Signals and Systems Review

8/25/15 M. J. Roberts - All Rights Reserved 1

 Continuous-Time Sinusoids
g ( t ) = Acos ( 2π t / T0 + θ ) = Acos ( 2π f0t + θ ) = Acos (ω 0t + θ )
↑ ↑ ↑ ↑ ↑
Amplitude Period Phase Shift Cyclic Radian
(s) (radians) Frequency Frequency
(Hz) (radians/s)
⎛ ⎞
ω
⎝ 0 =2 π f
0⎠

g ! t " # A cos ! 2 f0 t $ "

A

T0
0

8/25/15 M. J. Roberts - All Rights Reserved 2

 Continuous-Time Exponentials
g ( t ) = Ae−t /τ
↑ ↑
Amplitude Time Constant (s)

8/25/15 M. J. Roberts - All Rights Reserved 3

 Complex Sinusoids

Euler's Identity: e jx = cos ( x ) + j sin ( x )

e jx + e− jx e jx − e− jx
cos ( x ) = , sin ( x ) =
2 j2
8/25/15 M. J. Roberts - All Rights Reserved 4
 The Signum Function
⎧1 , t>0⎫
⎪ ⎪
sgn ( t ) = ⎨ 0 , t = 0 ⎬
⎪−1 , t < 0 ⎪
⎩ ⎭

sgn(t) sgn(t)
! !
t t
! !

The signum function, in a sense, returns an indication of

the sign of its argument.

8/25/15 M. J. Roberts - All Rights Reserved 5

 The Unit Step Function
⎧1 , t>0
⎪0 , t<0
⎪ 1
u (t ) = ⎨ , u ( t ) = ⎡⎣sgn ( t ) + 1⎤⎦ , t ≠ 0
⎪Undefined , t = 0 2
⎪⎩(but finite)
u(t) u(t)
1 1

t t

The product signal g ( t ) u ( t ) can be thought of as the signal g ( t )

“turned on” at time t = 0.

8/25/15 M. J. Roberts - All Rights Reserved 6

 The Unit Ramp Function

⎧t , t > 0 ⎫ t
ramp ( t ) = ⎨ ⎬ = ∫ u (λ ) dλ = t u (t )
⎩0 , t ≤ 0 ⎭ −∞

ramp(t)
1

t
1

8/25/15 M. J. Roberts - All Rights Reserved 7

 The Impulse

The continuous-time unit impulse is implicitly defined by

∞

g (0) = ∫ δ (t ) g (t ) dt
−∞

The unit step is the integral of the unit impulse and

the unit impulse is the generalized derivative of the
unit step.

8/25/15 M. J. Roberts - All Rights Reserved 8

 Properties of the Impulse
The Sampling Property
∞

∫ g (t )δ (t − t ) dt = g (t )
−∞
0 0

The sampling property “extracts” the value of a function at

a point. (In Ziemer and Tranter this is called the "sifting" property.)
The Scaling Property

δ ( a ( t − t 0 )) = δ ( t − t 0 )
1
a
This property illustrates that the impulse is different from
ordinary mathematical functions.

8/25/15 M. J. Roberts - All Rights Reserved 9

 The Unit Periodic Impulse
The unit periodic impulse is defined by
∞
δ T (t ) = ∑ δ (t − nT ) , n an integer
n=−∞
T
!(t) !(t)
T
1
... ... ... ...
(1) (1) (1) (1) (1)
t t
-2T -T T 2T -2T -T T 2T
The periodic impulse is a sum of infinitely many impulses
uniformly-spaced apart by T .

δ T ( a ( t − t 0 )) = δ T /a ( t − t 0 ) , n an integer
1
a
8/25/15 M. J. Roberts - All Rights Reserved 10
 The Unit Rectangle Function

⎧ 1 , t ≤1/ 2 ⎫ 1
Π ( t ) = rect ( t ) = ⎨ ⎬ = u (t + 1 / 2 ) − u (t − 1 / 2 ) , t ≠
⎩ 0 , t > 1 / 2⎭ 2

Π(t) = rect(t) Π(t) = rect(t)

1 1

t t
1 1 1 1
2 2 2 2

The product signal g ( t ) rect ( t ) can be thought of as the signal g ( t )

“turned on” at time t = − 1 / 2 and “turned back off” at time t = + 1 / 2.

8/25/15 M. J. Roberts - All Rights Reserved 11

 Random Signals
The sinusoid, exponential, signum, unit step, unit ramp, and unit
rectangle are all deterministic signals. The term deterministic
means that their values are specified at all times. Signals that are not
deterministic are random. The exact values of random signals are
unpredictable although their general behavior may be known to some
degree. x(t)

8/25/15 M. J. Roberts - All Rights Reserved 12

 Shifting and Scaling Functions
Amplitude Scaling, g(t ) → Ag(t )

8/25/15 M. J. Roberts - All Rights Reserved 13

 Shifting and Scaling Functions

Time shifting, t → t − t 0

8/25/15 M. J. Roberts - All Rights Reserved 14

 Shifting and Scaling Functions
Time scaling, t →t /a

8/25/15 M. J. Roberts - All Rights Reserved 15

 Differentiation
x(t) x(t)
4 1

4
t -4 4
t
-4 -1
dx/dt dx/dt
4 1

4
t -4 4
t
-4 -1

x(t) x(t)
1 1

-5 5
t 4
t
-1 -1
dx/dt dx/dt
1 1

-5 5
t 4
t
-1 -1

8/25/15 M. J. Roberts - All Rights Reserved 16

 Integration

8/25/15 M. J. Roberts - All Rights Reserved 17

 Even and Odd Signals
Even Functions Odd Functions
g ( t ) = g ( −t ) g ( t ) = − g ( −t )
Even Function Odd Function
g(t) g(t)

t t

Even Function Odd Function

g(t) g(t)

t t

8/25/15 M. J. Roberts - All Rights Reserved 18

 Even and Odd Parts of Functions
g ( t ) + g ( −t )
The even part of a function is g e ( t ) = .
2
g ( t ) − g ( −t )
The odd part of a function is g o ( t ) = .
2
A function whose even part is zero is odd and a function
whose odd part is zero is even.
The derivative of an even function is odd and the derivative
of an odd function is even.
The integral of an even function is an odd function, plus a
constant, and the integral of an odd function is even.

8/25/15 M. J. Roberts - All Rights Reserved 19

Integrals of Even and Odd Functions
Even Function Odd Function
g(t) g(t)
Area #1 Area #2
-a
Area #2
-a a t a t
Area #1
Area #1 = Area #2 Area #1 = - Area #2
a a a

∫ g (t ) dt = 2 ∫ g (t ) dt
−a 0 −a
∫ g (t ) dt = 0
The integral of an odd function, over limits that are
symmetrical about zero, is zero.
8/25/15 M. J. Roberts - All Rights Reserved 20
 Periodic Signals
If a function g(t) is periodic, g ( t ) = g ( t + nT ) where n is any integer
and T is a period of the function. The minimum positive value of T
for which g ( t ) = g ( t + T ) is called the fundamental period T0 of the
function. The reciprocal of the fundamental period is the fundamental
frequency f0 = 1 / T0 .
x(t) x(t) x(t)
... ... ... ... ...t
... t t

T0 T0 T0
A function that is not periodic is aperiodic.

8/25/15 M. J. Roberts - All Rights Reserved 21

 Signal Energy and Power

The signal energy of a signal x ( t ) is

∞

∫ x ( t ) dt
2
Ex =
−∞

8/25/15 M. J. Roberts - All Rights Reserved 22

 Signal Energy and Power

Some signals have infinite signal energy. In that case

it is more convenient to deal with average signal power.
The average signal power of a signal x ( t ) is
T /2
1
∫ x ( t ) dt
2
Px = lim
T →∞ T
−T /2

For a periodic signal x ( t ) the average signal power is

1
Px = ∫ x ( t ) dt
2

T T
where T is any period of the signal.

8/25/15 M. J. Roberts - All Rights Reserved 23

 Signal Energy and Power

A signal with finite signal energy is

called an energy signal.

A signal with infinite signal energy and

finite average signal power is called a
power signal.

8/25/15 M. J. Roberts - All Rights Reserved 24

 Sampling and Discrete Time
Sampling is the acquisition of the values of a continuous-time signal
at discrete points in time. x ( t ) is a continuous-time signal, x ⎡⎣ n ⎤⎦ is a
discrete-time signal.
x ⎡⎣ n ⎤⎦ = x ( nTs ) where Ts is the time between samples

Sampling Uniform Sampling

x (t ) x[n] x (t ) x[n]

ω s or fs
8/25/15 M. J. Roberts - All Rights Reserved 25
 Sampling and Discrete Time

8/25/15 M. J. Roberts - All Rights Reserved 26

 Exponentials
The form of the exponential is
x [ n ] = Aα n or x [ n ] = Aeβ n where α = eβ

Preferred

Real α Complex α
|z| < 1
Re(g[n]) Im(g[n])
-1 < z < 0
0 < z<1 n
n n

|z| > 1
z>1 z < -1 Re(g[n]) Im(g[n])
n
n n
n

8/25/15 M. J. Roberts - All Rights Reserved 27

 The Unit Impulse Function
δ[n]
⎧1 , n = 0
1 δ [n] = ⎨
n ⎩0 , n ≠ 0

The discrete-time unit impulse (also known as the “Kronecker

delta function”) is a function in the ordinary sense (in contrast
with the continuous-time unit impulse). It has a sampling property,
∞

∑ Aδ [ n − n ] x [ n ] = A x [ n ]
0 0
n=−∞

but no scaling property. That is,

δ [ n ] = δ [ an ] for any non-zero, finite integer a.
8/25/15 M. J. Roberts - All Rights Reserved 28
 The Unit Sequence Function
⎧1 , n ≥ 0
u[n] = ⎨
⎩0 , n < 0

u[n]
1
... ...
n

8/25/15 M. J. Roberts - All Rights Reserved 29

 The Signum Function
⎧1 , n > 0
⎪
sgn ⎡⎣ n ⎤⎦ = ⎨0 , n = 0 = 2u ⎡⎣ n ⎤⎦ − δ ⎡⎣ n ⎤⎦ − 1
⎪−1 , n < 0
⎩
sgn[n]
1
...
... n
-1

8/25/15 M. J. Roberts - All Rights Reserved 30

 The Unit Ramp Function
⎧n , n ≥ 0 ⎫ n
ramp [ n ] = ⎨ ⎬ = n u [ n ] = ∑ u [ m − 1]
⎩0 , n < 0 ⎭ m=−∞

ramp[n]

8
... 4 ...
n
4 8

8/25/15 M. J. Roberts - All Rights Reserved 31

 The Periodic Impulse Function
∞
δ N [n] = ∑ δ [ n − mN ]
m=−∞

N
[n]

!
... ...
n
-N N 2N

8/25/15 M. J. Roberts - All Rights Reserved 32

 Scaling and Shifting Functions
Time shifting n → n + n0 , n0 an integer

8/25/15 M. J. Roberts - All Rights Reserved 33

 Scaling and Shifting Functions

Time compression

n → Kn

K an integer > 1

8/25/15 M. J. Roberts - All Rights Reserved 34

 Scaling and Shifting Functions

Time expansion n → n / K, K > 1

For all n such that n / K is an integer, g [ n / K ] is defined.

For all n such that n / K is not an integer, g [ n / K ] is not defined.

8/25/15 M. J. Roberts - All Rights Reserved 35

 Scaling and Shifting Functions
There is a form of time expansion that is useful. Let
⎧x [ n / m ] , n / m an integer
y[n] = ⎨
⎩0 , otherwise
All values of y are defined.
This type of time expansion
is actually used in some
digital signal processing
operations.

8/25/15 M. J. Roberts - All Rights Reserved 36

 Differencing

8/25/15 M. J. Roberts - All Rights Reserved 37

 Accumulation

h[n] h[n]
2 2
n
n -5 20

g[ n ] = ∑ h[m] -2
g[n]
-10
g[n]
10
n

m=−∞
2 8
n
-5 20
-2 n
-10 10

8/25/15 M. J. Roberts - All Rights Reserved 38

 Even and Odd Signals
g [ n ] = g [ −n ] g [ n ] = − g [ −n ]

Even Function Odd Function

g[n] g[n]

... ...
n
... ...
n

g [ n ] + g [ −n ] g [ n ] − g [ −n ]
ge [n] = go [n] =
2 2

8/25/15 M. J. Roberts - All Rights Reserved 39

 Symmetric Finite Summation
Even Function Odd Function
g[n] g[n]
Sum #1
Sum #1 Sum #2
... -N
...
n
... ... N

-N N
n
Sum #2
Sum #1 = Sum #2
Sum #1 = - Sum #2
N N N

∑ g [ n ] = g [ 0 ] + 2∑ g [ n ] ∑ g[n] = 0
n=− N n=1 n=− N

8/25/15 M. J. Roberts - All Rights Reserved 40

 Periodic Functions
A periodic function is one that is invariant to the
change of variable n → n + mN where N is a period of the
function and m is any integer.

The minimum positive integer value of N for which

g [ n ] = g [ n + N ] is called the fundamental period N 0 .

8/25/15 M. J. Roberts - All Rights Reserved 41

 Signal Energy and Power

The signal energy of a signal x [ n ] is

∞

∑ x[n]
2
Ex =
n=−∞

8/25/15 M. J. Roberts - All Rights Reserved 42

 Signal Energy and Power
Some signals have infinite signal energy. In that case
It is usually more convenient to deal with average signal
power. The average signal power of a signal x [ n ] is
1 N −1
∑ x[n]
2
Px = lim
N →∞ 2N
n=− N

For a periodic signal x [ n ] the average signal power is

1
∑ x[n]
2
Px =
N n= N
⎛ The notation ∑ means the sum over any set of ⎞
n= N
⎜ ⎟
⎝ consecutive n 's exactly N in length. ⎠

8/25/15 M. J. Roberts - All Rights Reserved 43

 Signal Energy and Power

A signal with finite signal energy is

called an energy signal.

A signal with infinite signal energy and

finite average signal power is called a
power signal.

8/25/15 M. J. Roberts - All Rights Reserved 44

 Linearity and LTI Systems
• If a system is both homogeneous and additive
it is linear.
• If a system is both linear and time-invariant it
is called an LTI system
• Some systems that are non-linear can be
accurately approximated for analytical
purposes by a linear system for small
excitations

8/25/15
M. J. Roberts - All Rights Reserved
45

 Response of LTI Systems

An LTI system is completely characterized by its impulse response

h ( t ) . The response y ( t ) of an LTI system to an excitation x ( t ) is the
convolution of x ( t ) with h ( t ) .

y (t ) = x (t ) ∗ h (t ) = ∫ x ( λ ) h (t − λ ) d λ
−∞

8/25/15 M. J. Roberts - All Rights Reserved 46

 Convolution Integral Properties
x ( t ) ∗ Aδ ( t − t 0 ) = A x ( t − t 0 )
If g ( t ) = g 0 ( t ) ∗ δ ( t ) then g ( t − t 0 ) = g 0 ( t − t 0 ) ∗ δ ( t ) = g 0 ( t ) ∗ δ ( t − t 0 )
If y ( t ) = x ( t ) ∗ h ( t ) then y′ ( t ) = x′ ( t ) ∗ h ( t ) = x ( t ) ∗ h′ ( t )
and y ( at ) = a x ( at ) ∗ h ( at )
Commutativity
x (t ) ∗ y (t ) = y (t ) ∗ x (t )
Associativity
⎡⎣ x ( t ) ∗ y ( t ) ⎤⎦ ∗ z ( t ) = x ( t ) ∗ ⎡⎣ y ( t ) ∗ z ( t ) ⎤⎦
Distributivity
⎡⎣ x ( t ) + y ( t ) ⎤⎦ ∗ z ( t ) = x ( t ) ∗ z ( t ) + y ( t ) ∗ z ( t )

8/25/15
M. J. Roberts - All Rights Reserved
47

 The Unit Triangle Function
⎧1− t , t < 1
tri ( t ) = ⎨
⎩0 , t ≥1

The unit triangle, is the convolution of a unit rectangle with

Itself.

*8/25/15
M. J. Roberts - All Rights Reserved
48

 Systems Described by
Differential Equations

The transfer function:

bM s M + bM −1s M −1 +!+ b2 s 2 + b1s + b0
H(s) =
aN s N + aN −1s N −1 +!+ a2 s 2 + a1s + a0
This type of function is called a rational function because it is
a ratio of polynomials in s. The transfer function encapsulates
all the system characteristics and is of great importance in signal
and system analysis.

8/25/15
M. J. Roberts - All Rights Reserved
49

 Response of LTI Systems
If the excitation x ( t ) is a phasor or complex sinusoid of
frequency f0 , of the form
x ( t ) = Ax e jφx e j 2 π f0t
then the response y ( t ) is of the form
y ( t ) = H ( f0 ) x ( t ) = H ( f0 ) Ax e jφx e j 2 π f0t .
The response can also be written in the form
y ( t ) = Ay e y e j 2 π f0t where Ay = H ( f0 ) Ax and φy = φx + ! H ( f0 ) .
jφ

Applying this to real sinusoids, if x ( t ) = Ax cos ( 2π f0t + φx ) then

(
y ( t ) = Ay cos 2π f0t + φy . )
8/25/15 M. J. Roberts - All Rights Reserved 50
 The Convolution Sum

The response y [ n ] of an LTI system with impulse response h [ n ]

to an arbitrary excitation x [ n ] is
∞
y[n] = ∑ x[ m]h[n − m]
m=−∞

8/25/15
M. J. Roberts - All Rights Reserved
51

 Convolution Sum Properties

x [ n ] ∗ Aδ [ n − n0 ] = A x [ n − n0 ]
Let y [ n ] = x [ n ] ∗ h [ n ] then
y [ n − n0 ] = x [ n ] ∗ h [ n − n0 ] = x [ n − n0 ] ∗ h [ n ]
y [ n ] − y [ n − 1] = x [ n ] ∗ ( h [ n ] − h [ n − 1]) = ( x [ n ] − x [ n − 1]) ∗ h [ n ]
and the sum of the impulse strengths in y is the product of
the sum of the impulse strengths in x and the sum of the
impulse strengths in h.

8/25/15
M. J. Roberts - All Rights Reserved
52

Systems Described by Difference
Equations

The transfer function is

∑
M −k
b z b0 + b1z −1 + b2 z −2 +!+ bM z − M
H(z) = =
k=0 k

∑ a0 + a1z −1 + a2 z −2 +!+ aN z − N
N −k
a
k=0 k
z
or, alternately,

∑ k=0 k
M −k
b z b0 z M + b1z M −1 +!+ bM −1z + bM
H(z) = = z N −M
∑ a0 z N + a1z N −1 +!+ aN −1z + aN
N −k
k=0
a k z
The transfer function can be written directly from the system
( )
difference equation and vice versa. H e jΩ is the system's
frequency response. It is the transfer function H ( z ) with z
replaced by e jΩ .
8/25/15
M. J. Roberts - All Rights Reserved
53

 Continuous-Time Fourier Series
Definition

∞ t 0 +T
1
x (t ) = ∑ Xn e j 2 π nt /T and Xn =
T ∫ x ( t ) e− j 2 π nt /T dt .
n=−∞ t0

The signal and its harmonic function form a Fourier series

pair x ( t ) ←⎯⎯
FS
T
→ Xn where T is the representation time and,
therefore, the fundamental period of the continuous-time Fourier
series ( CTFS) representation of x ( t ) . If T is also a period of x ( t ) ,
the CTFS representation of x ( t ) is valid for all time. This is, by far,
the most common use of the CTFS in engineering applications. If T
is not a period of x ( t ) , the CTFS representation is generally valid
only in the interval t 0 ≤ t < t 0 + T .

8/25/15
M. J. Roberts - All Rights Reserved
54

 CTFS of a Real Function

It can be shown that the continuous-time Fourier series (CTFS)
harmonic function of any real-valued function x ( t ) has the property
that Xn = X−* n .

One implication of this fact is that, for real-valued functions,

the magnitudes of their harmonic functions are even functions
and their phases can be expressed as odd functions of harmonic
number k.

8/25/15
M. J. Roberts - All Rights Reserved
55

 The Sinc Function

Let x ( t ) = A rect ( t / w ) ∗ δ T0 ( t ) , w < T0 . Then
sin (π nw / T0 )
x ( t ) = A rect ( t / w ) ∗ δ T0 ( t ) ←⎯⎯
FS
→ Xn = A
T0
πn
sin (π x )
The mathematical form arises frequently enough
πx
sin (π t )
to be given its own name, "sinc". That is sinc ( t ) = .
πt

8/25/15
M. J. Roberts - All Rights Reserved
56

 The Uniqueness Property

If we find a Fourier series representation of a signal, it is
unique. That is, no other or alternate Fourier series
representation exists.
Example: Let x ( t ) = 3cos ( 8π t − π / 4 ) + 4 sin ( 4π t )
Using trigonometric identities, this can be rewritten as
x ( t ) = 3 ⎡⎣ cos ( 8π t ) cos (π / 4 ) − sin ( 8π t ) sin (π / 4 ) ⎤⎦ + 4 sin ( 4π t )
3 2
x (t ) = ⎡⎣ cos ( 8π t ) − sin ( 8π t ) ⎤⎦ + 4 sin ( 4π t )
2

8/25/15
M. J. Roberts - All Rights Reserved
57

 The Uniqueness Property

3 2 ⎡ e j 8 π t + e− j 8 π t e j 8 π t − e− j 8 π t ⎤ e j 4 π t − e− j 4 π t
x (t ) = ⎢ − ⎥ +4
2 ⎣ 2 j2 ⎦ j2

x (t ) =
3 2
4
(
⎡⎣(1+ j ) e j 8 π t + (1− j ) e− j 8 π t ⎤⎦ − j2 e j 4 π t − e− j 4 π t )
3 2 3 2
x (t ) = (1+ j ) e +
j 8π t
(1− j ) e− j 8π t − j2e j 4 π t + j2e− j 4 π t
4 4
This is THE ( complex ) CTFS representation of x ( t ) in which
∞
3 2
x (t ) = ∑Xe n
j 2 π nf0t
, f0 = 2 , X−2 =
4
(1− j ) , X−1 = j2 , X1 = − j2 ,
n=−∞

3 2
X2 = (1+ j ) and all other CTFS coefficients are zero.
4
8/25/15
M. J. Roberts - All Rights Reserved
58

 Some Common CTFS Pairs

1 ←⎯
FS
T
→ δ [ n ] , T arbitrary
⎧(1 / T0 ) , n / m an integer
δ T0 ( t ) ←⎯⎯
FS
mT0
→⎨
⎩0 , otherwise
e j 2 π qt /T0 ←⎯⎯
FS
mT0
→ δ [ n − mq ]
sin ( 2π qt / T0 ) ←⎯⎯
FS
mT0
→ ( j / 2 ) (δ [ n + mq ] − δ [ n − mq ])
cos ( 2π qt / T0 ) ←⎯⎯
FS
mT0
→ (1 / 2 ) (δ [ n − mq ] + δ [ n + mq ])
rect ( t / w ) ∗ δ T0 ( t ) ←⎯⎯
FS
mT0
→ ( w / T0 ) sinc ( wn / mT0 )δ m [ n ]
tri ( t / w ) ∗ δ T0 ( t ) ←⎯⎯
FS
mT0
→ ( 0)
w / T sinc 2
( wn / mT0 )δ m [ n ]
( m an integer )
8/25/15
M. J. Roberts - All Rights Reserved
59

 Definition of the CTFT

Forward f form Inverse
∞ ∞

X ( f ) = F ( x ( t )) = ∫ x ( t ) e − j 2 π ft
dt x (t ) = F ( X ( f )) = ∫ X ( f ) e
-1 + j 2 π ft
df
−∞ −∞

Forward ω form Inverse

∞ ∞

X ( jω ) = F ( x ( t )) = x ( t ) = F ( X ( jω )) =
1
∫ x ( t ) e − jω t
dt -1
∫ X ( jω ) e+ jω t dω
−∞
2π −∞

Commonly-used notation:
x ( t ) ←⎯
F
→ X( f ) or x ( t ) ←⎯
F
→ X ( jω )

8/25/15
M. J. Roberts - All Rights Reserved
60

 Some CTFT Pairs

()
δ t ←⎯
F
→1
()
e−α t u t ←⎯
F
(
→ 1/ jω + α ) , α >0 ( )
− e−α t u −t ←⎯
F
(
→ 1/ jω + α ) , α <0
te u ( t ) ←⎯→ 1/ ( jω + α ) , α > 0 − te u ( −t ) ←⎯→ 1/ ( jω + α ) , α < 0
−α t F 2 −α t F 2

t e u ( t ) ←⎯→ − t e u ( −t ) ←⎯→
n −α t n! n −α t n!
F
, α >0 F
, α <0
( jω + α ) ( jω + α )
n+1 n+1

ω ω
e−α t sin (ω t ) u ( t ) ←⎯→
F 0
, α >0 − e−α t sin (ω t ) u ( −t ) ←⎯→
F 0
, α <0
( jω + α ) + ω ( jω + α ) + ω
0 2 0 2
2 2
0 0

jω + α jω + α
e−α t cos (ω t ) u ( t ) ←⎯→
F
, α >0 − e−α t cos (ω t ) u ( −t ) ←⎯→
F
, α <0
( jω + α ) + ω ( jω + α ) + ω
0 2 0 2
2 2
0 0

−α t 2α
e ←⎯
F
→ , α >0
ω2 +α2

8/25/15
M. J. Roberts - All Rights Reserved
61

 More CTFT Pairs

()
δ t ←⎯
F
→1 1←⎯
F
( )
→δ f
sgn ( t ) ←⎯→ 1/ jπ f
F
u ( t ) ←⎯→ (1/ 2 )δ ( f ) + 1/ j2π f
F

rect ( t ) ←⎯→ sinc ( f )

F
sinc ( t ) ←⎯→ rect ( f )
F

tri ( t ) ←⎯→ sinc ( f )

F 2
sinc ( t ) ←⎯→ tri ( f )
2 F

δ ( t ) ←⎯→ f δ ( f ) , f = 1/ T
T0
F
0 f0 0 0
T δ ( t ) ←⎯→ δ ( f ) , T = 1/ f
0 T0
F
f0 0 0

cos ( 2π f t ) ←⎯→ (1/ 2 ) ⎡⎣δ ( f − f ) + δ ( f + f ) ⎤⎦

0
F
0 0
sin ( 2π f t ) ←⎯→ ( j / 2 ) ⎡⎣δ ( f + f ) − δ ( f − f ) ⎤⎦
0
F
0 0

8/25/15
M. J. Roberts - All Rights Reserved
62

Numerical Computation of the CTFT

It can be shown that the DFT can be used to approximate
samples from the CTFT. If the signal x ( t ) is a causal energy
signal and N samples are taken from it over a finite time
beginning at t = 0, at a rate fs then the relationship between the
CTFT of x ( t ) and the DFT of the samples taken from it is
X ( kfs / N ) ≅ Ts e− jπ k/N sinc ( k / N ) X DFT [ k ]
For those harmonic numbers k for which k << N
X ( kfs / N ) ≅ Ts X DFT [ k ]
As the sampling rate and number of samples are increased,
this approximation is improved.

8/25/15
M. J. Roberts - All Rights Reserved
63

 The Discrete-Time Fourier Series

The discrete-time Fourier series (DTFS) is similar to the CTFS.
A periodic discrete-time signal can be expressed as
1 n0 +N −1
x[n] = ∑ c x [ k ] e j 2 π kn/N cx [ k ] = ∑
N n=n0
x [ n ] e − j 2 π kn/N

k= N

where c x [ k ] is the harmonic function, N is any period of x [ n ]

and the notation, ∑ means a summation over any range of
k= N

consecutive k’s exactly N in length.

8/25/15
M. J. Roberts - All Rights Reserved
64

 The Discrete Fourier Transform

The discrete Fourier transform (DFT) is almost identical to the DTFS.
A periodic discrete-time signal can be expressed as
n0 + N −1
1
x[n] =
N
∑ X [ k ] e j 2 π kn/N X[ k ] = ∑ x [ n ] e− j 2 π kn/N
k= N n=n0

where X [ k ] is the DFT harmonic function and N is any period of x [ n ].

The main difference between the DTFS and the DFT is the location of
the 1/N term. So X [ k ] = N c x [ k ].

8/25/15
M. J. Roberts - All Rights Reserved
65

 The Discrete Fourier Transform

Because the DTFS and DFT are so similar, and because the DFT is
so widely used in digital signal processing (DSP), we will concentrate
on the DFT realizing we can always form the DTFS from
c x [ k ] = X [ k ] / N.

8/25/15
M. J. Roberts - All Rights Reserved
66

 The Discrete Fourier Transform

Notice that in
1
x[n] =
N
∑ X [ k ] e j 2 π kn/N
k= N

the summation is over N values of k, a finite summation. This is

because of the periodicity of the complex sinusoid, e− j 2 π kn/N
in harmonic number k. If k is increased by any integer
multiple of N the complex sinusoid does not change.
e− j 2 π kn/N = e− j 2 π ( k+mN )n/N = e− j 2 π kn/N e
− j 2 π mn
, m an integer
=1

This occurs because discrete time n is always an integer.

8/25/15
M. J. Roberts - All Rights Reserved
67

 The Dirichlet Function

drcl(t,4) drcl(t,5)
sin (π Nt ) 1 1
The functional form
N sin (π t ) t t
-2 2 -2 2
appears often in discrete-time
-1 -1
signal analysis and is given the
special name Dirichlet function. drcl(t,7) drcl(t,13)
That is 1 1

sin (π Nt )
drcl ( t, N ) = t t
N sin (π t )
-2 2 -2 2

-1 -1

8/25/15
M. J. Roberts - All Rights Reserved
68

 Response of LTI Systems

If the Fourier transform of the excitation x ( t ) is X ( f ) and the

Fourier transform of the response y ( t ) is Y ( f ) , then
Y ( f ) = H ( f ) X ( f ) and Y ( f ) = H ( f ) X ( f ) and
! Y( f ) = ! H ( f ) + ! X ( f ).
If x ( t ) is an energy signal (finite signal energy) then, from Parseval's
∞ ∞ ∞

∫ X( f ) ∫ Y( f ) ∫ H( f ) X ( f ) df
2 2 2 2
theorem Ex = df and Ey = df =
−∞ −∞ −∞

8/25/15 M. J. Roberts - All Rights Reserved 69

 Signal Distortion in Transmission
Distortion means changing the shape of a signal. Two
changes to a signal are not considered distortion, multiplying
it by a constant and shifting it in time. The impulse response
of an LTI system that does not distort is of the general form
h ( t ) = Kδ ( t − t d ) . where K and t d are constants. The corresponding
frequency response of such a system is H ( f ) = Ke− j 2 π ftd .
H ( f ) = K and !H ( f ) = −2π ft d . If H ( f ) ≠ K the system has
amplitude distortion. If !H ( f ) ≠ −2π ft d the system has delay
or phase distortion. Both of these types of distortion are classified
as linear distortions.
8/25/15 M. J. Roberts - All Rights Reserved 70
 Signal Distortion in Transmission
!H ( f )
If !H ( f ) = −2π ft d , then t d = − and t d is a constant
2π f
if !H ( f ) = −Kf (K a constant). If t d is not a constant,
phase distortion results. H( f )
A
f
Frequency response of a
→ !H( f )
distortionless LTI system
f = t1
d
f
-2π

8/25/15 M. J. Roberts - All Rights Reserved 71

 Signal Distortion in Transmission

Most real systems do not have simple delay. They have phases
that are not linear functions of frequency.

!H( f ) !H( f )

f f

8/25/15 M. J. Roberts - All Rights Reserved 72

 Signal Distortion in Transmission
For a bandpass signal with a small bandwidth W compared to
its center frequency fc , we can model the frequency response
phase variation as approximately linear over the frequency ranges
fc − W < f < fc + W , and the frequency response magnitude as
approximately constant, of the form
⎧e jφ0 , fc − W < f < fc + W
H ( f ) ≅ Ae
− j 2 π ftg
⎨ − jφ0
⎪⎩e , − fc − W < f < − fc + W
where φ0 = ! H ( fc ) . !H( f )
−φ0
2W fc
f
− fc 2W
φ0

8/25/15 M. J. Roberts - All Rights Reserved 73

 Signal Distortion in Transmission
If we now let the bandpass signal be
x ( t ) = x1 ( t ) cos ( 2π fct ) + x 2 ( t ) sin ( 2π fct )
Its Fourier transform is
⎧⎪ X1 ( f ) ∗ (1 / 2 ) ⎡⎣δ ( f − fc ) + δ ( f + fc ) ⎤⎦ ⎫⎪
X( f ) = ⎨ ⎬
⎪⎩+ X 2 ( f ) ∗ ( j / 2 ) ⎡⎣δ ( f + fc ) − δ ( f − fc ) ⎤⎦ ⎪⎭
{
X ( f ) = (1 / 2 ) ⎡⎣ X1 ( f − fc ) + X1 ( f + fc ) ⎤⎦ + j ⎡⎣ X 2 ( f + fc ) − X 2 ( f − fc ) ⎤⎦ }
The frequency response is modeled by
⎧e jφ0 , fc − W < f < fc + W
H ( f ) ≅ Ae
− j 2 π ftg
⎨ − jφ0
⎩⎪e , − fc − W < f < − fc + W
then the Fourier transform of the response y ( t ) is
⎧ X ( f − f ) e− j ( 2 π ftg −φ0 ) + X ( f + f ) e− j ( 2 π ftg +φ0 ) ⎫
⎪ 1 ⎪
Y( f ) ≅ H ( f ) X ( f ) = ( A / 2) ⎨
c 1 c

− j ( 2 π ftg +φ0 ) − j ( 2 π ftg −φ0 )

⎬
⎪⎩+ j X 2 ( f + fc ) e − j X 2 ( f − fc ) e ⎪⎭

8/25/15 M. J. Roberts - All Rights Reserved 74

 Signal Distortion in Transmission
⎧ X ( f − f ) e− j ( 2 π ftg −φ0 ) + X ( f + f ) e− j ( 2 π ftg +φ0 ) ⎫
⎪ 1 ⎪
Y( f ) ≅ H ( f ) X ( f ) = ( A / 2) ⎨
c 1 c

− j ( 2 π ftg +φ0 ) − j ( 2 π ftg −φ0 )

⎬
⎪⎩+ j X 2 ( f + fc ) e − j X 2 ( f − fc ) e ⎪⎭
Inverse Fourier transforming, using the time and frequency shifting properties,
( ) (
⎧⎪e jφ0 x1 t − tg e j 2 π fct + e− jφ0 x1 t − tg e− j 2 π fct
y (t ) ≅ ( A / 2 ) ⎨
) ⎫⎪
⎬
− jφ0
⎪⎩+ je x 2 t − tg e (
− j 2 π fc t
) jφ0
− je x 2 t − tg e ( )
j 2 π fc t
⎪⎭

⎪ (
⎣ )
⎧x1 t − tg ⎡ e j ( 2 π fct +φ0 ) + e− j ( 2 π fct +φ0 ) ⎤ ⎫
⎦ ⎪
y (t ) ≅ ( A / 2 ) ⎨ ⎬
(⎡
⎪⎩+ x 2 t − tg j ⎣ e )
− j ( 2 π fc t + φ 0 )
−e j ( 2 π fc t + φ 0 )
⎤
⎦ ⎪⎭
{ ( ) ( )
y ( t ) ≅ A x1 t − tg cos ( 2π fct + φ0 ) + x 2 t − tg sin ( 2π fct + φ0 ) }
y ( t ) ≅ A {x ( t − t ) cos ( 2π f ( t − t )) + x ( t − t ) sin ( 2π f ( t − t ))}
1 g c d 2 g c d

φ0 ! H ( fc )
where t d = − =− is known as the phase or carrier delay.
2π fc 2π fc

8/25/15 M. J. Roberts - All Rights Reserved 75

 Signal Distortion in Transmission
From the approximate form of the system frequency response
⎧e jφ0 , fc − W < f < fc + W
H ( f ) ≅ Ae
− j 2 π ftg
⎨ − jφ0
⎩e , − fc − W < f < − fc + W
we get
⎧−2π ftg + φ0 , fc − W < f < fc + W
!H ( f ) ≅ ⎨
⎩−2π ftg − φ0 , − fc − W < f < − fc + W
If we differentiate both sides w.r.t. f we get
d
df
( !H ( f )) ≅ −2π tg , fc − W < f < fc + W

or

tg ≅ −
1 d
2π df
( !H ( f )) , fc − W < f < fc + W

tg is known as the group delay.

8/25/15 M. J. Roberts - All Rights Reserved 76
 Signal Distortion in Transmission
Phase and Group Delay
Excitation
x(t) Excitation Modulated Carrier
1 Modulation

t t

-1

Phase Delay Group Delay

y(t) Response
Modulation
1

t t

-1
Modulated Carrier

Response

8/25/15 M. J. Roberts - All Rights Reserved 77

 Signal Distortion in Transmission

Linear distortion can be corrected (theoretically) by an equalization

network. If the communication channel's frequency response is H C ( f )
and it is followed by an equalization network with frequency response
H eq ( f ) then the overall frequency response is H ( f ) = H C ( f ) H eq ( f )
and the overall frequency response will be distortionless if
H ( f ) = H C ( f ) H eq ( f ) = Ke− jω td . Therefore, the frequency response
Ke− jω td
of the equalization network should be H eq ( f ) = . It is very rare
HC ( f )
in practice that this can be done exactly but in many cases an excellent
approximation can be made that greatly reduces linear distortion.

8/25/15 M. J. Roberts - All Rights Reserved 78

 Signal Distortion in Transmission
Communication systems can also have nonlinear distortion caused by
elements in the system that are statically nonlinear. In that case
the excitation and response are related through a transfer characteristic
of the form y ( t ) = T ( x ( t )) . For example, some amplifiers experience
a "soft" saturation in which the ratio of the response to the excitation
decreases with an increase in the excitation level.
y (t )

x (t )

8/25/15 M. J. Roberts - All Rights Reserved 79

 Signal Distortion in Transmission
The transfer characteristic is usually not a simple known function
but can often be closely approximated by a polynomial curve fit
of the form y ( t ) = a1 x ( t ) + a2 x 2 ( t ) + a3 x 3 ( t ) +!. The Fourier
transform of y ( t ) is

Y ( f ) = a1 X ( f ) + a2 X ( f ) ∗ X ( f ) + a3 X ( f ) ∗ X ( f ) ∗ X ( f ) +!

In a linear system if the excitation is bandlimited, the response has

the same band limits. The response cannot contain frequencies not
present in the excitation. But in a nonlinear system of this type
if X ( f ) contains a range of frequencies, X ( f ) ∗ X ( f ) contains
a greater range of frequencies and X ( f ) ∗ X ( f ) ∗ X ( f ) contains
a still greater range of frequencies.
8/25/15 M. J. Roberts - All Rights Reserved 80
 Signal Distortion in Transmission
If X ( f ) ∗ X ( f ) contains frequencies that are all outside the range of
X ( f ) then a filter can be used to eliminate them. But often X ( f ) ∗ X ( f )
contains frequencies both inside and outside that range, and those inside
the range cannot be filtered out without affecting the spectrum of X ( f ) .
As a simple example of the kind of nonlinear distortion that can occur
let x ( t ) = A1 cos (ω 1t ) + A2 cos (ω 2t ) and let y ( t ) = x 2 ( t ) . Then
y ( t ) = ⎡⎣ A1 cos (ω 1t ) + A2 cos (ω 2t ) ⎤⎦
2

= A12 cos 2 (ω 1t ) + A22 cos 2 (ω 2t ) + 2A1 A2 cos (ω 1t ) cos (ω 2t )

( ) ( )
= A12 / 2 ⎡⎣1 + cos ( 2ω 1t ) ⎤⎦ + A22 / 2 ⎡⎣1 + cos ( 2ω 2t ) ⎤⎦

( ) (
+ A1 A2 ⎡⎣ cos (ω 1 − ω 2 ) t + cos (ω 1 + ω 2 ) t ⎤⎦ )
8/25/15 M. J. Roberts - All Rights Reserved 81
 Signal Distortion in Transmission

( ) ( )
y ( t ) = A12 / 2 ⎡⎣1 + cos ( 2ω 1t ) ⎤⎦ + A22 / 2 ⎡⎣1 + cos ( 2ω 2t ) ⎤⎦

( ) (
+ A1 A2 ⎡⎣ cos (ω 1 − ω 2 ) t + cos (ω 1 + ω 2 ) t ⎤⎦ )
y ( t ) contains frequencies 2ω 1 , 2ω 2 , ω 1 − ω 2 and ω 1 + ω 2 . The
frequencies ω 1 − ω 2 and ω 1 + ω 2 are called intermodulation
distortion products. When the excitation contains more
frequencies (which it usually does) and the nonlinearity is of
higher order (which it often is), many more intermodulation
distortion products occur. All systems have nonlinearities
and intermodulation disortion will occur. But, by careful design,
it can often be reduced to a negligible level.

8/25/15 M. J. Roberts - All Rights Reserved 82

 Transmission Loss and Decibels
Communication systems affect the power of a signal. If the
signal power at the input is Pin and the signal power at the
output is Pout , the power gain g of the system is g = Pout / Pin .
It is very common to express this gain in decibels. A decibel
is one-tenth of a bel, a unit named in honor of Alexander
Graham Bell. The system gain g expressed in decibels would
be gdB = 10 log10 ( Pout / Pin ) .
g 0.1 1 10 100 1000 10, 000 100, 000
gdB −10 0 10 20 30 40 50
Because gains expressed in dB are logarithmic, they compress the
range of numbers. If two systems are cascaded, the overall power
gain is the product of the two individual power gains g = g1g2 .
The overall power gain expressed in dB is the sum of the two power
gains expressed in dB , gdB = g1,dB + g2,dB .

8/25/15 M. J. Roberts - All Rights Reserved 83

 Transmission Loss and Decibels
The decibel was defined based on a power ratio, but it is often
used to indicate the power of a single signal. Two common types
of power indication of this type are dBW and dBm. dBW is
the power of a signal with reference to one watt. That is, a one
watt signal would have a power expressed in dBW of 0 dBW. dBm
is the power of a signal with reference to one milliwatt. A 20 mW
signal would have a power expressed in dBm of 13.0103 dBm. Signal
power gain as a function of frequency is the square of the magnitude
of frequency response H ( f ) . Frequency response magnitude is often
2

(
expressed in dB also. H ( f ) dB = 10 log10 H ( f )
2
) = 20 log
10 ( H ( f ) ).

8/25/15 M. J. Roberts - All Rights Reserved 84

 Transmission Loss and Decibels
A communication system generally consists of components
that amplify a signal and components that attenuate a signal.
Any cable, optical or copper, attenuates the signal as it
propagates. Also there are noise processes in all cables and
amplifiers that generate random noise. If the power level
gets too low, the signal power becomes comparable to the noise
power and the fidelity of analog signals is degraded too far or the
detection probability for digital signals becomes too low. So,
before that signal level is reached, we must boost the signal power
back up to transmit it further. Amplifiers used for this purpose
are called repeaters.

8/25/15 M. J. Roberts - All Rights Reserved 85

 Transmission Loss and Decibels

On a signal cable of 100's or 1000's of kilometers many repeaters

will be needed. How many are needed depends on the attenuation
per kilometer of the cable and the power gains of the repeaters.
Attenuation will be symbolized by L = 1 / g = Pin / Pout or
LdB = −gdB = 10 log10 ( Pin / Pout ) , (L for "loss".) For optical and copper
cables the attenuation is typically exponential and Pout = 10 −α l /10 Pin where
l is the length of the cable and α is the attenuation coefficient in dB/unit
length. Then L = 10α l /10 and LdB = α l.

8/25/15 M. J. Roberts - All Rights Reserved 86

 Filters and Filtering
An ideal bandpass filter has the frequency response
⎧ Ke− jω td , fl ≤ f ≤ fh
H( f ) = ⎨
⎩0 , otherwise
where fl is the lower cutoff frequency and fh is the upper cutoff
frequency and K and t d are constants. The filter's bandwidth is
B = fh − fl . An ideal lowpass filter has the same frequency response
but with fl = 0 and B = fh . An ideal highpass filter has the same
frequency response but with fh → ∞ and B → ∞. These filters are
called ideal because they cannot actually be built. They cannot be
built because they are non-causal. But they are useful fictions for
introducing in a simplified way some of the concepts of communication
systems.

8/25/15 M. J. Roberts - All Rights Reserved 87

 Filters and Filtering
Strictly speaking a signal cannot be both bandlimited and timelimited.
But many signals are almost bandlimited and timelimited. That is,
many signals have very little signal energy outside a defined bandwidth
and, at the same time, very little signal energy outside a defined time
range. A good example of this is a Gaussian pulse
x ( t ) = e− π t ←⎯ → X ( f ) = e− π f
2 2
F

Strictly speaking, this signal is not bandlimited or timelimited. The

total signal energy of this signal is 1/ 2 . 99% of its energy lies
in the time range − 0.74 < t < 0.74 and in the frequency range
−0.74 < f < 0.74. So in many practical calculations this signal could
be considered both bandlimited and timelimited with very little error.

8/25/15 M. J. Roberts - All Rights Reserved 88

 Filters and Filtering
Real filters cannot have constant amplitude response and linear
phase response in their passbands like ideal filters.
|H( f )|
Passband Ripple

f
Stop Band Pass Band Stop Band

|H( f )|dB Transition Bands

Minimum Stopband
Attenuation

f
Stop Band

8/25/15 M. J. Roberts - All Rights Reserved 89

 Filters and Filtering
There are many types of standardized filters. One very common and
useful one is the Butterworth filter. The frequency response of a
1
lowpass Butterworth filter is of the form H ( f ) = where
1 + ( f / B)
2n

n is the order of the filter. As the order is increased, its magnitude

response approaches that of an ideal filter, constant in the passband
|H (jω)|
and zero outside the passband. (Below is a

1
illustrated the magnitude frequency response
of a normalized lowpass Butterworth filter 1
2
with a corner frequency of 1 radian/s.)
n=1

n=2

n=8 n=4
ω
-5 -4 -3 -2 -1 1 2 3 4 5

8/25/15 M. J. Roberts - All Rights Reserved 90

 Filters and Filtering
The Butterworth filter is said to be maximally flat in its passband. It is given
this description because the first n derivatives of its magnitude frequency
response are all zero at f = 0 (for a lowpass filter). The passband of a lowpass
Butterworth filter is defined as the frequency at which its magnitude frequency
response is reduced from its maximum by a factor of 1/ 2. This is also known
as its half -power bandwidth because, at this frequency the power gain of the filter
|H (jω)|
is half its maximum value. a

1
2

n=1

n=2

n=8 n=4
ω
-5 -4 -3 -2 -1 1 2 3 4 5

8/25/15 M. J. Roberts - All Rights Reserved 91

 Filters and Filtering
The step response of a filter is
∞ t

h −1 ( t ) = ∫ h ( λ ) u (t − λ ) d λ = ∫ h ( λ ) d λ
−∞ −∞

(g ( t ) in the book). That is, the step response is the integral of the impulse
response. The impulse response of a unity-gain ideal lowpass filter with
no delay is h ( t ) = 2Bsinc ( 2Bt ) where B is its bandwidth. Its step response
is therefore
t
⎡0 t
⎤
h −1 ( t ) = ∫ 2Bsinc ( 2Bλ ) d λ = 2B ⎢ ∫ sinc ( 2Bλ ) d λ + ∫ sinc ( 2Bλ ) d λ ⎥
−∞ ⎣ −∞ 0 ⎦
This result can be further simplified by using the definition of the sine
integral function
sin (α )
θ θ /π
Si (θ ) ! ∫ dα = π ∫ sinc ( λ ) d λ
0
α 0

8/25/15 M. J. Roberts - All Rights Reserved 92

 Filters and Filtering
The Sine Integral Function
2

1.5 π/2

0.5
Si(t)

-0.5

-1

-1.5 −π/2
-2
-30 -20 -10 0 10 20 30
t
8/25/15 M. J. Roberts - All Rights Reserved 93
 Filters and Filtering

⎡0 t
⎤
h −1 ( t ) = 2B ⎢ ∫ sinc ( 2Bλ ) d λ + ∫ sinc ( 2Bλ ) d λ ⎥
⎣ −∞ 0 ⎦
0 2 Bt

Let 2Bλ = α . Then h −1 ( t ) = ∫ sinc (α ) dα + ∫ sinc (α ) dα .

−∞ 0
0 ∞

Using the fact that sinc is an even function, ∫ sinc (α ) dα = ∫ sinc (α ) dα .

−∞ 0
θ /π
Then, using Si (θ ) = π ∫ sinc (α ) dα and Si ( ∞ ) = π / 2, we get
0

Si ( ∞ ) 1 1 1
h −1 ( t ) = + Si ( 2π Bt ) = + Si ( 2π Bt )
π π 2 π

8/25/15 M. J. Roberts - All Rights Reserved 94

 Filters and Filtering
1 1
h −1 ( t ) = + Si ( 2π Bt )
2 π
This step response has precursors, overshoot, and oscillations
(ringing). Risetime is defined as the time required to move from
10% of the final value to 90% of the final value. For this ideal
lowpass filter the rise time is 0.44/B. The rise time for a single-pole,
lowpass filter is 0.35/B. 1.5

h (t)
0.5
Step response of an Ideal
→ -1
Lowpass Filter with B = 1
0

-0.5
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
t
8/25/15 M. J. Roberts - All Rights Reserved 95
 Filters and Filtering
The response of an ideal lowpass filter to a rectangular pulse of width τ is

y ( t ) = h −1 ( t ) − h −1 ( t − τ ) = ⎡⎣Si ( 2π Bt ) − Si ( 2π B ( t − τ )) ⎤⎦ .
1
π
From the graph (in which B = 1) we see that, to reproduce the
rectangular pulse shape, even very crudely, requires a bandwidth
much greater than 1/τ . If we have a pulse train with pulse widths
τ and spaces between pulses also τ and we 1.5
τ = 1/4
τ = 1/2
want to simply detect whether or not a pulse τ=2
1

is present at some time, we will need at least

B ≥ 1 / 2τ . If the bandwidth is any lower the

y(t)
0.5

overlap between pulses makes them very 0

hard to resolve.
-0.5
-2 -1 0 1 2 3 4
t

8/25/15 M. J. Roberts - All Rights Reserved 96

 Pulse Width and Bandwidth
Pulses (and their Fourier transforms) can have many shapes

1.5 1.5

Infinite pulse width 1 1

|X( f )|
x(t)
0.5 0.5

and bandwidth −0.5

0

−0.5
0

−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
t (s) f (Hz)
1.5 1.5

1 1
Finite pulse width and

|X( f )|
x(t)
0.5 0.5

0 0
infinite bandwidth −0.5
−4 −3 −2 −1 0 1 2 3 4
−0.5
−4 −3 −2 −1 0 1 2 3 4
t (s) f (Hz)
1.5 1.5

1 1

|X( f )|
Infinite pulse width and
x(t)

0.5 0.5

0 0

finite bandwidth −0.5

−4 −3 −2 −1 0
t (s)
1 2 3 4
−0.5
−4 −3 −2 −1 0
f (Hz)
1 2 3 4

8/25/15 M. J. Roberts - All Rights Reserved 97

 Pulse Width and Bandwidth
We need a practical general relationship between pulse width
and bandwidth.
x(t) |X( f )|
x(0) X(0)
Equal Equal
Areas Areas

t f
T T −W W
2 2
Let the rectangular pulse approximate the general pulse with the
∞ ∞

same height and area. Then T x ( 0 ) = ∫ x (t ) dt ≥ ∫ x (t ) dt = X ( 0 ).

−∞ −∞

Let the rectangular bandwidth approximate the general pulse bandwidth

∞ ∞

with the same height and area. Then 2W X ( 0 ) = ∫ X ( f ) df ≥ ∫ X ( f ) df = x ( 0 ).

−∞ −∞

8/25/15 M. J. Roberts - All Rights Reserved 98

 Pulse Width and Bandwidth

Now we have the relationships

x (0) 1 x (0)
≥ and 2W ≥
X(0) T X(0)

1 1
which combine to 2W ≥ or W ≥ . This is a handy, practical
T 2T
rule of thumb for the approximate bandwidth of a pulse.

8/25/15 M. J. Roberts - All Rights Reserved 99

 Quadrature Filters and Hilbert Transforms
A quadrature filter is an allpass network that shifts the phase of
positive frequency components by − 90° and negative frequency
components by + 90°. Its frequency response is therefore
⎧− j , f > 0 ⎫
HQ ( f ) = ⎨ ⎬ = − j sgn ( f ) .
⎩ j , f < 0⎭
Its magnitude is one at all frequencies, therefore an even function
of f and its phase is an odd function of f . The inverse Fourier
transform of H Q ( f ) is the impulse response hQ ( t ) = 1 / π t. The Hilbert
transform x̂ ( t ) of a signal x ( t ) is defined as the response of a
1 x (λ )
∞

quadrature filter to x ( t ) . That is x̂ ( t ) = x ( t ) ∗ hQ ( t ) = ∫ d λ.

π −∞ t − λ
F ( x̂ ( t )) = − j sgn ( f ) X ( f )

8/25/15 M. J. Roberts - All Rights Reserved 100

Quadrature Filters and Hilbert Transforms
The impulse response of a quadrature filter hQ ( t ) = 1 / π t is non-causal.
That means it is physically unrealizable. Some important properties of
the Hilbert transform are
1. The Fourier transforms of a signal and its Hilbert transform have
the same magnitude. Therefore the signal and its Hilbert transform
have the same signal energy.
2. If x̂ ( t ) is the Hilbert transform of x ( t ) then − x ( t ) is the Hilbert
transform of x̂ ( t ) .
3. A signal x ( t ) and its Hilbert transform are orthogonal on the entire
∞

real line. That means for energy signals ∫ x ( t ) x̂ ( t ) dt = 0 and for

−∞
T
1
power signals lim
T →∞ 2T ∫ x (t ) x̂ (t ) dt = 0.
−T

8/25/15 M. J. Roberts - All Rights Reserved 101

Quadrature Filters and Hilbert Transforms
g (t ) ĝ ( t )
a1 g1 ( t ) + a2 g 2 ( t ); a1 , a2 ∈! a1 ĝ1 ( t ) + a2 ĝ 2 ( t )
h (t − t0 ) ĥ ( t − t 0 )
h ( at );a ≠ 0 sgn ( a ) ĥ ( at )
d
dt
( h ( t )) d
dt
( )
ĥ ( t )

1
δ (t )
πt
e jt − je jt
e− jt je− jt
cos ( t ) sin ( t )
1 2t + 1
rect ( t ) ln
π 2t − 1
sinc ( t ) (π t / 2 ) sinc 2 (t / 2 ) = sin (π t / 2 ) sinc (t / 2 )
1 t
1+ t2 1+ t2

8/25/15 M. J. Roberts - All Rights Reserved 102

 Analytic Signals and Complex Envelopes
An analytic signal x p ( t ) corresponding to a real signal x ( t ) is
defined by x p ( t ) = x ( t ) + j x̂ ( t ) . The envelope of a signal x ( t ) is
defined as the magnitude of the analytic signal x p ( t ) . It follows
that
X p ( f ) = X ( f ) + j × ( − j ) sgn ( f ) X ( f ) = X ( f ) ⎡⎣1+ sgn ( f ) ⎤⎦ = 2 X ( f ) u ( f )
⎧2 X ( f ) , f > 0
Therefore X p ( f ) = ⎨ . Similarly,
⎩0 ,f <0
⎧0 , f >0
x n ( t ) = x ( t ) − j x̂ ( t ) and X n ( f ) = 2 X ( f ) u ( − f ) = ⎨ .
⎩2 X ( f ) , f < 0

8/25/15 M. J. Roberts - All Rights Reserved 103

Analytic Signals and Complex Envelopes
The complex envelope of a real signal x ( t ) is defined as
x! ( t ) = x p ( t ) e− j 2 π f0t where f0 is a reference frequency chosen for
convenience. Therefore x p ( t ) = x! ( t ) e j 2 π f0t = x ( t ) + j x̂ ( t )
( ) ( )
and x ( t ) = Re x! ( t ) e j 2 π f0t and x̂ ( t ) = Im x! ( t ) e j 2 π f0t .
x ( t ) = Re ( x! ( t ) ( cos ( 2π f t ) + j sin ( 2π f t )))
0 0

⎛ Re ( x! ( t )) cos ( 2π f0t ) + j Im ( x! ( t )) cos ( 2π f0t ) ⎞

x ( t ) = Re ⎜ ⎟
⎜ + jRe ( x! ( t )) sin ( 2π f0t ) + "
j × j Im ( x! ( t )) sin ( 2π f0t )⎟
⎝ =−1 ⎠
x ( t ) = x R ( t ) cos ( 2π f0t ) − x I ( t ) sin ( 2π f0t )
where x R ( t ) = Re ( x! ( t )) and x I ( t ) = Im ( x! ( t )) ,
x! ( t ) = x R ( t ) + j x I ( t ) , x R ( t ) is the "in-phase" component of x ( t )
and x I ( t ) is the "quadrature" component of x ( t ) .

8/25/15 M. J. Roberts - All Rights Reserved 104

 Analytic Signals and Complex Envelopes

It can be shown (page 89 in the text) that if a system has a

bandpass response with impulse response h ( t ) and it is
excited by a bandpass signal x ( t ) , that the complex envelope
of the system response is y! ( t ) = x! ( t ) ∗ h! ( t ) = F −1 X(
! ( f )H
!(f) )
1
(
and the system response is y ( t ) = Re y! ( t ) e j 2 π f0t
2
). (The term
"bandpass" means that there is a finite-width band of frequencies,
including f = 0, in which the Fourier magnitude spectrum is zero
or, as a practical matter, small enough to be considered negligible.)

8/25/15 M. J. Roberts - All Rights Reserved 105

 Analytic Signals and Complex Envelopes
In the previous two slides a real signal x ( t ) was related to its complex
( )
envelope x! ( t ) by x ( t ) = Re x! ( t ) e j 2 π f0t and a real system impulse
response h ( t ) was related to its complex envelope h! ( t ) by
( )
h ( t ) = Re h! ( t ) e j 2 π f0t . But then when x ( t ) is applied to the system
and the response is y ( t ) , we found y! ( t ) = x! ( t ) ∗ h! ( t ) and related it
1
(
to y ( t ) by y ( t ) = Re y! ( t ) e
2
j 2 π f0 t
) 1
. Where did the factor of come
2
from? It can be seen in the derivation on page 89. But it can also be
seen in concept by looking at what happens when we convolve a
bandpass signal and a bandpass impulse response and compare that
to convolving the corresponding complex envelopes.
8/25/15 M. J. Roberts - All Rights Reserved 106
 Analytic Signals and Complex Envelopes
Let x ( t ) = Π ( t ) sin ( 8π t ) and h ( t ) = −Π ( t ) sin ( 8π t ) . The complex
envelope of x ( t ) has twice the signal energy of x ( t ) . The same is
true for h ( t ) . As Bandpass Signals and Impulse Response Complex Envelopes
a result, the 1 1

x(t)

~x(t)
0 0
complex envelope −1 −1

of y ( t ) has four −2 −1.5 −1 −0.5 0

t
0.5 1 1.5 2 −2 −1.5 −1 −0.5 0
t
0.5 1 1.5 2

1 1
times the signal
h(t)

h(t)
0 0

energy of y ( t ) .

~
−1 −1

−2 −1.5 −1 −0.5 0 0.5 1 1.5 2 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2

t t

1 1
y(t)

y(t)
0 0

~
−1 −1

−2 −1.5 −1 −0.5 0 0.5 1 1.5 2 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2

t t

8/25/15 M. J. Roberts - All Rights Reserved 107

 Analytic Signals and Complex Envelopes
Example
Let x ( t ) = sinc ( t ) cos ( 4π t ) .
1
Then X ( f ) = rect ( f ) ∗ ⎡⎣δ ( f − 2 ) + δ ( f + 2 ) ⎤⎦
2
1
X ( f ) = ⎡⎣ rect ( f − 2 ) + rect ( f + 2 ) ⎤⎦ and X̂ ( f ) = − j sgn ( f ) X ( f ) .
2
j j
X̂ ( f ) = − rect ( f − 2 ) + rect ( f + 2 )
2 2
j
= rect ( f ) ∗ ⎡⎣δ ( f + 2 ) − δ ( f − 2 ) ⎤⎦
2
x̂ ( t ) = sinc ( t ) sin ( 4π t ) .

8/25/15 M. J. Roberts - All Rights Reserved 108

Analytic Signals and Complex Envelopes
x ( t ) = sinc ( t ) cos ( 4π t ) x̂ ( t ) = sinc ( t ) sin ( 4π t )
x p ( t ) = sinc ( t ) ⎡⎣ cos ( 4π t ) + j sin ( 4π t ) ⎤⎦
x p ( t ) is the envelope of x ( t ) . The concept of an envelope will
be very useful later in the exploration of modulation techniques.
1

x(t)
0.5
x(t)
x(t)

−0.5

t (s)
−1
−4 −3 −2 −1 0 1 2 3 4
t (s)

1 1

0.9

0.8
0.5
0.7

0.6
|xp(t)|
x(t)

0 0.5

0.4

0.3
−0.5
0.2

0.1

−1 0
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4

t (s) t (s)

8/25/15 M. J. Roberts - All Rights Reserved 109

 Analytic Signals and Complex Envelopes
Example 2.32 in the text :
⎛t⎞
Let x ( t ) = Π ⎜ ⎟ cos ( 2π f0t ) and let h ( t ) = α e−α t u ( t ) cos ( 2π f0t ) .
⎝τ ⎠
Find the system output signal y ( t ) using complex envelope techniques.
⎛t⎞ ⎛t⎞
x p ( t ) = x ( t ) + j x̂ ( t ) = Π ⎜ ⎟ cos ( 2π f0t ) + jΠ ⎜ ⎟ sin ( 2π f0t )
⎝τ ⎠ ⎝τ ⎠
⎛ t ⎞ j 2 π f0 t
x p (t ) = Π ⎜ ⎟ e
⎝τ ⎠
(
⇒ x! ( t ) = Re x p ( t ) e − j 2 π f0 t
)
⎛ t⎞
= Π ⎜ ⎟ . Similarly,
⎝τ ⎠
!h ( t ) = α e−α t u ( t ) . Therefore y! ( t ) = x! ( t ) ∗ h! ( t ) = Π ⎛⎜ t ⎞⎟ ∗ α e−α t u ( t )
⎝τ ⎠
∞
⎛ λ ⎞ −α (t−λ )
y! ( t ) = ∫−∞ Π ⎜⎝ τ ⎟⎠ α e u (t − λ ) d λ

8/25/15 M. J. Roberts - All Rights Reserved 110

 Analytic Signals and Complex Envelopes
⎡∞ ⎤
⎢ ∫ u (λ + τ / 2) e u (t − λ ) d λ ⎥
− α ( t− λ )

y! ( t ) = α ⎢ ∞ ⎥
−∞

⎢ ⎥
⎢− ∫ u (λ − τ / 2) e u (t − λ ) d λ ⎥
− α ( t− λ )

⎢⎣ −∞ ⎥⎦

( ) ( )
y! ( t ) = ⎡⎣ 1− e−α (t+τ /2 ) u ( t + τ / 2 ) − 1− e−α (t−τ /2 ) u ( t − τ / 2 ) ⎤⎦
1
(
y ( t ) = Re y! ( t ) e j 2 π f0t
2
)
1
((( ) ( ) )
y ( t ) = Re 1− e−α (t+τ /2 ) u ( t + τ / 2 ) − 1− e−α (t−τ /2 ) u ( t − τ / 2 ) e j 2 π f0t
2
)
1⎡
( ) ( )
y ( t ) = ⎣ 1− e−α (t+τ /2 ) u ( t + τ / 2 ) − 1− e−α (t−τ /2 ) u ( t − τ / 2 ) ⎤⎦ cos ( 2π f0t )
2

8/25/15 M. J. Roberts - All Rights Reserved 111

 Energy Spectral Density
According to Parseval's Theorem, the signal energy of a signal can be
∞

∫ X( f ) df . X ( f )
2
found directly from its Fourier transform, Ex =
−∞

indicates the variation of the amplitudes of the complex sinusoidal

components of x ( t ) as a function of their frequencies, f . So the units
V
(if x ( t ) is a voltage signal). Therefore the units of X ( f )
2
are
Hz
2
⎛ V⎞
. When we integrate X ( f ) over all frequencies we
2
must be ⎜ ⎟
⎝ Hz ⎠
2
⎛ V⎞
get signal energy whose units are ⎜ ⎟ Hz or V 2
⋅s, the units of
⎝ Hz ⎠
signal energy.
8/25/15 M. J. Roberts - All Rights Reserved 112
 Energy Spectral Density

Therefore X ( f ) is the density of signal energy as a function of

2

frequency. It is known as Energy Spectral Density (ESD). The

term "spectral density" means "variation with respect to frequency".
A common symbol for ESD is G ( f ) = X ( f ) .
2

8/25/15 M. J. Roberts - All Rights Reserved 113

 Power Spectral Density

Energy Spectral Density is the variation of signal energy with

frequency. It applies to energy signals. The corresponding quantity
that applies to power signals is Power Spectral Density (PSD).
∞

Power spectral density S ( f ) is defined by P = ∫ S( f ) df . That is,

−∞

its integral over all frequency yields total average signal power.
Therefore S ( f ) indicates the variation of average signal power as
a function of frequency.

8/25/15 M. J. Roberts - All Rights Reserved 114

 Autocorrelation
Suppose we take the inverse Fourier transform of energy spectral
density

( G ( f )) ⎛
(
X ( f ) ⎞ = F −1 X ( f ) X * ( f ) )
2
−1 −1
F =F
⎝ ⎠
( )
= F −1 X ( f ) ∗ F −1 X * ( f )
!#"#$ !# #"## $
( )
=x t () =x −t( )
∞

So F −1
(G ( f )) = x (t ) ∗ x ( −t ) = ∫ x (τ ) x (t + τ ) dτ .
−∞

We can exchange the meanings of t and τ to form

∞

φ (τ ) = x (τ ) ∗ x ( −τ ) = ∫ x (t ) x (t + τ ) dt
−∞

This integral is the area under the product of a function x and a

version of x that has been shifted to the left by τ , as a function of
the shift amount. This function is called autocorrelation.

8/25/15 M. J. Roberts - All Rights Reserved 115

 Autocorrelation
The autocorrelation function indicates how similar a signal is to
∞

itself when shifted. When the shift is zero (τ =0), φ ( 0 ) = ∫ x 2 ( t )dt

−∞

which is the signal energy. If the shift τ is small and the value of
φ (τ ) does not change much, we say there is a strong correlation
between x and the shifted version of x for small shifts. So a slowly
changing φ (τ ) indicates that the signal still looks like itself even
when shifted a significant amount. A quickly changing φ (τ ) indicates
that even a small shift makes the signal look very different.

8/25/15 M. J. Roberts - All Rights Reserved 116

 Autocorrelation
∞

The definition φ (τ ) = ∫ x (t ) x (t + τ ) dt applies to energy signals.

−∞

For power signals, the definition of autocorrelation is

R (τ ) = x ( t ) x ( t + τ ) and R (τ ) ←⎯
F
→ S ( f ) . Some properties of
autocorrelation are
∞

1. R ( 0 ) = ∫ S( f ) df = total average signal power

−∞

2. R ( 0 ) ≥ R (τ ) , autocorrelation can never exceed the signal power

3. R (τ ) is always an even function, that is R (τ ) = R ( −τ )

( )
4. F R (τ ) is everywhere non-negative
5. If x ( t ) is periodic then R x (τ ) is also, with the same period

6. If x ( t ) contains no periodic components lim R x (τ ) = x ( t )

2

τ →∞

8/25/15 M. J. Roberts - All Rights Reserved 117

 Autocorrelation Examples
Autocorrelations of a cosine and sine "burst".
They are very similar but not exactly the same.
Notice that both are even functions, even though
cosine is even and sine is odd.
 Autocorrelation Examples
Three random power signals with different frequency content

and their autocorrelations.

8/26/15 M. J. Roberts - All Rights Reserved 119

 Autocorrelation Examples

Four Different

Random Signals

with Identical

Autocorrelations

8/26/15 M. J. Roberts - All Rights Reserved 120

 Autocorrelation Examples

Four Different

Random Signals

with Identical

Autocorrelations

8/26/15 M. J. Roberts - All Rights Reserved 121

 Sampling
Uniform sampling of a continuous-time signal can be represented by
multiplying the signal by a periodic impulse, forming a signal
consisting only of impulses.
∞
xδ (t ) = ∑ x ( nT )δ (t − nT )
s s
n=−∞

where Ts is the time between samples. When sampling a signal, the

salient question is always whether the original continuous-time signal
can be recovered from the samples. The sampling theorem says that
if the signal is sampled for all time at a rate greater than twice the
highest frequency in the signal, the original signal can be recovered
exactly from the samples.

8/26/15 M. J. Roberts - All Rights Reserved 122

 Sampling
Sampling Theorem: If the signal is sampled for all time at a rate
greater than twice the highest frequency in the signal, the original
signal can be recovered exactly from the samples.
Practically speaking, a signal can never be sampled for all time. Also
if a signal is not bandlimited its highest frequency is infinite, requiring
an infinite sampling rate, and no real signal can be bandlimited because
all real signals are time limited. Therefore the sampling theorem can
never quite be satisfied. The sampling theorem really just serves as a
limiting requirement to be approached but never reached in practice.
Practically, sampling always yields an approximation, but one which
can often be very good.

8/26/15 M. J. Roberts - All Rights Reserved 123

 Sampling
The sampling rate that is twice the highest rate in a signal is called
the Nyquist rate in honor of Harry Nyquist, one of the earliest
contributors to sampling theory.
There is a variation on the sampling theorem for signals that are
narrowband. That is, signals whose center frequency is much greater
than the bandwidth. If the bandwidth is W and the highest frequency
is f u and the signal is sampled at a rate f s = 2 f u / m where m is the
greatest integer in f u / W , then the signal can be recovered from the
samples by passing it through a bandpass filter.

8/26/15 M. J. Roberts - All Rights Reserved 124