ELEC2100: Signals and Systems

Fall 2011

Lecturer: Albert K. Wong

Office: Room 2455
Tel: 2358 7065
Email: [email protected]

AKW Fall 2011 Chapter 1-2 1

 ELEC2100: Signals and Systems

Textbook: A. V. Oppenheim, et. al., Signals and Systems, 2nd edition,

Prentice-Hall, 1997

Syllabus: Chapters 1-9 of the above text. We will learn the basic
techniques for analyzing systems in time and frequency
domains through Fourier analysis and Laplace transform.

Please consult LMES for details for demonstrator and TA office

hours

AKW Fall 2011 Chapter 1-2 2

 ELEC2100: Signals and Systems
Homework: 5 homework assignments (10%)

Lab: 4 laboratory assignments with prelab exercises (10%)

PRS: 6 unannounced PRS short quizzes with one lowest score

dropped (5%)

Mid-term 2-hour mid-term on November 10 (Thurs), 7-9 pm. (30%)

Final Exam 3-hour final (45%)

Grading: Absolute grading will be used based on attainment of

intended learning outcomes.
Guaranteed A+, A, A- for scores above 92%, 85%, 78%
Guaranteed B+, B, B- for scores above 72%, 66%, 60%
Guaranteed F for scores <40%

AKW Fall 2011 Chapter 1-2 3

 Why is this course essential?
• This course introduces the basic framework for the study of
“signals” and “systems”

• Signals and systems are fundamental to all disciplines of engineering

• In this course, you will be introduced to the language and a comprehensive,

consistent, and elegant set of principles for the study of Linear Time Invariant
(LTI) systems. These language and principles are fundamental to the study of
many problems in communication, circuits, signal processing, control theory,
and physics.

AKW Fall 2011 Chapter 1-2 4

 Lecture 1
Chapter 1 - Introduction

I) What are signals and systems? Continuous-Time and

Discrete-Time Signals and Systems

II) Key Math Concepts needed in this course: Complex

Number, Space and Projection

AKW Fall 2011 Chapter 1-2 5

 I. What is a signal?

A vague definition: Signal is some physical quantity or

measurement that varies with time or space, carrying
information and bringing interaction between systems.

Examples: - Voltage waveform in a circuit

- Acoustic pressure carrying speech signal
- A color picture
- Radio wave signal received by your cell phone
- Share price of HSBC on the Hong Kong Stock Exchange

• Mathematically: Function of one or more independent variables

- x(s1, s2, s3)
Independent variables: time, space, etc.

AKW Fall 2011 Chapter 1-2 6

 Example 1: Speech signals
Speech Signal: Variation in acoustic pressure as a function of time
Variation in the air pressure reaching our ears leads to our
 Figure 1.3 perception of sound. This variation is very small.

time

AKW Fall 2011 Chapter 1-2 7

 Example 2: Digital Images
• Black-and-White • Color Pictures:
(Monochromatic) Pictures: Color and brightness of pixels (picture
Brightness as a function of 2- elements) over 2-dimensional space
dimensional space
Lena – A standard reference
Figure 1.4 picture for study of image
processing

AKW Fall 2011 Chapter 1-2 8

 Signal as a Function of Time
• In this course, we will focus on 1-dimensional signal, where
there is only one independent variable. With no loss of generality,
we assume this independent variable is “time”

x(t) Share Price

1 single independent variable

 a 1-dimensional signal
t

• Hence, in ELEC2100, signal is just a function of time.

• In the physical world, “time” is often regarded as a continuous
variable. Today, signals are often processed by the computer,
which performs computation only at discrete time instances. We
need to develop both the continuous-time and discrete-time
models for signals and systems
AKW Fall 2011 Chapter 1-2 9
 Continuous Time and Discrete Time signals
When t is continuous, we have a Continuous-Time (CT) signal. When t is
discrete, we have a Discrete-Time (DT) signal. For CT signals, the
independent variable time is a real number. For DT signals, the
independent variable time is an integer.

Hang Seng Index at any

7800 moment
CT Signal A Continuous curve

7400

Aug 10 Aug 20 time

Hang Seng Index at end of
A discrete sequence
7800 day
DT Signal of values, plot as a
stick diagram
7400

Aug 10 Aug 20 time

AKW Fall 2011 Chapter 1-2 10
• Modern systems often involve interactions between continuous-time
physical systems and discrete-time computer systems. Therefore we often
need to go back-and-forth between the CT and DT formulations.

Physical Systems Computers

Sampling
Continuous-time Discrete-time
signals signals
Interpolation

In this course, we need to learn two sets of mathematical formulations, one

for CT and one for DT. We need to understand the linkage between the two
formulations and comfortably switch between them.

AKW Fall 2011 Chapter 1-2 11

 Calculus - A Continuous Formulation
• Calculus, based on two concepts – differentiation and integration – is one of the
greatest intellectual achievements of mankind. Calculus led to the extensive use of
functions of continuous variables (polynomials, sinusoids, exponentials, logarithms,
etc.) to model real-world problems and to derive closed-form solutions. In the
1930’s the differential machine was developed to solve many different problems
that can be modeled as differential equations.
Example: We can model human height as a continuous variable
while addressing problems related to human height distribution
• Total population on planet is:

Height Distribution
of Human Beings

f (h)dh  6 billions
f(h) • Number of human beings between 1.7m
billion/m and 1.8m tall: 1.8
1.7
f (h)dh

h: height

AKW Summer 2010 Chapter 1-2 12

• Calculus and differential equations allows human to represent and
manipulate many problems in a concise and systematic way. But treating
height as a continuous variable may lead to some oddities:
Example:
What is your height? (answer: Say 1.8m)
What is the number of human beings who is 1.8m tall? (answer: 0!?)
- The formulation can become cumbersome when the distribution may
take on some discrete values (say, need to use the delta function)
Example: The emperor declares that every human being taller than 1.8m
must be chopped to 1.8m.
Height Distribution With chopping, some human beings will
of Human Beings indeed be exactly 1.8m tall. We have to
billion/m f(h) model that using a delta function (more
about the delta function later)

h: height
AKW Fall 2011 Chapter 1-2 13
 Computation by Computer – A Discrete Formulation
• Computers typically do not do integration. To handle problems related to
human heights using the computer, typically we will sample the human
height distribution to convert it into a discrete distribution (i.e., as a
histogram). Then we can address many of the same problems numerically
Example we can also model human height as a discrete distribution,
with granularity of, say, 1cm
• Total population on the planet is:
f[h]
billion  f [n]  6 billion
n
0.2
• Number of human beings at least 1.7m
0.1
and less than 1.8m tall: n 1.8/ 1

n 1.7 / 
f [ n]
1.7m 1.71m h: height
So in the continuous model, we do a lot of
integrations, and in the discrete model, we do a lot of
summations. But they are just two nearly equivalent
formulations for solving equivalent problems.

AKW Fall 2011 Chapter 1-2 14

 Real-Valued and Complex-Valued Signals
• A physical signal (share price, temperature, acoustic pressure, etc.)
should have a simple real value. But, as we will soon understand, it is
more convenient to allow, mathematically, that signals can be complex
valued. Use of complex numbers will make our math far simpler and
far more concise.

• Thus, a 1-dimensional signal is a mapping of a real number or an

integer (the CT or DT “time” variable) to a complex number (the
signal value). In ELEC2100 we deal with deterministic signals which
means that the mapping is one-to-one and is a function. We will use
the word signal and mapping/function interchangeably in ELEC2100.
Signal as mapping/function
Domain:
A set of real numbers “time” signal value Range:
or integers A set of complex numbers
(for 2-D signal: A set of a pair
of real numbers or integers)

AKW Fall 2011 Chapter 1-2 15

 CT and DT Signals as Mappings
Parenthesis means
CT Signal: Mapping of a real Continuous-Time
number to a complex number x(t)

x(t)

t
Square bracket means Discrete-
Time. Often, we use n instead of
DT Signal: Mapping of an t to notate integer time

integer to a complex number x[n]

x[t] or x[n]
-2 -1 0 1 2 3 4 5
n
AKW Fall 2011 Chapter 1-2 16
 Finite Duration Signal and Infinite Duration Signal
• A signal may last for a very long time, in which case we say that it lasts
forever, or has infinite duration. Or a signal may last only for a finite
period of time, in which case we say that it is of finite duration. (We also say
that it has finite support; i.e., the domain is finite). A finite duration signal may simply
represent one segment of a longer duration signal that we extract and
process at a time.
A finite duration CT signal with
A finite duration DT signal duration T as a segment of a longer
with duration N=6 signal

x[n] xT(t)

-2 -1 0 1 2 3 4 5 T t
N-1=5
n x(t)

AKW Fall 2011 Chapter 1-2 17

 What is a system?
• A system takes input signals and produces output signals

Input Output
Signals System Signals

So everything we deal with can be regarded as a system

• Examples
- Cell phone (radio wave (in), audible speech signal (out))

- Robot (inertial measurements (in), feedback control to keep robot walking (out))

- ELEC2100 course work (Your effort (in), final grade (out))

• In this course, we will focus on single-input single-output systems

AKW Fall 2011 Chapter 1-2 18

 CT and DT System
• A system that processes CT signals is a Continuous-Time System.
A system that processes DT signals is a Discrete-Time System

CT Input CT Output
x(t) CT System y(t)

DT Input DT Output
x[n] DT System y[n]

• Many of the concepts/properties that we will discuss apply to both

CT and DT signals/systems.

AKW Fall 2011 Chapter 1-2 19

 Math Needed in this course
• Mathematics is a language that allows us to represent “things” and the
associated mathematical logic allows us to process problems in a precise
way.
Example:
The number 38 for the caveman:
The number 38 for Romans: XXXVIII
The number 3 for a toddler:
The number 38 for a modern man: 38

The Arabic numeral system allows modern man to process far more
complicated problems more concisely than the Romans.

AKW Summer 2010 Chapter 1-2 20

 II. Key Math Concepts 1 - Complex Number
• The use of complex number allows us to handle many problems far more
concisely. Complex number is particularly effective when dealing with
problems related to waves and oscillations.

In secondary school, you learned to represent sine waves and exponentials

by:
sin t , cos t , e t , e t sin t

In ELEC2100, we will represent sine waves and exponentials by:

st where s is a complex number, and
e
jt
the special case e where the exponent is purely imaginary

To use the new mathematical convention requires that we be very familiar

with the complex number system.
AKW Summer 2010 Chapter 1-2 21
 Review of Complex Number
• In Rectangular form, a complex number z C is represented as:

z  a  bj; where j  1, and a , b are real numbers

a  Re{ z} is called the real part of z
b  Im{ z} is called the imaginary part of z
complex plane
• A complex number can be viewed as
a vector in a 2-dimenional plane known
as the complex plane.
b
• The magnitude of z is | z | a b
2 2

a
• The angle or phase of z is the angle it forms
with the real axis which is
∠ arctan

AKW Fall 2011 Chapter 1-2 22

Euler’s Relation and Complex Number in Exponential Form
• Euler’s Relation is one of the most elegant relations in mathematic. It
states that for a real number  :
j we will use Euler’s
e  cos   j sin  Relation without proof

• From Euler’s Relation, we can represent a complex number in

Polar Form j
z  re
where r,  are real numbers and are the z  re j
magnitude and angle/phase of z respectively r

From the magnitude and angle of z we can
find its real part and imaginary part of :

Again,
b
r | z | a  b
2 2
  z  tan 1

a
AKW Fall 2011 Chapter 1-2 23
(1) When we add two complex numbers, we add the real
parts and the imaginary parts respectively
+ ; +
+ +

(2) When we multiply two complex numbers, we

multiply the magnitudes and add the phases:
z1 z2  r1e j1 r2 e j 2 =r1r2 e j (1  2 )
Or we can multiply the rectangular form:
+ +
So, a complex number is just like a 2-vector (a vector with two real elements). So why
don’t we just use 2-vectors? The answer is that the polar form of complex number makes
working with angle, phase shift, and magnitude changes easy! An N-complex vector
would be like a 2N-real vector.
AKW Fall 2011 Chapter 1-2 24
 Complex Conjugate
• For a complex number z  a  bj  re j

Its complex conjugate is z   a  bj  re  j

• z*and z is called a conjugate pair. z  a  bj  re j

They have the same real part but r
opposite imaginary parts. 
Alternatively said, they have the
same magnitude but opposite phase.
z  a  bj  re j

• A complex number and its conjugate often appear like twins, so it is

important for you to be familiar with the basic properties of the
complex conjugate:

AKW Fall 2011 Chapter 1-2 25

 Complex Conjugate - Properties
1. Adding a complex number with its conjugate produces 2 times the real
part  
z  z  2 a  2 Re{ z} or Re{ z}  ( z  z ) / 2
2. Subtracting a complex number by its conjugate produces 2j times the
imaginary part
z  z   2bj  2 j Im{ z} or Im{ z}  ( z  z  ) / 2 j
3. Multiplying a complex number with its conjugate produces its
magnitude square
zz   ( a  bj )( a  bj )  a 2  (bj ) 2  a 2  b 2  re j re  j  r 2

 
1/ 2

or | z | zz
4. Conjugate of a sum is the sum of the conjugates and conjugate of a
product is the product of the conjugates. That is, for complex numbers zi’s:

 z  =z z  =z

   
i i ; i i

AKW Fall 2011 Chapter 1-2 26

 Key Math Concepts 2 – Space & Projection

• You should have learned about vector and vector space


• Assume you have a real-valued N-vector x  [ x1 , x2 ,....x N ]

You can think of x as a point in a space with a given coordinate.

You can also think of x as a weighted sum of a set of basic vectors
    
That is: x  x1u1  x2 u2  x3u3 ..   x N u N
Example of a 2-vector

u1  [1, 0, 0...., 0]
where  x2 
u 2  [0,1, 0...., 0]
x
: 
 u2
u N  [0, 0, 0....,1] 
u1 x1
• The magnitude (length) of
the vector is:   2
1/ 2

| x |   xi 
 i 
AKW Fall 2011 Chapter 1-2 27
 Dot Product and Projection
• The dot product of two real vectors is: ∙ ∑ | | | | cos ∠
The dot product is the product of the lengths of the two vectors times the
cosine of the angle between them.
The dot product of a vector with itself is its magnitude square: ∙

• The dot product allows us to compute the projection of onto , which is

⋅
. Projection can be interpreted as the part of that is in the

direction of . In the figure below, p xy is the projection of onto .
 
 x y   
p xy   2 y x
| y| y

⋅ p xy
is a projection coefficient which
describes how much of there is in
AKW Fall 2011 Chapter 1-2 28
 Orthogonal Decomposition
• Two vectors are orthogonal if their dot product is zero: ∙ 0. The
   
basic vectors u1 , u 2 , u 3 .....u N two slides earlier are all orthogonal
• We can decompose vector into a different set of orthogonal basic
vectors:     
x  X 1v1  X 2 v2  X 3v3  ..  X N v N
 
x  vk 
where: Xk   2 x
| vk | 
y
   
with: vn  vk  0 n  k v2 v1

 
x  vk
The projection coefficients X k  | v |2 tell us how much of the new basic
k
vector there is in .

AKW Fall 2011 Chapter 1-2 29

 Dot Product for Complex-Valued Vectors

• For complex-valued vectors, the definition of the dot product needs

to be revised:     *
x  y   xi yi *   y  x 
i

Note that we need to take the conjugate of each component of y .
    *  
The dot product is no longer commutative: x  y   y  x   y  x

• So the magnitude square of a complex vector is

 
x  x   xi xi *
i

• This is important because all the transforms that we will learn in

this course are simply projections. We want to express signals as a
sum of basic signals. The transforms, or projections, allows us to
compute how much of each basic signal there is in the given signal.

AKW Fall 2011 Chapter 1-2 30

 Signal Space
• A finite duration DT signal x[n] of duration N is simply a sequence of N
numbers and can thus be viewed an N-vector. An infinite duration DT
signal can be viewed as an N-vector in the limit N ∞
• A finite or infinite duration CT signal x(t) can be viewed as an N-vector
with N ∞, i.e., infinite dimension
• Thus in advanced signal processing and estimation, we frequently make
use of the term “signal space”. All signals can be viewed as a vector in a
finite-dimension or infinite dimension space.
• To find the dot product of two DT signals, we conjugate-multiply the
individual terms in the two signal sequences, and add up the products
x[ n ]  y[ n ]   x[ k ]y *[ k ]
k Over N terms if finite duration;
Infinite number of terms if infinite duration
• To find the dot product of two CT signals, we conjugate-multiply the
two signals and integrate
x (t )  y (t )   x (t ) y * (t )dt
Over an interval T if finite duration;
over the infinite interval if infinite duration
AKW Fall 2011 Chapter 1-2 31
 Useful Trigonometric Relations
Nevertheless, we will often go back to the high school convention of
representing waves: cost.

We will also make use of these trigonometric relations occasionally:

cos( A  B)  cos A cos B  sin A sin B

sin( A  B )  sin A cos B  cos A sin B
1
cos A cos B  (cos( A  B )  cos( A  B ))
2

AKW Fall 2011 Chapter 1-2 32

 Self-Test
To test your familiarity with complex numbers, plot the followings on
the complex plane:
Im complex plane
A. 2 j
j
B. e j

C. e j / 2
j 1
D. 2e Re

E. e 0.5 j
0.5 j
F. z ;z  e
2

G. z  ; z  e
2 * 0.5 j

AKW Fall 2011 Chapter 1-2 33

 Supplemental Slides

Learning Outcomes - Abilities we expect You to obtain

1. To describe Basic Continuous Time and Discrete Time signals and

different ways to make use of and manipulate them
- Complex exponential, impulse, unit step, window, impulse train
2. To list the Properties of LTI systems and to determine the output of an
LTI system
- In time domain using impulse response with convolution sum/integral
- In frequency domain using frequency response
3. To describe the four basic variants for Fourier Analysis (FS, DTFS, FT
DTFT), their properties, and their relationships to Laplace Transform and
z-Transform
4. To state and prove the sampling theorem using the multiplication and
convolution property

Chapter 1-2 34
 Learning Outcomes - Abilities we expect You to achieve

5. To explain the principles behind practical systems including

Amplitude Modulation, Frequency Division Multiplexing, Envelop
detector, SAW filter, mechanical systems using theories learned
6. To analyze differential and difference equations as causal LTI
systems and to realize them in different block diagram forms
7. To characterize LTI systems base on pole-zero locations and ROC
of their Laplace Transforms
8. To use Matlab as a Software Tools to manipulate, process, analyze
and plot signals

Chapter 1-2 35
 ELEC2100 Application Example I - Telephony
• In 1887, Alexander Graham Bell invented the telephone, which converted
the acoustic pressure from human speech into an electrical signal that was
transmitted down a pair of wires and converted back into an acoustic signal

Acoustic signal Electrical signal Acoustic signal

• How do we model the telephony signal and the

channel? How can we preserve the quality of the
speech signal? How can we transmit many speech
signals at the same time in a cost effective way?
• Many of the early advances in theories on signals
and systems theories are driven by applications in
the telephony system
AKW Fall 2011 Chapter 1-2 36
 ELEC2100 Example II – Radio Communications
• For about 100 years, human has used radio waves to transmit information.
Today, our airwave is filled by transmissions from hundreds of applications
– radio and TV broadcasting, mobile phone, cordless phone, WiFi,
Bluetooth, GPS, walkie talkie. Each application may transmit information
over tens to hundreds of channels
• How is information carried by radio waves? How do all these different
signals share the airwave?

Small
• While it seems so easy
today for us to make a
Power
Signal RF RF

mobile phone call, the

mobile phone and the
Power
Management network is the most
complex system ever built
Analog
by mankind.
Baseband

Digital Baseband
(DSP + MCU)

The radio spectrum allocation table for the US

AKW Fall 2011 Chapter 1-2 37
 ELEC2100 Example III – Control Systems

• How does an automobile suspension system provides a comfortable ride?

How do we keep a moving robot or Segway from falling down? How do
we analyze whether a system is stable or not?

At the end of the course, you will be able address some of these
questions…..
AKW Fall 2011 Chapter 1-2 38
 3 Centuries of Mathematic and Scientific Developments
• Publications on Calculus by Newton (1693) and Leibniz (1684)
• Around 1740, Euler’s relation for complex numbers
• Around 1770, Laplace Transform developed for solving ODE (Ordinary
Differential Equations) with Initial Value Problems
• Around 1800s, Fourier Series and Fourier Transform developed.
• Maxwell’s wave equations in 1864 and radio communications in 1900
• 1930’s, differential machine as analog calculator
• In 1927, Sampling Theorem providing linkage to discrete-time system
• 1970 onwards, widespread use of computer and digital processing of signals

Newton Leonhard Euler Laplace Fourier

1643-1727 1707-1783 1749-1827 1768-1830
AKW Fall 2011 Chapter 1-2 39
 Lecture 2
Chapter 1 - Basic Manipulation and
Characterization of Signals
I) Plotting and Mathematical Expression of Signals

II) Transformation of Independent Variable

III) Even/odd signals

IV) Periodic Signals

V) Energy and Power of Signal

AKW Fall 2011 Chapter 1-2 40

 I. Plotting and Expressing Signals
We need to know how to plot and express different signals. Let’s start
from some simple signals.
Waveforms
1. Unit Sine Wave
 Mathematical expression

2 in angular
cos(0t )

frequency
1
in ordinary
0
cos( 2f 0t ) frequency
-1
T=2/0
-2 Re{e j0t } in complex notation
0 0.5T T 1.5T time period

2. Rectangular Pulse Signal pulse width

1 0  t  T

1
p (t )  
T t 0 Otherwise
AKW Fall 2011 Chapter 1-2 41
3. Square Wave
1 | t  kT | T1
x (t )  
0 T1 | t  kT | T / 2
T1 T/2 T 2T where k is an integer

4. Discrete-Time Unit Impulse (DT Delta Function)

1 n0
Discrete-Time Unit Impulse x[ n ]   [ n ]  
 [n ]  0 otherwise

0 n

AKW Fall 2011 Chapter 1-2 42

 Expressing a Complicated Signal as Sum of
Simpler Signals

5. A Telecommunication Data Signal x (t )  d
k 
k p (t  kT )
2

1 0t T
Data

1
p (t )  
 0 Otherwise
0

-1

-2 dk  1,  1

The telecommunication data signal can be viewed as a superposition, or

summation, of shifted rectangular pulse signals. The coefficient dk for each
shifted rectangular pulse signal can be +1 or -1 depending on whether a 1 bit
or a 0 data bit is transmitted at the k-th bit position.
Decomposition of a signal into a collection of simpler signals is the essence of
this course.

AKW Fall 2011 Chapter 1-2 43

 II. Transformation of the Independent Variable
To decompose and combine signals requires that we become
familiar with shifting and scaling of signals through transformations
of the independent variable:

1. Shifting the signal in time

x(t )  x(t  t0 ); x[n]  x[n  n0 ]
2. Reflecting/Reversing the signal in time
x(t )  x(t ); x[n]  x[n]

3. Time Scaling (compressing/dilating) for CT signals

x(t )  x(at )
Time scaling of a DT signal cannot be done in a straightforward
manner and will be discussed as needed

AKW Fall 2011 Chapter 1-2 44

 Transformation of the independent variable
1. Time shift: x’[n]=x[n-n0], or x’(t)=x(t-t0)
Minus – shift to the right = time delay
Plus – shift to the left = time advance

x’(t)=x(t-t0)
x’[n]=x[n-n0] n0 > 0 t0 < 0

AKW Fall 2011 Chapter 1-2 45

2. Reflection (or Time Reversal): x’[n]=x[-n], or x’(t)=x(-t)

x’[n]=x[-n] x’(t)=x(-t)

AKW Fall 2011 Chapter 1-2 46

3. Time scaling x’(t)=x(t)
time reversal is equivalent to =-1

compression ||>1

expansion ||<1

AKW Fall 2011 Chapter 1-2 47

 Combination of shifting and scaling

Example 1.1-1.3

x1(t) = x(t+1)
shift of x(t) to the left: advance

x2(t) = x1(-t) =x(-t+1)

time reversal of x1(t)

x3(t) = x(t)
|1, compression of x(t)

x4(t) = x(t+1)
shift of x3(t) to the left

AKW Fall 2011 Chapter 1-2 48

 Combination of shifting and scaling
When combining time shifting and scaling, you will find it very helpful
to visualize graphically how you are transforming the time variable t.
x(t)
Example: x’(t) = x(at+b)
 t
Let = at+b. We have x’(t) = x()
Plot  against t. The -intercept is b and the t-intercept is –b/a

= at+b = at+b, = at+b

a>0: no time reversal
|a|<1: dilation x()
x()

 x’(t) x’(t) b
b
(-b)/a
(-b)/a
-b/a t
-b/a t = at+b,
a<0: time reversal
|a|>1: compression
AKW Fall 2011 Chapter 1-2 49
 III. Even and Odd Signals
Sometimes we will exploit symmetry in signals:

 Even signals: x(-t) = x(t) or x[-n] = x[n]

x(t) x[n]

t n

 Odd signals: x(-t) = -x(t) or x[-n] = -x[n]

x(t) x[n]

t n

AKW Fall 2011 Chapter 1-2 50

 Decomposition of Signal into Even and Odd Parts
• Any signal can be written as a sum of an even part and an odd
part:
 
x(t )  xeven (t )  xodd (t )   {x(t )}  d {x(t )}

where the even and odd parts can be created from x(t) and its
time reversed version x(-t)
x(t )  x(t ) x(t )  x(t )
xeven (t )  ; xodd (t ) 
2 2
We can easily check that
x(t )  x(t )
xeven (t )   xeven (t ) is indeed even
2
x(t )  x(t ) is indeed odd
xodd (t )    xodd (t )
2
AKW Fall 2011 Chapter 1-2 51
 IV. Periodic Signals
• Definition: For a CT signal x(t), if there is a positive value T such that
x(t) = x(t+T) for all t, then x(t) is periodic with period T , or T-periodic

x( t )
… ……. We can shift x’(t) by T
… again and again and the
-2T -T 0 T 2T t
signal remains the same.
Shift by T
x’(t) =x(t+T) Hence, it is also true that
… ……. x(t) = x(t+kT)
…
-2T -T 0 T 2T t
• Definition: For a DT signal x[n], if there is a positive value N such that
x[n]=x[n+N] for all n, there x[n] is periodic with period N, or N-periodic

• Aperiodic signals: signals that are not periodic (one cannot find a T
such that x(t) = x(t+T) for all t)
AKW Fall 2011 Chapter 1-2 52
 More on Periodic Signals – the sinuosoid

• The most common periodic signal is the sinusoid cos(0t )

The period is T  2 / 0 where 0 is the angular frequency
cos(o (t  2 / 0 ))  cos(0t  2 )  cos(0t )
We could use sine instead of cosine, but the two really are the
same except for a phase shift – cosine is 90o ahead of sine:
cos(o t )  sin(0t   / 2)

• A periodic signal is a mathematical idealization of a real physical signal

because no physical signal extends to time t   . We are simply assuming
that the signal is periodic over the time interval of interest, implying that any
initial, or start-up effects, are gone.

AKW Fall 2011 Chapter 1-2 53

 More on Periodic Signals – Finite Duration Signal, Poisson Sum

• We know everything about a periodic signal if we know what it is over

T T
one period, say ( , ) . In this sense, a periodic signal is “the same” as a
2 2
finite-duration signal.
• We can take any signal x(t) and generate a periodic signal by forming
what is known as a Poisson sum, which is the adding together of an
infinite number of shifted copies of x(t):

x (t ) 
,
 x(t  kT )
k 

We will come back to the Poisson sum during the study of sampling. We
can easily verify that a Poisson sum is periodic:
for all t
  
x, (t  nT )   x((t  nT )  kT ) 
k 
 x(t  (k  n)T ) 
k 
 x(t
k 
 kT )  x,
(t) t

AKW Fall 2011 Chapter 1-2 54

 V. Power and Energy of Physical Signals
• For physical signals that represent voltage and current, their square is
proportional to its instantaneous power. And the time integral of power is
“energy”.
• As an example, consider the RC circuit below, the power and energy across
the resistor is found by:
1 2
power p (t )  v (t )i (t )  v (t )  Ri 2 (t )
R
C t2 t2
1
+
Energy 
t1
p (t ) dt 
R 
t1
v 2 (t ) dt
+ V(t)
Vs i(t) t2 t2
R
 
_
_
Energy p (t ) dt  R i 2 (t ) dt
t1 t1

AKW Fall 2011 Chapter 1-2 55

 Signal Power & Energy for Complex Signal
But mathematical signals may be complex-valued. So instead of relating
power and energy to the square of the signal, we relate to the square of its
magnitude instead, which is equal to the signal multiplied by its complex
conjugate. We define:

• Definition Complex conjugate

2 2
Power : x(t )  x(t ) x *(t ) Power : x[n]  x[n]x *[n]
CT DT
n2


t2

Energy   | x (t ) | dt
2
Energy  | x [n] |2
t1 nn1

In the future, you will see that these mathematical definitions are indeed
consistent with the physical definitions of power and energy. We observe that
these definitions are valid also for the case when x(t) or x[n] is real. Observe
that the energy in a signal is the self-dot product of the signal.
AKW Fall 2011 Chapter 1-2 56
 Average Power of a Signal over an Interval
• The average power of a signal over an interval (t1, t2) is
t
1 2


2
Pave (t1 , t 2 )  x (t ) dt CT
t1  t2  t1
1 n2

n1  n2  1 n
2
Pave (n1 , n2 )  x[n] DT
 n1

• Signals with practical interest should have a non-zero average power

over some finite interval.

AKW Fall 2011 Chapter 1-2 57

 Signals with No Energy over any Finite Time Interval
Consider the mathematical signals below.

1 t  integer 1 t  rational
 (t )    (t )  
0 otherwise 0 otherwise
They are not zero, but they are zero almost everywhere and they do not
have any power over any finite interval. So practical they are zero.

You can define these signals mathematically. But do you think they exist in
the lab?

More importantly, consider two signals x(t) and y(t) and let e(t) = x(t) - y(t)
be their difference. If e(t) has no power over any finite interval, then
practically x(t) and y(t) are the same and we can mathematically represent
x(t) by y(t).

AKW Fall 2011 Chapter 1-2 58

 Lecture 3
Chapter 1 – Two Fundamental Signals

I) Complex Exponential

II) Impulse

III) Extending to Discrete-Time

AKW Fall 2011 Chapter 1-2 59

 Introduction – Real Exponentials and Sinusoids
We first start with signals that you should be familiar with: the real
exponential and the real sinusoidal
1. Real Exponential Signals: x(t) = eat a is real

a>0 a<0
x( t )
x( t )

t t
A reason why the exponential is so prevalent in natural phenomena
(examples: discharge of a capacitor, radio active decay, etc.) is that it is
unchanged under differentiation except for multiplication by a constant.
de at d 2 e at d k at
e
 ae at , 2
 a 2 at
e ,........, k
 a k at
e
dt dt dt
With the exponential, differentiation is equivalent to multiplication, which
is very nice!
AKW Fall 2011 Chapter 1-2 60
2. Real Sinusoidal Signals: A

x(t )  A cos(o t   ) t


2 0 -A
Period: T0  2
0 T0 
0

Like the exponential, the sinusoid is everywhere in nature – electromagnetic

waves, pendulum, spring mass system. The derivative of a sinusoid is a
sinusoid at the same frequency but there is a phase shift:
d cos(t )
  sin(t )   cos(t   / 2)
dt
d sin(t )
  cos(t )   sin(t   / 2)
dt

AKW Fall 2011 Chapter 1-2 61

 Fundamental Signal I: The Complex Exponential

Central to the study of Signals and System is the Complex

Exponential Signal:
x(t) = est
where s is a complex numbers and is called the complex frequency

The complex exponential allows us to handle real exponentials,

sinusoids, and sinusoids within exponential envelops:
• When s is real, x(t) is the real exponential described in the previous slide
• When s is purely imaginary, we have a complex-valued periodic signal
which is also called the complex sinusoid.
• When s is complex, we have a general complex exponential

AKW Fall 2011 Chapter 1-2 62

3. Complex Sinusoids:
For x ( t )  e where the exponent s  j0 is purely imaginary ( 0 is
st

real), we have a Periodic Complex Exponential Signals, also known as

the Complex Sinusoid.
This signal is periodic because from Euler’s Relation:
e j0t  cos(0t )  j sin(0t )
Which means that x(t) consists of a real and an imaginary sinusoid, both
varying with time at angular frequency 0 , or a period of 2 / 0 .
Im
The Complex e j 0 t e j 0t is a point on the unit circle of
Plane the complex plane. As t increases, e j 0 t
moves around the unit circle in the
counterclockwise direction
Re
The Unit Circle

Chapter 1-2 63
4. General Complex Exponential Signals: For est where s  r  j0
is complex. We have:
e st  e( r  j0 )t  e rt e j0t  e rt cos(0t )  e rt j sin(0t )

The real part is either a growing sinusoid or a decaying sinusoid as shown

below:
x(t )  e rt cos(0t ); r  0 x(t )  e rt cos(0t ); r  0

t t

The signal both oscillates and decays/grows. The imaginary part is the
same except the sinusoidal is 900 out of phase.
Chapter 1-2 64
 Advantages of using the Complex Exponential
- We can combine sinusoids, exponentials, and products of the two into one
general class of signals.
We can visualize est in a number of ways:

- We can easily represents many operations such as differentiation, phase

changes, time, shift, etc., in algebraic forms.

Chapter 1-2 65
 Representing Differentiation: multiplication by s

Example: The derivative of the growing/decaying sinusoid is:

de rt cos( 0 t )
 re rt cos( 0 t )  e rt  0 sin( 0 t )
dt
Working with the complex exponential, we have:

d e st
 se st
dt

Which of the two do you like better??

Chapter 1-2 66
 Representing Magnitude and Phase Change: by
Multiplication with Complex Number
Consider multiplying est by a complex number C

Expressing C in polar form: C | C | e j

we have: Ce st | C | e j e( r  j0 )t | C | e rt e j (0t  )

 | C | e rt cos(0t   )  j | C | e rt sin(0t   )

Multiplication by C has added a magnitude of |C| and a phase of  to

the real and the imaginary sinusoids. This is why the magnitude of C is
called the magnitude and the phase of C is called the phase!

Example: Let C = j
The phase of C is 90o. The magnitude of C is just 1.
Hence multiplication by j is equivalent to a phase advance of 90o.

Chapter 1-2 67
 Prelude to the Impulse – the Unit Step
Before we introduce the impulse signal, let’s introduce the unit step which
is easier to understand
5. Unit step signal, denoted u(t), is defined as

0, t  0
u (t )  
1 t  0
t
The unit step is something that is switched on at time zero.

What is u(t) at t=0? u(0)=0? u(0)=1? u(0)=0.5?

Mathematically, u(0) is undefined because u(t) is discontinuous at t=0.

Physically, the unit step is an idealization because nothing can be switched
on truly in no time.

Chapter 1-2 68
 One-Sided Complex Exponential
Consider again the complex exponential with Re{s}= r ≠ 0. They do not seem to
represent any realistic signals because they grow to infinity either at t = ∞ or t = -∞.
A more “realistic” signal is the one-sided decaying or growing complex
exponential, which we can represent making use of the unit step signal.

6. One-sided Complex Exponential

We can multiply the complex exponential with the unit step:
e st u (t ) is a right-sided complex exponential
Re{e u (t )}
st 8 Re{e st u (t )}
1

r  0.2,   2
6

r  0.2,   2
0.8

0.6 4

0.4
2
0.2

0 0

-0.2
-2
-0.4

-0.6 -4

-0.8
-6
-1
-2 0 2 4 6 8 10
-8
-2 0 2 4 6 8 10

Chapter 1-2 69
 Left-Sided Complex Exponential
Note that u(-t) is the time-reversed version of u(t)

1, t  0
u (t )  
0 t  0
t

Hence, e st u (t ) is a left-sided complex exponential

8 Re{e st u (t )} 1
Re{e st u (t )}
r  0.2,   2
0.8
6
r  0.2,   2 0.6

4
0.4

2 0.2

0
0

-0.2
-2
-0.4
-4
-0.6

-6
-0.8

-8 -1
-10 -8 -6 -4 -2 0 2 -10 -8 -6 -4 -2 0 2

Chapter 1-2 70
 Fundamental Signal II – The Impulse
Finally, we introduce the Continuous-Time Unit Impulse Signal

6. Impulse signal Denoted “(t)” and also called the delta function, it is
defined as the signal whose integral is the unit step:
t
u (t )    ( )d

du (t )
This suggest that (t) is the derivative of u(t):  (t )  .
dt
But this definition is problematic to mathematicians because u(t) is not
continuous at t = 0 and its derivative is undefined there. So we stick
with the integral definition.
(t) belongs to a class of functions known as generalized functions. They
are not defined by their values, but by what they do.

AKW Fall 2011 Chapter 1-2 71

 The Unit Impulse
• So, how to interpret (t)?
One way is to say that (t) is the limit of the derivative of the following
“approximate” step u  (t )
1
limit as slope goes to 
as  goes to 0.

d
• Thus,   (t )  u  (t ) can be viewed as a very narrow and very tall
dt
pulse with an area of 1 under it   (t )
1/
and  (t )  lim   (t )
 0

Our intuition tells us that we should not care too much about the exact
waveform of   (t ) as long as it is very narrow and has an area of 1
under it so that it can integrate to give a close approximation to u(t).
Chapter 1-2 72
• (t) as the following properties:
- zero elsewhere  (t )  0, when t  0

- infinite amplitude at t=0: lim  (t )   

t 0

- unit area 
 ( )d 1

• (t) is represented graphically as

• (t) is the most important conceptual tool in this course. But it contains
infinite energy, within a zero interval of time!
  1 2 1
T otal Energy   |  ( ) | d   lim  | | d   lim  
2
 0 0  0 

So (t) cannot physically exist, but it is one of the most important

mathematical tool or idealization in engineering
AKW Fall 2011 Chapter 1-2 73
Physical Meaning of the Continuous-Time Unit Impulse
In pure math we study math for the sake of math. In applied math we use
math to model practical things.
We can study (t) for its mathematical properties and definitions.
We can think about what (t) means for modeling practical things.

We can fill a pool using a garden hose. We can plot the flow rate of water
into the pool as a function of time
Flow rate of water into the pool
(liter/sec )

time

Chapter 1-2 74
Imaging now that we suddenly empty a large bucket of water into the
swimming pool.
How do we plot the flow rate of water into the swimming pool?

Flow rate
(liter/sec )
Which curve is correct?

t=0 time

Does it matter which curve is correct?

Chapter 1-2 75
Impulse as an Idealization of Things that Happen very quickly

If things happen so quickly, it is infeasible and also meaningless to try to

plot the flow rate exactly. We just assumed that the water is emptied into
the pool instantaneously in no time!

Then we represent the flow rate by an impulse. The magnitude of the

impulse is the amount of water in the bucket

Flow rate
(liter/sec ) k

t=0 time

So the impulse is a mathematical idealization of something that happens so

quickly that we assume it is instantaneous. An impulse of magnitude k causes
the water in the swimming pool to increase by k units.

Chapter 1-2 76
• So we explained what the impulse signal intends to mean physically.
Now we need to be familiar with how to work with (t) mathematically.
• There are two ways to represent u(t) as an integration of the impulse

t 
1 u (t )  

 ( )d 2 u (t )    (t   )d
0

AKW Fall 2011 Chapter 1-2 77

 Scaling the Impulse Signal

• The area under the unit impulse is 1:


 ( )d 1
• We can scale an impulse by multiplying it with a constant k: k (t )

Clearly


k ( )d  k where k is the area of the impulse

We label a scaled impulse by its

area/magnitude as below:

AKW Fall 2011 Chapter 1-2 78

 Time Scaling the Impulse Signal
• Time Scaling  (at )
This property is seldom used in this course, but we state it here only for
curiosity. An impulse is defined by its area, so if we scale an impulse in
time, we obtain the following interesting result:

1
 (at )   (t )
|a|
The intuitive reasoning for this property, however, should be clear to
you. A unit impulse is a very narrow pulse with a total area of 1. If we
compress the unit impulse in time (time scaling with |a| > 1), the area is
reduced proportionally. We apply the absolute value to a because time
reversal (a<0) will not affect the sign of area. Similarly, if we dilate the
unit impulse in time (time scaling with |a| < 1), the area is expanded
proportionally.

AKW Fall 2011 Chapter 1-2 79

 The Sampling Property of (t)
• When we multiple any signal x(t) with (t), we find that the value of x(t)
at anywhere other than t = 0 is irrelevant. Only the value of x(0) matters.
That is:
x(t ) (t )  x(0) (t )

This is because (t) is zero everywhere t  0 , while the area of pulse at

t =0 is scaled by x(0), as visualized below:
 (t)
x(t) 1
x(t) (t)  x(0) (t)

x (0 )


0 
• More generally, if we multiply any signal x(t) with a shifted impulse:

x(t ) (t  t0 )  x(t0 ) (t  t0 )
AKW Fall 2011 Chapter 1-2 80
Summary – Mathematical Properties of (t)

t 
u (t )    ( )d    (t   )d
 0

x(t ) (t )  x(0) (t )

x(t ) (t  t0 )  x(t0 ) (t  t0 )

What is 

x(t ) (t  t0 )dt ?

AKW Fall 2011 Chapter 1-2 81

 III. Basic Discrete-time Signals
Now we move on to DT signals
1. Real Exponential Signal: x(n) = n where  is real

>1 growing exponential

 < 1 decaying exponential

AKW Fall 2011 Chapter 1-2 82

1. Real Exponential Signals (Cont): x(n) = n
If  is negative, the signal alternates between positive and
negative values
 < 0, || < 1

 < 0, || > 1

AKW Fall 2011 Chapter 1-2 83

 Basic DT Signals
2. Discrete-Time Sinusoidal signals x[ n ]  cos(0 n )

• It looks like the CT sinusoid except that

now the independent variable “time” is
discrete and takes on only an integer
value
• The above gives rise to two note-worthy
facts about the DT sinusoid:
1. (not so important) It is not always
strictly-speaking periodic
2. (very important!)
cos( 0 n ) and cos(( 0  m 2 ) n ) are the
same sinusoid for any integer m

AKW Fall 2011 Chapter 1-2 84

 Two facts about the DT sinusoid
1.While the envelop of a DT sinusoid is always periodic with period 2 / 0 .
For x[n] to be periodic, we need an N such that x[n]=x[n+N] for all n, or
cos(0 n)  cos(0 ( n  N ))  cos(0 n  0 N ) n
This implies 0 N  m  2 for some integer m , or  0  m 2 / N

Since  is irrational, this means  0 must be a rational number multiplied by 2 .

Alternative stated, the frequency f0  0 / 2 must be a rational number

Question – which of the three examples in the previous page is/are periodic?

2. Since m, n are integers, cos((0  m 2 ) n )  cos(0 n  mn 2 )  cos(0 n ) .

DT sinusoids with angular frequencies 0 and 0  2 refer to the same
DT sinusoids. The two frequencies are completely equivalent for DT
signals.

AKW Fall 2011 Chapter 1-2 85

 Some Questions to think about concerning the DT sinusoid

1. What is the minimum possible period N for a DT signal?

2. What is the highest frequency possible for a DT sinusoid?

3. How many distinct DT sinusoids there are that is periodic with period N?

AKW Fall 2011 Chapter 1-2 86

3. Complex Sinusoid:
As in CT, this is the case of a complex exponential where the exponent is
purely imaginary: j0 n
x[ n ]  e  cos( 0 n )  j sin( 0 n )
• Again, x[n] is also called the complex sinusoid. x[n] has an envelop that
is periodic with period 2 / 0 .
• x[n] is strictly speaking periodic, however, only if there are integers N, m
such that No  m2 , in which case the period is N:
cos(0 ( n  N ))  cos(0 n  m 2 )  cos(0 n)
sin(0 ( n  N ))  sin(0 n  m 2 )  sin(0 n)
j n j (  2 ) n
• Also, e 0  e 0 and the maximum angular frequency is 
(maximum ordinary frequency is 1/2)

AKW Fall 2011 Chapter 1-2 87

4. General Complex Exponential Signals
C n where C and  are generally complex
Expressing C and  in polar form: C | C | e j ;  |  | e j o

C n | C ||  |n e j (o n  ) | C ||  |n [cos(o n   )]  j | C ||  |n [sin(o n   )]

|  | 1
Focusing on the real part only,
we plot:
|  |n [cos( o n   )]
in the figure to the right
|  | 1

AKW Fall 2011 Chapter 1-2 88

 Basic DT Signals – Unit Step
5. Unit step signal/sequence:
Unit step Signal becomes 1 for n > 0
u[n]
0, n  0
u[n]  
1, n  0
n

As in the CT case, u[-n] is the time-reversal of u[n]

u[-n]
0, n  0
u[n]  
1, n  0
n

AKW Fall 2011 Chapter 1-2 89

 Basic DT Signals – One-Sided Complex Exponential
6. One-sided complex exponential
Right-sided Complex Exponential  n u[ n ]
 e 0.1 j 0.4 
 1.105e j 0.4    e 0.1 j 0.4  0.905e j 0.4
5

0.8

0.6

0.4

0.2
0 0

-0.2

-0.4

-0.6

-0.8
-5 0 5 10 15 20

-5
-5 0 5 10 15 20

Left-sided Complex Exponential  n u[  n ]

  e 0.1 j 0.4  1.105e j 0.4 5   e 0.1 j 0.4  0.905e j 0.4
1

0.8

0.6

0.4

0.2 0
0

-0.2

-0.4

-0.6

-0.8
-20 -15 -10 -5 0 5

-5
-20 -15 -10 -5 0 5

AKW Fall 2011 Chapter 1-2 90

 Basic DT Signals –Unit Impulse
Now, unlike the CT case, the unit impulse in DT is straightforward
7. Unit Impulse signal/sequence (also called the unit
sample signal)
Unit impulse/sample
[n] Signal is 0 everywhere
0, n  0 except at n = 0
 [ n]  
1, n  0
0 n

AKW Fall 2011 Chapter 1-2 91

 Relationship of [n] and u[n] in DT
1. The unit impulse sequence is the difference between two
unit step sequences: [n] = u[n] – u[n-1]
2. The unit step can be written as a sum function of the unit
impulse sequence in two ways:


  [n  k ]
n
1 u[ n ] 
  [m]
m  
2 u[ n ] 
k 0

n0
m

n0
m

AKW Fall 2011 Chapter 1-2 92

 The Sampling Property of [n]
3. x[n]  [n] = x[0]  [n]
4. x[n]  [n-n0] = x[n0]  [n-n0]
As in CT, these two are known as the sampling property of the unit
sample signal. Only the one sampled value of x[n] matters.
x[n]
Independent of n, A function of n,
a constant
x[1][n-1]
-2 -1 0 1 2 3 4 5
n Multiply
[n-1]
n

AKW Fall 2011 Chapter 1-2 93

 Lecture 4
Chapter 1 – Basic Characterization of
Systems

• Six Basic Characterization of Systems

• Linear Time Invariant (LTI) Systems

AKW Fall 2011 Chapter 1-2 94

 CT and DT Systems
Now we move on to discuss properties of systems. Our discussion
applies to both CT and DT systems.

• Continuous-time systems

x(t) y(t)

• Discrete-time systems

x[n] y[n]

AKW Fall 2011 Chapter 1-2 95

 Example 1: A RC circuit
System described by a simple differential equation:

• Question: Relationship between Vs (input) and Vc (outputs) in the

following RC circuit?

R Vs (t )  i (t ) R  Vc (t )
dVc (t )
i (t )  C
+
Vc dt
Vs i(t)
_ C

dVc (t )
Vs (t )  CR  Vc (t )
dt
AKW Fall 2011 Chapter 1-2 96
 Example 2: Balance in a bank account
from month to month
DT system described by a difference equation:

x[n] + y[n]
Input: Unit delay
Net deposit 1.01
Output:
y[ n  1] Bank balance
in month n Interest rate = Bank balance at month n
1% per month at month n-1

y [ n ]  1 .0 1 y [ n  1]  x [ n ]
 y [ n ]  1 .0 1 y [ n  1]  x [ n ]

AKW Fall 2011 Chapter 1-2 97

Example 3: A Spring/Damper System
The following is a commonly found 2nd order
spring/damper system. Examples:
1. Suspension systems in automobiles to provide
comfortable rides
2. Hinges in self-closing doors
3. Accelerometers in your notebook PCs, cell phones,
etc., to detect motions and falls

AKW Fall 2011 Chapter 1-2 98

 Simple systems  complicated systems
• Simple systems can be
interconnected to build a Serial

more complicated system

• Interconnections
Parallel
a) Serial or cascade
b) Parallel
c) Serial-Parallel Combination Combination

d) Feedback Input
+ System 1
Output

System 2

(d) Feedback

AKW Fall 2011 Chapter 1-2 99

 Characterization of Systems
Now we define 6 characterizations of systems:
1. Memory or Memoryless
• If the output of a system at time t (or n) depends on the past or future
input signal, then we say the system has memory.
• If the output of a system at time t (or n) depends only on the input
signal at time t, then we say the system is memoryless.
• Examples (classify the following as memoryless system or not)

y (t )  5 x (t ) t
y (t )   x( )d
y (t )  5 x (t  1) 

y (t )  x (t ) cos( 2t  5) y (t )  5 x(t  1)

AKW Fall 2011 Chapter 1-2 100

 Invertible systems

2. Invertibility

- A system is said to be invertible if distinct inputs lead to

distinct outputs; otherwise it is non-invertible.

- Examples (Classify the following as invertible system or not )

dx (t )
y (t )  5 x (t ) y (t ) 
dt
y (t )  5 x (t  1) y (t )  5 x 2 (t )
t
y (t )   x( )d

AKW Fall 2011 Chapter 1-2 101
 Causal systems
3. Causality
- A system is causal if the output at any time depends only on values of
the input at the present and the past, but not the future.
- We expect that real physical systems with time as the independent
variable should be causal
- Examples - Classify the followings as causal systems or not:

y(t )  5x(t )
y(t )  x(t  5)
t 5
y(t )   x( )d
t
y(t )   x( )d



y(t )  x(t ) cos(t  5)

AKW Fall 2011 Chapter 1-2 102
 Stable systems
4. Stability
- A system is stable if bounded inputs lead to bounded outputs.
A bounded signal means that there is a finite constant B such
that the magnitude of the signal is no greater than B for all t.
If B   such that x (t )  B t ,
Means
“there is” then B '  , y (t )  B ' t.

- Examples Stable? t

1. y (t )  5 x (t ) 2. y (t )   x ( ) d

3. Example 2 of systems (bank balance)
dy (t ) d 2 y (t )
4. y (t )  2 3 2
 x (t )
dt dt
We will come back to this in greater details
AKW Fall 2011 Chapter 1-2 103
 Time-invariant systems
5. Time-invariance: a system is time-invariant (TI) if the behavior
and characteristics of the system are fixed over time. This implies
that shifted input will give shifted output. That is, if y1(t) is the
output to x1(t) , then y1(t-t0) must be the output to x1(t -t0) .
• Example 1.14. Consider a CT system defined by y ( t )  sin[ x ( t )]

Is it TI? To check, This operator itself is

let x (t )  Not a function of time
2 x1 (t  t 0 )
then y 2 (t )  sin[ x2 (t )]  sin[ x1 (t  t 0 )]  y1 (t  t 0 )

Since shifted input gives shifted output, this system is time-invariant

AKW Fall 2011 Chapter 1-2 104

Example 1.15. Is the following DT system TI?
This is a function of time
y[ n ]  nx [ n ]
1 n0
Consider the input x1 [ n ]   [ n ]  
0 n0
The output is y1 [ n ]  0 n
1 n 1
Now consider the input x 2 [ n ]   [ n  1]  
0 n 1
The output is y 2 [ n ]   [ n  1]  y1 [ n ]

Hence the system is not time-invariant

Any system that is a time-dependent operation on the input is in

general not time-invariant
AKW Fall 2011 Chapter 1-2 105
 Time-invariant systems (cont.)

• Example 1.16

y (t )  x ( 2t )

Again, the system is not

time-invariant

not equal

AKW Fall 2011 Chapter 1-2 106

 Time-invariant systems (cont.)

It is not difficult to convince ourselves that any system that compresses

or dilates a signal in time is not time-invariant. This is because time
compression or dilation is centered around a fixed time-axis (t=0).

Other systems that is centered around a fixed time-axis are also not
time-invariant:

Examples: y (t )  x (  t )
y (t )  Ev{x (t )} Not Time-Invariant!

y (t )  Od {x (t )}

AKW Fall 2011 Chapter 1-2 107

 Linear systems
6. Linearity a system is linear if the superposition property holds. It means
that if the input is a weighted sum (superposition) of some inputs, the output
must be the same weighted sum of the individual outputs
CT: ax1 (t )  bx2 (t )  ay1 (t )  by2 (t )
DT: ax1[n]  bx2 [n]  ay1[n]  by2 [n]
Where “  ” means producing the output. The weighting factors a, b can
generally be complex.
• Superposition consists of two sub-properties:
i. The additive property: x1 (t )  x2 (t )  y1 (t )  y2 (t )
ii. The homogeneity (or scaling) property: ax1 (t )  ay1 (t )

AKW Fall 2011 Chapter 1-2 108

 Linearity
• Note that to satisfy the homogeneity property, the system must satisfy the
zero-input/zero-output property – if input is 0, output must be 0
• Mathematically, one can prove that the homogeneity property follows
from the additive property for all a that is a rational number. The
homogeneity property is almost redundant but not quite
Example (1.17) y (t )  tx(t )
You can easily verify that this is linear

The superposition property for linear systems forms the basis of all of the analysis
that we will do in this course. With linearity, we can decompose a signal into a sum
of individual components, determine how a system would response to them, and can
add the individual responses together at the end. Nonlinear systems are often very
hard to analyze. (Read the story about Harold Black’ invention of using feedback to make amplifiers
linear. It is one key invention that has made the modern electronic age possible.)

AKW Fall 2011 Chapter 1-2 109

• Example 1.18 y (t )  x 2 (t )
Let’s apply the linearity checking procedure:
If x3(t) = ax1(t) + bx2(t) , does y3(t) = ay1(t) + by2(t) ??

AKW Fall 2011 Chapter 1-2 110

• Example 1.19 y[n]  e{x[n]}

AKW Fall 2011 Chapter 1-2 111

Linear systems (cont.)

Does not satisfy zero-in/zero-out

AKW Fall 2011 Chapter 1-2 112

 Linear time-invariant (LTI) systems
ELEC2100 will focus on the study of systems for which both
linearity and time-invariance hold:
• Linearity: Superposition holds
x ( t )   a i xi ( t )  y (t )   ai yi (t )
• Time Invariance: Independent of time shift
x (t )  x1 (t  t 0 )  y (t )  y1 (t  t 0 )
• Such systems are called Linear Time-Invariant (LTI) systems.
- LTI systems are amenable to analysis, and many physical
systems are LTI or can be approximated as LTI.
• In Chapter 2 we will begin the study of LTI systems.

AKW Fall 2011 Chapter 1-2 113

Supplemental Slides on LTI Systems

AKW Fall 2011 Chapter 1-2 114

 Linear and Time Invariant System
• This course focuses on the study of Linear and Time Invariant Systems
(LTI).
• Time invariance means that the property of the system does not
change with time. So if we delay the input to the system, the output
will be delayed correspondingly but otherwise unchanged.
Time Invariance
input output Delayed Input Delayed Output
implies

• Most physical laws, such as Newton’s F = ma and Einstein’s E = mc2,

do not change with time. Say, if we have a piece of electronic circuit
sitting in the lab, we also do not expect the behavior of this circuit to
change (at least not quickly) over time
Chapter 1-2 115
 Linear and Time Invariant System
• Linearity means that if any input x1(t) produces output y1(t) and
any input x2(t) produces output y2(t) , an input that is ax1(t) +bx2(t)
will produce the output ay1(t) +by2(t) .
Linearity
input x1(t) Output y1(t)
Input ax1(t) Output ay1(t)

implies

Input Output
input x2(t) Output y2(t)
ax1(t) + bx2(t) ay1(t) + by2(t)

As an example, let x(t) be the rate at which you deposit money into your
bank account, and let y(t) be the account balance at time t. We expect this
system to satisfy linearity
Chapter 1-2 116
 Linearity in Real World Systems
• Within a limited range of operation, many physical and physical laws are
linear. For many engineering systems, it is often very important to keep
the system operating within the linear range.
• As an example, in the early part of the 20th century, people used the
vacuum tube to amplify telephone signals that were attenuated as they
were transmitted over long distances. But the vacuum tube was not very
linear, and the telephone signal was distorted as a result. Engineers tried
very hard to make vacuums tube more linear.
• In 1927, a Bell Lab engineer named Harold Black, while commuting to
work on the New York Ferry, came up with and scribbled on a piece of
newspaper the idea of using negative feedback to build approximately
linear amplifier. This simple idea has dramatically changed the world!

Chapter 1-2 117

 Negative Feedback Amplifier
• For an amplifier, the output y at time t is an amplified version of the input
x at the same time instance, i.e., y = A(x) and A(x) >> x
• We want the amplifier to be linear; i.e., we want y to be some constant a
times x, or y = ax. That means if we plot output y against input x, we want
the curve to be a straight line (hopefully a steep one)
• But for the vacuum tube, the curve is not a straight line. For the transistor
(not invented until 1947), A(x) is not only highly nonlinear, it is also
highly uncertain and varies from transistor to transistor
Ideal Amplifier
x100 A(x) = ax
Vacuum tube:
input Nonlinear Output y Nonlinear A(x)
Amplifier
x(t) y(t) = A(x(t))
Transistor:
x y=A(x) Nonlinear and
Uncertain A(x)

y=x
(y on x100 scale)

x Chapter 1-2 118

 Negative Feedback Amplifier
• Black came up with the idea of using negative feedback as shown. A
portion of the output y is subtracted from the input x before the input is
sent into the amplifier:
Nonlinear
Amplifier Output
Input x z=
y=A(z)
z = x - bA(z)  x = z + bA(z)
x - by
+ y A(z)
-  =
R1 x z + bA(z)
by = bA(z) b=
R2/(R1+R2)
R1 and R2 are R2
resistances of resistors

If A(z) is so large such that bA(z) is much larger than z, then we can ignore z in
the denominator and the gain of the entire system becomes:
The gain is determined only by the passive
y A(z) 1 R1+R2
= = = values of the two resistors, and the actual
x bA(z) b R2 value of A(z) does not matter!
Chapter 1-2 119
 ELEC2100: Signals and Systems

Lecture 5

Chapter 2 – Impulse Response

I) LTI System and Impulse Response

II) Convolution sum and convolution integral

AKW Fall 2011

Chapter 1-2 120
Characterizing Discrete-Time (DT) LTI systems
• Given an input to an LTI system, what can we say about
the output or response of the system?
• How can we specify the input/output relationship of an
LTI system?

Let’s start with the Discrete-Time case

Given an input What is the response

x[n] System (DT) y[n] ???

AKW Fall 2011

Chapter 1-2 121
 I. Unit Impulse Response of (DT) LTI systems
• The simplest DT input signal is the unit impulse signal [n].
Let h[n] be the response to [n]. We call h[n] the unit
impulse/sample response, or simply the impulse response.

unit Impulse h[n ]

 [n] impulse response
-2 -1 0 1 2 n n

• h[n] completely specifies a DT LTI system. That is, if we know

what h[n] is, we can determine the response of this system under
any input. This is because all DT input signals can be viewed as a
superposition of shifted unit impulse signals.

AKW Fall 2011

Chapter 1-2 122
 Representation of DT Signals as Impulses
The signal x[n] on the left below can be regarded as a sum of the
weighted and shifted unit impulse signals on the right

x[1] [n 1]
… … k  1
x[n] -2 -1 0 1 2 n
+
x[0] [n]
…
-2 -1 0 1 2
…
n = … … k 0
-2 -1 0 1 2 n

Mathematically: shifted impulse +

x[1] [n 1]

x[n]   x[k ] [n  k ]
k  
…
-2 -1 0 1 2
…
n
k 1

AKW Fall 2011

Chapter 1-2 123
 Output of LTI (DT) systems (cont.)
• Now, we return to the question:

x[n] system What is the response ???

• Let hk [n] be the system’s responses the shifted impulse

function  [n  k ] . That is,  [n  k ]  hk [n]

• So, if the system is linear, the output is the sum of the

responses to the weighted and shifted impulse functions!

 
x[n]   x[k ] [n  k ]
k  
Linear System y[n]   x[k ]h [n]
k  
k

Eq. (2.3)

AKW Fall 2011

Chapter 1-2 124
 Output of LTI (DT) systems (cont.)
• Next, if the system is Time Invariant, then hk [ n]  h0 [ n  k ]
where h0 [n]  h[n] which is the unit impulse response

• Replacing hk[n] by h[n – k] in Eq. (2.3), we have:

Eq. (2.6)
Linear 

x[n]   x[k ] [n  k ]
k  
Time-Invariant
(LTI)
y[n]   x[k ]h[n  k ]
k  
System

• Eq. (2.6) is called the convolution sum. It means that y[n], the
output of an LTI system at any time n, can be founded by
summing up the responses of the system due to the inputs at all
different times - at all values of k, and the response at time n due
to the input at time k is x[k]h[n-k].
AKW Fall 2011
Chapter 1-2 125
 The Convolution Sum
• We also say that the output of an LTI system is the convolution of the
input with the system’s impulse response. We notate the convolution of
x[n] and h[n] as x[n]*h[n]

y [ n ]  x[ n ]  h[ n ]   x[ k ]h [ n  k ]
k  

Conversely, if the output of a system is given by a convolution sum of

the input, the system is an LTI system! This is obvious from the meaning
of the convolution sum, or it can be shown mathematically as below:


Linearity:
( ax 1 [ n ]  bx 2 [ n ])  h [ n ]   ( ax
k  
1 [ k ]  bx 2 [ k ]) h [ n  k ]
 
a  x [ k ]h [ n  k ]  b  x
k  
1
k  
2 [ k ]h [ n  k ]

 k ' k  m 

TI: x[ n  m ]  h[ n ]   x[ k  m ]h [ n  k ]
k  
  x[ k ' ]h [( n  m )  k ' ]  y[ n  m ]
k '  
AKW Fall 2011
Chapter 1-2 126
 Meaning of The Convolution Sum
• Example: Imagine there is a couponed US government bond that
provides the following payout schedule:
10% rebate immediately  h[0]=0.1
Thus, you invest $1,
8% after one year  h[1]=0.08 you get $1.26 back
8% after two years  h[2]=0.08 in total over 3 years.
100% at the end of the third year  h[3]=1
• To help pay for your HKUST education, your parents buy for you
$10 of this bond in 2005, $20 of this bond in 2006, $40 of this
bond in 2007. How much of a payout will you receive in 2009?
y [2 0 0 9 ]  x [ 2 0 0 5]  h [ 4 ] 2009-2005=4 2009-2007=2
 x [ 2 0 0 6 ]  h [3]  x [ 2 0 0 7 ]  h [2 ]
 x [ 2 0 0 8]  h [1]  x [ 2 0 0 9 ]  h [0 ]
 $ 0 + $ 2 0  1 0 0 % + $ 4 0  8 % + $ 0 + $ 0  $ 2 3 .2
AKW Fall 2011
Chapter 1-2 127
Output of LTI (DT) systems – Another Look
Unit Impulse Response
y[n] h[n]
 [n] DT system

-2 -1 0 1 2
n
Time Invariance
 [n  2] y[n] h[n  2]
TI- DT system
-2 -1 0 1 2
n
Linearity
 [ n]  2 [ n  2] h[ n ]  2 h[ n  2 ]
y[n]
LTI- DT system
-2 -1 0 1 2
n

AKW Fall 2011

Chapter 1-2 128
 Convolution with [n]
• Consider convolving a DT signal x[n] with [n]:
0 n k
 [n  k ]  
 1 nk
x[ n ]   [ n ]  
k  
x [ k ] [ n  k ]  x [ n ]

This means that [n] is the identity function under convolution.

Convolving any signal x[n] with [n] results in x[n] unchanged.

So an LTI system with h[n] = [n] is a do-nothing system

h[n] = A[n] is an amplifier that multiplies every input sample by A

h[n] = [n-m] means a delay system that delays the input by m

x[ n ]   [ n  m ]  
k  
x [ k ] [( n  m )  k ]  x [ n  m ]
equals 1 only if k=n-m
AKW Fall 2011
Chapter 1-2 129
 Example DT Impulse Responses
What do these DT systems do?
1. h[n]   [n] 4. h[n]  u[n]

……
-2 -1 0 1 2 n -2 -1 0 1 2 3 4

1
2. h[n]   [n  2] 5. h [ n ]   u [ n ]  u [ n  3] 
3
1/ 3
-2 -1 0 1 2 n -2 -1 0 1 2 3 4

3. h[n]   [n]  0.3 [n  4] 6. h[n]   [n]   [n  1]

-2 -1 0 1 2 3 4 n -2 -1 0 1 2

AKW Fall 2011

Chapter 1-2 130
A Pretty Picture of Lena and its Thumbnail

Chapter 1-2 131

 Lena with lower resolution

Enlarging
the thumbnail

Chapter 1-2 132

 Lower Resolution Photo Smoothed

We can obtain a decent

picture of Lena by
enlarging and smoothing
the thumbnail.

The perimeter of the

picture is dark because
those pixels do not have
real data next to them for
smoothing

Chapter 1-2 133

 Lena After First-Difference

First difference
highlight the edges
in the picture

Chapter 1-2 134

 Impulse response – Continuous Time Systems
• Next, we consider the Continuous-Time system. Now, assume that the
response of the CT system to the CT impulse function (t) is h(t).
• h(t) is called the impulse response.
Impulse response
h(t )
 (t ) h(t )
CT system
t t

• We argued in Chapter 1 that (t) can be approximated by (t), a

narrow pulse function with area of 1 and width . We will show that
any CT input can be approximated by a weighted sum of shifted (t)
functions. This is called a staircase approximation. Then, when we take
limit of  0, the staircase approximation become exact and the
weighted sum becomes an integral.
AKW Fall 2011
Chapter 1-2 135
 Impulse representation of CT Signals
Define an approximate impulse:
x(0) x()
x()
x ( 2 )   (t )
x(t)
… … 1

  3 t
 t
x()  (t  ) 
xˆ (t )   x(k)
k  
 (t  k)
 t
x(0)  (t )
xˆ (t ) is a staircase approximation of x(t)
xˆ (t )
t
x()  (t  )

t x(t) can be approximated by a sum of

AKW Fall 2011 shifted approximate impulses
Chapter 1-2 136
When  becomes very small, the staircase approximation becomes exact,
and the summation becomes an integral:

x ( t )  lim xˆ ( t )  lim
0  0

k  
x ( k  )   ( t  k  )


the convolution integral
 

x ( )  ( t   ) d 

Observe that the integral looks very similar to the convolution sum.
In fact it is called a convolution integral. In this case it is the
convolution of x(t) with (t) and it has two meanings:
1. Any signal x(t) can be viewed as weighted sum of infinitely many
shifted impulses with infinitesimally small weights
2. The impulse signal is the identity under convolution. Any x(t)
convolving with (t) produces x(t).
AKW Fall 2011
Chapter 1-2 137
 Output of LTI (CT) systems
Because of LTI, if the input is a weighted sum of shifted impulses, the
output must be the same weighted sum of the shifted impulse responses
h(t )
 (t ) h(t )
CT system
t
t

 (t  t 0 ) h(t  t0 )
h(t )
TI- CT system
t0 t
t

x(0) (t )  x(t0 ) (t  t0 ) x(0)h(t )  x(t0 )h(t  t0 )

h(t )
LTI- CT system
t
t
AKW Fall 2011
Chapter 1-2 138
Since the input can be viewed as infinitely many shifted, weighted impulses added
(integrated, superposed) together, the output must be infinitely many shifted,
weighted impulse responses integrated together Where h (t ) is the system’s response to the

approximate impulse   (t )


 x(k)h

yˆ (t )  (t  k)
xˆ (t )   x(k)
k  
 (t  k )
LTI-CT system k  


x(t )  lim


 x(k)  (t  k)
y (t )  lim
 0
 x(k)h
k  
 (t  k )
 0
k   LTI-CT system



x(t )   x( ) (t   )d LTI-CT system
y (t )   x( )h(t   )d



• Convolution integral 
y (t )  x(t )  h(t )   x( )h(t   )d

AKW Fall 2011
Chapter 1-2 139
 Convolving with (t)
• Now, what do you think is a CT LTI system with h(t)=(t)?
Let x(t) be the input
sifting property of ():
y (t )  x(t)  h(t)  x(t)   (t) x( ) (t  )  x(t) (t  )
  
  x( ) (t   )d   x(t ) (t   )d x(t )   (t   )d x(t )
  

Again, convolving any signal x(t) with (t) yields x(t) unchanged
• What if h(t)=(t – t0)?
 
y ( t )  x(t)   (t  t 0 )  

x ( ) ( t  t 0   ) d   

x ( ) (( t  t 0 )   ) d 

 
= 

x ( t  t 0 ) (( t  t 0 )   ) d   = x ( t  t 0 )   (( t  t 0 )   ) d   x ( t  t 0 )


Again, convolving with a shifted impulse yields a shifted version of the

input AKW Fall 2011
Chapter 1-2 140
 Example CT Impulse Responses
1. Time-shifter h ( t )   ( t  t 0 )  y ( t )  x ( t  t 0 )

h(t) = (t – t0) h(t) = (t – t0)

t0 > 0 delay system
t0 < 0 time advance system
1 1

0 t0 t t0 0 t

2. Echo System:
h ( t )   ( t )  0.3 ( t  4)  y ( t )  x ( t )  0.3 x ( t  4)
h(t) = (t) + 0.3 (t - 4)

0 t

Chapter 1-2 141

3.Integrator: h ( t )  u ( t )

h(t) = u(t) u(t – )=0 for > t

1
 t
y (t )   
x ( ) u ( t   ) d    
x ( ) d 
0 t
Thus, an integrator is an LTI system!

4. Smoother (averager): h ( t )   u ( t )  u ( t  W )  / W

h(t) = (u(t) – u(t-3))1/3

h(t – )=0 for > t or < t – 3
 1 t
y ( t )   x ( ) h ( t   ) d    x ( ) d 
1/3

0 3 t  3 t 3

Chapter 1-2 142

 CT Differentiator
How about a system that will take the derivative of the input (a
differentiator)? Recall x(t) can be viewed as its convolution with (t):

x (t )   
x ( ) ( t   ) d 

Taking the derivative of both sides with respect to t:

dx ( t ) d    d 
dt

dt  x ( )  ( t   ) d    x ( ) 
 dt
 ( t   )  d

Thus, dx ( t ) is the convolution of x(t) with the derivative of (t).
dt
So, the differentiator, like the integrator, is also an LTI system. The
differentiator has an impulse response that is the derivative of (t) –
which we call the unit doublet.
The unit doublet is discussed in Section 2.5.3 of the text

Chapter 1-2 143

 Summary of LTI systems

• Impulse responses x[n]   [n]  y[n]  h[n]

x(t )   (t )  y (t )  h(t )

• Output of LTI systems x[n]  y[n]  x[n]  h[n]

x(t )  y(t )  x(t )  h(t )


• Convolution y[n]  x[n]  h[n]   x[k ]h [n  k ]
k  

y (t )  x(t )  h(t )   x( )h(t   )d


AKW Fall 2011

Chapter 1-2 144
 Lecture 6
Chapter 2 – Properties and Characterization of LTI
Systems
• Commutative, Distributive and Associative Properties of LTI
systems

• Characterization of LTI systems from Impulse Response

• Convolution Integral Example

AKW Fall 2011

Chapter 1-2 145
 Properties of LTI systems -
1. Commutative property x(t )  h(t )  h(t )  x(t )
x[n]  h[n]  h[n]  x[n]
1. Substitution of variable
Proof:   't  
x(t)  h(t)   x( )h(t   )d

  x(t   ')h( ')d ( ')

2. Limits of integration
 changed correspondingly
  x(t   ')h( ')d '  h(t )  x(t )
 3. Reverse limits of integration with change of sign

The commutative symmetry is obvious from our couponed bond example. We can compute
the payout in 2009 by enumerating over the years in which a bond purchase was made:
y [2009]  x [2005]  h [2009  2005]  x [2006]  h [2009  2006]
 x [2007 ]  h [2009  2007 ]  x [2008]  h [2009  2008]  x [2009]  h [2009  2009]
Or we can enumerate over h[n]
y [2009]  h [0]  x [2009  0]  h [1]  x [2009  1]
 h [2]  x [ 2009  2]  h [3]  x [2009  3]  h [4]  h [2009  4]
Chapter 1-2 146
The commutative property means that the followings are equivalent:

x(t) y(t)=x(t)*h(t)
System h(t)

h(t) y(t)=h(t)*x(t)
System x(t)

From now on, we will say “x convolved with h” or “h convolved with x”

interchangeably

AKW Fall 2011

Chapter 1-2 147
 Properties of LTI systems -
2. Distributive property
x(t )  [h1 (t )  h2 (t )]  x(t )  h1 (t )  x(t )  h2 (t )
x[n]  {h1 [n]  h2 [n]}  x[n]  h1 [n]  x[n]  h2 [n]

Proof and meaning:


x(t)[h1 (t )  h2 (t )]   x( ){h1 (t  )  h2 (t  )}d

 
  x( )h1 (t  )d   x( )h2 (t  )d
 

AKW Fall 2011

Chapter 1-2 148
 The distributive property also follows from the commutative
property and linearity

x(t) h1(t)+h2(t)
commutative
h1(t)+h2(t) x(t)
x(t )  [h1 (t )  h2 (t )]
linearity  [h1 (t )  h2 (t )]  x(t )
h1(t) x(t)  h1 (t )  x(t )  h2 (t )  x(t )
+  x(t )  h1 (t )  x(t )  h2 (t )
h2(t) x(t)
commutative
x(t) h1(t)
+
x(t) h2(t) Isn’t this fun ?!
AKW Fall 2011
Chapter 1-2 149
 Properties of LTI systems -
3. Associative property
x(t )  [h1 (t )  h2 (t )]  [ x(t )  h1 (t )]  h2 (t )
x[n]  {h1 [n]  h2 [n]}  {x[n]  h1 [n]}  h2 [n]
We use the symbol  because
Proof: the symbol  is already used
x(t) [h1 (t )  h2 (t )]  x(t ) [h2 (t )  h1 (t )] Then this is g(t-)
Regard as g(t)




 x(t )   h2 ( )h1 (t  )d   x( )   h2 ( )h1 (t    )d  d
   

   
  h2 ( )   x( )h1 (t    )d d  h2 (t )    x( )h1 (t   )d 
    
 h2 (t ) [ x(t )  h1 (t )]  [ x(t )  h1 (t )]  h2 (t )
AKW Fall 2011
Chapter 1-2 150
 Properties of LTI systems (cont.)

• Meaning?

x(t )  [h1 (t )  h2 (t )]
 [ x(t )  h1 (t )]  h2 (t )

x[n] {h1[n]  h2 [n]}

 {x[n]  h1[n]}  h2 [n]

Implication:
When cascading two LTI systems, the
order of the cascade does not matter

AKW Fall 2011

Chapter 1-2 151
 Other Properties - Memory

• If h(t) = 0 for t  0 , then the LTI system is memoryless

• If h[n] = 0 for n  0 , then the LTI system is memoryless

Proof:

y (t )  

x ( ) h (t   ) d   x (t )  k
 k (t ) t0
if h (t )  
 0 t0

AKW Fall 2011

Chapter 1-2 152
 Other Properties - Invertibility

• For an LTI with impulse response h(t), if there is an hInv (t) such
that h(t )  hInv (t )   (t ) , then the LTI system is invertible.

Proof and meaning:

 
x(t) (t)   x( ) (t  )d  x(t)   (t  )d  x(t)
 

x(t)*h(t) x(t )  h(t )  hInv (t )  x(t )

x(t) h(t) hIn(t)

AKW Fall 2011

Chapter 1-2 153
• Inverse System Example:

Consider the echo system:

h(t) = (t) + 0.3 (t - 4)
Echo System:
1
y ( t )  x ( t )  0.3 x ( t  4)
t
0

Below is the impulse response of the corresponding echo cancellation :

hInv(t) = (t) - 0.3 (t - 4) + 0.09 (t - 4) – 0.027 (t - 8) + 0.0081 (t - 12) -…..


   0.3   (t )
n
1 hInv ( t ) 
n0

0
t

AKW Fall 2011

Chapter 1-2 154
 Other Properties - Causality

• If h(t)=0 for t<0, or h[n]=0 for n<0,

then the LTI system is causal

Proof :
h() = 0 for  < 0
 
y (t )   x(t   )h( )d   x (t   )h( ) d
 0

This means y(t) does not depends on any x(t –) with t –> t

AKW Fall 2011

Chapter 1-2 155
 Other Properties - Stability
 
• If  h[k ]  
k  
or  
h( ) d  
then the system is stable

BIBO: Bounded Input Bounded Output stability means if

| x(t ) | B t then | y (t ) | B' t where B, B'  
Proof: 
y (t )  

x(t   )h( )d

 x(t   ) h( ) d


 B  h( ) d  


AKW Fall 2011

Chapter 1-2 156
 Unit step response of LTI systems
DT
 [n ] h[n] h[n ] u[n ] h[n] s[ n ]  ?
  n
Unit Step
Response
s[ n ]   u[k ]h[n  k ]   h[n  k ]   h[k ]
k   k 0 k  

CT u (t ) h(t) s (t )  ?

Unit Step  t
Response
s (t )   h( )u (t   ) d   h( ) d
 

Or to show alternatively
 t t
u (t )    ( )u (t   ) d    ( ) d  s (t )   h( ) d
  

AKW Fall 2011

Chapter 1-2 157
 Unit step response from Properties of LTI
Instead of mathematical derivation, we can also deduce the unit step
response in the following ways:

u (t ) h(t) s (t ) unit step response

u(t) is the integral
of (t)
u (t ) s (t )
 (t )  h(t)
integrator
associative
s (t )
 (t )  h(t)
integrator
commutative
h(t ) t
 (t ) h(t)  s (t )   h(t )dt

integrator

AKW Fall 2011

Chapter 1-2 158
 Response to derivative/integral of input
Let y(t) be the output of an LTI system when the input is x(t). Let x(k)(t) represents
the k-th derivative of x(t) (k=1 means first derivative, k=-1 means integral, etc.). We
can replace the input in the previous slide by an arbitrary input x(t), and replace the
integrator by a k-th order differentiator (negative k means integrator).

x ( k ) (t ) r (t )  x ( k ) (t )  h(t )
x(t ) k-th
differentiator
h(t)

x(t ) h(t) k-th r (t )  x (t )  h ( k ) (t )

differentiator

y (t ) r (t )  y ( k ) (t )
x(t ) h(t) k-th
differentiator

This means that we can determine the output by taking the k-th derivative
of the original output , or by passing x(t) through a system with impulse
response that is the k-th derivative of the original impulse response!
Chapter 1-2 159
Example. y(t) is the output of an LTI system when the input is x(t).
y(t)

1/3

0 3 t

t
Sketch the output when the input is x ( 1) (t )  2 x(t  2) or 

x( )d  2 x(t  2)

Sketch the output when the input is x (1) (t ) or x '(t )

Chapter 1-2 160

 Summary of LTI systems

• Impulse responses x(t )   (t )  y (t )  h(t )

x[n]   [n]  y[n]  h[n]

• Output of LTI systems x(t )  y (t )  x(t )  h(t )

x[n]  y[n]  x[n]  h[n]

y[n]  x[n]  h[n]   x[k ]h [n  k ]
• Convolution k  

y (t )  x(t)  h(t)   x( )h(t   )d


• Properties: commutative, distributive, associative

AKW Fall 2011
Chapter 1-2 161
 How to compute convolution sum
• The convolution sum for DT system is relatively easy to do

y[n]  x[n]  h[n]   x[k ]h[n  k ]
k  

 Step 1: draw h[n] and replace “n” by “k”

 Step 2: flip h[k] to get h[-k]
 Step 3: Shift h[-k] by n to obtain h[n-k]
 Step 4: Multiple with x[k] and add up the sum

AKW Fall 2011

Chapter 1-2 162
 coupon bond example :
coupon bond example :
h[0]  0.1
x[0]  $10 n  0 means year 2005
• Example h[n] h[1]  0.08 x[n] x[1]  $20
h[2]  0.08
x[2]  $40
h[3]  1

-2 -1 0 1 2 3 4 n -2 -1 0 1 2 3 4 n

What is the output at n=4 (year 2009)?

What is the output at n=5?
We multiple x[k] with h[4 – k]
and add up the sum We shift h[– k] to h[5 – k]
x[k ] x[1] x[2]
We can also plot x[k] and h[5-k]
together for ease of visualization:
-2 -1 0 1 2 3 4

h[4  k ] x[k ] h[5  k ]

h[4  k ]

-2 -1 0 1 2 3 4 5 k
-2 -1 0 1 2 3 4 k

y[4]  x[1]h[3]  x[2]h[2] y[5]  x[2]h[3]

Examples at http://www.jhu.edu/~signals/discreteconv2/index.html Joy of Convolution

AKW Fall 2011
Chapter 1-2 163
 How to compute convolution integral
We can computing the convolution integral following a similar
procedure. Sometimes things can become quite scary when we
try to come up with closed-form integration results!


y (t )  x(t)  h(t)   x( )h(t   )d


 step 1 : draw h(t) and replace 't' axes by '’ axes

 step 2 : flip h() to get h(-)
 step 3 : shift h(-) by t to get h(t-  )
 step 4 : multiply and compute the integral

First we go through some easier examples:

AKW Fall 2011
Chapter 1-2 164
 convolution integral example 1.
x(t) h(t) Again, h(t) here indicates a smoother,
which takes the average of x(t) over
1 * 1 the interval of one time unit
t t
2 1
We change the dummy variable from t to , time reverse h(), shift it by t,
multiply, and integrate (find area)

x() x() x()

h(t   ); h(t   ); h(t   );
t  0.4 1 t  1.1 1 t  2.2 1

  
2 2 2
Area=0.4 Area=1 Area=?

y(t) In general form:  t 0  t 1

y(0.4) = 0.4, y (t )   1 1 t  2
1
y(1.1) = 1, 3  t 2t 3
y(2.2) = 0.8 t 
0.4 11.1 2 2.2 3  0 otherwise
Chapter 1-2 165
 Variations on example:
x(t) h(t) y(t)

1 * 1 1

t t t
1 1 2
1

x(t) h(t) y(t)

1.5 1.5
1 *
t t t
1 -0.5 0.5 -0.5 0.5 1.5

x(t) h(t) y(t)

1.5 1.5
1
1 * 1
t t t
1 1 2 1 2 3

Chapter 1-2 166

 convolution integral example 2.
x(t) h(t )  et u (t ) Here, h(t) is a one-sided exponential
1 * 1

t t
2 1
h(t   )  e (t  )u (t   )
e.g. h(2   )  e (2 )u (2   )
x() x() x()
h (t   )
h(0.4   ) 1 1 1
h(1   )
  
2 2 2 t
Area=1-e-0.4 Area=1-e-1 Area=?

y(t) In general form:

y (0.4)  1  e0.4  0 t0
1
y (2)  1  e2 
y (t )   1  e t 0t 2
y (3)  e1  e3 t   (t  2)  e t
1 2 3 e t2
Chapter 1-2 167