Control

The four key subspaces

Let the SVD of the matrix A be given as,

  
h i Σ1 | 0 V1T
A = U1 | U2    with Σ1 > 0
0 |0 V2T



then, since Ax = U1 Σ1 (V1T x) ,

range(A) = {y ∈ Rm : y = Ax for some x ∈ Rn } = span(U1 )

n
Further, since for x = V2 α : R
R
m
"A"
Ax = U1 Σ1 V1T V2 α = 0, V1
U1
V
2
U2
n 0 0
ker(A) = {x ∈ R : Ax = 0} = span(V2 )

29/39

Delft Center for Systems and Control

Overview Linear Algebra (LA)

• The matrix concept!

• The Usefull matrix factorization: The SVD
• A Quick view on its potential
• “Matrix-crimes”
• The Useful matrix Lemma

30/39

Delft Center for Systems and Control

“Optimal” low rank approximation

Theorem: Let the SVD in the SVD-theorem be

given and let k < rank(A) and let the following
approximation Ak of A be given:
k
X
Ak = σi ui viT
i=1

then,
min kA − Bk2 = kA − Ak k2 = σk+1
rank(B)=k

31/39

Delft Center for Systems and Control

 Spiegelman.m

32/39

Delft Center for Systems and Control

Overview Linear Algebra (LA)

• The matrix concept!

• The Usefull matrix factorization: The SVD
• A Quick view on its potential
• ‘‘Matrix-crimes”
• The Useful matrix Lemma

33/39

Delft Center for Systems and Control

Matrix-crimes: Syntax crimes
1. Non-compatibility of dimensions: A + B when
A ∈ R2×3 and B ∈ R3×3 and the same for
AT B.
2. Matrix products do (in general) not commute:
AB 6= BA.
3. Matrix inverse of the product of matrices:
(AB)−1 6= A−1 B −1 in stead of
(AB)−1 = B −1 A−1 - provided inverses exist!
4. (A + B)2 6= A2 + 2AB + B 2 !
Typical violations of Stanford students [S. Boyd - EE 263], our TUD students
“too often” join the club ... 34/39

Delft Center for Systems and Control

Matrix-crimes: Semantic crimes

Matrix expressions that simply do not make

sense. Examples:
1. Let x ∈ Rn , then xxT exists but (xxT )−1 not,
why?
2. If the matrix Q ∈ Rm×n for m > n, then

QQT
can never be the identity matrix.

35/39

Delft Center for Systems and Control

Overview Linear Algebra (LA)

• The matrix concept!

• The Usefull matrix factorization: The SVD
• A Quick view on its potential
• “Matrix-crimes”
• The Useful matrix Lemma

36/39

Delft Center for Systems and Control

Lemma 2.3 p. 19

Schur Complements: Block Triangular Factorizations

Let the block matrix A ∈ Rn×n (symmetric) be invertible, then a
very useful matrix factorization of matrix consisting of different
blocks is the following (C ∈ Rm×m ):
     
A B I 0 A 0 I A−1 B
 =   
BT C B T A−1 I 0 C − B T A−1 B 0 I

Therefore the following holds,

 
A B
  ≥ 0 ⇔ A > 0 and C − B T A−1 B ≥ 0
BT C
37/39

Delft Center for Systems and Control

Lemma 2.3 p. 19

Schur Complements: Block Triangular Factorizations

When C is invertible, then we have:
     
A B I BC −1 A − BC −1 B T 0 I 0
 =   
BT C 0 I 0 C C −1 B T I

Therefore the following holds,

 
A B
  ≥ 0 ⇔ C > 0 and A − BC −1 B T ≥ 0
BT C

The condition M atrix ≥ 0 among others means that a square

root of the matrix exists: M atrix = M atrix1/2 M atrixT /2
38/39

Delft Center for Systems and Control

Summary of Lecture 1

To start the discovery tour for retrieving system information from

measured data records:

What we just have done is a brief review of linear algebra. Next

we briefly review probability theory and filtering of stochastic
processes! We will also start with analysing the derterministic
least squares problem !

Reading of the course book of first Day Lecture:

Study Chapters 1, 2(2.1-2.5), 3, 4(4.1-4.3)
39/39

Delft Center for Systems and Control