DIGITAL SIGNAL PROCESSING

LECTURE 4
Fall 2010
2K8-5th Semester
Tahir Muhammad
[email protected]
Content and Figures are from Discrete-Time Signal Processing, 2e by Oppenheim, Shafer, and
Buck, ©1999-2000 Prentice Hall Inc.
FREQUENCY-DOMAIN REPRESENTATION

input x[n] = e jωn for − ∞ < n < ∞ (complex exponential)

∞
⎛ ∞ ⎞
y[n] = ∑ h[k ]e
k = −∞
jω ( n − k )
=e jωn
⎜ ∑ h[k ]e − jωk
⎝ k =−∞ ⎠
Define
∞
e jωn : eigenfunct
H (e ) = ∑ h[k ]e
jω − jωk i f tion i off the
th system
t
k = −∞ H (e jω ) : eigenvalue
⇒ y[ n] = H (e jω )e jωn

H (e jω ) : frequency response, complex number

H (e ) = H (e ) + jH (e ) = H (e )e
jω jω
R
jω jω j∠H (e jω )
I

Digital Signal Processing Lecture 4 2

EXAMPLE

y[n] = x[n − nd ] : ideal delay

y[n] = e jω ( n − nd ) = e − jωnd e jωn (when e jωn is inputed)
( )
⇒ H e jω = e − jω n d
h[n] = δ [n − nd ]

( ) = ∑ δ [n − n ]e = e
∞
jω − jω n − j ωn d
⇒H e d
n = −∞

H (e ) = cos(ωn ) H (e ) = − sin (ωn )

R
jω
d I
jω
d

H (e ) = 1 ∠H (e ) = −ωn
jω jω
d

Digital Signal Processing Lecture 4 3

 General input :

x[n] = ∑ α k e jωk n
k

y[n] = ∑ α k H e ( )e
jω k jω k n

Digital Signal Processing Lecture 4 4

EXAMPLE
A jφ jω 0 n A − jφ − jω 0 n
x[n] = A cos(ω 0 n + φ ) = e e + e e
2 2
y[n] = e H (e )e + e H (e − jω0 )e − jω 0n
A jφ jω 0 jω 0 n A − jφ
2 2
If h[n] is real ⇒ H (e − jω 0 ) = H * (e jω 0 )

[ ]
y[n] = H (e jω 0 )e jφ e jω 0n + H (e − jω 0 )e − jφ e − jω 0n
A
2
= A H (e jω 0 ) cos(ω 0 n + φ + θ ) where θ = ∠H (e jω 0 )
For ideal delay H (e jω 0 ) = 1 , θ = ∠H (e jω0 ) = −ω 0 nd
y[n] = A cos[ω 0 (n − nd ) + φ ]
Digital Signal Processing Lecture 4 5
EIGENFUNCTIONS OF LTI SYSTEMS
| Complex exponentials are eigenfunctions of LTI systems:
x[n] = e jωn
| Let’s see what happens if we feed x[n] into an LTI system:
y
∞ ∞
y[n] = ∑ h[k ]x[n − k ] = ∑ h[k ]e jω(n − k )

k = −∞ k = −∞

| The eigenvalue is called the frequency response of the

system
y[n] = ⎜⎜ ∑ h[k ]e − jωk e jωn = H(e jω )e jωn
⎛ ∞ ⎞ Eigenfunction
⎝ k = −∞ ⎠
( )
j ω is a complex function of frequency Eigenvalue
|
H e
( ) = ∑ h[k ]e
∞
jω − jωk
He
k = −∞

Digital Signal Processing Lecture 4 6

DISCRETE-TIME FOURIER TRANSFORM
| Many sequences can be expressed as a weighted sum of
complex exponentials as

( ) = ∑ x [n ]e
∞
jω − jω n
X e (forward transform)
n = −∞

| Where the weighting is determined as

x [n ] =
1 π
2π ∫− π
X (e jω
)e jω n
dω (inverse transform)

y ( )
X e j ω is the Fourier spectrum of the sequence x[n]
y The phase wraps at 2π hence is not uniquely specified
y The frequency response of a LTI system is the DTFT of the
impulse response
( ) = ∑ h[k ]e 1 π
( )
∞
He jω − j ωk
and h[n] = ∫ H e jω
e jωn
dω
k = −∞ 2π − π

Digital Signal Processing Lecture 4 7

ABSOLUTE AND SQUARE SUMMABILITY
| For a g
given sequence
q if the infinite sum convergence,
g
the DTFT exist

( )
X e jω < ∞ for all ω

( ) = ∑ x[n]e
∞ ∞ ∞
Xe jω

n = −∞
− j ωn
≤ ∑ x[n] e
n = −∞
− jωn
= ∑ x[n] < ∞
n = −∞

| All stable systems are absolute summable and have

q y response
finite and continues frequency p

Digital Signal Processing Lecture 4 8

ABSOLUTE AND SQUARE SUMMABILITY
| Absolute summability is sufficient condition for DTFT
| Some sequences may not be absolute summable but only
square summable ∞

∑ x[n] < ∞
2

n = −∞

| Such sequences can be represented by fourier transform if

( ) = ∑ x[n]e
∞
( ) = ∑ x [ n ]e
M
jω − jωn jω − j ωn
Xe XM e
n = −∞ n =− M
π

∫ X(e ) − X (e )
2
jω jω
lim M =0
M→ ∞
−π

| In other words, the error X e jω

− X M e jω
may not ( ) ( )
approach
pp zero at each value of ω as M → ∞ but the total
energy in the error does.
Digital Signal Processing Lecture 4 9
EXAMPLE: IDEAL LOWPASS FILTER
| The periodic DTFT of the ideal lowpass filter is
⎧⎪1 ω < ωc
( )
Hlp e jω
=⎨
⎪⎩0 ωc < ω ≤ π
| Th inverse
The i can beb written itt as
H lp (e ) e d ω =
1 π 1 ωc j ω n
hlp [ n ] = ∫
jω j ωn
∫ e dω
2π − π 2π c − ω

sin ωc n
( )
1 ωc 1 j ωc n − j ωc n
= ⎡⎣e j ω n ⎤⎦ = e − e =
2π jn − ωc 2π jn πn

| Not causal, Not absolute summable but it has a DTFT, The

DTFT converges in the mean-squared sense
| Role of Gibbs phenomenon

Digital Signal Processing Lecture 4 10

| The impulse
p response
p is not causal,, Not absolutely
y
summable, but squarely summable,
| Since sequence values approach zero as n->infinity,
| B t only
But l as 1/n
1/

H M (e ) = ∑ sin ω n e
M
jω c − jωn
Let
n=− M πn

Digital Signal Processing Lecture 4 11

ROLE OF GIBBS PHENOMENON

Digital Signal Processing Lecture 4 12

 Is absolute summabality a necessary condition?
C
Consider
id theh Fourier
F i Transform:
T f
∞
X (e jω
) = ∑ 2πδ (ω − ω 0 + 2πr )
r = −∞

Assume that − π < ω 0 ≤ π .

1 π
∫
jωn
x[n] = 2πδ (ω − ω ) e dω
2π
0
−π

= e jω0n (Not absoultely summable nor square summable!!)

absolute summabality is not a necessary condition.

If x[n] = ∑ ak e jω k n
k
∞
X (e jω
) = ∑∑ 2πa δ (ω − ω
k k + 2πr )
r = −∞ k

Digital Signal Processing Lecture 4 13

EXAMPLE

Unit step sequence u[n]

∞
U (e ) =
1
jω
− jω
+ ∑ πδ (ω + 2πr )
1− e r = −∞

Digital Signal Processing Lecture 4 14

EXAMPLE: GENERALIZED DTFT
| DTFT of x[n]
[ ]=1
| Not absolute summable, Not even square summable
But we define its DTFT as a pulse train
( ) = ∑ 2πδ(ω + 2πr )
∞
jω
Xe
r = −∞

| Let’ss place into inverse DTFT equation

Let
x [n ] =
2π
1 π
∫− π
X e jω
( )
e j ωn
dω

1 π ⎡ ∞ ⎤ j ωn
=
2π ∫− π ⎢ ∑
⎣ r =−∞
2πδ (ω + 2π r ) ⎥ e d ω
⎦
π
= ∫ δ (ω ) e j ω n d ω = e j 0 n = 1
−π

Digital Signal Processing Lecture 4 15

 SYMMETRIC SEQUENCE AND FUNCTIONS

Conjugate-
Conjugate-symmetric
antisymmetric

Sequence x e [n] = x*e [− n] x o [n] = −x*o [− n]

x[n] = x e [n] + x o [n] x e [n] =

1
2
(
x[n] + x* [− n] ) x o [n] =
1
2
(
x[n] − x* [− n] )

Function ( ) (
X e e jω = X*e e − jω ) ( )
X o e jω = −X*o e − jω ( )
( ) ( ) ( )
X e jω = X o e jω + X e e jω ( )
X e e jω =
1
[( ) (
X e jω + X* e − jω )] X (e ) = 12 [X(e ) − X (e )]
o
jω jω * − jω

Digital Signal Processing Lecture 4 16

PROPERTIES OF DTFT
Sequence x[n] Discrete-Time Fourier Transform X(ejω)
x*[n] X*(e-jω)
x*[-n] X*(ejω)
Re{x[n]} Xe(ejω) (conjugate-symmetric part)
jIm{x[n]} Xo(ejω) (conjugate-antisymmetric part)
xe[n] XR(ejω)= Re{X(ejω)}
xo[n] jXI(ejω)= jIm{X(ejω)}
Any real x[n] X(ejω)=X*(e-jω) (conjugate symmetric)
Any real x[n] XR(ejω)=XR(e-jω) (real part is even)
Any real x[n] XI(ejω)=-XI(e-jω) (imaginary part is odd)
Any real x[n] |X(ejω)|=|X(e-jω)| (magnitude is even)
Any real x[n] ∠X(ejω)=-∠X(e-jω) (phase is odd)
xe[n] XR(ejω)
xo[n] jXI(ejω)

Digital Signal Processing Lecture 4 17

EXAMPLE: ILLUSTRATION OF SYMMETRY
PROPERTIES
| DTFT of the real sequence
q x[n]=a
[ ] nu[n]
[ ]
( )
Xe =jω 1
1 − ae − jω
if a < 1
| Some properties are
( )
X e jω =
1
1 − ae − j ω
= X *
(
e − jω
)
1 − a cos ω
( )
X R e jω =
1 + a 2 − 2a cos ω
= X R (e − jω
)
−a sin ω
( )
X I e jω =
1 + a 2 − 2a cos ω
= − X I (e − jω
)
( )
X e jω =
1
(
= X e − jω )
1 + a 2 − 2a cos ω
⎛ −a sin ω ⎞
( )
∠X e j ω = tan −1 ⎜
⎝ 1 − a cos ω ⎠
⎟ = −∠ X e(− jω
)
Digital Signal Processing Lecture 4 18
FOURIER TRANSFORM THEOREMS
x[n] X(ejω)
y[n] Y(ejω)
ax[n]+by[n] aX(ejω)+bY(ejω)
x[n-n
[ d] ( )
e − jωnd X e jω
x[-n] X(e-jω)
nx[n]
j
dX e jω ( )
dω
x[n]∗y[n] X(ejω)Y(ejω)
π
x[n]y[n] 1
∫
2π − π
( )( )
X e jθ Y e j ( ω − θ ) d θ
π
1
( )
∞

∑ x[n]
2 2

2π −∫π
jω
Parseval' s Theorem : = X e dω
n = −∞
π
1
( ) ( )
∞
Parseval' s Theorem : ∑ x[n]y [n] = ∫
*
X e jω Y * e jω dω
n = −∞ 2π − π

Digital Signal Processing Lecture 4 19

FOURIER TRANSFORM PAIRS
Sequence DTFT
δ[n-no] e − jωno
1 ∞
u[n] + ∑ πδ(ω + 2πk )
1 − e − jω k = −∞
sin(ωcn) ⎧⎪1 ω < ωc
( )
jω
Xe = ⎨
πn ⎪⎩0 ωc < ω ≤ π
⎧1 0≤n≤M sin[ω(M + 1) / 2] − jωM / 2
x[n] = ⎨ e
⎩0 otherwise sin(ω / 2)
∞

e jωon ∑ 2πδ(ω − ω
k = −∞
o + 2πk )

cos((ωon+φ) ∑ [πe φδ (ω − ω
k = −∞
j
o + 2πk ) + πe − jφ δ (ω + ωo + 2πk )]

Digital Signal Processing Lecture 4 20

EXAMPLE
Determining
g an inverse fourier transform

jω 1
X (e )=
(1 − ae − j ω )(1 − be
b − jω )

jω a /(a − b ) b /(a − b )
X (e )= − jω
−
1 − ae 1 − be − j ω

⎡ a b ⎤
x [n ] = ⎢( )a − (
n
)b n ⎥ u [n ]
⎣ a −b a −b ⎦

Digital Signal Processing Lecture 4 21

EXAMPLE
Determining
g the Impulse response from the frequency
y
response

Digital Signal Processing Lecture 4 22

EXAMPLE
Determining
g the Impulse response for a Difference
Equation 1 1
y [n ] − y [n − 1] = x [n ] − x [n − 1]
2 4

T find
To fi d the
h iimpulse
l response h[n],
h[ ] we set
1 1
h [n ] − h [n − 1] = δ [n ] − δ [n − 1]
2 4

Applying the DTFT to both sides of equation. We

obtain 1 1
H (e j ω ) − e − j ω H (e j ω ) = 1 − e − j ω
2 4
1
1− e − jω
H (e j ω ) = 4
1
1− e − jω
2
Digital Signal Processing Lecture 4 23
EXAMPLE

x[n] = a nu[n − 5]
1
⇒ a u[n] ⇔
n

1 − ae − jω
− j 5ω
e
⇒ a n −5u[n − 5] ⇔
1 − ae − jω
a 5e − j 5ω
⇒ a u[n − 5] ⇔
n

1 − ae − jω

Digital Signal Processing Lecture 4 24