Lecture 3 State-Space Solutions and Realization - v1

State-Space Solutions and Realization
Dr. Wenjie Dong
Department of Electrical and Computer Engineering,

The University of Texas Rio Grande Valley
W. Dong (UTRGV, Dept. of ECE) 1 / 75

Solution of LTI State Equations
Consider the linear time-invariant (LTI) state-space equation
ẋ(t) = Ax(t) + Bu(t) (1)
y (t) = Cx(t) + Du(t) (2)
where A, B, C , and D are, respectively, n × n, n × p, q × n, and q × p
constant matrices. The problem is to find the solution excited by the
initial state x(0) and the input u(t).
d −At
(e x(t)) = e −At Bu(t)
Z t dt Z t
−Aτ
d(e x(τ )) = e −Aτ Bu(τ )dτ
0 0
Z t
At
x(t) = e x(0) + e A(t−τ ) Bu(τ )dτ
0
Z t
y (t) = Ce At x(0) + C e A(t−τ ) Bu(τ )dτ + Du(t)
0

The solution of (1)-(2) can be computed with the aid of Laplace transform.
x̂(s) = (sI − A)−1 [x(0) + B û(s)]

ŷ (s) = C (sI − A)−1 [x(0) + B û] + D û(s)
x(t) = L−1 [x̂(s)] = L−1 [(sI − A)−1 [x(0) + B û(s)]]
= L−1 [(sI − A)−1 ]x(0) + L−1 [(sI − A)−1 B û(s)]
y (t) = L−1 [ŷ (s)]
One has
e At = L−1 [(sI − A)−1 ]

Methods for finding e At :

1 First, compute the eigenvalues of A; next, find a polynomial h(λ) of
degree n − 1 that equals e λt on the spectrum of A; then e At = h(A).
2 Using Jordan form of A: Let A = Q ÂQ −1 ; then e At = Qe Ât Q −1 ,
where Â is in Jordan form and e Ât can readily be obtained.
3 e At = L−1 [(sI − A)−1 ].

Example
Find e At , where
0 −1
A=
1 −2

Example
Solution: Method 1: The eigenvalues of A are −1, −1. Let

h(λ) = β0 + β1 λ. If h(λ) equals f (λ) = e λt on the spectrum of A, then
f (−1) = h(−1) : e −t = β0 − β1
f 0 (−1) = h0 (−1) : te −t = β1
Thus we have
h(λ) = e −t + te −t + te −t λ
So
e At = h(A) = (e −t + te −t )I + te −t A
−t
e + te −t −te −t

=
te −t e −t − te −t

Example
Method 2:
−1
−1 s 1 1 s + 2 −1
(sI − A) = = 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
(1 + t)e −t −te −t

At −1 −1
e = L [(sI − A) ]=
te −t (1 − t)e −t

Example
Method 3: The eigenvalues of A are −1, −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
" #
s+2 1
(s+1)2
− (s+1) 2
= 1 s
(s+1)2 s+1)2
(1 + t)e −t −te −t

−1 −1
L [(sI − A) ] =
te −t (1 − t)e −t
Example
Find the solution of

0 −1 0
ẋ = x+ u
1 −2 1

Example
Solution:
Z t
At
x(t) = e x(0) + e A(t−τ ) Bu(τ )dτ
0
(1 + t)e −t −te −t

= x(0)
te −t (1 − t)e −t
Rt
− 0 (t − τ )e −(t−τ ) u(τ )dτ

+ R t −(t−τ ) u(τ )dτ
0 [1 − (t − τ )]e

Based on the solution of a system, the following conclusions can be

obtained:
1 If A has an eigenvalue, simple or repeated, with a negative real part,
then every zero input response will approach zero as t → ∞. If A has
an eigenvalue, simple or repeated, with a positive real part, then most
zero-input responses will grow unbounded as t → ∞.
2 If A has some eigenvalues with zero real part and all with index 1 and
the remaining eigenvalues all have negative real parts, then no
zero-input response will grow unbounded. However, if the index is 2
or higher, then some zero-input response may become unbounded.

Discretization
Consider the continuous-time state equation
ẋ(t) = Ax(t) + Bu(t) (3)
y (t) = Cx(t) + Du(t) (4)
If the set of equations is to be computed on a digital computer, it must be
discretized.
Method 1:
x(t + T ) − x(t)
ẋ(t) = lim
T →0 T
x(t + T ) = x(t) + Ax(t)T + Bu(t)T (5)
If t = kT for k = 0, 1, . . ., then
x((k + 1)T ) = (I + TA)x(kT ) + TBu(kT )
y (kT ) = Cx(kT ) + Du(kT )
This discretization is the easiest to carry out but yields the least accurate
results for the same T .
Discretization
Method 2: Let
u(t) = u(kT ) =: u[k], x(t) = x(kT ) =: x[k] for kT ≤ t ≤ (k + 1)T
one has
Z kT
x[k] = x(kT ) = e AkT x(0) + e A(kT −τ ) Bu(τ )dτ
0
Z (k+1)T
x[k + 1] = e AT x[k] + e A(kT +T −τ ) Bdτ u[k]
kT
ZT
= e AT x[k] + e Aα Bdαu[k], α = kT + T − τ
0
x[k + 1] = Ad x[k] + Bd u[k] (6)
y [k] := y (kT ) = Cx(kT ) + Du(kT ) = Cd x[k] + Dd u[k] (7)

Discretization
where
Ad = e AT (8)
Z T Z T
Aτ −1
Bd = e dτ B = A de Aτ B
0 0
= A−1 (e AT − I )B = A−1 (Ad − I )B, if A is nonsingular (9)
Cd = C (10)
Dd = D (11)
Matlab command [ad,bd]=c2d(a,b,T)

Solution of Discrete-Time Equations
Consider the discrete-time state-space equation
x[k + 1] = Ax[k] + Bu[k] (12)
y [k] = Cx[k] + Du(k] (13)
It is given x[0] and u[k], k = 0, 1, . . . ,, we want to find the solution.
Noting
x[1] = Ax[0] + Bu[0]
x[2] = Ax[1] + Bu[1] = A2 x[0] + ABu[0] + Bu[1]
proceeding forward, one can obtain, for k > 0,
k−1
X
k
x[k] = A x[0] + Ak−1−m Bu[m] (14)
m=0
k−1
X
y [k] = CAk x[0] + CAk−1−m Bu[m] + Du[k] (15)
m=0

Equivalent State Equations
Consider the n-dimensional state equation
ẋ(t) = Ax(t) + Bu(t) (16)
y (t) = Cx(t) + Du(t) (17)
where A ∈ R n×n and B ∈ R n×p , and C ∈ R q×n .
Definition 1
Let P be an n × n real nonsingular matrix and let x̄ = Px. Then the state
equation,
˙
x̄(t) = Āx̄(t) + B̄u(t) (18)
y (t) = C̄ x̄(t) + D̄u(t) (19)
where
Ā = PAP −1 , B̄ = PB, C̄ = CP −1 , D̄ = D (20)
is said to be (algebraically) equivalent to (16)-(17) and x̄ = Px is called
an equivalence transformation.
The characteristic equation of (18) is

¯ = det(λI − Ā) = det(λPP −1 − PAP −1 ) = det[P(λI − A)P −1 ]
∆
= det(P)det(λI − A)det(P −1 ) = det(λI − A) = ∆(s)
The transfer function of the system (18)-(19) is
ˆ (s) = C̄ (sI − Ā)−1 B̄ + D̄ = C (sI − A)−1 B + D = Ĝ (s)

Ḡ (21)
• Thus equivalent state equations have the same characteristic polynomial

and, consequently, the same set of eigenvalues and same transfer matrix.
• Matlab command: [ab,bb,cb,db]=ss2ss(a,b,c,d,p)

Two state 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 2
Two linear time-invariant state equations {A, B, C , D} and {Ā, B̄, C̄ , D̄}
are zero-state equivalent or have the same transfer matrix if and only if
D = D̄ and
CAm B = C̄ Ām B̄, m = 0, 1, 2, . . .

Proof: Noting
(1 − s −1 λ)−1 = 1 + s −1 λ + s −2 λ2 + · · ·
s −1 (1 − s −1 λ)−1 = s −1 + s −2 λ + s −3 λ2 + · · ·
(sI − A)−1 = s −1 (I − s −1 A)−1 = s −1 I + s −2 A + s −3 A2 + · · ·
D+CBs −1 +CABs −2 +CA2 Bs −3 +· · · = D̄+C̄ B̄s −1 +C̄ ĀB̄s −2 +C̄ Ā2 B̄s −3 +· · ·
for any s. Compare the coefficients of s −i , the theorem is proved.

It is clear that (algebraic) equivalence implies zero-state equivalence. In

order for two state equations to be equivalent, they must have the same
dimension.

Example
Consider the circuit in Fig. 1.
Figure 1: A circuit
Let x1 be the current through the inductor and x2 be the voltage over the
capacitor. For the two loops, by KVL,
−u + ẋ1 + x2 = 0, 1(ẋ2 − x1 ) + x2 = 0
So

ẋ1 0 1 1
= x+ u = Ax + Bu
ẋ2 1 −1 0
y = [0, 1]x = Cx
Example
If we choose two mesh currents as x̄1 and x̄2 , then
−u + x̄˙ 1 + 1(x̄1 − x̄2 ) = 0,

Z
1(x̄2 − x̄1 ) + x̄2 dt = 0 ⇒ x̄˙ 2 − x̄˙ 1 = −x̄2 ⇒ x̄˙ 2 = −x̄1 + u
So
x̄˙ 1

−1 1 1
= x̄ + u = Āx̄ + B̄u
x̄˙ 2 −1 0 0
y = [1, −1]x̄ = C̄ x̄
Noting that x̄1 = x1 and x2 = 1(x̄1 − x̄2 ) = x1 − x̄2 , then

1 0
x̄ = x = Px
1 −1
It can be verified that Ā = PAP −1 , B̄ = PB, C̄ = CP −1 .

Canonical Forms
For simplicity, we consider a 4-th order SISO LTI causal system with the
equation
ẋ = Ax + bu
y = Cx + du
Two canonical forms will be considered: companion form and modal form.

Companion form
If
|λI − A| = λ4 + β4 λ3 + β3 λ2 + β2 λ + β1
and [b, Ab, A2 b, A3 b] is nonsingular, let
P −1 = Q = [b, Ab, A2 b, A3 b].
The transformation x̄ = Px transforms the system to
x̄˙ = Āx̄ + b̄u, y = C̄ x̄ + du
where  
0 0 0 β1
1 0 0 β2 
Ā = PAP −1 = Q −1 AQ = 


 0 1 0 β3 
0 0 1 β4
is the companion form.
The MATLAB command is: [ab,bb,cb,db,P]=canon(a,b,c,d,’companion’)
Modal form
Assume the eigenvalues of A are: λ1 , λ2 , α + jβ, α − jβ, where λ1 , λ2 , α,

and β are real.
Let q1 , q2 , q3 , and q4 are the corresponding eigenvalues of the four
eigenvalues. It is obvious that q1 and q2 are real and q3 = q4∗ are complex.
Let P −1 = Q = [q1 , q2 , q3 , q4 ], then
 
λ1 0 0 0
 0 λ2 0 0
Ā = PAP −1 = Q −1 AQ = 


 0 0 α + jβ 0 
0 0 0 α − jβ
Q and Ā can be obtained using MATLAB command: [Q,Ab]=eig(A)

Modal form
If we let P −1 = Q = [q1 , q2 , Re(q3 ), Im(q3 )], then

 
λ1 0 0 0
0 λ2 0 0
Ā = PAP −1 = Q −1 AQ = 
 

 0 0 α β 
0 0 −β α
and Ā is in the modal form.

The MATLAB command is: [ab,bb,cb,db,P]=canon(a,b,c,d,’modal’) or
[ab,bb,cb,db,P]=canon(a,b,c,d)

Realizations
Every linear time-invariant (LTI) system can be described by the

input-output description
ŷ (s) = Ĝ (s)û(s)
and, if the system is lumped as well, by the state-space equation
description
ẋ(t) = Ax(t) + Bu(t) (22)

y (t) = Cx(t) + Du(t) (23)
If the state equation is known, the transfer matrix can be computed as

Ĝ (s) = C (sI − A)−1 B + D which is unique. For a given transfer function,
it is possible to find a state-space equation. This is called the realization
problem.

Realization
A transfer matrix Ĝ (s) is said to be realizable if there exists a

finite-dimensional state equation (22)-(23) 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).
Remarks:
• An LTI distributed system can be described by a transfer matrix, but not
by a finite-dimensional state equation. Thus not every Ĝ (s) is realizable.
• If Ĝ (s) is realizable, then it has infinitely many realizations, not
necessarily of the same dimension. Thus the realization problem is fairly
complex.
We study here only the realizability condition. The other issues will be
studied in later chapters.

Realization
Theorem 3
A transfer matrix C(s) is realizable if and only if C(s) is a proper rational
matrix.

Realization
Proof: ”⇒”: If Ĝ is realizable, there exists {A, B, C , D} such that
1
Ĝ (s) = C (sI − A)−1 B + D = C [Adj(sI − A)]B + D (24)
det(sI − A)
Since each entry of Adj(sI − A) is the determinant of an (n − 1) × (n − 1)

submatrix of (sI − A), it has at most the degree (n − 1). So,
Ĝ (∞) = D (25)
Since D is a constant matrix, Ĝ (s) is a proper rational matrix.

Realization
”⇐”: If Ĝ (s) is a q × p proper rational matrix, then Ĝ (s) can be written

as
Ĝ (s) = Ĝ (∞) + Ĝsp (s)
where Ĝsp is the strictly proper part of Ĝ (s). Let
d(s) = s r + α1 s r −1 + · · · + αr −1 s + αr
be the least common denominator of all entries of Ĝsp (s). Here we require
d(s) to be monic; that is, its leading coefficient is 1. Then Ĝsp (s) can be
expressed as
1 1
Ĝsp (s) = [N(s)] = [N1 s r −1 + N2 s r −2 + · · · + Nr −1 s + Nr ] (26)
d(s) d(s)
where Ni are q × p constant matrices.

Realization
Now we claim that the set of equations

   
−α1 Ip −α2 Ip · · · −αr −1 Ip −αr Ip Ip
 Ip
 0 ··· 0 0  
  0 

ẋ = 
 0 I p · · · 0 0 +
  0  u (27)

 .. .. . .. .
.. ..   .. 
 . . .   . 
0 0 ··· Ip 0 0
y = [N1 , N2 , . . . , Nr −1 , Nr ]x + Ĝ (∞)u (28)
is a realization of Ĝ (s). The matrix Ip is the p × p unit matrix and every 0

is a p × p zero matrix. The A matrix is said to be in block companion
form; it consists of r rows and r columns of p × p matrices; thus the A
matrix has order rp × rp. The B matrix has order rp × p. Because the C
matrix consists of r number of Ni , each of order q × p, the C matrix has
order q × rp.

Realization
Next, we show that (27)-(28) is a realization of Ĝ (s). Let us define
Z = [Z10 , Z20 , . . . , Zr0 ]0 := (sI − A)−1 B (29)
where Zi is p × p and Z is rp × p. Then the transfer matrix of (27)-(28)

equals
C (sI − A)−1 B + Ĝ (∞) = N1 Z1 + N2 Z2 + · · · + Nr Zr + Ĝ (∞) (30)
By (29),
sZ = AZ + B (31)
which means that
sZ1 = −αZ1 − α2 Z2 − · · · − αr Zr + Ip
sZ2 = Z1 , sZ3 = Z2 , . . . , sZr = Zr −1

Realization
So,
α2 αr
sZ1 = −α1 Z1 − Z1 − · · · − r −1 Z1 + Ip
s s
α2 αr
= − α1 + + · · · + r −1 Z1 + Ip
s s
α2 αr d(s)
s + α1 + + · · · + r −1 Z1 = r −1 Z1 = Ip
s s s
s r −1 s r −2 1
Z1 = Ip , Z2 = Ip , . . . , Zr = Ip ,
d(s) d(s) d(s)
So,
1
C (sI −A)−1 B + Ĝ (∞) = [N1 s r −1 +N2 s r −2 +· · ·+Nr ]+ Ĝ (∞) = Ĝsp (s)
d(s)

Realization
Procedure for realization of Ĝ (s) ∈ R q×p :
1 Decompose Ĝ (s) as
Ĝ (s) = Ĝ (∞) + Ĝsp (s)
where Ĝsp is the strictly proper part of Ĝ (s) and

Ĝ (∞) = lims→∞ Ĝ (s).
2 Let
d(s) = s r + α1 s r −1 + · · · + αr −1 s + αr
be the least common denominator of all entries of Ĝsp (s).
3 Express Ĝsp (s) as
1 1
Ĝsp (s) = [N(s)] = [N1 s r −1 + N2 s r −2 + · · · + Nr −1 s + Nr ](32)
d(s) d(s)
where Ni are q × p constant matrices.

Realization
4 The state space equations are

   
−α1 Ip −α2 Ip ··· −αr −1 Ip −αr Ip Ip
 Ip
 0 ··· 0 0  
  0 

ẋ = 
 0 Ip ··· 0 0 +
  0  u (33)

 .. .. .. .. ..   .. 
 . . . . .   . 
0 0 ··· Ip 0 0
y = [N1 , N2 , . . . , Nr −1 , Nr ]x + Ĝ (∞)u (34)
The realization in (33)-(34) has dimension rp and is said to be in

controllable canonical form.

Example
Consider the proper rational matrix

" #
4s−10 3
2s+1 s+2
Ĝ (s) = 1 s+1
(2s+1)(s+2) (s+2)2
find its realization.

Example
Solution:
−6(s + 2)2 3(s + 2)(s + 0.5)

2 0 1
Ĝ (s) = +
0 0 s 3 + 4.5s 2 + 6s + 2 0.5(s + 2) (s + 1)(s + 0.5)

2 0 1 −6 3
= + 3 s 2+
0 0 s + 4.5s 2 + 6s + 2 0 1

−24 7.5 −24 3
s+
0.5 1.5 1 0.5

Example
A realization is
.. ..
 
 −4.5 0 . −6 0 . −2 0   
. . 1 0
−4.5 .. 0 −6 .. 0 −2   0 1
 
 0 
 . . 
· · · .. · · · · · · .. · · · · · ·  
  . . 
 ···

  .. .. 

 .. ..   
 1 0 . 0 0 . 0 0   0 0  u1

ẋ =  .. .. + 
 0 0 . 0 0 . 0 0 
 0 0  u2
  
 .. ..

. .

 . .
· · · .. · · · · · · .. · · · · · · 
 
 ··· 
  
  0 0

.. . 
0 .. 0

 0 0 . 1 0 
  0 0
.. ..
0 0 . 0 1 . 0 0
 
. .
−6 3 .. −24 7.5 .. −24 3 
y = 
. .
0 1 .. 0.5 1.5 .. 1 0.5
This is a six dimensional realization.
Realization
Other methods to realize a proper transfer matrix Ĝ (s) ∈ R q×p .
Method 1: Let Ĝ (s) = [Ĝc1 (s), . . . , Ĝcp (s)] and u = [u1 , . . . , up ]> ,
then
ŷ (s) = Ĝc1 (s)û1 (s)+· · ·+ Ĝcp (s)ûp (s) =: ŷc1 (s)+ ŷc2 (s)+· · ·+ ŷcp (s)
as shown in Figure 2(a).
Method 2: Let  
Ĝr 1 (s)
Ĝ (s) = 
 .. 
. 
Ĝrq (s)
and let y = [y1 , . . . , yq ]> be the ith component of the output vector
y . Then    
ŷ1 (s) Ĝr 1 (s)û(s)
 .. ..
ŷ (s) =  . =
  
. 
ŷq (s) Ĝrq (s)û(s)
as shown in Figure 2(b).
Realization
Figure 2: Realization of Ĝ (s) by columns and by rows
MATLAB function [a,b,c,d]=tf2ss(num,den) generates the

controllable-canonical-form for any single-input multiple-output transfer
matrix Ĝ (s).

Example
Consider the proper rational matrix

" #
4s−10 3
2s+1 s+2
Ĝ (s) = 1 s+1
(2s+1)(s+2) (s+2)2

Example
Solution:
" #
4s 2 −2s−20 3s+6
Ĝ (s) = 2s 2 +5s+2 s 2 +4s+4 = [Ĝc1 , Ĝc2 ]
1 s+1
2s 2 +5s+2 s 2 +4s+4
The realization of the first column of Ĝ (s) is

−2.5 −1 1
ẋ1 = A1 x1 + B1 u1 = x1 + u1 (35)
1 0 0

−6 −12 2
yc1 = C1 x1 + d1 u1 = x1 + u1 (36)
0 0.5 0
The realization of the second column of Ĝ (s) is

−4 −4 1
ẋ2 = A2 x2 + B2 u2 = x2 + u2 (37)
1 0 0

3 6 0
yc2 = C2 x2 + d2 u2 = x2 + u2 (38)
1 1 0
Example
MATLAB commands:
num1=[4,-2,-20;0,0,1]; den1=[2,5,2]; [a1,b1,c1,d1]=tf2ss(num1,den1)
num2=[3,6;1,1]; den2=[1,4,4]; [a2,b2,c2,d2]=tf2ss(num2,den2)
sys1=ss(a1,b1,c1,d1); sys2=ss(a2,b2,c2,d2);
Other commands: sys=parallel(sys1,sys2), sys=series(sys1,sys2),

sys=append(sys1,sys2,...sysN)

Example
The two realization can be combined as

ẋ1 A1 0 x1 b1 0 u1
= + (39)
ẋ2 0 A2 x2 0 b2 u2
y = yc1 + yc2 = [C1 , C2 ]x + [d1 , d2 ]u (40)
or
   
−2.5 −1 0 0 1 0
 1 0 0 0    0 0 
ẋ = = x + u (41)
 0 0 −4 −4   0 1 
0 0 1 0 0 0

−6 −12 3 6 2 0
y = x+ u (42)
0 0.5 1 1 0 0

Solution of Linear Time-Varying (LTV) Equations
Consider
ẋ = A(t)x (43)
where A(t) is n × n with continuous functions of t as its entries.
Lemma 4
For each initial state xi (t0 ), there exists a unique solution xi (t), for
i = 1, 2, . . . , n. If x1 (t0 ), x2 (t0 ), . . . , xn (t0 ) are linearly independent, then
the columns of X (t) = [x1 (t), x2 (t), . . . , xn (t)] are linear independent for
any t ≥ t0 .

Proof: Prove it by contradiction. Assume at t1 x1 (t1 ), x2 (t1 ),. . . ,xn (t1 )

are linearly dependent. Then, there exists a constant vector v such that
x(t1 ) = X (t1 )v = 0
which in turn implies x(t) := X (t)v ≡ 0 for all t, in particular, at t = t0 .

This is a contradiction. Thus, X (t) is nonsingular for all t.

• If X (t0 ) is nonsingular or the n initial states are linearly independent,

then X (t) is called a fundamental matrix of (43).
Remark: The fundamental matrix is not unique.
• Let X (t) be any fundamental matrix of ẋ = A(t)x, then
Φ(t, t0 ) := X (t)X −1 (t0 )
is called the state transition matrix of ẋ = A(t)x. The state transition

matrix is also the unique solution of
∂
Φ(t, t0 ) = A(t)Φ(t, t0 ) (44)
∂t
with the initial condition Φ(t0 , t0 ) = I .
Remark: The state transition matrix is unique.

Properties of Φ(t, t0 ):
1 Φ(t, t) = I
2 Φ−1 (t, t0 ) = Φ(t0 , t)
3 Φ(t, t0 ) = Φ(t, t1 )Φ(t1 , t0 )
4 x(t) = Φ(t, t0 )x(t0 ) for any t and t0 .

Theorem 5
Consider the system
ẋ(t) = A(t)x(t) + B(t)u(t) (45)

y (t) = C (t)x(t) + D(t)u(t) (46)
It is assumed that its state transition matrix is Φ(t, t0 ). The solution of

the system is
Z t
x(t) = Φ(t, t0 )x(t0 ) + Φ(t, τ )B(τ )u(τ )dτ (47)
t0
Z t
y (t) = C (t)Φ(t, t0 )x(t0 ) + C (t) Φ(t, τ )B(τ )u(τ )dτ
t0
+D(t)u(t) (48)

Proof:
Z t
∂ ∂
ẋ(t) = Φ(t, t0 )x(t0 ) + Φ(t, τ )B(τ )u(τ )dτ
∂t ∂t t0
Z t
∂
= A(t)Φ(t, t0 )x(t0 ) + Φ(t, τ )B(τ )u(τ ) dτ
t0 ∂t
+Φ(t, t)B(t)u(t)
Z t
= A(t)Φ(t, t0 )x(t0 ) + (A(t)Φ(t, τ )B(τ )u(τ )) dτ + B(t)u(t)
t0
= A(t)x(t) + B(t)u(t)
y (t) = C (t)x(t) + D(t) = · · ·

If the initial state x(t0 ) = 0, then

Z t
y (t) = C (t) Φ(t, τ )B(τ )dτ + D(t)u(t)
t0
Z t
= [C (t)Φ(t, τ )B(τ ) + D(t)δ(t − τ )]u(τ )dτ
t
Z 0t
= G (t, τ )dτ
t0
where
G (t, τ ) = C (t)Φ(t, τ )B(τ ) + D(t)δ(t − τ ) (49)
G (t, τ ) is the impulse response matrix and is the output at time t excited
by an impulse input applied at time τ .

• If A(t) has the commutative property

Z t Z t
A(t) A(τ )dτ = A(τ )dτ A(t)
t0 t0
for all t and t0 , then

∞ Z t k
Rt
A(τ )dτ
X 1
Φ(t, t0 ) = e t0
= A(τ )dτ (50)
k! t0
k=0
• If A(t) is a constant,
Φ(t, τ ) = e A(t−τ ) = Φ(t − τ )
and X (t) = e At .

Example
Consider the system

0 0
ẋ(t) =
t 0
Find its fundamental matrix and the transition matrix.

Example
Solution: Solve the equation, we have x1 (t) = x1 (0) and
x2 (t) = 0.5t 2 x1 (0) + x2 (0).
Choose an initial state

x1 (0) 1 1
x(0) = = ⇒ x(t) =
x2 (0) 0 0.5t 2
and another initial state

x1 (0) 1 1
x(0) = = ⇒ x(t) =
x2 (0) 2 0.5t 2 + 2
The two initial states are linearly independent; thus

1 1
X (t) =
0.5t 2 0.5t 2 + 2
is a fundamental matrix.
0.25t 2 + 1 −0.5

−1
X (t) =
−0.25t 2 0.5
Example
Thus the state transition matrix is given by
0.25t 2 + 1 −0.5

−1 1 1
Φ(t, t0 ) = X (t)X (t) =
0.5t 2 0.5t 2 + 2 −0.25t 2 0.5

1 0
=
0.5(t 2 − t02 ) 1

Discrete-Time Case
Consider the discrete time system
x[k + 1] = A[k]x[k] + B[k]u[k] (51)

y [k] = C [k]x[k] + D[k]u[k] (52)
with initial state x[k0 ] and input u[k] for k ≥ k0 .

The state transition matrix Φ[k, k0 ] is defined by
Φ[k + 1, k0 ] = A[k]Φ[k, k0 ] with Φ[k0 , k0 ] = I
for k = k0 , k0 + 1, . . . , .
It can be shown that
Φ[k, k0 ] = A[k − 1]A[k − 2] · · · A[k0 ] (53)
and
Φ[k, k0 ] = Φ[k, k1 ]Φ[k1 , k0 ]
Discrete-Time Case
The solution of the system is

k−1
X
x[k] = Φ[k, k0 ]x[k0 ] + Φ[k, m + 1]B[m]u[m]
m=k0
k−1
X
y [k] = C [k]Φ[k, k0 ]x[k0 ] + C [k] Φ[k, m + 1]B[m]u[m] + D[k]u[k]
m=k0

Discrete-Time Case
If the initial state x[k0 ] = 0, then

k−1
X
y [k] = C [k] Φ[k, m + 1]B[m]u[m] + D[k]u[k]
m=k0
k−1
X
= C [k]Φ[k, m + 1]B[m] + D[m]δ[k − m])u[m]
m=k0
k−1
X
= G [k, m]u[m]
m=k0
where
G [k, m] = C [k]Φ[k, m + 1]B[m] + D[m]δ[k − m]
is the impulse response sequence.

Equivalent Time-Varying Equations
Consider the n-dimensional linear time-varying state equation
ẋ(t) = A(t)x(t) + B(t)u(t) (54)

y (t) = C (t)x + D(t)u (55)
Let P(t) be an n × n matrix. It is assumed that P(t) is nonsingular and

both P(t) and Ṗ(t) are continuous for all t. Let x̄ = P(t)x, then
d
[P(t)x(t)] = Ṗ(t)x(t) + P(t)ẋ(t) = Ṗ(t)x(t) + P(t)A(t)x(t)
dt
+P(t)B(t)u(t)
= [Ṗ(t) + P(t)A(t)]x(t) + P(t)B(t)u(t)
= [Ṗ(t) + P(t)A(t)]P −1 (t)x̄(t) + P(t)B(t)u(t)
y (t) = C (t)P −1 (t)x̄(t) + D(t)u(t)

So,
˙
x̄(t) = Ā(t)x̄(t) + B̄(t)u(t) (56)
y (t) = C̄ (t)x̄(t) + D̄(t)u(t) (57)
where
Ā(t) = [Ṗ(t) + P(t)A(t)]P −1 (t)

B̄(t) = P(t)B(t)
C̄ (t) = C (t)P −1
D̄(t) = D(t)

Let X (t) be a fundamental matrix of (54), then
X̄ (t) := P(t)X (t) (58)
is a fundamental matrix of (56). It can be verified by
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)]X (t)
= [Ṗ(t) + P(t)A(t)]P −1 (t)X̄ (t) = Ā(t)X̄ (t)
Theorem 6
Let A0 be an arbitrary constant matrix. Then there exists an equivalence
transformation that transforms (54) into (56) with Ā(t) = A0 .

Proof: Let X be a fundamental matrix of ẋ = A(t)x. Since

X −1 (t)X (t) = I , then
Ẋ −1 X + X −1 Ẋ = 0
Ẋ −1 = −X −1 Ẋ X −1 = −X −1 A
Let
P(t) := e A0 t X −1
then
Ā = [PA + Ṗ]P −1
= [e A0 t X −1 A + A0 e A0 t X −1 + e A0 t Ẋ −1 ]Xe −A0 t
= A0 e A0 t X −1 Xe −A0 t = A0

If A0 = 0, then
Ā = 0, B̄ = X −1 B, C̄ = CX , D̄ = D (59)
The block diagrams of (54) with A(t) 6= 0 and A(t) = 0 are plotted in
Figure 3. The block diagram with A(t) = 0 has no feedback and is
considerably simpler. Every time-varying state equation can be
transformed into such a block diagram. However, in order to do so. its
fundamental matrix should be known.

Figure 3: Block diagrams with feedback and without feedback.

The impulse response matrix of (56) is, using (58) and (59),
Ḡ (t, τ ) = C̄ (t)X̄ (t)X̄ −1 (τ )B̄(τ ) + D̄(t)δ(t − τ )

= C (t)P −1 (t)P(t)X (t)X −1 (t)P −1 (t)P(τ )B(τ ) + D(t)δ(t − τ )
= C (t)X (t)X −1 (t)B(τ ) + D(t)δ(t − τ ) = G (t, τ )
• Thus the impulse response matrix is invariant under any equivalence

transformation.
• The property of the A-matrix. however, may not be preserved in
equivalence transformations.
• In the time-invariant case, equivalence transformations will preserve all
properties of the original state equation. Thus the equivalence
transformation in the time-invariant case is not a special case of the
time-varying case.

Definition 7
A matrix P(t) is called a Lyapunov transformation if P(t) is nonsingular,
P(t) and Ṗ(t) are continuous, and P(t) and P −1 are bounded for all t.
Equations (54) and (56) are said to be Lyapunov equivalent if P(t) is a
Lyapunov transformation.
• If P(t) = P is a constant matrix, then it is a Lyapunov transformation.

Thus the (algebraic) transformation in the time-invariant case is a special
case of the Lyapunov transformation.
• Not every time-varying state equation can be Lyapunov equivalent to a
state equation with a constant A-matrix. However, this is true if A(t) is
periodic.

Theorem 8
Consider (54) with A(t) = A(t + T ) for all t and some T > 0. Let X (t)
be a fundamental matrix of ẋ = A(t)x. Let Ā be the constant matrix
computed from e ĀT = X −1 (0)X (T ). Then (54) is Lyapunov equivalent to
˙
x̄(t) = Āx̄(t) + P(t)B(t)u(t) (60)
−1
ȳ (t) = C (t)P (t)x̄(t) + D(t)u(t) (61)
where P(t) = e Āt X −1 (t).

Proof: Since
Ẋ (t + T ) = A(t + T )X (t + T ) = A(t)X (t + T )
X (t + T ) is also a fundamental matrix. Furthermore,
X (t + T ) = X (t)X −1 (0)X (T ) = X (t)e ĀT
P(t) is periodic because
P(t + T ) = e Ā(t+T ) X −1 (t + T ) = e Āt e ĀT [e −ĀT X −1 ] = e Āt X −1 = P(t)
The matrix P(t) satisfies all conditions of Lyapunov transformation. The

rest of the theorem follows directly from Theorem 6.

Time-varying Realization
Every linear time-varying system can be described by the input-output
description Z t
y (t) = G (t, τ )u(τ )dτ
t0
and, if the system is lumped as well, by the state equation
x(t) = A(t)x(t) + B(t)u(t) (62)

y (t) = C (t)x(t) + D(t)u(t) (63)
If the state equation is available, the impulse response matrix can be
computed from
G (t, τ ) = C (t)X (t)X −1 (τ )B(τ ) + D(t)δ(t − τ ) for t ≥ τ (64)
where X (t) is a fundamental matrix of ẋ = A(t)x. The converse problem
is to find a state equation from a given impulse response matrix. An
impulse response matrix G (t, τ ) is said to be realizable if there exists
{A(t), B(t), C (t), D(t)} to meet (64).
Theorem 9
A q × p 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 ) (65)
for all t ≥ τ . where M, N, and D are, respectively, q × n, n × p, and

q × p matrices for some integer n.

Proof:
If G (t, τ ) is realizable, there exists a realization that meets (64).
Identifying M(t) = C (t)X (t) and N(τ ) = X −1 (τ )B(τ ) establishes the
necessary part of the theorem.
If G (t, τ ) can be decomposed as in (65), then the n-dimensional state
equation
ẋ(t) = N(t)u(t) (66)

y (t) = M(t)x(t) + D(t)u(t) (67)
is a realization. Indeed, a fundamental matrix of ẋ = 0 is X (t) = I . Thus

the impulse response matrix of the above system is
M(t)I × I −1 N(τ ) + D(t)δ(t − τ ) = G (t, τ )

Example
Consider g (t) = te λt or
g (t, τ ) = g (t − τ ) = (t − τ )e λ(t−τ )

Example
Solution: It is straightforward to verify that
−τ e −λτ

λt λt
g (t − τ ) = [e , te ]
e −λτ
One time-varying realization is
−te −λt

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

Example
Another realization can be obtained as follows. The Laplace transform of

g (t) is
1 1
ĝ (s) = 2
= 2
(s − λ) s − 2λs + λ2
Its realization is
2λ −λ2

1
ẋ = x(t) + u(t)
1 0 0
y (t) = [1, 0]x(t)
This LTI state equation is clearly more desirable because it is easy to be

implemented.

Summary
For time-invariant systems:

1 Solutions of continuous-time LTI systems
2 Discretization: two methods.
3 Solutions of discrete-time systems
4 Equivalent systems: same transfer matrix
5 Canonical forms: Companion form, modal form
6 Realizations
For time-varying systems:
1 Solutions of LTV systems: fundamental matrix, state transition matrix
2 Solutions of discrete-time systems
3 Equivalent systems: algebraic transformation, Lyapunov
transformation
4 Time-varying realizations

Lecture 3 State-Space Solutions and Realization - v1

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lecture 3 State-Space Solutions and Realization - v1

Uploaded by

Copyright:

Available Formats

State-Space Solutions and Realization

Dr. Wenjie Dong

Department of Electrical and Computer Engineering,

W. Dong (UTRGV, Dept. of ECE) 1 / 75

W. Dong (UTRGV, Dept. of ECE) 1 / 75

x̂(s) = (sI − A)−1 [x(0) + B û(s)]

W. Dong (UTRGV, Dept. of ECE) 2 / 75

Methods for finding e At :

W. Dong (UTRGV, Dept. of ECE) 3 / 75

W. Dong (UTRGV, Dept. of ECE) 4 / 75

Solution: Method 1: The eigenvalues of A are −1, −1. Let

W. Dong (UTRGV, Dept. of ECE) 5 / 75

W. Dong (UTRGV, Dept. of ECE) 6 / 75

Find the solution of

W. Dong (UTRGV, Dept. of ECE) 8 / 75

W. Dong (UTRGV, Dept. of ECE) 9 / 75

Based on the solution of a system, the following conclusions can be

W. Dong (UTRGV, Dept. of ECE) 10 / 75

u(t) = u(kT ) =: u[k], x(t) = x(kT ) =: x[k] for kT ≤ t ≤ (k + 1)T

W. Dong (UTRGV, Dept. of ECE) 12 / 75

Matlab command [ad,bd]=c2d(a,b,T)

W. Dong (UTRGV, Dept. of ECE) 13 / 75

W. Dong (UTRGV, Dept. of ECE) 14 / 75

The characteristic equation of (18) is

The transfer function of the system (18)-(19) is

ˆ (s) = C̄ (sI − Ā)−1 B̄ + D̄ = C (sI − A)−1 B + D = Ĝ (s)

• Thus equivalent state equations have the same characteristic polynomial

W. Dong (UTRGV, Dept. of ECE) 16 / 75

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

W. Dong (UTRGV, Dept. of ECE) 17 / 75

W. Dong (UTRGV, Dept. of ECE) 18 / 75

It is clear that (algebraic) equivalence implies zero-state equivalence. In

W. Dong (UTRGV, Dept. of ECE) 19 / 75

−u + x̄˙ 1 + 1(x̄1 − x̄2 ) = 0,

Noting that x̄1 = x1 and x2 = 1(x̄1 − x̄2 ) = x1 − x̄2 , then

It can be verified that Ā = PAP −1 , B̄ = PB, C̄ = CP −1 .

W. Dong (UTRGV, Dept. of ECE) 22 / 75

P −1 = Q = [b, Ab, A2 b, A3 b].

The transformation x̄ = Px transforms the system to

x̄˙ = Āx̄ + b̄u, y = C̄ x̄ + du

Assume the eigenvalues of A are: λ1 , λ2 , α + jβ, α − jβ, where λ1 , λ2 , α,

Q and Ā can be obtained using MATLAB command: [Q,Ab]=eig(A)

W. Dong (UTRGV, Dept. of ECE) 24 / 75

If we let P −1 = Q = [q1 , q2 , Re(q3 ), Im(q3 )], then

and Ā is in the modal form.

W. Dong (UTRGV, Dept. of ECE) 25 / 75

Every linear time-invariant (LTI) system can be described by the

ẋ(t) = Ax(t) + Bu(t) (22)

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

W. Dong (UTRGV, Dept. of ECE) 26 / 75

A transfer matrix Ĝ (s) is said to be realizable if there exists a

W. Dong (UTRGV, Dept. of ECE) 27 / 75

W. Dong (UTRGV, Dept. of ECE) 28 / 75

Proof: ”⇒”: If Ĝ is realizable, there exists {A, B, C , D} such that

Since each entry of Adj(sI − A) is the determinant of an (n − 1) × (n − 1)

Since D is a constant matrix, Ĝ (s) is a proper rational matrix.

W. Dong (UTRGV, Dept. of ECE) 29 / 75

”⇐”: If Ĝ (s) is a q × p proper rational matrix, then Ĝ (s) can be written

where Ni are q × p constant matrices.

W. Dong (UTRGV, Dept. of ECE) 30 / 75

Now we claim that the set of equations

is a realization of Ĝ (s). The matrix Ip is the p × p unit matrix and every 0

W. Dong (UTRGV, Dept. of ECE) 31 / 75

Consider the system