ECE/ME 2646: Linear System Theory (3 Credits, Fall 2017)

Lecture 3: LTI Systems; Rational Transfer

Functions and State-space Equations;
Linearization

September 12, 2017

Zhi-Hong Mao
Associate Professor of ECE and Bioengineering
University of Pittsburgh, Pittsburgh, PA
1

Outline of this lecture

• Review of last lecture

• LTI systems
• Rational transfer functions and state-space
equations
• Linearization

Review of last lecture

• Categories of systems
– Causal system and noncausal system

Remark: Causality describes the relationship between

causes and effects, and is fundamental to all natural
science, especially physics. In classical physics, the
cause simply had to precede its effect; in modern
physics, the effect must belong to the future light cone
of its cause, even if the spacetime is curved. Causality
is also studied from the perspectives of philosophy,
computer science, statistics, and psychology.

1
 Review of last lecture
• Categories of systems
– Causal system and noncausal system

Infant causal perception

Review of last lecture

• Categories of systems
– Causal system and noncausal system

Question: Filtering can be classified into three different

problems: estimation, prediction, and smoothing. Estimation
provides an estimate of the input based upon all data up to
the current time. Prediction provides an estimate of what the
value ahead of the observed data. Smoothing provides an
estimate using observations that are ahead of in time of the
estimate. Which one of these can be realized by a physical
system (in real time)?

Review of last lecture

• Categories of systems
– Causal system and noncausal system

Question: Consider a system modeled as a

map from u(t) to y(t), where


e
|t  |
y (t )  u( )d .

Is this system causal?

2
 Review of last lecture
• Categories of systems
– Causal system and noncausal system

– Concept of state
• The concept of state was proposed to address the
following question: Given the input at some time, how
much information do we need about past inputs in order
to determine the present output?
– This question addresses the issue of memory in the system
– The answer gives us an idea of the complexity, or number
of degrees of freedom, associated with the dynamic
behavior of the system
– In a control application, the answer to the above question
suggests the required degree of complexity of the controller

Review of last lecture

• Categories of systems
– Causal system and noncausal system

– Concept of state
• The concept of state was proposed to address the following question: Given the input at some time, how
much information do we need about past inputs in order to determine the present output?

• The state x(t0) of a system at time t0 is the information at

t0 that, together with the input u(t), for t  t0, determines
uniquely the output y(t) for all t  t0

Review of last lecture

• Categories of systems
– Causal system and noncausal system

– Concept of state
• The concept of state was proposed to address the following question: Given the input at some time, how
much information do we need about past inputs in order to determine the present output?
• The state x(t0) of a system at time t0 is the information at t0 that, together with the input u(t), for t  t0,
determines uniquely the output y(t) for all t  t0

• State is not unique; length of state is not unique

– In the example of F = ma, you can add to the set of state
variables (position and velocity) the temperature, your
age, the number of students in our class, etc.—the choice
of state can be very subjective. For this case, you could
also have used all atomic velocities and positions as state
variables.… We will see in this course how to determine
whether a set of state variables is “just enough” or
redundant

3
 Review of last lecture
• Categories of systems
– Causal system and noncausal system

– Concept of state
• The concept of state was proposed to address the following question: Given the input at some time, how
much information do we need about past inputs in order to determine the present output?
• The state x(t0) of a system at time t0 is the information at t0 that, together with the input u(t), for t  t0,
determines uniquely the output y(t) for all t  t0
• State is not unique; length of state is not unique

• In the text book by Kuo (1995), the state variables of a

system are defined as a minimal set of variables such
that knowledge of these variables at any time t0, and
information on the input excitation subsequently applied,
are sufficient to determine the state of the system at any
time t  t0

10

Review of last lecture

• Categories of systems
– Causal system and noncausal system

– Concept of state

Question: How many state variables are needed to

describe the following systems:
(1) y(t) = u(t) + 2
(2) dy(t) / dt = u(t) (in general, how about a n-th order
linear ODE?)
(3) y(t) = u(t − 1)?

11

Review of last lecture

• Categories of systems
– Causal system and noncausal system

– Concept of state

Question: How can the following system be

approximated by a system with finite state variables: y(t)
= u(t−1)?

12

4
 Review of last lecture

• Categories of systems
– Causal system and noncausal system
– Concept of state

– Lumped system and distributed system

13

Linear time-invariant systems

• Definition
– A system is said to be time invariant if for every state-
input-output pair
x(t0 )  x 0 
  y (t ), t  t0
u(t ), t  t0 
and any T, we have
x(t0  T )  x 0 
  y (t  T ), t  t0  T
u(t  T ), t  t0  T 
– The above definition means that if the initial state is
shifted to time t0+T and the same input waveform is
applied from t0+T, the output waveform will be the
same except that it starts to appear from time t0+T
14

Linear time-invariant systems

• Definition
– A system is said to be time invariant if for every state-input-output pair
x(t0 )  x 0 
  y (t ), t  t0
u(t ), t  t0 

and any T, we have x(t0  T )  x 0 

  y (t  T ), t  t0  T
u(t  T ), t  t0  T 
– The above definition means that if the initial state is shifted to time t0+T and the same input
waveform is applied from t0+T, the output waveform will be the same except that it starts to
appear from time t0+T

– Examples of time-variant systems: burning rocket and

brain

15

5
 Linear time-invariant systems
• Definition

• Input-output description

For a causal, SISO, LTI

system relaxed at 0:
g (t , )  g (t  T ,  T )  g (t   ,0)  g (t   )
t t
y (t )   g (t , )u( )d y (t )   g (t   )u( )d
0
0

Convolution integral

16

Linear time-invariant systems

• Definition

• Input-output description

Question: If an LTI system is causal, what is

the value of g(t) for t < 0?

17

Linear time-invariant systems

• Definition

• Input-output description

The condition for an LTI system to be

causal is g(t) = 0 for t < 0

Question: The impulse response of an ideal

lowpass filter is given by
sin 2 (t  t0 )
g (t )  2 .
2 (t  t0 )
Is the ideal lowpass filter causal?

18

6
 Linear time-invariant systems
• Definition
• Input-output description

• Transfer-function matrix

For a causal, MIMO, LTI

system (with p input and ˆ ( s)uˆ ( s)
yˆ ( s)  G
q output) relaxed at 0:

 gˆ11( s ) gˆ12 ( s )  gˆ1 p ( s ) 

 
gˆ 21( s ) gˆ 22 ( s )  gˆ 2 p ( s )
G ( s )  
ˆ

  
 
 gˆ q1 ( s ) gˆ q 2 ( s )  gˆ qp ( s) 
19

Rational transfer functions and state-

space equations
• If an LTI system is lumped, then its transfer function
is a rational function of s: gˆ (s)  N (s) / D(s)

– gˆ ( s) is proper, if deg (degree of) D(s)  deg N (s)

– gˆ ( s) is strictly proper if deg D(s)  deg N (s)
– gˆ ( s) is biproper deg D(s)  deg N (s)
– gˆ ( s) is improper if deg D(s)  deg N (s)

Question: Improper rational transfer functions rarely

arise in practice, since they will amplify high-frequency
noise. Why?

20

Rational transfer functions and state-

space equations
• If an LTI system is lumped, then its transfer function is a rational function of s:
gˆ (s)  N (s) / D(s)
• Poles and zeros
– A real or complex number  is called a pole if D() = 0, and
 is called a zero if N() = 0
– If N(s) and D(s) have no common factors of degree 1 or
higher, they are called coprime

21

7
 Rational transfer functions and state-
space equations
• If an LTI system is lumped, then its transfer function is a rational function of s:
gˆ (s)  N (s) / D(s)
• Poles and zeros

• State-space equations
– Deriving transfer-function matrix from state-space equation
Laplace
x (t )  Ax (t )  Bu(t ) transform sxˆ ( s)  x(0)  Axˆ ( s)  Buˆ ( s)
y (t )  Cx(t )  Du(t ) yˆ ( s)  Cxˆ ( s)  Duˆ ( s)

xˆ ( s)  ( sI  A) 1 x(0)  ( sI  A) 1 Buˆ ( s)
yˆ ( s)  C( sI  A) 1 x(0)  C( sI  A) 1 Buˆ ( s)  Duˆ ( s)
For zero initial state

yˆ (s)  [C(sI  A) 1 B  D] uˆ (s) ˆ (s)  C(sI  A) 1 B  D

G
22

Linearization

• Many physical systems can be described by

nonlinear differential equations
x (t )  h( x(t ), u(t ), t )
y(t )  f ( x(t ), u(t ), t )

• Some nonlinear equations can be approximated by

linear equations (but how?)

23

Linearization
• Many physical systems can be described by nonlinear differential equations
x (t )  h( x(t ), u(t ), t )
y(t )  f ( x(t ), u(t ), t )

• Some nonlinear equations can be approximated by

linear equations

d 2 (t )
ML  Mg sin  (t )  0
L dt 2
θ
For small value of θ
M
d  (t )
2
Mg L  g (t )  0
dt 2
24

8
 Linearization
• Many physical systems can be described by nonlinear differential equations
x (t )  h( x(t ), u(t ), t )
y(t )  f ( x(t ), u(t ), t )

• Some nonlinear equations can be approximated by

linear equations

25

Linearization
• Many physical systems can be described by nonlinear differential equations
x (t )  h( x(t ), u(t ), t )
y(t )  f ( x(t ), u(t ), t )

• Some nonlinear equations can be approximated by

linear equations
y
Slope

f(x0)
x f(  ) y
x0 x
Operating
point

26

Linearization
• Many physical systems can be described by nonlinear differential equations
x (t )  h( x(t ), u(t ), t )
y(t )  f ( x(t ), u(t ), t )
• Some nonlinear equations can be approximated by
linear equations
– For some input u0(t) and some initial state, x0(t) is the
solution of the above equations
x 0 (t )  h(x0 (t ), u0 (t ), t )

– Suppose x(t )  x0 (t )  x(t ) for slightly perturbed input

u(t )  u0 (t )  u(t ) and initial state

27

9
 Linearization
• Many physical systems can be described by nonlinear differential equations

• Some nonlinear equations can be approximated by

linear equations
x 0 (t )  x (t )  h( x 0 (t )  x (t ), u0 (t )  u (t ), t )
x (t )  h( x(t ), u(t ), t )
h h
x 0 (t )  h( x 0 (t ), u0 (t ), t )  h( x 0 (t ), u0 (t ), t )  x u 
x u
x(t )  x 0 (t )  x (t ) h  [h1  hn ]' , x  [ x1  xn ]' , u  [u1 u p ]'
u(t )  u0 (t )  u(t ) h1 / x1 h1 / x2  h1 / xn 
 
h h2 / x1 h2 / x2  h2 / xn 
A (t )  
x     
 
hn / x1 hn / x2  hn / xn 
x (t )  A(t )x(t )  B(t )u h1 / u1 h1 / u2  h1 / u p 
Jacobians
 
h h2 / u1 h2 / u2  h2 / u p 
B( t )  
u     
 
hn / u1 hn / u2  hn / u p  28

Linearization
• Many physical systems can be described by nonlinear differential equations
• Some nonlinear equations can be approximated by linear equations

• An example

x1  1 , x2  1 , x3   2 , x4  2

x1  x2 1 l1
g mg
x 2   sin x1  2 cos x3 sin( x3  x1 )
l1 m1l1
1 2 l2
 sin x3 sin( x3  x1 )  u
m1l1 u(t)
m1g
x3  x4
g 1
x 4   sin x3  (cos x3 ) u
l2 m2l2 m2g
29

Linearization
• Many physical systems can be described by nonlinear differential equations
• Some nonlinear equations can be approximated by linear equations

• An example
x1  x2
g mg 1
x 2   sin x1  2 cos x3 sin( x3  x1 )  sin x3 sin( x3  x1 )  u
l1 m1l1 m1l1
x3  x4
g 1
x 4   sin x3  (cos x3 ) u
l2 m2l2

 0 1 0 0  0 
 ( m  m ) g /( m l ) 0 m2 g / m1l1 0   0 
x   x u
1 2 11

 0 0 0 1  0 
   
 0 0  g/l 2 0 1 / m2l2 
30

10
 Linearization
• Many physical systems can be described by nonlinear differential equations
• Some nonlinear equations can be approximated by linear equations
• An example

• In some cases, the linearized model is a very poor

approximation to the physical system

31

References
• K. J. Astrom and R. M. Murray. Feedback Systems: An Introduction for
Scientists and Engineers. Manuscript, 2007.
• C.-T. Chen. Linear System Theory and Design, 3rd Edition, Oxford
University Press, 1999.
• M. Dahleh, M. A. Dahleh, and G. Verghese. Lecture Notes for 6.241
Dynamic Systems and Control. Massachusetts Institute of Technology,
2003.
• E. Feron. Lecture Notes for 16.31 Feedback Control. Massachusetts
Institute of Techology, 1998.
• G. F. Franklin, J. D. Powell, and A. Emami-Naeni. Feedback Control of
Dynamic Systems, Addison-Wesley, 2002.
• T. Hu. Lecture Notes for 16.513 Control Systems. University of
Massachusetts at Lowell, 2006.
• B. C. Kuo. Automatic Control Systems. Prentice-Hall, 1995.
• C. L. Phillips and R. D. Harbor. Feedback Control Systems, 4th Edition,
Prentice Hall, 2000.
• http://en.wikipedia.org/wiki/Causality_%28physics%29
• http://www.ensc.sfu.ca/people/faculty/cavers/ENSC380/
• http://www.math.ku.edu/%7Ebyers/ode/
32

11