EE61209: Linear Systems and Control

Post Mid-sem Lecture 2: System Interconnections

Bhargav Jha

October 1, 2024

Bhargav Jha October 1, 2024 1 / 12

System Interconnections Using State-Space

One major advantage of using transfer function in Laplace domain:

▶ Interconnections between transfer function can be performed using
algebraic manipulations.
▶ More transparency compared to time-domain inpulse response.
Example:
u1 G1 G2 G2 G 1 u 1

Bhargav Jha October 1, 2024 2 / 12

System Interconnections Using State-Space

One major advantage of using transfer function in Laplace domain:

▶ Interconnections between transfer function can be performed using
algebraic manipulations.
▶ More transparency compared to time-domain inpulse response.
Example:
u1 G1 G2 G2 G 1 u 1

Let there be two LTI systems G1 (s) and G2 (s) with realizations:
( (
ẋ1 = A1 x1 (t) + B1 u1 (t) ẋ2 = A2 x2 (t) + B2 u2 (t)
G1 : G2 :
y1 = C1 x1 (t) + D1 u1 (t) y2 = C2 x2 (t) + D2 u2 (t)

Find state-space realization of G1 G2 , G1−1 G2 , G1 + G2 , etc?

Frequently switching between transfer function domain to state-space
becomes tedious.

Bhargav Jha October 1, 2024 2 / 12

Parallel Interconnection (Addition)
u1 y1
G1 G1

u1 (G1 + G2 )u1 u y
u2 y2
G2 G2

Figure: Parallel Connection

Given
   
A1 B1 A2 B2
G1 : u1 → y1 := and G2 : u2 → y1 := .
C1 D 1 C2 D 2

To find  
A B
G : u → y := = G1 + G2
C D
Note that, y = y1 + y2 and u1 = u2 = u
Bhargav Jha October 1, 2024 3 / 12
Parallel Interconnection (Addition)
Substituting this in the given state-space equations:

ẋ1 = A1 x1 + B1 u (1)
ẋ2 = A2 x2 + B2 u (2)
y = y1 + y2 = C1 x1 + C2 x2 + (D1 + D2 )u (3)

By reorganizing in matrix form we get,

      
ẋ1 A 0 x1 B
= 1 + 1 u (4)
ẋ2 0 A2 x2 B2
 
  x1
y = C1 C2 + (D1 + D2 )u (5)
x2
 
A1 0 B1
Therefore, G = G1 + G2 =  0 A2 B2 
C1 C2 D1 + D2
Bhargav Jha October 1, 2024 4 / 12
Cascaded Interconnection (Multiplication)

u1 u1 y1 u2 y2
G1 G2 G2 G 1 u 1 G1 G2

Figure: Cascaded Interconnection

To find  
A B
G : u → y := = G2 G1
C D
Note that, y1 = u2 , y2 = y and u1 = u

Bhargav Jha October 1, 2024 5 / 12

Cascaded Interconnection (Multiplication)
Substituting u1 = u, u2 = y1 , and y2 = y , in the given state-space
equations:

ẋ1 = A1 x1 + B1 u (6)
ẋ2 = A2 x2 + B2 (C1 x1 + D1 u) (7)
y = y2 = C2 x2 + D2 (C1 x1 + D1 u) (8)

By reorganizing in matrix form we get,

      
ẋ1 A1 0 x1 B1
= + u (9)
ẋ2 B2 C1 A2 x2 B2 D1
 
  x1
y = D2 C1 C2 + (D2 D1 )u (10)
x2
 
A1 0 B1
Therefore, G = G2 G1 =  B2 C1 A2 B2 D1 
D2 C1 C2 D2 D1
Bhargav Jha October 1, 2024 6 / 12
Cascaded Interconnection (Multiplication)

Q. Compute G : u → y for the block diagram below:

u G2 G1 y

Bhargav Jha October 1, 2024 7 / 12

Inverse of a System
 
A B
Given G : u → y be a system with realization . Find the
C D
realization of G −1 .

Bhargav Jha October 1, 2024 8 / 12

Inverse of a System
 
A B
Given G : u → y be a system with realization . Find the
C D
realization of G −1 .
Clearly, y is the input to G −1 and u is the output. From the realization of
G
ẋ = Ax + Bu u = D −1 (y − Cx)
=⇒ (11)
y = Cx + Du ẋ = Ax + B(D −1 (y − Cx))

Reorganizing,

ẋ = (A − BD −1 C )x + BD −1 y (12)
−1 −1
u = −D Cx + D y (13)

For existence of inverse D −1 ̸= 0. What kind of system does it represent?

- Bi-proper system.

Bhargav Jha October 1, 2024 8 / 12

Partial Fraction Expansion

as+b d1 d2
Q. Express the transfer function (s−2)(s−3) as s−2 + s−3 .
How to do the same in state-space form? That is, express G1 (s)G2 (s)
as H1 (s) + H2 (s) such that ”A” matrix of Hi is same as Gi .
Reminder
▶ Cascaded ▶ Parallel
Interconnection interconnection
   
A1 0 B1 A1 0 B1
 B2 C1 A2 B2 D1   0 A2 B2 
D2 C1 C2 D2 D1 C1 C2 D1 + D2
Convert
 the block-triangular
  matrix
 to block-diagonal matrix such
A1 0 A1 0
that and are similar (same eigenvalues).
0 A2 B2 C1 A2

Bhargav Jha October 1, 2024 9 / 12

Partial Fraction Expansion

Note, we don’t desire to change the behavior of the system. We just

want to represent it differently.
Way out?

Bhargav Jha October 1, 2024 10 / 12

Partial Fraction Expansion

Note, we don’t desire to change the behavior of the system. We just

want to represent it differently.
Way out? Reminder: Similarity transformation with a non-singular T
preserves eigenvalues.

TAT −1 T −1 B
   
A B
→
C D CT D
 
I 0
Now consider T = . Check that the block matrix is
X I
non-singular and compute the inverse.

Bhargav Jha October 1, 2024 10 / 12

Partial Fraction Expansion: Roth’s Removal Rule

Apply the similarity transformation to the A matrix of the cascaded

system.
     
I 0 A1 0 I 0 A1 0
=
X I B2 C1 A2 −X I XA1 + B2 C1 − XA2 A2
The term in blue should be zero for the desired parallel connection.
   
A1 0 A1 0
Roth-rule Matrices and are similar iff there exist
Q A2 0 A2
a solution X to

XA1 + Q − XA2 = 0, Q = B2 C1 (in this context)

There exist a unique solution iff spec(A1 ) ∩ spec(A2 ) = ∅. If this is

not the case, then there can be infinite or no solution, depending on
Q.
This equation is called Sylvester equation.

Bhargav Jha October 1, 2024 11 / 12

Partial Fraction Expansion

Apply the similarity transform and verify that

      
A1 B1 A2 B2 A1 B1 A2 B2 D1 + XB1
C D C D
= D C −C X 0 + C 0
+ D2 D1
1 1 2 2 2 1 2 2

Bhargav Jha October 1, 2024 12 / 12