Chapter 3.

Controllability and Observability

Modern Control Theory (Course Code: 10213403)

Professor Jun WANG (

Department of Control Science & Engineering School of Electronic & Information Engineering Tongji University

Spring semester, 2012

Outline
1

3.1 Introduction 3.2 Analysis of controllability 3.3 Analysis of observability 3.4 Principle of duality 3.5 Obtaining controllable and observable canonical forms 3.6 Canonical decomposition 3.7 Simulations with MATLAB

Outline
1

3.1 Introduction 3.2 Analysis of controllability 3.3 Analysis of observability 3.4 Principle of duality 3.5 Obtaining controllable and observable canonical forms 3.6 Canonical decomposition 3.7 Simulations with MATLAB

3.1 Introduction

Who introduced the concepts?

Rudolf E. Kalman
(Hungary, 1821- )
Nationality Hungarian-born American Fields Elec Engn; Mathematics; Applied Engn Systems Theory Institutions Stanford Univ; Univ of Florida; Swiss Federal Inst of Tech Alma mater MIT; Columbia Univ Notable awards IEEE Medal of Honor; National Medal of Science; Charles Stark Draper Prize ; Kyoto Prize
Prof J Wang (Tongji Uni)

Obama awards National Medals of Science, Techonology and Innovation

U.S. President Barack Obama (R) presents a 2008 National Medal of Science to Rudolf Kalman (L) of Swiss Federal Institute of Technology in Zurich during an East Room ceeremony October 7, 2009 at the White House in Washington, DC.

Chap 3. Controllability and observability

Spring 2012

4 / 52

3.1 Introduction

What are controllability and observability?

Controllability determines whether the state of a state equation can be controlled by the input. Observability determines whether the initial state can be observed from the output. The conditions of controllability and observability govern the existence of a solution to the control system design problem.

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

5 / 52

3.1 Introduction

How to understand the concepts?

Lets rst look at some examples Example (Simple circuits)
L R V (t ) R 1F R x R y V (t ) RC x2 I (t ) C x1 R R1 R2 C x1

x2

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

6 / 52

Example (A numerical example) Suppose a system can be described by the following state-space model 1 x x 4 0 1 + u 1 = 2 2 0 5 x2 x x1 y = 0 6 x2

What can we observe from these equations? to zero by a proper control input u;

1 = 4x1 + u x 2 = 5x2 + 2u x y = 6x
2

Both the state x1 and x2 can be moved from the their initial values The output y is associated with the state x2 , and is completely isolated from the state x1 . Therefore, the system is completely controllable, but NOT completely observable.

3.1 Introduction

Denitions of controllability and observability?

Denition (Controllability) A system is completely controllable if there exists an unconstrained control u(t) that can transfer any initial state x(t0 ) to any other desired location x(t) in a nite time, t0 Denition (Observability) A system is completely observable if, given the control u(t), every state x(t0 ) can be determined from the observation of y(t) over a nite time interval, t0 t T. t T.

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

8 / 52

Outline
1

3.1 Introduction 3.2 Analysis of controllability 3.3 Analysis of observability 3.4 Principle of duality 3.5 Obtaining controllable and observable canonical forms 3.6 Canonical decomposition 3.7 Simulations with MATLAB

How to judge the controllability of an LTI system?

Consider the continuous-time single-input system x = Ax + bu where the state x(t) Rn1 , control input u(t) R, and the matrices A Rnn and b Rn1 . The solution of the equation is x(t) = eAt x(0) +
t 0

eA(t) bu() d

Applying the denition of complete state controllability, we have x(t1 ) = 0 = eAt1 x(0) + or x(0) =
0 t1 t1 0

eA(t1 ) bu() d

eA bu() d

As the matrix exponential function eA can be written as

n1

eA =
k=0

k ()Ak

we have
t1 n1

x(0) =

0 k=0 n1 t1

k ()Ak bu() d k ()Ak bu() d

t1 0

=
k=0 n1 0

=
k=0 n1

Ak b

k ()u() d
t1 0

=
k=0

Ak bk , where k =

k ()u() d

we are now in a position to get the result by the following steps

n1

x(0) =
k=0

Ak bk

= b1 Ab2 An1 bn1 1 2 n 1 = b Ab A b . . . n1 If the system is completely state controllable, the above equation must be satised for any initial state x(0), which requires that rank b Ab An1 b = n It can be extended to a general case where u is a vector.

3.2 Analysis of controllability

Algebraic controllability criterion

Theorem The system (A, B) is completely controllable if and only if the rank of the controllability matrix Qc = B AB A2 B An1 B is n, i.e. rank(Qc ) = n

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

13 / 52

3.2 Analysis of controllability

Example Consider the system described by

1 x 2 x 3 x

2 2 0 0 0 1 0 3 4

x1 x2 x3

10 01 11

u1 u2

Solutions For this case, Qc = B 1 = 0 1 AB A2 B 0 2 2 1 1 1 1 4 7 2 13 2 25 4 7

Since rank(Qc ) = 3, the system is completely state controllable.

Prof J Wang (Tongji Uni) Chap 3. Controllability and observability Spring 2012 14 / 52

3.2 Analysis of controllability

Example Prove that an SISO system must be completely controllable if it can be described by a state-space model of the controllable canonical form. Solutions Since the model is in controllable canonical form, 1 0 0 . . . .. . . . . A= , b = 0 0 1 1 a0 a1 an1

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

15 / 52

3.2 Analysis of controllability

We have the controllability matrix Qc = b Ab An1 b 0 0 0 1 . . . 0 . . . 1 . .. .. . = . . . . . . . . . 0 1 1 It is clear that rankQc = n Therefore the system is completely controllable.

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

16 / 52

3.2 Analysis of controllability

PBH controllability criterion

Theorem The system (A, B) is completely state controllable if and only if all the eigenvalues i of the state matrix A satisfy rank [i I A, B] = n,
Prof J Wang (Tongji Uni)

i = 1, 2, , n
Spring 2012 17 / 52

Chap 3. Controllability and observability

3.2 Analysis of controllability

Corollary (Diagonal canonical criterion) If the eigenvalues of A are distinct and the corresponding diagonal canonical form after similarity transformation is (t) = P1 APx x (t) + P1 Bu(t) where .. . n

1 2 1 = P AP = A

The original system (A, B) is completely state controllable if and only if there = P1 B. is no zero row in B
Prof J Wang (Tongji Uni) Chap 3. Controllability and observability Spring 2012 18 / 52

3.2 Analysis of controllability

Corollary (Jordan canonical criterion) Suppose the system matrix A has repeated eigenvalues, and the corresponding Jordan canonical form after similarity transformation is (t) = P1 APx x (t) + P1 Bu(t) where = P1 AP = A J
1

J2 .. . Jl

, Ji =

i 1 i 1 .. .. . . i 1 i

and each distinct eigenvalue is associated with only one Jordan block. Then the original system (A, B) is completely state controllable if and only if = P1 B corresponding to the last row of each Jordan block is not the row of B zero row.
Prof J Wang (Tongji Uni) Chap 3. Controllability and observability Spring 2012 19 / 52

3.2 Analysis of controllability

Example Consider the system described by

1 x 2 x 3 x 4 x 5 x 6 x 1 1 1

x1 x x2 2 1 3 x4 2 1 x 5 2 x6 3

b21 b51 b61

b22 b52 b62

u1 u2

where denotes any real constant value. Solutions The system is completely state controllable if and only if
1 2 3

b21 and b22 are not all zero; b51 and b52 are not all zero; b61 and b62 are not all zero;
Prof J Wang (Tongji Uni) Chap 3. Controllability and observability Spring 2012 20 / 52

Outline
1

3.1 Introduction 3.2 Analysis of controllability 3.3 Analysis of observability 3.4 Principle of duality 3.5 Obtaining controllable and observable canonical forms 3.6 Canonical decomposition 3.7 Simulations with MATLAB

3.3 Analysis of observability

Algebraic observability criterion

Theorem The system (A, C) is completely observable if and only if the observability matrix C CA 2 CA Qo = . . . n1 CA or its transpose has rank n, i.e. rank(Qo ) = rank(QT o) = n

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

22 / 52

3.3 Analysis of observability

Example Consider the system described by

1 x 2 x

1 1 2 1 x1 x2

x1 x2

0 1

y = [1 0]

Is the system controllable and observable? Solutions Calculating the controllability and observability matrices yields Qc = [ b Ab ] = Qo = controllable and observable.
Prof J Wang (Tongji Uni) Chap 3. Controllability and observability Spring 2012 23 / 52

0 1 1 1 10 11

c cA

Since rank(Qc ) = rank(Qo ) = 2, the system is completely state

3.3 Analysis of observability

PBH observability criterion

Theorem The system (A, C) is completely state observable if and only if all the eigenvalues i of the state matrix A satisfy i I A = n, i = 1, 2, , n rank C

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

24 / 52

3.3 Analysis of observability

Corollary (Diagonal canonical criterion) If the eigenvalues of A are distinct and the corresponding diagonal canonical form after similarity transformation is (t) = P1 APx x (t) + P1 Bu(t) y(t) = CPx (t) + Du(t) where .. . n

1 2 1 P AP = A =

The original system (A, C) is completely state observable if and only if there is = CP. no zero column in C
Prof J Wang (Tongji Uni) Chap 3. Controllability and observability Spring 2012 25 / 52

3.3 Analysis of observability

Corollary (Jordan canonical criterion) If the system matrix A has repeated eigenvalues, and the corresponding Jordan canonical form after similarity transformation is (t) = P1 APx x (t) + P1 Bu(t) y(t) = CPx (t) + Du(t) where = P1 AP = A
J1 J2 .. . Jl

, Ji =

i 1 i 1 .. .. . . i 1 i

and each distinct eigenvalue is associated with only one Jordan block, then the original system (A, C) is completely state observable if and only if the column = CP corresponding to the rst column of each Jordan block is not zero of C column.
Prof J Wang (Tongji Uni) Chap 3. Controllability and observability Spring 2012 26 / 52

Example Are the following systems (A, B, C) state controllable and observable? (1) 1 : A = (2) 2 : A = (3) 3 : A = Solutions (1) 1 is uncontrollable, but observable; (2) 2 is controllable and observable; (3) How about 3 ? Can we get the answer directly? rank [ B AB A2 B ] = rank
015 137 112 1 0 0 4

,B=
0 0 2 0 0 0 0 1 2 0

0 1

4 0 0 0 0

1 4 0 0 0

0 0 0 , 1 2 0 1 1

, C = [1 2] B= , C = [1 0 1]

0 1 0 , 2 1

C=

10101 00110

210 021 102

,B=

= 3,

rank

C CA CA2

= rank

101 311 752

=3

3 is controllable and observable.

Outline
1

3.1 Introduction 3.2 Analysis of controllability 3.3 Analysis of observability 3.4 Principle of duality 3.5 Obtaining controllable and observable canonical forms 3.6 Canonical decomposition 3.7 Simulations with MATLAB

3.4 Principle of duality

Principle of duality
Denition (Dual systems) x (t) = Ax(t) + Bu(t) 1 : y(t) = Cx(t) x Rn , u Rl , y Rm . (t) = AT (t) + CT (t) 2 : (t) = BT (t)

Rn , Rm , Rl .

The LTI systems 1 (A, B, C) and 2 (AT , CT , BT ) are dual of each other and the dimensions of the input and the output are exchanged between dual systems. u
+

B
+

x A

BT

CT

AT
Chap 3. Controllability and observability Spring 2012 29 / 52

Prof J Wang (Tongji Uni)

Theorem (Principle of duality) The system 1 (A, B, C) is completely controllable (observable) if and only if its dual system 2 (AT , CT , BT ) are completely observable (controllable). For system 1 (A, B, C) Controllability criterion: rank B AB An1 B = n Observability criterion: C CA =n rank . . . CAn1 For system 2 (AT , CT , BT ) Controllability criterion: rank CT AT CT AT
n1

CT = n

Observability criterion: BT B T AT =n rank . . . n1 B T AT

Outline
1

3.1 Introduction 3.2 Analysis of controllability 3.3 Analysis of observability 3.4 Principle of duality 3.5 Obtaining controllable and observable canonical forms 3.6 Canonical decomposition 3.7 Simulations with MATLAB

3.5 Obtaining controllable and observable canonical forms

Controllable canonical form for SISO systems

Given a completely controllable SISO system (A, b, c), nd a similarity transformation matrix Pc , which can transform it into the controllable canonical form (Ac , bc , cc ), i.e.
1 Ac = P c APc , 1 bc = P c b,

cc = cPc 0 . . . bc = 0 1
Spring 2012 32 / 52

1 0 . .. . . . , Ac = 0 1 a0 a1 an1
Prof J Wang (Tongji Uni) Chap 3. Controllability and observability

From the denition, we have

1 Ac = P c APc

1 1 P c A = Ac P c

Expanding the above equation and comparing the two sizes, we have pc1 A = pc2 pc2 A = pc3 . . . pc(n1)A = pcn pcn A = a0 pc1 a1 pc2 an1 pcn If pc1 is known, then we can work out pc2 , pc3 , ,pcn successively.

1 Let pci denotes the ith row vector of P c . Hence 1 pc1 pc1 0 . . .. pc2 pc2 . . . A = . . . 0 . 1 . pcn pcn a0 a1 an1

1 Next, we will use equation bc = P c b to obtain pc1 . p c1 0 . pc2 . 1 b = . Pc b = bc . . 0 . pcn 1

Then we have pc1 b = 0 pc2 b = pc1 Ab = 0 . . . pc(n1) b = pc1 An2 b = 0 pcn b = pc1 An1 b = 1 pc1 b Ab A2 b An1 b = 0 0 0 1

3.5 Obtaining controllable and observable canonical forms

Therefore, we have pc1 = 0 0 0 1 b Ab A2 b An1 b

1

1 = 0 0 0 1 Q c

pc1 A = pc2 pc2 A = pc3 . . . pc(n1) A = pcn =

1 P c

p A c1 2 p A = c1 . . . n 1 pc1 A

pc1

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

35 / 52

3.5 Obtaining controllable and observable canonical forms

Similarity transformation matrix for controllable canonical form Given a completely controllable SISO system : (A, b, c), the following similarity transformation matrix Pc can transform into the controllable canonical form c : (Ac , bc , cc ). pc1 p A c1 2 1 p A Pc = c1 . . . n 1 pc1 A where
1 pc1 = 0 0 0 1 Q c

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

36 / 52

3.5 Obtaining controllable and observable canonical forms

Example Consider a system described by

1 x 2 x 3 x

1 0 0 1 2 0 0 5 0

x1 x2 x3

1 0 0

Find the similar transformation matrix Pc which can transform the system into the controllable canonical form. Solutions (1) Check the controllability of the system Qc = [ b Ab A2 b ] = rank(Qc ) = 3 The system is controllable and can be transformed into the controllable canonical form.
Prof J Wang (Tongji Uni) Chap 3. Controllability and observability Spring 2012 37 / 52

1 1 1 0 1 3 0 0 5

3.5 Obtaining controllable and observable canonical forms

(2) Compute pc1 .

1 Q c = 1 5 552 053 001

11 0 1 00

2 5 3 5 1 5

1 1 pc1 = [ 0 0 1 ] Q c = [0 0 5 ] 1 (3) Compute P c and Pc . 1 P c = pc1 pc1 A pc1 A2

1 0 0 5 0 1 0 1 2 0

Pc =

021 010 500

(4) Compute the controllable canonical form.

1 c = P A c APc = 0 1 0 0 0 1 0 2 3

1 bc = P c b=

0 0 1

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

38 / 52

3.5 Obtaining controllable and observable canonical forms

Observable canonical form for SISO systems

Similarity transformation matrix for observable canonical form Given a completely observable SISO system : (A, b, c), the following similarity transformation matrix Po can transform into the observable canonical form o : (Ao , bo , co ). Po = po1 Apo1 A2 po1 An1 po1 where 1 0 c 0 . . . cA . 1 . . = Qo = . 0 . . 0 1 cn1 A 1
Chap 3. Controllability and observability Spring 2012 39 / 52

po1

Prof J Wang (Tongji Uni)

Outline
1

3.1 Introduction 3.2 Analysis of controllability 3.3 Analysis of observability 3.4 Principle of duality 3.5 Obtaining controllable and observable canonical forms 3.6 Canonical decomposition 3.7 Simulations with MATLAB

3.6 Canonical decomposition

Decomposition according to controllability

Suppose an nth-order SISO system : (A, b, c) is not completely state controllable, say rank Qc = rank b Ab A2 b An1 b = n1 < n Dene a nonsingular matrix P = b Ab A2 b An1 1 b q1 qnn1 p1 p2 pn where (n n1 ) columns q1 , , qnn1 are entirely arbitrary as long as the matrix P is nonsingular.
Prof J Wang (Tongji Uni) Chap 3. Controllability and observability Spring 2012 41 / 52

3.6 Canonical decomposition

By using the similar matrix P dened above, the system can be ,c : (A ,b transformed into ). b x A A12 x c + c u c= c nc nc 0 x nc 0 A x x c y= c c c nc x nc c Rn1 1 , c c Rn1 n1 , b c R1n1 . where A c Note that, the n1 dimensional subsystem cu cx c = A x c + b y=c cx c is completely state controllable.
Prof J Wang (Tongji Uni) Chap 3. Controllability and observability Spring 2012 42 / 52

Features of controllability decomposition

: System
c x nc x

c: System
c A 12 A nc 0 A c nc ] x c x nc x c x nc

c b 0

cu cx c = A x c + b y=c cx c

c y = [c

c b

+ +

x c
+

c c
y1 y2
+

c A 12 A x nc
+

y
+

G (s) = G (s) = G c (s) c c )1 b =c c (sI A

c nc

nc A

3.6 Canonical decomposition

Example Consider the following state-space model x =

110 010 011

x+

0 1 0

y = [1 1 1]x Determine whether the system is controllable. If not, make a decomposition according to controllability. Solutions Since rank(Qc ) = 2 < 3, the system is not completely controllable. The similar matrix can be constructed as follows P = [ b Ab q1 ] =
Prof J Wang (Tongji Uni)

011 110 010

P1 =

0 1 1 00 1 1 0 1
Spring 2012 44 / 52

Chap 3. Controllability and observability

Letting x = Px yields = P1 AP = A
0 1 0 1 2 0 0 0 1

= P1 b = b

1 0 0

c = cP = [ 1 3 1 ]

The controllable subsystem is c = x

0 1 1 2

x c +

1 0

y = [1 3]x c The transfer function of the original system G (s) = [ 1 1 1 ]

s1 1 0 0 s1 0 0 1 s1 1 0 1 0

(s 1)(s + 1) s+1 = (s 1)3 (s 1)2 s+1 (s 1)2

The transfer function of the controllable subsystem is G c (s) = [ 1 3 ] It is apparent that G (s) = G c (s)
s 1 1 s2 1 1 0

Decomposition according to observability

Suppose an nth-order LTI system (A, b, c) is not completely state observable, i.e. rank Qo = rank
cT AT cT

(AT )

2 T c

(AT )

n 1 T c

= n1 < n

u The system can be transformed into

o x no x

o + b

c o
+

o A +
o b no b

o 0 A 21 A no A x o x no

x o x no

u no b
+

21 A

o 0 ] y = [c

no A

Canonical structure of system

If a system is neither completely controllable nor completely observable, the system can be decomposed as follows co 0 A 13 0 co A x x co x 21 A co 23 A 24 x co co A A = + x c 0 Ac 0 x o o o 0 c 43 A c c x c x 0 0 A o o o y= co 0 C c C o 0 x

co B co B u 0 0

where x co is controllable and observable, x co is controllable but not observable, x c o is not controllable, but observable, x c o is neither controllable nor observable.

3.6 Canonical decomposition

Block diagram of the canonical decomposition

co B u co B co 24 A c o co co )1 b G(s) = c co (sI A 21 A 23 A 43 A co 13 A c o c C o co C
+ +

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

48 / 52

Outline
1

3.1 Introduction 3.2 Analysis of controllability 3.3 Analysis of observability 3.4 Principle of duality 3.5 Obtaining controllable and observable canonical forms 3.6 Canonical decomposition 3.7 Simulations with MATLAB

3.7 Simulations with MATLAB

Matlab commands used in this chapter

Command canon ctrb obsv rank Description Canonical state-space realizations Compute the controllability matrix Compute the observability matrix Matrix rank

Prof J Wang (Tongji Uni)

Chap 3. Controllability and observability

Spring 2012

50 / 52

A numerical example
Example
Examine the controllability and observability of the system (A, B, C) where A = B=
0 0 1 0 1 0 0 0 1 6 11 6

, C = [ 5 6 1 ].

Solutions
The following MATLAB Program computes the controllability and observability of this system, which turns out to be controllable, but not observable.

>> A = [0 1 0;0 0 1;-6 -11 -6]; >> B = [0;0;1]; >> C = [5 6 1]; >> D = [0]; >> CONT = ctrb(A,B) CONT = 0 0 1 0 1 -6 1 -6 25 >> rank(CONT) ans = 3 >> OBSER = obsv(A,C) OBSER = 5 6 1 -6 -6 0 0 -6 -6 >> rank(OBSER) ans = 2

Chapter 3. Controllability and Observability

Modern Control Theory (Course Code: 10213403)

Professor Jun WANG (

Department of Control Science & Engineering School of Electronic & Information Engineering Tongji University

Spring semester, 2012

Go to next chapter!