Linear System Theory

Dr. Vali Uddin

Hamdard University

[email protected]

Lecture 1
 Introduction
 Motivation  Course Overview

– Math. Descriptions of Systems ~ Review

 Classification of Systems
 Linear Systems

 LTI Systems

Lecture 1
 INTRODUCTION
Motivation

 What is a "system"?
– A physical process or a mathematical model of a
physical process that relates a set of input signals to
yield another set of output signals
y(t)
Input Output
System t

– Examples: Cars, circuits, bank accounts, stock markets

• Two general categories of signals/systems:
– Continuous-time signals/systems
• Examples: Signals in cars and circuits
• Described by differential eqs., e.g., dy/dt = ay(t) + bu(t)
Lecture 1
 • Signals themselves could be discontinuous
– Discrete-time signals/systems y(k)

• Examples: Money in a bank account, quarterly profit

• No derivative exists
• Signals described by difference equations,
e.g., y[k+1] = ay[k] + bu[k]
–They are quite similar, and shall be treated in parallel
• What is "System Theory"?
–Understanding the physical system under consideration
–Describing the system mathematically
–Analyzing the properties
–Controlling it to meet certain criteria
Lecture 1
 Why are you in the class?
– Foundation for most control/communication courses
– Included in MS and Ph.D. examinations
 What are the prerequisites?
– Signals and Systems: Working knowledge of
 Laplace transform
 z-transform
 Differential equations
 Linear algebra, and
 Modeling of electrical systems

Lecture 1
Example: A simple electric circuit
i(t) ~ Output

R L
u(t) ~ C
~ Input

 Understanding the components and

interconnections:
– What are the components? How to model
them?
di L dv C
v R  Ri R , v L  L , iC  C
dt dt
Lecture 1
 -Interconnections:
• KVL: Voltage across a loop = 0
• KCL: Current to a node = 0
i(t) ~ Output

R L
u(t) ~ C
~ Input

di 1 t
KVL : Ri  L   i( )d  v0  u( t )
dt C 0
d 2i( t ) di( t ) 1 du ( t )
L  R  i ( t ) 
dt 2 dt C dt

– An integral-differential or differential equation

– Input-output description or external description
Lecture 1
 Analyzing the properties/responses
– For example, find the output i(t) given u(t)
and IC.

 For now, shall use Laplace transform

(Effective for LTI systems)

Lecture 1
Laplace Transform, A Quick
Review

F ( s)   f (t )e  st dt
0

 Key Properties
– Linearity: a1f1(t) + a2f2(t)  a1F1(s) +
a2F2(s)
– Derivative theorem: f'(t)  sF(s) - f(0-); f(-
1)(t)  F(s)/s

Lecture 1
Laplace Transform, A Quick
Review
 Converting linear constant coefficient
differential equations into algebraic
equations
– Differentiation in the frequency domain:
tf(t)  (-1)F'(s)
– Convolution: h(t)f(t)  H(s)F(s)
– Time and frequency shifting: f(t-t0)u(t-t0)
 e-st0 F(s); es0t f(t)  F(s - s0)
Lecture 1
Example (Continued)

di 1 t
KVL : Ri  L   i( )d  v0  u( t )
dt C 0

 
Rî (s)  L sî (s)  i0 
î (s) v 0
Cs

s
 û(s)

Lecture 1
Laplace Transform, A Quick
Review

– Time and frequency scaling: f(at)  1/a

F(s/a) for a > 0
– Initial Value Theorem: f(0+) = lims sF(s)
– Final Value Theorem: f() = lims0 sF(s) if
all the poles of sF(s) have strictly negative
real parts

Lecture 1
Laplace Transform, A Quick
Review
Example i(t) ~ Output
(Continued) R L
u(t) ~ C
~ Input

di 1 t
KVL : Ri  L   i( )d  v0  u( t )
dt C 0

 
Rî (s)  L sî (s)  i0 
î (s) v 0
Cs

s
 û(s)

Lecture 1
• An algebraic equation as opposed to an
integral-differential equation. Solution:

 Ls  R  1 î (s)  û(s)  Li  v0 

   0 s 
 cs 

cs LCsi 0  cv 0
î (s)  û(s) 
LCs 2  RCs  1 LCs 2  RCs  1

 What can we say about it?

– It has two components, one caused by
input, and the other by IC
Lecture 1
 How about the voltage across the capacitor?
î (s) v0 1 Li 0  LCs  RC v0
v̂(s)    û ( s ) 
Cs s LCs 2  RCs  1 LCs 2  RCs  1
• What is the system's transfer function?
• Assume that the ICs are zero, then
^u(s) ^ ^i(s) = g(s)
^ u(s)
^
g(s)

Cs Cs
î (s)  û(s) ĝ(s) 
2
LCs  RCs  1 LCs 2  RCs  1

– Frequency domain analysis

Lecture 1
 How to obtain the response in time domain?

 
i( t )  L1 î (s)
– Suppose that L = C = 1, R = 2, v0 = i0 = 0, and u(t) =
U(t) (unit step function). Then
1 Cs û(s) 1 1
û(s)  î (s)   
s LCs 2  RCs  1 s 2  2s  1 s  12
0.4
0.35
0.3
t
i( t )  te
0.25
i(t) 0.2
0.15
0.1
0.05
0
0 5 10 15
t

Lecture 1
 Does this make sense for the circuit?

+
R L
u(t) C v(t)
-

 Laplace transform is not effective for time

varying systems

Lecture 1
Example

Solve y'(t) + t y(t) = f(t). How?

– Take Laplace transform and recall that
tf(t)  (-1)F'(s)
– (sY(s) - y(0-)) + (- Y'(s)) = F(s)
– It is still a differential equation, not an algebraic
equation
– The use of Laplace transform is restricted to LTI
systems
Lecture 1
 i(t) ~ Output

R L
u(t) ~ C v(t)
~ Input

• State-Space Description
– What are the state variables?
– Voltage across C and current through L
– What is the state equation?
dv
C i  v   0 1/ C  v  0 
 i    1 / L  R / L  i   1 / L u
dt
di       
L  u  Ri  v
dt
Lecture 1
– A set of first-order differential equations
• It describes the behaviors inside the system by using the
state variables v(t) and i(t)
• How to describe the output? i(t) ~ Output

v
y  i  0 1   R L
v(t)
i u(t) ~ C
~ Input

– The output equation

– Combined with the state equation, we have the
state-space description or internal description

Lecture 1
• How to analyze the system?
- Can also use Laplace transform

 v   0 1   v  0
 i    1  2  i   1 u
      

sv̂ (s)  v 0   0 1   v̂ (s) 0

 sî (s)  i    1  2  î (s)   1
 0      s 

s  1   v̂ (s) 0
1 s  2  î (s)   1
    s 

Lecture 1
 v̂ (s) 1 s  2 1 0 1 1
 î (s)   2   1 s 1  2 s 
  s  2s  1    s  s  2s  1 1

i( t )  te  t as expected

v ( t )  u( t )  e  t  te  t  Quantitative analysis

Lecture 1
• What else can be said about this system?
– Is the system controllable? observable? stable?
– These are "qualitative analysis" as opposed to the
previous "quantitative analysis"
• Analysis is one of our major emphases
• What happens if the performance of a system is not
satisfactory?
• Design ~ How to realize a system, adjust system
parameters (e.g., the resistance R), or design
feedback control to meet certain specifications
Lecture 1
 Design is our final goal
– System realization
– State feedback and state estimators
– Pole placement and model matching
– Introduction to optimal control
 The focus will be on linear systems

Lecture 1
 Course Overview

 Textbooks:
– Chi-Tsong Chen, Linear System Theory and
Design, 3rd Edition, Oxford University Press,
1999

– William L. Brogan, Modern Control Theory,

3rd edition, Prentice Hall, 1990

Lecture 1
 Goals: To provide a thorough
understanding about systems theory and
multivariable system design
 Tentative Outline:
– Introduction
– Modeling: How to model a physical system
– The fundamentals of linear algebra
– Analysis:
 Quantitative: How to derive response for a given
input
 Qualitative: How to analyze controllability,
observability, and stability
Lecture 1
– Design:
 How to realize a system given its mathematical
description
 How to design a control law so that system
response satisfies certain criteria
 How to design an observer to estimate the state
of the system
 Pole Placement and Model Matching

– Shall treat continuous-time and discrete-time

systems in parallel

Lecture 1
 General Rules:
 Homework can be done individually or in teams of
two.
 Homework should be clear, concise, and complete
 Not all the problems will be graded. Grading will be
based on some randomly selected problems
 Late assignments will be discounted 10% a day, up
to 5 days

Lecture 1
 Mathematical Descriptions of
Systems

u(t) y(t)
System
u[k] y[k]
px1 qx1

– Classification of Systems
– Linear Systems
– Linear time invariant (LTI) Systems
Lecture 1
 Classification of Systems
 Basic assumption: When an input signal is applied to
the system, a unique output is obtained
Q. How do we classify systems?
– Number of inputs/outputs; with/without memory;
causality; dimensionality; linearity; time invariance
 The number of inputs and outputs
– When p = q = 1, it is called a single-input single-
output (SISO) system
– When p > 1 and q > 1, it is called a multi-input
multi-output (MIMO) system

Lecture 1
• Memoryless vs. with Memory

– If y(t) depends on u(t) only, the system is said to be

memoryless, otherwise, it has memory
– An example of a memoryless system?
+
+ R1 y(t)
u(t) R2
- -
A purely resistive circuit
R2
y( t )  u( t ) ~ Memoryless
R1  R 2
Lecture 1
– An example of a system with memory?

R i

+
u(t) L
-

di di R 1
Ri  L  u or  i u
dt dt L L

R R
 t  t 0  1 t  t  
i( t )  e L i( t 0)  e
L u( )d
Lt
0

Lecture 1
 – i(t) depends on i(t0) and u() for t0    t, not just u(t)
– A system with memory
• Causality: No output before an input is applied
Input Output
System

– A system is causal or non-anticipatory if y(t0) depends

only on u(t) for t  t0 and is independent of u(t) for t > t0
– Is the circuit discussed last time causal?
u(t)
R i 1
t
+
u(t) L y(t)
-
1
t
Lecture 1
– An example of a non-causal system?
– y(t) = u(t + 2) u(t)
1
t
y(t)
1 t

– Can you truly build a physical system like this?

– What is an example of a non-causal system in
practice?
– If you can invent such a system, let me know. We will
be rich through Connecticut Lottery

Lecture 1
 The Concept of State
– The state of a system at t0 is the information at t0
that, together with u[t0,), uniquely determines
the behavior of the system for t  t0
– The number of state variables = the number of
ICs needed to solve the problem
– For an LRC circuit, the number of state variables
= the number of C + the number of L (except for
degenerated cases)
– A natural way to choose state variables as what
we have done earlier: {vc} and {iL}
Lecture 1
 – Is this the unique way to choose state
variables?
– Any invertible transformation of the above can
serve as a state, e.g.,
 x1( t )  2 1  v( t ) 2 v( t )  i( t )
 x ( t )  0 1  i( t )    
 2      i ( t ) 

– Although the number of state variables = 2, there are

infinite numbers of representations
• Order of dimension of a system: The number of state
variables
– If the dimension is a finite number  Finite dimensional
(or lumped) system
– Otherwise, an infinite dimensional (or distributed) system
Lecture 1
Q. Give an example of an infinite dimensional system

u(t) y(t) = u(t-1)

System A delay line

– Given u(t) for t  0, what information is needed to know y(t)

for t  0?
u(t)
t
? y(t)
t
1
?
– We need an infinite amount of information  An infinite
dimensional system
Lecture 1
 Linear Systems
Linearity
 Double the efforts double the outcome?
– Suppose we have the following state-input-output
pairs:
x1( t 0 ) 
  y1( t ), t  t 0
u1( t ), t  t 0 

x2 (t0 ) 
  y 2 ( t ), t  t 0
u 2 ( t ), t  t 0 
– What would be the output of
x1( t 0 )  x 2 ( t 0 ) 
 y1( t )  y 2 ( t ), t  t 0
u1( t )  u 2 ( t ), t  t 0 
Lecture 1
– If it is true ~ Additivity
– How about
 x1( t 0 ) 
  y1( t ), t  t 0
 u1( t ), t  t 0 

– If it is true ~ Homogeneity
– Combined together to have:
1x1( t 0 )   2 x 2 ( t 0 ) 
 1y1( t )  2 y 2 ( t )
1u1( t )   2 u 2 ( t ), t  t 0 

– If it is true ~ Superposition or linearity property

– A system with such a property: a Linear System

Lecture 1
• Are R, L, and C linear elements?
di dv
v R  Ri R , v L  L L , iC  C C
dt dt
– Yes (differentiation is a linear operation)

v Affine v Nonlinear
v = Ri
Linear
i i

 Also, KVL and KCL are linear constraints. When

put together, we have a linear system

Lecture 1
  The additivity property implies that
 x1( t 0 )  x1( t 0 )  x1( t 0 )  0
y( t ) due to  y( t ) due to  y( t ) due to
u1( t ), t  t 0 u1( t )  0 u1( t ), t  t 0

– Response = zero-input response + zero-state response

Response of a Linear System

u(t) Linear y(t)

System
• How to obtain the response of a linear system to a given u(t)
with zero IC?
• Use the linearity property. How?
Lecture 1
• Let (t-ti) be a pulse at time ti with width  and
height 1/
t-ti
1/
Area = 1
t
ti ti+
– Let the system response to (t-ti) at time t be g(t, ti)
– Then what?
– A general input u(t) can be approximated as a sum of
such pulses
– The response y(t) would then be the sum of such
responses based on linearity
Lecture 1
   ( t  t i ) u ( t i )  ( t  t i )  g  ( t, t i )

  ( t  t i ) u ( t i )  g  ( t , t i ) u ( t i )
ti t
u ( t )     ( t  t i ) u ( t i )   g  ( t , t i ) u ( t i )
i i
– What is (t-ti) in the limit as  0?
t-ti t-ti
1/

ti ti+ ti

 ( t  t i )  ( t  t i ) ~ A shifted unit impulse
Lecture 1
 g  ( t, t i )  g( t, t i ) ~ Unit impulse response

y( t )   g  ( t, t i )u( t i )    g( t, )u( )d
i 

• Thus far, we have used linearity

– What if the system is causal?
– g(t,) ~ Response at t from a unit impulse at 
t
g( t, )  0 for t   or for   t y( t )   g( t, )u( )d

– A system is said to be relaxed at t0 if the initial
state at t0 is 0
– In this case, y(t) for t  t0 is caused exclusively by
u(t) for t  t0 t
y( t )   g( t, )u( )d
t0
Lecture 1
 • How about for a system with p inputs and q outputs?
– Have to analyze the relationship for input/output pairs
t
y( t )   G( t, )u( )d
t0
 g11( t , ) g12 ( t, ) g1p ( t, ) 
g ( t ,  ) g ( t ,  ) gij(t,): The impulse
g 2 p ( t , )
G ( t , )   
21 22 response between the
  jth input and ith output
 g ( t , ) g ( t , ) g qp ( t , ) 
 q1 q2

State-Space Description
 A linear system can be described by
x ( t )  A( t ) x( t )  B( t )u( t )
y( t )  C( t ) x( t )  D( t )u( t )
– The derivation of solutions will be done later
Lecture 1
 Linear Time-Invariant (LTI) Systems
 Time Invariance: The characteristics of a system do not change
over time
– What are some of the LTI examples? Time-varying
examples?
– What happens for an LTI system if u(t) is delayed by T?
Have to watch out ICs
u(t) y(t)
t t

u(t-T) y(t-T)

Lecture 1
– If the initial state is also shifted to time
t0 + T, then the two responses should
be the same, only shifted by T:
x( t 0 )  x 0 
  y( t ), t  t 0
u( t ), t  t 0 

x( t 0  T)  x 0 
  y( t  T ), t  t 0  T
u( t  T ), t  t 0  T 

Lecture 1
 What happens to the unit impulse
response when the system is LTI?
g( t, )  g( t  T,   T) for any T

g( t, )  g( t  ,   )  g( t  ,0)  g( t  )

– Only the difference between t and  matters

– What happens
t to y(t)?
y( t )   g( t, )u( )d
t0
t t
y( t )   g( t  )u( )d   g( )u( t  )d
t0 t0
~ Convolution integral
Lecture 1
THE END

Lecture 1