EE 580: Linear Control Systems

Chapter 4: State-space solutions and realizations

Prof. Minghui Zhu

EE, Penn State
Outline

1 Continuous-time LTI systems

General solutions
Equivalent state-space equations
Realizations

2 Continuous-time LTV systems

General solutions
Equivalent time-varying equations
Time-varying realizations

EE580 2 / 42
General solutions

ẋ(t) = Ax(t) + Bu(t), y(t) = Cx(t) + Du(t) (1)

Fact: The solution of (1) is given by
Z t
At
x(t) = e x(0) + eA(t−τ ) Bu(τ )dτ, (2)
0
Z t
At
y(t) = Ce x(0) + C eA(t−τ ) Bu(τ )dτ + Du(t).
0
Verification one: Differentiating (2) yields
Z t
d  At
eA(t−τ ) Bu(τ )dτ

ẋ(t) = e x(0) +
dt 0
Z t
At
= Ae x(0) + AeA(t−τ ) Bu(τ )dτ + eA(t−τ ) Bu(τ )|τ =t
0
Z t
 At
eA(t−τ ) Bu(τ )dτ + eA·0 Bu(t)

= A e x(0) +
0
= Ax(t) + Bu(t).
EE580 3 / 42
General solutions (cont’)

Verification two: Premultiplying e−At on both sides of (1) yields

e−At ẋ(t) − e−At Ax(t) = e−At Bu(t),

which implies
d −At
x(t) = e−At Bu(t).


e
dt
Its integration from 0 to t yields
Z t
−Aτ
e t
x(τ )|τ =0 = e−Aτ Bu(τ )dτ ;
0

Z t
e−At x(t) − e0 x(0) = e−Aτ Bu(τ )dτ ;
0
Z t
At
x(t) = e x(0) + eA(t−τ ) Bu(τ )dτ.
0

EE580 4 / 42
Computation of eAt
Method one (Using Theorem 3.5 in Chen’s book): First compute the
eigenvalues of A; next, find a polynomial h(λ) of degree n − 1 that equals
eλt on the spectrum of A; then
eAt = h(A).

Method two (Using Jordan form of A): Let A = QÂQ−1 ; then

eAt = QeÂt Q−1

where  is in Jordan form and eÂt can be readily obtained by using (3.48)
in Chen’s book.
Method three (Using the infinite power series):
∞
t2 2 X 1
eAt = I + tA + A + ··· = t k Ak .
2! k!
k=0

Method four (Using inverse Laplace transformation):

eAt = L−1 (sI − A)−1 .
EE580 5 / 42
Example one
 
−1 0 −1
We use Methods 1 and 4 to compute (sI − A) where A = .
1 −2
Method 4:
 −1  
s 1 1 s+2 −1
(sI − A)−1 = = 2
−1 s+2 s + 2s + 1 1 s
" #
s+2 −1
(s+1)2 (s+1)2
= 1 s .
(s+1)2 (s+1)2

Method 1: The eigenvalues of A are −1 and −1. Let h(λ) = β0 + β1 λ. If h(λ)

equals f (λ) = (s − λ)−1 on the spectrum of A, then

f (−1) = h(−1), (s + 1)−1 = β0 − β1 ,

f 0 (−1) = h0 (−1), (s + 1)−2 = β1 .

Thus we have h(λ) = (s + 1)−1 + (s + 1)−2
+ (s + 1)−2 λ and

(sI − A)−1 = h(A) = (s + 1)−1 + (s + 1)−2 I + (s + 1)−2 A.

EE580 6 / 42
Example two
Consider the system
   
0 −1 0
ẋ(t) = x(t) + u(t).
1 −2 1

We first compute (sI − A)−1 as follows:

−1 " #
 s+2 −1
s 1 (s+1)2 (s+1)2
(sI − A)−1 = = 1 s .
−1 s+2 (s+1)2 (s+1)2

Then we have
(1 + t)e−t −te−t
 
eAt = L−1 (sI − A)−1 =
te−t (1 − t)e−t

and the solution

" Rt #
(1 + t)e−t −te−t −(t − τ )e−(t−τ ) u(τ )dτ
 
x(t) = x(0) + Rt 0 .
te−t (1 − t)e−t (1 − (t − τ ))e−(t−τ ) u(τ )dτ
0

EE580 7 / 42
Outline

1 Continuous-time LTI systems

General solutions
Equivalent state-space equations
Realizations

2 Continuous-time LTV systems

General solutions
Equivalent time-varying equations
Time-varying realizations

EE580 8 / 42
Example 4.4.1

Choose the inductor current x1 and capacitor voltage x2 as state variables

      
ẋ1 (t) 0 −1 x1 (t) 1
= + u(t)
ẋ2 (t) 1 −1 x2 (t) 0
y(t) = [0 1]x(t)
Choose the loop currents x̄1 and x̄2 as state variables
      
x̄˙ 1 (t) −1 1 x̄1 (t) 1
= + u(t)
x̄˙ 2 (t) −1 0 x̄2 (t) 1
y(t) = [1 − 1]x̄(t)
One circuit, two equivalent state-space equations.
EE580 9 / 42
Equivalence transformation

Consider the n-dimensional state-space equation:

ẋ(t) = Ax(t) + Bu(t)

y(t) = Cx(t) + Du(t). (3)

Let P be an n × n real non-singular matrix and let x̄ = P x. Then the

state-space equation

˙
x̄(t) = Āx̄(t) + B̄u(t)
y(t) = C̄ x̄(t) + D̄u(t), (4)

where Ā = P AP −1 , B̄ = P B, C̄ = CP −1 and D̄ = D, is said to be equivalent

to (3) and x̄ = P x is called an equivalence transformation.

EE580 10 / 42
Discussion

Equivalent state-space equations have the same characteristic polynomial

¯
∆(λ) = det(λI − Ā) = det(λP P −1 − P AP −1 ) = det(P (λI − A)P −1 )
= det(P )det(λI − A)det(P −1 ) = det(λI − A) = ∆(λ);

Equivalent state-space equations have the same set of eigenvalues;

Equivalent state-space equations have the same transfer matrix
ˆ (s) = C̄(sI − Ā)−1 B̄ + D̄ = CP −1 (P (sI − A)P −1 )−1 P B + D
Ḡ
= CP −1 P (sI − A)−1 P −1 P B = C(sI − A)−1 B + D = Ĝ(s);

In fact, all properties of (3) are preserved under any equivalence

transformation;
Matlab function [ab, bb, cb, db] = ss2ss(a, b, c, d, p) carries out equivalence
transformation.

EE580 11 / 42
Revisiting Example 4.4.1

    
x1 1 0 x̄1
=
x2 1 −1 x̄2
    
x̄1 1 0 x1
=
x̄2 1 −1 x2

EE580 12 / 42
Zero-state equivalence

Two state-space equations are said to be zero-state equivalent if they have

the same transfer matrix or

D + C(sI − A)−1 B = D̄ + C̄(sI − Ā)−1 B̄.

Theorem
Two LTI state-space equations {A, B, C, D} and {Ā, B̄, C̄, D̄} are zero-state
equivalent if and only if D = D̄, and CAm B = C̄ Ām B̄ for m = 0, 1, , 2, · · · .

This can be verified by using

∞
X
(s−1 A)k = I + s−1 A + s−2 A2 + · · · = (I − s−1 A)−1
k=0
1 1
D + C(I − s−1 A)−1 B = D + C(I + s−1 A + s−2 A2 + · · · )B.
s s

EE580 13 / 42
Example 4.4.2

The transfer functions for two circuits:

s
s+1 1 1
ĝ1 (s) = s = , ĝ2 (s) = .
s + s+1 s+2 s+2
The state-space equation for the LHS circuit:
   
−1 1 1
ẋ(t) = x(t) + u(t) = Ax(t) + Bu(t)
1 −1 0
y(t) = [1 − 1]x(t) = Cx(t) + D · u(t)
The state-space equation for the RHS circuit:
˙
x̄(t) = −2x̄(t) + u(t) = Āx̄(t) + B̄u(t)
y(t) = x̄(t) = C̄ x̄(t) + D̄u(t)
EE580 14 / 42
Example 4.4.2 (cont’)

The RHS circuit: D̄ = 0 and C̄ Ām B̄ = (−2)m

The LHS circuit: D = 0 and
  
−1 1 1
CAm B = [1 − 1](−2)m−1 = (−2)m .
1 −1 0

So two state-space equations are zero-state equivalent.

Remarks:
The eigenvalues of A are 0 and −2;
Let f (λ) = λm and h(λ) = β0 + β1 λ;
On the spectrum of A, we have f (0) = 0 = h(0) = β0 and
f (−2) = (−2)m = h(−2) = 0 + β1 (−2) which implies β1 = (−2)m−1 ;
Thus we have Am = β0 I + β1 A = (−2)m−1 A for any positive integer m.

EE580 15 / 42
Outline

1 Continuous-time LTI systems

General solutions
Equivalent state-space equations
Realizations

2 Continuous-time LTV systems

General solutions
Equivalent time-varying equations
Time-varying realizations

EE580 16 / 42
Realizability

Every SISO LTI system can be described by the transfer function

ŷ(s) = ĝ(s)û(s),

and the state-space equation:

ẋ(t) = Ax(t) + bu(t)

y(t) = cx(t) + du(t).

If the state-space equation is known, the transfer matrix can be computed

as ĝ(s) = c(sI − A)−1 b + d;
A transfer function ĝ(s) is said to be realizable if there is a state-space
equation, or simply {A, b, c, d} such that ĝ(s) = c(sI − A)−1 b + d and
{A, b, c, d} is called a realization of ĝ(s);
LTI distributed systems are not realizable;
Realizations are not unique if exist.

EE580 17 / 42
Realizability theorem

Theorem
A transfer function ĝ(s) is realizable if and only if ĝ(s) is a proper rational
function.

Sufficiency:
ĝsp (s) = c(sI − A)−1 b = 1
det(sI−A) c[adj(sI − A)]b;
det(sI − A) has degree n;
Each entry of adj(sI − A) is the determinant of an (n − 1) × (n − 1)
submatrix of sI − A and has at most degree n − 1;
c(sI − A)−1 b is a strictly proper rational function;
If d 6= 0, then c(sI − A)−1 b + d is proper.

EE580 18 / 42
Realizability theorem (cont’)

Necessity: We consider a special case but the procedure applies to any proper
rational transfer function.
b̄1 s4 + b̄2 s3 + b̄3 s2 + b̄4 s + b̄5
ĝ(s) =
ā1 s4 + ā2 s3 + ā3 s2 + ā4 s + ā5
b1 s3 + b2 s2 + b3 s + b4
=d+ 4 .
s + a2 s3 + a3 s2 + a4 s + a5
The controllable-form realization is
   
−a2 −a3 −a4 −a5 1
 1 0 0 0   0 
ẋ(t) = 
 0
 x(t) +   u(t),
1 0 0   0 
0 0 1 0 0
y(t) = [b1 b2 b3 b4 ]x(t) + du(t), (5)

where x(t) = [x1 (t) x2 (t) x3 (t) x4 (t)]T .

EE580 19 / 42
Realizability theorem (cont’)

We proceed to show (5) is a realization of ĝ(s).

Applying the Laplace transform and assuming zero initial conditions yield

sx̂1 (s) = −a2 x̂1 (s) − a3 x̂2 (s) − a4 x̂3 (s) − a5 x̂4 (s) + û(s)
sx̂2 (s) = x̂1 (s), sx̂3 (s) = x̂2 (s), sx̂4 (s) = x̂3 (s).
x̂1 (s) x̂1 (s) x̂1 (s)
So we have x̂2 (s) = s , x̂3 (s) = s2 , x̂4 (s) = s3 . This implies

s3
x̂1 (s) = û(s),
s4 + a2 s3 + a3 s2 + a4 s + a5
b1 s3 + b2 s2 + b3 s + b4

ŷ(s) = d + 4 û(s).
s + a2 s3 + a3 s2 + a4 s + a5

EE580 20 / 42
Realizability theorem (cont’)

Necessity:

ĝ(s) = ĝ(s)T = [c(sI − A)−1 b + d]T

= bT (sT I T − AT )−1 cT + dT
= bT (sI − AT )−1 cT + d.

The observable-form realization is

   
−a2 1 0 0 b1
 −a3 0 1 0   b2 
ẋ(t) = 
 −a4 0 0
 x(t) +   u(t),
1   b3 
−a5 0 0 0 b4
y(t) = [1 0 0 0]x(t) + du(t). (6)

EE580 21 / 42
Example

3s4 + 5s3 + 24s2 + 23s − 5

ĝ(s) =
2s4 + 6s3 + 15s2 + 12s + 5
1.5s4 + 2.5s3 + 12s2 + 11.5s − 2.5
=
s4 + 3s3 + 7.5s2 + 6s + 2.5
−2s3 + 0.75s2 + 2.5s − 6.25
= 4 + 1.5.
s + 3s3 + 7.5s2 + 6s + 2.5
Its controllable-form realization is
   
−3 −7.5 −6 −2.5 1
 1 0 0 0   0 
ẋ(t) = 
 0
 x(t) +   u(t)
1 0 0   0 
0 0 1 0 0
y(t) = [−2 0.75 2.5 − 6.25]x(t) + 1.5u(t).

EE580 22 / 42
Example (cont’)

3s4 + 5s3 + 24s2 + 23s − 5

ĝ(s) =
2s4 + 6s3 + 15s2 + 12s + 5
1.5s4 + 2.5s3 + 12s2 + 11.5s − 2.5
=
s4 + 3s3 + 7.5s2 + 6s + 2.5
−2s3 + 0.75s2 + 2.5s − 6.25
= 4 + 1.5.
s + 3s3 + 7.5s2 + 6s + 2.5
Its observable-form realization is
   
−3 1 0 0 −2
 −7.5 0 1 0 
 x(t) +  0.75  u(t)
 
ẋ(t) = 
 −6 0 0 1   2.5 
−2.5 0 0 0 −6.25
y(t) = [1 0 0 0]x(t) + 1.5u(t).

Remark: The Matlab function [A, B, C, D] = tf 2ss(b, a) returns the

controllable-form realization given a transfer function.
EE580 23 / 42
Outline

1 Continuous-time LTI systems

General solutions
Equivalent state-space equations
Realizations

2 Continuous-time LTV systems

General solutions
Equivalent time-varying equations
Time-varying realizations

EE580 24 / 42
Solutions of linear systems

Scalar LTI systems: ẋ(t) = ax(t) and its solution is x(t) = eat x(0)

Vector LTI systems: ẋ(t) = Ax(t) and its solution is x(t) = eAt x(0) by
using the property AeAt = eAt A
Remark: In general, AB 6= BA and e(A+B)t 6= eAt eBt
Rt
a(τ )dτ
Scalar LTV systems: ẋ(t) = a(t)x(t) and its solution is x(t) = e 0 x(0)
Rt
A(τ )dτ
Vector LTV systems: ẋ(t) = A(t)x(t) and usually x(t) = e 0 x(0) is
not its solution

t
1 t
Rt
Z Z Z t
A(τ )dτ
e 0 =I+ A(τ )dτ + A(τ )dτ A(s)ds + · · ·
0 2 0 0
Z t
1 t
Z
d R t A(τ )dτ 1
e0 = A(t) + A(t) A(s)ds + A(τ )dτ A(t) + · · ·
dt 2 0 2 0
Rt
A(τ )dτ
6= A(t)e 0

EE580 25 / 42
Fundamental matrix

ẋ(t) = A(t)x(t), x(t) ∈ Rn (7)

For i = 1, · · · , n, pick initial state x[i] (t0 ) and denote its solution as x[i] (t).
Denote X(t) = [x[1] (t) x[2] (t) · · · x[n] (t)] and then Ẋ(t) = A(t)X(t).
If X(t0 ) is non-singular, then X(t) is called a fundamental matrix of (7).
Since one can arbitrarily choose non-singular X(t0 ), the fundamental
matrix is not unique.

EE580 26 / 42
Example

 
0 0
ẋ(t) = x(t) (8)
t 0

The solution of ẋ1 (t) = 0 is x1 (t) = x1 (0).

The solution of ẋ2 (t) = tx1 (t) = tx1 (0) is x2 (t) = 12 t2 x1 (0) + x2 (0).
   
1 1
The solution of (8) when starting from x(0) = is x(t) = 1 2 .
0 t
   2 
1 1
The solution of (8) when starting from x(0) = is x(t) = 1 2 .
2 2t + 2
Since these
 two initial states
 are linearly independent, so
1 1
X(t) = 1 2 1 2 is a fundamental matrix.
2t 2t + 2

EE580 27 / 42
Non-singularity of fundamental matrix

Fact: The fundamental matrix X(t) is non-singular for all t.

Verification by contradiction:
If x(t0 ) = 0, then x(t) ≡ 0 for t ≥ t0 is the unique solution.
Assume X(t) is singular at some t1 > t0 .
There is a non-zero vector v such that x(t1 ) = X(t1 )v = 0.
Since x(t1 ) = 0, then x(t) ≡ 0 for all t ≥ t0 is the unique solution.
x̄(t) = X(t)v = 0 for all t ≥ t0 , in particular, at t0 .
This contradicts the assumption that X(t0 ) is non-singular.
X(t) is non-singular for all t ≥ t0 .

EE580 28 / 42
State transition matrix
Let X(t) be any fundamental matrix of ẋ(t) = A(t)x(t). Then

Φ(t, t0 ) = X(t)X(t0 )−1

is called the state transition matrix of ẋ(t) = A(t)x(t). The transition

matrix is also the unique solution of
∂
Φ(t, t0 ) = A(t)Φ(t, t0 )
∂t
with the initial condition Φ(t0 , t0 ) = I.
Three important properties:

Φ(t, t) = I
Φ(t, t0 )−1 = [X(t)X(t0 )−1 ]−1 = X(t0 )X(t)−1 = Φ(t0 , t)
Φ(t, t0 ) = Φ(t, t1 )Φ(t1 , t0 )

for every t, t0 and t1 .

EE580 29 / 42
Example

Consider the homogeneous equation

 
0 0
ẋ(t) = x(t).
t 0
 
1 1
X(t) = 1 2 1 2 is a fundamental matrix.
2t 2t + 2
 1 2
t + 1 − 12

Its inverse is X(t)−1 = 4 1 2 1 .
−4t 2
The state transition matrix is given by 
1 0
Φ(t, t0 ) = X(t)X(t0 )−1 = 1 2 2 .
2 (t − t0 ) 1

EE580 30 / 42
General solutions

ẋ(t) = A(t)x(t) + B(t)u(t), y(t) = C(t)x(t) + D(t)u(t) (9)

Fact: The solution of (9) is given by:

Z t
x(t) = Φ(t, t0 )x(t0 ) + Φ(t, τ )B(τ )u(τ )dτ
t0
Z t
y(t) = C(t)Φ(t, t0 )x(t0 ) + C(t) Φ(t, τ )B(τ )u(τ )dτ + D(t)u(t)
t0

where Φ(t, τ ) is the state transition matrix of ẋ(t) = A(t)x(t)

EE580 31 / 42
Solution verification

Z t0
x(t0 ) = Φ(t0 , t0 )x(t0 ) + Φ(t, τ )B(τ )u(τ )dτ = Ix(t0 ) + 0 = x(t0 )
t0
Z t
d ∂ ∂
x(t) = Φ(t, t0 )x(t0 ) + Φ(t, τ )B(τ )u(τ )dτ
dt ∂t ∂t t0
Z t
∂
= A(t)Φ(t, t0 )x(t0 ) + Φ(t, t)B(t)u(t) + (Φ(t, τ )B(τ ))u(τ )dτ
t0 ∂t
Z t
= A(t)Φ(t, t0 )x(t0 ) + Φ(t, t)B(t)u(t) + A(t)Φ(t, τ )B(τ )u(τ )dτ
t0
= A(t)x(t) + B(t)u(t)

EE580 32 / 42
Computing state transition matrices

Computing state transition matrix is generally difficult other than the following
special cases.
Assume A(t) has the following commutative property:
Z t

Z t

A(t) A(τ )dτ = A(τ )dτ A(t)
t0 t0

for all t0 and t; e.g., when A(t) is diagonal or constant. Then we have
∞ Z t
Rt
A(τ )dτ
X 1
k
Φ(t, t0 ) = e t0
= A(τ )dτ .
k! t0
k=0

For A(t) constant, we further have

Φ(t, τ ) = eA(t−τ ) = Φ(t − τ ).

EE580 33 / 42
Outline

1 Continuous-time LTI systems

General solutions
Equivalent state-space equations
Realizations

2 Continuous-time LTV systems

General solutions
Equivalent time-varying equations
Time-varying realizations

EE580 34 / 42
Equivalence transformation

Consider the n-dimensional state-space equation:

ẋ(t) = A(t)x(t) + B(t)u(t)

y(t) = C(t)x(t) + D(t)u(t). (10)

Let P (t) be an n × n real non-singular matrix with P (t) and Ṗ (t) being
continuous and let x̄(t) = P (t)x(t). Then the state-space equation

˙
x̄(t) = Ā(t)x̄(t) + B̄(t)u(t)
y(t) = C̄(t)x̄(t) + D̄(t)u(t), (11)

where Ā(t) = (P (t)A(t) + Ṗ (t))P (t)−1 , B̄(t) = P (t)B(t), C̄(t) = C(t)P (t)−1
and D̄(t) = D(t), is said to be equivalent to (10) and x̄(t) = P (t)x(t) is called
an equivalence transformation.

EE580 35 / 42
Fundamental matrix

Fact: If X is a fundamental matrix of (10), then X̄(t) = P (t)X(t) is a

fundamental matrix of (11).
Verification:
d
(P (t)X(t)) = Ṗ (t)X(t) + P (t)Ẋ(t) = Ṗ (t)X(t) + P (t)A(t)X(t)
dt
= (Ṗ (t) + P (t)A(t))(P (t)−1 P (t))X(t) = Ā(t)(P (t)X(t)).

EE580 36 / 42
Outline

1 Continuous-time LTI systems

General solutions
Equivalent state-space equations
Realizations

2 Continuous-time LTV systems

General solutions
Equivalent time-varying equations
Time-varying realizations

EE580 37 / 42
Realizability

Every LTV system can be described by the input-output relation

Z t
y(t) = G(t, τ )u(τ )dτ,
t0

and the state-space equation:

ẋ(t) = A(t)x(t) + B(t)u(t)

y(t) = C(t)x(t) + D(t)u(t).

If the state-space equation is known, the impulse response matrix can be

computed as G(t, τ ) = C(t)X(t)X(τ )−1 B(τ ) + D(t)δ(t − τ ) for t ≥ τ .
An impulse response matrix G(t, τ ) is said to be realizable if there is
{A(t), B(t), C(t), D(t)} such that
G(t, τ ) = C(t)X(t)X(τ )−1 B(τ ) + D(t)δ(t − τ ) for t ≥ τ and
{A(t), B(t), C(t), D(t)} is called a realization of G(t, τ ).

EE580 38 / 42
Realizability theorem

Theorem
An impulse response matrix G(t, τ ) is realizable if and only if G(t, τ ) can be
decomposed as G(t, τ ) = M (t)N (τ ) + D(t)δ(t − τ ).

Necessity: Let M (t) = C(t)X(t) and N (τ ) = X(τ )−1 B(τ )

Sufficiency: We proceed to verify the following is a realization:

ẋ(t) = N (t)u(t)
y(t) = M (t)x(t) + D(t)u(t). (12)

A fundamental matrix of ẋ = 0 · x is X(t) = I. The impulse response matrix

of (12) is
M (t)I · I −1 N (τ ) + D(t)δ(t − τ )
which equals G(t, τ ).

EE580 39 / 42
Example

Consider g(t, τ ) = (t − τ )eλ(t−τ ) . It is easy to verify

−τ e−λτ
 
λt λt
g(t − τ ) = [e te ] .
e−λτ

Thus the two-dimensional time-varying state-space equation

−te−λt
 
ẋ(t) = 0x(t) + u(t)
e−λt
y(t) = [eλt teλt ]x(t)

is a realization of the impulse response of g(t) = teλt .

EE580 40 / 42
Appendix: Op-Amp circuit implementation

EE580 41 / 42
Appendix: Op-Amp circuit implementation
(cont’)

      
ẋ1 (t) −0.3 2 x1 (t) −2
= + u(t)
ẋ2 (t) −8 1 x2 (t) 0
 
x1 (t)
y(t) = [−2 3] + 5u(t).
x2 (t)
EE580 42 / 42