0% found this document useful (0 votes)
43 views6 pages

1 Minimal Realizations: Co o C C o 1 2 Co 3 C o 1 Co C o

1) The document describes how to compute a minimal realization from a given non-minimal realization (A, B, C, D). 2) It first discusses computing a controllable realization by transforming (A, B, C) using singular value decompositions of the controllability matrix C(A,B). 3) It then discusses computing an observable realization by similarly transforming (A, B, C) using singular value decompositions of the observability matrix O(A,C). 4) Finally, it provides an algorithm to compute a minimal realization by applying both the controllable and observable transformations sequentially.

Uploaded by

Mouli
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
43 views6 pages

1 Minimal Realizations: Co o C C o 1 2 Co 3 C o 1 Co C o

1) The document describes how to compute a minimal realization from a given non-minimal realization (A, B, C, D). 2) It first discusses computing a controllable realization by transforming (A, B, C) using singular value decompositions of the controllability matrix C(A,B). 3) It then discusses computing an observable realization by similarly transforming (A, B, C) using singular value decompositions of the observability matrix O(A,C). 4) Finally, it provides an algorithm to compute a minimal realization by applying both the controllable and observable transformations sequentially.

Uploaded by

Mouli
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

1 Minimal Realizations

Definition: A realization (A, B, C) is minimal if (A, B) is controllable and


(A, C) is observable.

Theorem: For any (A, B, C) there exists a nonsingular similarity


transformation T that transforms the original system into the form
   
Aco 0 Ao 0 Bco
Ac Acō A1 A2  Bcō  ,
Ā = T −1 AT =  −1
   
 0 , B̄ = T B =
0 Ac̄o 0   0 
0 0 A3 Ac̄ō 0
£ ¤
C̄ = CT = Cco 0 Cc̄o 0 .

In this form:
1) The subsystem (Aco , Bco , Cco ) is controllable and observable and
C(sI − A)−1 B = Cco (sI − Aco )−1 Bco ;
2) The subsystem µ· ¸ · ¸ ¶
Aco 0 Bco £ ¤
, , Cco 0
Ac Acō Bcō
is controllable;
3) The subsystem µ· ¸ · ¸ ¶
Aco Ao Bco £ ¤
, , Cco Cc̄o
0 Ac̄o 0
is observable;
4) The subsystem (Ac̄ō , 0, 0) is neither controllable nor observable.

Problem: Given a non-minimal realization (A, B, C, D) find an equivalent


realization (Ā, B̄, C̄, D̄) that is minimal.

MAE 280A 1 Maurı́cio de Oliveira


2 The Singular Value Decomposition
Theorem: For any matrix A ∈ Rm×n with rank(A) = r < n there exist real
square matrices
£ ¤ £ ¤
U = U1 U2 , V = V1 V2 ,

and a diagonal real matrix


· ¸
Σ+ 0
Σ= ,
0 0
such that
¤ Σ+ 0 V1T
· ¸· ¸
A = U ΣV T = U1
£
U2 ,
0 0 V2T
where

U U T = U1 U1T + U2 U2T = I, V V T = V1 V1T + V2 V2T = I,

and Σ+ > 0.

Proof: There exists orthogonal U and V such that

AAT = U Λ1 U T , AT A = V Λ 2 V T ,

where Λ1 ≥ 0, Λ2 ≥ 0. Also, if A = U ΣV then

AAT = U ΣV T V ΣT U T = U ΣΣT U T ,
AT A = V ΣT U T U ΣV T = V ΣT ΣV T .

One can prove that the choice

Λ1 = ΣΣT , Λ2 = ΣT Σ,

is possible.

MAE 280A 2 Maurı́cio de Oliveira


3 Computing Minimal Realizations
3.1 Computing Controllable Realizations
Let (A, B, C) be a non-minimal realization.
1) Compute C(A, B).
2) Compute the singular value decomposition
¤ Σ+ 0 V1T
· ¸· ¸
T
£
C(A, B) = U1 U2 T = U 1 Σ+ V 1 ,
0 0 V2
where Σ+ > 0 and

U1 U1T + U2 U2T = I, V1 V1T + V2 V2T = I.

3) Set
r = rank C(A, B) = dim Σ+ .

4) Compute
−1/2 1/2
R 1 = U 1 Σ+ T1 = U 1 Σ+ .

Theorem: The realization

(Ā, B̄, C̄) = R1T AT1 , R1T B, CT1


¡ ¢

of order r is controllable and B(sI − A)−1 C = B̄(sI − Ā)−1 C̄.

Proof: First note that


· T¸ · ¸
U I 0
U T U = 1T U1 U2 = U1T U1 = I,
£ ¤
, ⇒
U2 0 I
so that for
−1/2 1/2
P = P T = R1 T1T = U1 Σ+ Σ+ U1T = U1 U1T
we have

P C(A, B) = (U1 U1T )(U1 Σ+ V1T ) = U1 Σ+ V1T = C(A, B).

MAE 280A 3 Maurı́cio de Oliveira


That is
P C(A, B) = P B P AB · · · P An−1 B = B AB · · · An−1 B .
£ ¤ £ ¤

Therefore
(AP )B = A(P B) = AB,
(AP )2 B = AP (AP B) = AP (AB) = A(P AB) = A2 B,
..
.
(AP )n B = AP An−1 B = An B.
Consequently
Āi B̄ = (R1T AT1 )i R1T B,
= (R1T AT1 )(R1T AT1 ) · · · (R1T AT1 )R1T B,
= R1T (AT1 R1T )(AT1 R1T ) · · · (AT1 R1T )B,
= R1T (AP )i B,
= R1T Ai B,
so that
C(Ā, B̄) = B̄ ĀB̄ · · · Ān−1 B̄ ,
£ ¤

= R1T B AB · · · An−1 B ,
£ ¤

= R1T C(A, B)
As R1 has full-column, rank C(Ā, B̄) = r and therefore (Ā, B̄) is controllable.
Finally remember that

X
−1
C (sI − A) B= Hi , Hi = CAi−1 B.
i=1

Since
H̄i = C̄(Āi−1 B̄),
= (CT1 )(R1T Ai−1 B),
= C(P Ai−1 B),
= C(Ai−1 B) = Hi ,
we conclude that ¢−1
C (sI − A)−1 B = C̄ sI − Ā
¡
B̄.

MAE 280A 4 Maurı́cio de Oliveira


3.2 Computing Observable Realizations
Let (A, B, C) be a non-minimal realization.
1) Compute O(A, C).
2) Compute the singular value decomposition
¤ Σ+ 0 V1T
· ¸· ¸
T
£
O(A, C) = U1 U2 T = U 1 Σ+ V 1 ,
0 0 V2
where Σ+ > 0 and

U1 U1T + U2 U2T = I, V1 V1T + V2 V2T = I.

3) Set
r = rank O(A, C) = dim Σ+ .

4) Compute
1/2 −1/2
R 1 = V 1 Σ+ T1 = V 1 Σ+ .

Theorem: The realization

(Ā, B̄, C̄) = R1T AT1 , R1T B, CT1


¡ ¢

of order r is observable and B(sI − A)−1 C = B̄(sI − Ā)−1 C̄.

Proof: Note that


1/2 −1/2
P = P T = R1 T1T = V1 Σ+ Σ+ V1T = V1 V1T

and

P O(A, C)T = (V1 V1T )(U1 Σ+ V1T )T = O(A, C)T

and follow the same steps as in the previous proof.

MAE 280A 5 Maurı́cio de Oliveira


3.3 Computing Minimal Realizations
Let (A, B, C) be a non-minimal realization.
1) Compute C(A, B) and O(A, C).
2) Compute the singular value decomposition
¤ Σc 0 VcT
· ¸· ¸
T
£
C(A, B) = Uc Uc̄ T = U c Σ+ V c ,
0 0 Vc̄
where Σc > 0 and

Uc UcT + Uc̄ Uc̄T = I, Vc VcT + Vc̄ Vc̄T = I.

3) Compute the singular value decomposition


¤ Σco 0 VcoT
· ¸· ¸
1/2 T
£
O(A, C)Uc Σ = Uco Uco¯ T = Uco Σ+ Vco ,
0 0 Vco¯
where Σco > 0 and
T T
Uco Uco ¯ = I,
+ Uco¯ Uco Vco VcoT + Vco¯ Vco¯T = I.

4) Set
r = rank O(A, C)Tc = dim Σco .

5) Compute

R1 = Uc Σ−1/2
c Vco Σ1/2
co T1 = Uc Σ1/2 −1/2
c Vco Σco .

Theorem: The realization

(Ā, B̄, C̄) = R1T AT1 , R1T B, CT1


¡ ¢

of order r is a minimal realization.

Proof: Combine the two previous theorems.

MAE 280A 6 Maurı́cio de Oliveira

You might also like