Digital Control Module 7 Lecture 3

Module 7: Discrete State Space Models

Lecture Note 3

1 Characteristic Equation, eigenvalues and eigen vectors

For a discrete state space model, the characteristic equation is defined as
|zI − A| = 0
The roots of the characteristic equation are the eigenvalues of matrix A.
1
1. If det(A) 6= 0, i.e., A is nonsingular and λ1 , λ2 , · · · , λn are the eigenvalues of A, then, λ1
,
1
λ2
, · · · , λ1n will be the eigenvalues of A−1 .

2. Eigenvalues of A and AT are same when A is a real matrix.

3. If A is a real symmetric matrix then all its eigenvalues are real.
The n × 1 vector vi which satisfies the matrix equation
Avi = λi vi (1)
where λi , i = 1, 2, · · · , n denotes the ith eigenvalue, is called the eigen vector of A associated
with the eigenvalue λi . If eigenvalues are distinct, they can be solved directly from equation (1).

Properties of eigen vectors

1. An eigen vector cannot be a null vector.
2. If vi is an eigen vector of A then mvi is also an eigen vector of A where m is a scalar.
3. If A has n distinct eigenvalues, then the n eigen vectors are linearly independent.
Eigen vectors of multiple order eigenvalues

When the matrix A an eigenvalue λ of multiplicity m, a full set of linearly independent may not
exist. The number of linearly independent eigen vectors is equal to the degeneracy d of λI − A.
The degeneracy is defined as
d=n−r
where n is the dimension of A and r is the rank of λI − A. Furthermore,
1≤d≤m

I. Kar 1
Digital Control Module 7 Lecture 3

2 Similarity Transformation and Diagonalization

Square matrices A and Ā are similar if

AP = P Ā
or, Ā = P −1 AP
and, A = P ĀP −1

The non-singular matrix P is called similarity transformation matrix. It should be noted that
eigenvalues of a square matrix A are not altered by similarity transformation.
Diagonalization:
If the system matrix A of a state variable model is diagonal then the state dynamics are
decoupled from each other and solving the state equations become much more simpler.

In general, if A has distinct eigenvalues, it can be diagonalized using similarity transforma-

tion. Consider a square matrix A which has distinct eigenvalues λ1 , λ2 , . . . λn . It is required
to find a transformation matrix P which will convert A into a diagonal form
 
λ1 0 . . . 0
Λ =  0 λ2 . . . 0 
0 0 . . . λn
through similarity transformation AP = P Λ. If v1 , v2 , . . . , vn are the eigenvectors of matrix
A corresponding to eigenvalues λ1 , λ2 , . . . λn , then we know Avi = λi vi . This gives
 
λ1 0 . . . 0
A [v1 v2 . . . vn ] = [v1 v2 . . . vn ]  0 λ2 . . . 0 
0 0 . . . λn

Thus P = [v1 v2 . . . vn ]. Consider the following state model.

x(k + 1) = Ax(k) + Bu(k)

If P transforms the state vector x(k) to z(k) through the relation

x(k) = P z(k), or, z(k) = P −1 x(k)

then the modified state space model becomes

z(k + 1) = P −1 AP z(k) + P −1 Bu(k)

where P −1 AP = Λ.

3 Computation of Φ(t)
We have seen that to derive the state space model of a sampled data system, we need to know
the continuous time state transition matrix Φ(t) = eAt .

I. Kar 2
Digital Control Module 7 Lecture 3

3.1 Using Inverse Laplace Transform

For the system ẋ(t) = Ax(t) + Bu(t), the state transition matrix eAt can be computed as,
eAt = L −1 (sI − A)−1


3.2 Using Similarity Transformation

If Λ is the diagonal representation of the matrix A, then Λ = P −1 AP . When a matrix is in
diagonal form, computation of state transition matrix is straight forward:
 λt 
e 1 0 ... 0
eΛt =  0 eλ2 t . . . 0 
0 0 . . . e λn t
Given eΛt , we can show that
eAt = P eΛt P −1

1
eAt = I + At + A2 t2 + . . .
 2! 
−1 At −1 1 22
⇒P e P = P I + At + A t + . . . P
2!
1
= I + P −1 AP t + P −1 AP P −1 AP t2 + . . .
2!
1 22
= I + Λt + Λ t + . . .
2!
Λt
= e
⇒e At
= P eΛt P −1

3.3 Using Caley Hamilton Theorem

Every square matrix A satisfies its own characteristic equation. If the characteristic equation is
△(λ) = |λI − A| = λn + α1 λn−1 + · · · + αn = 0
then,
△(A) = An + α1 An−1 + · · · + αn I = 0
Application: Evaluation of any function f (λ) and f (A)
f (λ) = a0 + a1 λ + a2 λ2 + · · · + an λn + · · · order ∞
f (λ) g(λ)
= q(λ) +
△(λ) △(λ)

f (λ) = q(λ)△(λ) + g(λ)

= g(λ)
= β0 + β1 λ + · · · + βn−1 λn−1 order n − 1

I. Kar 3
Digital Control Module 7 Lecture 3

If A has distinct eigenvalues λ1 , · · · , λn , then,

f (λi ) = g(λi ), i = 1, · · · , n

The solution will give rise to β0 , β1 , · · · , βn−1 , then

f (A) = β0 I + β1 A + · · · + βn−1 An−1

If there are multiple roots (multiplicity = 2), then

f (λi ) = g(λi ) (2)

∂ ∂
f (λi ) = g(λi ) (3)
∂λi ∂λi
Example 1:  
0 0 −2
If A = 0 1 0 
1 0 3
then compute the state transition matrix using Caley Hamilton Theorem.

λ 0 2
△(λ) = |λI−A| = 0 λ−1 0 = (λ−1)2 (λ−2) = 0 ⇒ λ1 = 1 (with multiplicity 2), λ2 = 2
−1 0 λ−3

Let f (λ) = eλt and g(λ) = β0 + β1 λ + β2 λ2

Then using (2) and (3), we can write

f (λ1 ) = g(λ1 )
∂ ∂
f (λ1 ) = g(λ1 )
∂λ1 ∂λ1
f (λ2 ) = g(λ2 )

This implies

et = β0 + β1 + β2 (λ1 = 1)
tet = β1 + 2β2 (λ1 = 1)
e2t = β0 + 2β1 + 4β2 (λ2 = 2)

Solving the above equations

β0 = e2t − 2tet , β1 = 3tet + 2et − 2e2t , β2 = e2t − et − tet

Then

eAt = β0 I + β1 A + β2 A2
2e − e2t 0 2et − 2e2t
 t 

=  0 et 0 
2t t 2t t
e −e 0 2e − e

I. Kar 4
Digital Control Module 7 Lecture 3

 
1 1
Example 2 For the system ẋ(t) = Ax(t) + Bu(t), where A = . compute eAt using 3
−1 1
different techniques.

Solution: Eigenvalues of matrix A are 1 ± j1.

Method 1
 −1
At −1 −1 s − 1 −1
−1
e = L (sI − A) = L =
1 s−1
 
1 s−1 1
= L −1 2
s − 2s + 2 −1 s − 1
" #
s−1 1
−1 (s−1)2 +1 (s−1)2 +1
=L −1 s−1
(s−1)2 +1 (s−1)2 +1

et cos t et sin t
 
=
−et sin t et cos t
Method 2
e(1+j)t
 
At Λt Λt 0
e = Pe P −1
where e = . Eigen values are 1 ± j. The corresponding
0 e(1−j)t
eigenvectors are found by using equation Avi = λi vi as follows:
    
1 1 v1 v
= (1 + j) 1
−1 1 v2 v2
 
1
Taking v1 = 1, we get v2 = j. So, the eigenvector corresponding to 1 + j is and the one
  j
1
corresponding to 1 − j is . The transformation matrix is given by
−j
   
1 1 −1 1 1 −j
P = [v1 v2 ] = ⇒P =
j −j 2 1 j
Now,
eAt = P eΛt P −1
   (1+j)t  
1 1 1 e 0 1 −j
=
2 j −j 0 e(1−j)t 1 j
 (1+j)t
e(1−j)t
 
1 e 1 −j
=
2 je(1+j)t −je(1−j)t 1 j
 (1+j)t
+ e(1−j)t
−j e(1+j)t − e(1−j)t


1 e
=
2 j e(1+j)t − e(1−j)t e(1+j)t + e(1−j)t
2et cos t −j(j)et 2 sin t
 
1
=
2 et (j)(j)2 sin t 2et cos t
 t
e cos t et sin t

=
−et sin t et cos t

I. Kar 5
Digital Control Module 7 Lecture 3

Method 3: Caley Hamilton Theorem

The eigenvalues are λ1,2 = 1 ± j.

eλ1 t = β0 + β1 λ1
eλ2 t = β0 + β1 λ2

Solving,
1 1
β0 = (1 + j)e(1+j)t + (1 − j)e(1−j)t
2 2
1  (1+j)t t

β1 = e − e(1−j)
2j
Hence,

eAt = β0 I + β1 A
 t
e cos t et sin t

=
−et sin t et cos t

We will now show through an example how to derive discrete state equation from a contin-
uous one.

Example: Consider the following state model of a continuous time system.

   
1 1 0
ẋ(t) = x(t) + u(t)
0 2 1
y(t) = x1 (t)

If the system is under a sampling process with period T , derive the discrete state model of
the system.
To derive the discrete state space model, let us first compute the state transition matrix of
the continuous time system using Caley Hamilton Theorem.

λ − 1 −1
△(λ) = |λI − A| = = (λ − 1)(λ − 2) = 0 ⇒ λ1 = 1, λ2 = 2
0 λ−2
Let f (λ) = eλt
This implies

et = β0 + β1 (λ1 = 1)
e2t = β0 + 2β1 (λ2 = 2)

Solving the above equations

β1 = e2t − et β0 = 2et − e2t

I. Kar 6
Digital Control Module 7 Lecture 3

Then

eAt = β0 I + β1 A
 t 2t
e e − et

=
0 e2t

Thus the discrete state matrix A is given as

eT e2T − eT
 
A = Φ(T ) =
0 e2T

The discrete input matrix B can be computed as

Z T  
′ 0
B = Θ(T ) = Φ(T − t ) dt′
0 1
Z T  T −t′ 2T −2t′ ′  
e .e e .e − eT .e−t 0
= 2T −2t′ dt′
0 e .e 1
0 T
e − 1 0.5e2T − eT + 0.5 0
 
=
0 0.5e2T − 0.5 1
0.5e2T − eT + 0.5
 
=
0.5e2T − 0.5

The discrete state equation is thus described by

 T 2T
e e − eT 0.5e2T − eT + 0.5
  
x((k + 1)T ) = x(kT ) + u(kT )
0 e2T 0.5e2T − 0.5
 
y(kT ) = 1 0 x(kT )

When T = 1, the state equations become

   
2.72 4.67 1.48
x(k + 1) = x(k) + u(k)
0 7.39 3.19
 
y(k) = 1 0 x(k)

I. Kar 7