Professional Documents
Culture Documents
CH 5
CH 5
CH 5
5.0 Introduction
• There are many similarities and strong parallels in analyzing continuous-time and discrete-
time signals.
• There are also important differences. For example, the Fourier series representation of a
discrete-time periodic signal is finite series, as opposed to the infinite series representation
required for continuous-time period signal.
• In this chapter, the analysis will be carried out by taking advantage of the similarities
between continuous-time and discrete-time Fourier analysis.
Consider a general sequence that is a finite duration. That is, for some integers N 1 and N 2 , x[n]
equals to zero outside the range N 1 ≤ n ≤ N 2 , as shown in the figure below.
We can construct a periodic sequence ~ x [n] using the aperiodic sequence x[n] as one period. As
~
we choose the period N to be larger, x [n] is identical to x[n] over a longer interval, as N → ∞ ,
~
x [n] = x[n] .
Based on the Fourier series representation of a periodic signal given in Eqs. (3.80) and (3.81), we
have
1/5 Yao
ELG 3120 Signals and Systems Chapter 5
~
x [ n] = ∑a e k
jk (2π / N ) n
, (5.1)
k =< N >
ak = ∑ ~x [n]e − jk (2π / N ) n
. (5.2)
k= N
N2 ∞
1 1
ak =
N
∑ x[n]e − jk ( 2π / N ) n = N
∑ x[n]e − jk ( 2π / N ) n
, (5.3)
k = N1 k = −∞
∞
jω
X (e ) = ∑ x[n ]e
n = −∞
− j ωn
, (5.4)
So a k can be written as
1
ak = X (e jkω 0 ) , (5.5)
N
Then ~
x [n] can be expressed as
1 1
~
x [n] = ∑
k =< N > N
X (e jk ω 0 )e jk (2π / N ) n =
2π
∑ X (e jk ω0
)e jk ( 2π / N ) nω 0 . (5.6)
k =< N >
As N → ∞ ~
x [n] = x[n] , and the above expression passes to an integral,
1
x[n ] =
2π ∫ 2π
X (e jω )e jωn dω , (5.7)
1
x[n ] =
2π ∫ 2π
X (e jω )e jωn dω , (5.8)
∞
X (e ) = jω
∑ x[n]e
n = −∞
− jω n
. (5.9)
2/5 Yao
ELG 3120 Signals and Systems Chapter 5
Eq. (5.8) is referred to as synthesis equation, and Eq. (5.9) is referred to as analysis equation
and X (e jkω 0 ) is referred to as the spectrum of x[n] .
∞ ∞ ∞
X (e jω ) = ∑ x[n]e − jωn = ∑ a n u[n ]e − jωn = ∑ ae − jω ( ) −n
=
1
1 − ae − jω
. (5.11)
n = −∞ n= −∞ n= 0
The magnitude and phase for this example are show in the figure below, where a > 0 and a < 0
are shown in (a) and (b).
∞ −1 ∞
X (e jω ) = ∑ a u[n ]e − jωn =
n = −∞
n
∑ a −n e − jωn + ∑ a n e − jωn
n = −∞ n =0
∞ ∞ ∞
X (e jω ) = ∑ a u[ n]e − jωn = ∑ a m e jωm + ∑ a n e − jωn
n
n = −∞ m =1 n =0
. (5.13)
jω
ae 1 1− a 2
= jω
+ − jω
=
1 − ae 1 − ae 1 − 2 a cos ω + a 2
3/5 Yao
ELG 3120 Signals and Systems Chapter 5
1, n ≤ N1
x[n ] = , (5.14)
0, n > N1
sin ω (N1 + 1 / 2 )
N1
X ( jω ) = ∑e
n= − N
− j ωn
=
sin (ω / 2 )
. (5.15)
1
5.1.3 Convergence
∞
The equation X (e jω ) = ∑ x[n ]e
n = −∞
− j ωn
converges either if x[n] is absolutely summable, that is
∑ x[n] < ∞ ,
n = −∞
(5.16)
∞ 2
4/5 Yao
ELG 3120 Signals and Systems Chapter 5
And there is no convergence issues associated with the synthesis equation (5.8).
1
∫
W
xˆ[ n] = X ( e jω )e jω n dω , (5.18)
2π −W
and xˆ[n ] = x[n ] for W = π . Therefore, the Gibbs phenomenon does not exist in the discrete-time
Fourier transform.
5/5 Yao
ELG 3120 Signals and Systems Chapter 5
x[n ] = e jω 0 n , (5.19)
its Fourier transform of this signal is periodic in ω with period 2π , and is given
+∞
jω
X (e ) = ∑ 2πδ (ω − ω
l = −∞
0
− 2πl ) . (5.20)
Now consider a periodic sequence x[n] with period N and with the Fourier series representation
x[n ] = ∑a e k
jk ( 2π / N ) n
. (5.21)
k =< N >
+∞
2πk
jω
X (e ) = ∑ 2πa δ (ω −
k = −∞
k
N
). (5.22)
1 jω0n 1 − jω0 n 2π
x[n ] = cos ω0 n = e + e , with ω 0 = , (5.23)
2 2 3
is given as
2π 2π
X (e jω ) = πδ ω − + πδ ω + , −π ≤ ω < π . (5.24)
3 3
6/5 Yao
ELG 3120 Signals and Systems Chapter 5
+∞
x[n ] = ∑δ [n − kN ] .
k = −∞
(5.25)
ak = ∑ x[n ]e − jk (2π / N ) n
. (5.26)
n =< N >
1
ak = . (5.27)
N
2π ∞
2πk
X (e jω ) =
N
∑ δ ω −
k = −∞ N
. (5.28)
7/5 Yao
ELG 3120 Signals and Systems Chapter 5
Notations to be used
X (e jω ) = F {x[n]},
{ }
x[n ] = F −1 X (e jω ) ,
x[n ] ←→
F
X (e jω ) .
( ) ( )
X e j (ω +2π ) = X e jω . (5.29)
5.3.2 Linearity
If x1 [n] ←→
F
X 1 (e j ω ) , and x 2 [n] ←→
F
X 2 (e j ω ) ,
then
If x[n ] ←→
F
X (e jω ) ,
then
x[ n − n0 ] ←→
F
e − jωn0 X ( e jω ) (5.31)
and
e jω0n x[ n] ←→
F
X ( e j (ω− ω0 ) ) (5.32)
8/5 Yao
ELG 3120 Signals and Systems Chapter 5
If x[n ] ←→
F
X (e jω ) ,
then
x *[n]←→
F
X * (e− jω ) (5.33)
X (e jω ) = X * (e− jω ) (5.34)
{ } { }
From this, it follows that Re X (e jω ) is an even function of ω and Im X (e j ω ) is an odd
function of ω . Similarly, the magnitude of X (e jω ) is an even function and the phase angle is
an odd function. Furthermore,
Ev{x[n]}←→
F
{
Re X (e jω , } (5.35)
and
Od {x[ n]}←→
F
{
j Im X (e jω . } (5.36)
If x[n ] ←→
F
X (e jω ) ,
then
For signal
n
y[ n] = ∑ x[ m] ,
m = −∞
(5.38)
9/5 Yao
ELG 3120 Signals and Systems Chapter 5
n +∞
1
∑
m = −∞
x[ m] ←→
1−e
F
− jω
X ( e ) + πX (e ) ∑ δ (ω − 2πk ) .
jω j0
m =−∞
(5.39)
The impulse train on the right-hand side reflects the dc or average value that can result from
summation.
For example, the Fourier transform of the unit step x[n ] = u[n ] can be obtained by using the
accumulation property.
n +∞ +∞
1 1
x[n ] = ∑ g[m] ←→
m = −∞
F
(
1 − e − jω)
G (e jω
) + π G ( e j0
) ∑
k = −∞
δ (ω − 2π k
(
) =
)
1 − e − jω
+ π ∑ δ (ω − 2πk ) .
k = −∞
(5.40)
If x[n ] ←→
F
X (e jω ) ,
then
x[−n] ←→
F
X (−e jω ) . (5.41)
1 jω
x( at) ←→
F
X . (5.42)
a a
For discrete-time signals, however, a should be an integer. Let us define a signal with k a
positive integer,
x[n / k ], if n is a multiple of k
x( k ) [n] = . (5.43)
0, if n is not a multiple of k
x( k ) [n] is obtained from x[n] by placing k − 1 zeros between successive values of the original
signal.
10/5 Yao
ELG 3120 Signals and Systems Chapter 5
+∞ +∞ +∞
X ( k ) (e j ω ) = ∑ x( k ) [n ]e − jωn =
n = −∞
∑ x ( k ) [rk ]e − jωrk =
r = −∞
∑ x[r ]e
r = −∞
− j ( kω ) r
= X (e jk ω ) . (5.44)
That is,
For k > 1 , the signal is spread out and slowed down in time, while its Fourier transform is
compressed.
Example: Consider the sequence x[n] displayed in the figure (a) below. This sequence can be
related to the simpler sequence y[n] as shown in (b).
x[n ] = y( 2 ) [n ] + 2 y (2 ) [ n − 1] ,
where
As can be seen from the figure below, y[n] is a rectangular pulse with N 1 = 2 , its Fourier
transform is given by
sin( 5ω / 2)
Y (e jω ) = e − j 2ω .
sin(ω / 2)
11/5 Yao
ELG 3120 Signals and Systems Chapter 5
sin( 5ω )
y ( 2) [n] ←→
F
e − j 4ω
sin(ω )
sin( 5ω )
2 y ( 2) [n − 1] ←→
F
2e − j 5ω
sin(ω )
sin( 5ω )
X (e jω ) = e − j 4ω (1 + 2e − jω ) .
sin(ω )
If x[n ] ←→
F
X (e jω ) ,
∞
Differentiate both sides of the analysis equation X (e jω ) = ∑ x[n ]e
n = −∞
− jωn
jω
dX (e ) +∞
= ∑ − jnx[n]e − jωn .
dω n= −∞
(5.46)
The right-hand side of the Eq. (5.46) is the Fourier transform of − jnx[n] . Therefore, multiplying
both sides by j , we see that
dX (e jω )
nx[n]←→ j F
dω . (5.47)
If x[n ] ←→
F
X (e jω ) , then we have
+∞
1
∑ x[n] = ∫
2
X (e jω ) dω
2
n = −∞ 2π 2π (5.48)
12/5 Yao
ELG 3120 Signals and Systems Chapter 5
Example: Consider the sequence x[n] whose Fourier transform X (e jω ) is depicted for
− π ≤ ω ≤ π in the figure below. Determine whether or not, in the time domain, x[n] is periodic,
real, even, and /or of finite energy.
• The periodicity in time domain implies that the Fourier transform has only impulses located
at various integer multiples of the fundamental frequency. This is not true for X (e jω ) . We
conclude that x[n] is not periodic.
• Since real-valued sequence should have a Fourier transform of even magnitude and a phase
function that is odd. This is true for X (e jω ) and ∠X (e jω ) . We conclude that x[n] is real.
• If x[n] is real and even, then its Fourier transform should be real and even. However, since
X (e jω ) = X (e jω ) e − j 2ω , X (e jω ) is not real, so we conclude that x[n] is not even.
2
• Based on the Parseval’s relation, integrating X (e j ω ) from − π to π will yield a finite
quantity. We conclude that x[n] has finite energy.
If x[n] , h[n] and y[n] are the input, impulse response, and output, respectively, of an LTI
system, so that
then,
Y (e jω ) = X (e jω ) H (e jω ) , (5.50)
13/5 Yao
ELG 3120 Signals and Systems Chapter 5
where X (e jω ) , H (e j ω ) and Y (e jω ) are the Fourier transforms of x[n] , h[n] and y[n] ,
respectively.
Example: Consider the discrete-time ideal lowpass filter with a frequency response H (e j ω )
illustrated in the figure below. Using − π ≤ ω ≤ π as the interval of integration in the synthesis
equation, we have
1 π
h[ n] =
2π ∫ −π
H (e j ω )e jωn dω
1 π sin ω c n
=
2π ∫
−π
e jωn dω =
πn
1
H (e j ω ) = ,
1 − α e − jω
and
1
X (e jω ) = ,
1 − β e − jω
so that
1
Y (e jω ) = H (e jω ) X (e jω ) = − jω
.
(1 − α e )(1 − β e − j ω )
14/5 Yao
ELG 3120 Signals and Systems Chapter 5
α β
−
A B α−β α −β
Y (e jω ) = + = + ,
(1 − α e ) (1 − β e ) (1 − α e ) (1 − β e − jω )
− jω − jω − jω
α β
y[ n] =
α−β
α n u[n] −
α−β
β n u[n ] =
1
α−β
( )
α n +1u[n ] − ββ n +1u[n ] .
For α = β ,
1
Y (e jω ) = , which can be expressed as
(1 − α e − j ω ) 2
j jω d 1
Y (e jω ) = e .
− jω
α dω 1 − α e
Using the frequency differentiation property, we have
d 1
nα n u[n] ←→ j
F
,
dω 1 − α e − j ω
d 1
(n + 1)α n +1u[n + 1] ←→ je jω
F
,
dω 1 − α e − jω
y[ n] = (n + 1)α n u[n + 1] .
y[ n] = (n + 1)α n u[n] .
Example: Consider the system shown in the figure below. The LTI systems with frequency
response H lp (e j ω ) are ideal lowpass filters with cutoff frequency π / 4 and unity gain in the
passband.
15/5 Yao
ELG 3120 Signals and Systems Chapter 5
⇒ W1 (e jω ) = X (e j (ω −π ) ) .
• W2 (e jω ) = H lp (e j ω ) X (e j (ω − π ) ) .
• w3 [ n] = (−1) n w2 [n ] = e jπn w2 [ n]
⇒ W3 (e j ω ) = W2 (e j (ω −π ) ) = H lp (e ( j ω − π ) ) X (e j (ω − 2π ) ) .
• W4 (e jω ) = H lp (e j ω ) ) X (e jω ) .
• [ ]
Y (e jω ) = W3 (e jω ) + W4 (e j ω ) = H lp (e ( j ω −π ) ) + H lp (e jω ) X (e j ω ) .
[ ]
H lp (e jω ) = H lp (e ( jω −π ) ) + H lp (e jω ) X (e jω ) ,
It is important to note that not every discrete-time LTI system has a frequency response. If an
LTI system is stable, then its impulse response is absolutely summable; that is,
+∞
∑ h[n] < ∞ ,
n = −∞
(5.51)
Consider y[n] equal to the product of x1 [n] and x 2 [n] , with Y (e jω ) , X 1 (e jω ) , and X 2 (e jω )
denoting the corresponding Fourier transforms. Then
16/5 Yao
ELG 3120 Signals and Systems Chapter 5
1
y[ n] = x1 [n] x2 [n ] ←→ ∫ X 1 (e jω ) X 2 (e j ( ω −θ ) ) dθ
F
(5.52)
2π 2π
Example: Consider the Fourier transform of a signal x[n] which the product of two signals; that
is
where
sin( 3πn / 4)
x1 [n] = , and
πn
sin(πn / 2)
x 2 [n ] = .
πn
1 π
X (e jω ) =
2π ∫
−π
X 1 (e jω ) X 2 (e j (ω −θ ) )dθ . (5.53)
Eq. (5.53) resembles aperiodic convolution, except for the fact that the integration is limited to
the interval of − π < θ < π . The equation can be converted to ordinary convolution with
integration interval − ∞ < θ < ∞ by defining
Then replacing X 1 ( e jω ) in Eq. (5.53) by Xˆ 1 (e jω ) , and using the fact that Xˆ 1 (e jω ) is zero for
− π < ω < π , we see that
1 π 1 ∞
X (e jω ) =
2π ∫
−π
X 1 (e jω ) X 2 (e j (ω −θ ) )dθ =
2π ∫
−∞
Xˆ 1 (e j ω ) X 2 (e j (ω −θ ) )dθ .
Thus, X (e jω ) is 1 / 2π times the aperiodic convolution of the rectangular pulse Xˆ 1 (e jω ) and the
periodic square wave X 2 (e jω ) . The result of thus convolution is the Fourier transform X (e jω ) ,
as shown in the figure below.
17/5 Yao
ELG 3120 Signals and Systems Chapter 5
18/5 Yao
ELG 3120 Signals and Systems Chapter 5
19/5 Yao
ELG 3120 Signals and Systems Chapter 5
5.7 Duality
For continuous-time Fourier transform, we observed a symmetry or duality between the analysis
and synthesis equations. For discrete-time Fourier transform, such duality does not exist.
However, there is a duality in the discrete-time series equations. In addition, there is a duality
relationship between the discrete-time Fourier transform and the continuous-time Fourier
series.
Consider the periodic sequences with period N, related through the summation
1
f [m ] = ∑ g (r )e − jr ( 2π / N ) m .
N r =< N >
(5.54)
1
f [ n] = ∑ N
g (− r )e jr ( 2π / N ) n . (5.55)
k =< N >
x[n] = ∑a e
k= N
k
jk ( 2π / N ) n
, (3.80)
1
ak =
N
∑ x[n]e
k= N
− jk ( 2π / N ) n
. (3.81)
1
we fond that g (− r ) corresponds to the sequence of Fourier series coefficients of f [n ] . That is
N
1
f [n] ←→
FS
g[− k ] . (5.56)
N
This duality implies that every property of the discrete-time Fourier series has a dual. For
example,
x[n − n 0 ] ←→
FS
a k e − jk ( 2π / N ) n 0 (5.57)
e jm ( 2π / N ) n ←→
FS
a k −m (5.58)
20/5 Yao
ELG 3120 Signals and Systems Chapter 5
are dual.
1 sin( 5πn / 9)
9 sin(πn/9) , n ≠ multiple of 9
x[n ] = (5.59)
5 , n = multiple of 9
9
We know that a rectangular square wave has Fourier coefficients in a form much as in Eq. (5.59).
Duality suggests that the coefficients of x[n] must be in the form of a rectangular square wave.
1, n ≤2
g[ n ] = , (5.60)
0, 2< n ≤4
The Fourier series coefficients b k for g[n] can be given (refer to example on page 27/3)
1 sin( 5πk / 9)
9 sin(πk/9) , k ≠ multiple of 9
bk = . (5.61)
5 , k = multiple of 9
9
1 2
bk = ∑ (1)e − j 2πnk / 9 . (5.62)
9 n =−2
Interchanging the names of the variable k and n and noting that x[n ] = bk , we find that
1 2
x[n ] = ∑
9 k =−2
(1)e − j 2πnk / 9 .
1 2
x[n ] = ∑
9 k =−2
(1)e + j 2πnk ' / 9 .
21/5 Yao
ELG 3120 Signals and Systems Chapter 5
Finally, moving the factor 1 / 9 inside the summation, we see that the right side of the equation
has the form of the synthesis equation for x[n] . Thus, we conclude that the Fourier coefficients
for x[n] are given by
1 / 9, k ≤2
ak = ,
0, 2< k ≤4
with period of N = 9 .
A general linear constant-coefficient difference equation for an LTI system with input x[n] and
output x[n] is of the form
N M
∑a
k =0
k y[n − k ] =∑ bk x[n − k ] ,
k =0
(5.63)
• The first way is to apply an input x[n ] = e jωn to the system, and the output must be of the
form H (e j ω )e j ωn . Substituting these expressions into the Eq. (5.63), and performing some
algebra allows us to solve for H (e j ω ) .
• The second approach is to use discrete-time Fourier transform properties to solve for
H (e j ω ) .
Y (e j ω )
jω
H (e ) = . (5.64)
X (e jω )
Applying the Fourier transform to both sides and using the linearity and time-shifting properties,
we obtain the expression
N M
∑ a k e − jkω Y (e jω ) = ∑ bk e − jkω X (e jω ) .
k =0 k =0
(5.65)
22/5 Yao
ELG 3120 Signals and Systems Chapter 5
or equivalently
∑
M
jω Y (e jω ) k =0
bk e − jk ω
H (e ) = = . (5.66)
X (e jω ) ∑ a e − jkω
N
k =0 k
Example: Consider the causal LTI system that is characterized by the difference equation,
Y (e jω ) 1
H (e j ω ) = = .
X (e ) 1 − ae jω
jω
h[n ] = a n u[n] .
Example: Consider a causal LTI system that is characterized by the difference equation
3 1
y[ n] − y[n − 1] + y[n − 2] = 2 x[n] .
4 8
1 2
H (e j ω ) = = .
1 − 12 e − jω
+ 18 e − j 2 ω
(1 − 4 e )(1 − 14 e − jω )
3 − jω
4 2
H (e j ω ) == − ,
1 − 12 e − jω
1 − 14 e − jω
n n
1 1
h[n ] = 4 u[n] − 2 u[n ] .
2 4
23/5 Yao
ELG 3120 Signals and Systems Chapter 5
2 1
Y (e jω ) = H (e jω ) X (e jω ) = 3 − jω 1 − jω 1 − jω
(1 − 4 e )(1 − 4 e ) 1 − 4 e
.
2
= − jω
(1 − e
3
4
)(1 − e − jω )(1 − 14 e − jω )
1
4
4 2 8
Y (e jω ) = H (e jω ) X (e jω ) = − − +
1 − 14 e − jω
(
1 − 14 e − jω )
2
1 − 12 e − j ω
1 n 1
n
1
2
y[ n] = − 4 − 2( n + 1) + 8 u[n ] .
2 4 2
24/5 Yao