seoul national university 1. discrete-time signals and systems
TRANSCRIPT
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
1. Discrete-time signals and systems
• Discrete - time signals: sequences
– Discrete-time signals are represented as sequences of numbers
– A sequence is a function whose domain is the set of integers
– Define the nth number in the sequence by , a set of numbers can be represented
by
– Note that
discrete-time signal analog signal
where T denote the sampling period.
– A delayed or shifted version of :
where is an integer
[ ]x n
[ ]{ },x x n n= −∞ < < ∞
[ ] ( )ax n x nT=
x
x n
n
x −1x 0
x 1
x 2
[ ]x n
[ ] [ ]0y n x n n= −
0n
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
– An unit sample sequence is defined by
which is also referred to as a discrete-time impulse or simply as an impulse
– An arbitrary sequence can be represented as a sum of weighted and delayed
impulses
Ex)
– The unit step sequence is defined by-3 -2 -1 0 1 2 3
4
5 6
m1
m−3
m4
[ ] [ ] [ ]k
x n x k n kδ∞
=−∞
= −∑
[ ] [ ] [ ] [ ]3 1 43 1 4x n m n m n m mδ δ δ−= + + − + −
[ ] 0; 01; 0
nn
nδ
=⎧= ⎨ ≠⎩
[ ]
[ ] [ ] [ ] [ ]
[ ] [ ] [ ]0
1, 00, 0
1 ...
1k
nu n
n
u n n n n k
n u n u n
δ δ δ
δ
∞
=
≥⎧= ⎨ <⎩
⇒ = + − + = −
⇒ = − −
∑-1 0 1 2 3
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
• Let be an exponential sequence given by
– For complex and
– Complex exponential sequence
where is called the frequency of the complex sinusoid and is called the phase
– Since
complex exponential sequences with frequencies , where k is an integer,
are indistinguishable from one another
– A sinusoidal sequence can be represented by
– Since
we consider in a frequency interval of length
or
[ ]x n
[ ] [ ]nx n A u nα=
[ ]( ) ( )
0
0cos sin
n j nj
no
x n A e e
A n j n
ωφ α
α ω φ ω φ
=
= + + +⎡ ⎤⎣ ⎦
0je ωα α= ,jA A e φ=
( )1α =
[ ] ( ) ( )0 0cos sinx n A n j A nω φ ω φ= + + +
0ω φ
[ ] ( ) 0 02 2 ,oj n j n j nj nx n Ae Ae e Aeω π ω ωπ+= = =
( )0 2 kω π+
[ ] 0cos( ), x n A n nω φ= + ∀
[ ] ( ) ( )0 0cos 2 cos ,x n A k n A nω π φ ω φ= + + = +⎡ ⎤⎣ ⎦
2 ,π0π ω π− < ≤ 00 2 .ω π≤ <
ω
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
– A sequence is called periodic with an integer period of N, if
– In order for a sinusoidal sequence to be
it requires for
– A sinusoidal sequence is not necessarily periodic with a period of and may not
be periodic at all, depending on the value of
Ex
• High and low frequency concept: A continuous time sinusoidal signal
oscillates more rapidly as increases. However, oscillates
more rapidly as increases from 0 toward .
But, as increases from to , the oscillation becomes slower due to symmetry with respect to .
[ ] [ ], x x x n N n= + ∀
( ) ( )0 0 cos cosA n A n Nω φ ω φ+ = + +⎡ ⎤⎣ ⎦
0 2 .N kω π=
0
0
1 2 3 631 2 No integer
kN k
N k N
πω ππ
ω π
= ⇒ = ⋅ =
= ⇒ = ⇒
0
2πω
0ω
[ ] ( )0cosx n A t φ= Ω +
0Ω [ ] ( )0cosx n A nω φ= +
0ω π
0ω 2πππ
15 cos cos
8 8A n A n
πφ π φ+ = +⎛ ⎞ ⎛ ⎞
⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠
Ex :
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
• Discrete-time systems
– Ideal delay system
– Moving averager
– Memoryless system
A system is referred to as memoryless if the output depends only on the input
– Linear system : principle of superposition
Ex: accumulator
x n y n{}T ⋅ [ ] [ ]{ }y n T x n=
[ ] [ ] [ ] [ ]{ } [ ]1
0
1 11 ... 1N
k
y n x n x n x n N x n kN N
−
=
= + − + + − + = −∑
[ ] [ ], d dy n x n n n= − is a positive integer
[ ]y n [ ]x n
[ ] [ ]2 ,y n x n n= ∀Ex :
[ ] [ ]{ } [ ]{ } [ ]{ }1 2 1 2
T ax n bx n aT x n bT x n=+ +
[ ] [ ]n
k
y n x k=−∞
= ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
– Time-invariant system (TIS): A time shift or delay of the input sequence results in a
corresponding shift in the output sequence
– Causality: A system is causal if the output sequence at depends only on the
input sequences for
Ex: non causal
causal
– Stability: A system is stable in the BIBO sense iff every bounded input sequence
produces a bounded output sequence
[ ] [ ]{ } [ ] [ ][ ] [ ]
[ ] [ ]
1 1
2
d d
n
k
y n T x n n y n n y n
y n x n
y n x k=−∞
= − ⇒ − =
= →
= →∑
Ex : TIS
TIS
[ ] [ ][ ] [ ]
[ ] ( ) [ ]1 1
0 0 1
:
o
y n x Mn
y x Mn x Mn n
y n n x m n n y n
=
= = −
− = − ≠ →⎡ ⎤⎣ ⎦
compressor
notTIS
[ ] [ ] [ ][ ] [ ] [ ]
1
1
y n x n x n
y n x n x n
= + − ⇒
= − − ⇒
0, 0n n n≤ ∀on n=
[ ] [ ] , X y
x n B y n B n≤ < ∞ ⇒ ≤ < ∞ ∀
[ ] [ ] 0 01 0
n
k
ny n u k
n n=−∞
<⎧= = ⎨ + ≥⎩∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
• Linear time-invariant systems
– Let be the response of the system to the input , i.e.,
– Time invariance implies
– Discrete-time convolution
– To compute first reflect about the origin to obtain and then shift
the origin to
[ ]kn n [ ]n kδ − [ ] [ ]{ }kh n T n kδ= −
[ ] [ ]{ }
[ ] [ ] [ ] [ ]{ }
[ ] [ ]
k k
kk
y n T x n
T x k n k x k T n k
x k h n
δ δ∞ ∞
=−∞ =−∞
∞
=−∞
=
⎧ ⎫= − = −⎨ ⎬
⎩ ⎭
=
∑ ∑
∑
[ ] [ ] [ ] [ ] [ ] kn h n n k h n k h nδ δ→ ⇒ − ⇒ − =
[ ] [ ] [ ] [ ] [ ] k
y n x k h n k x n h n∞
=−∞
= − Δ ∗∑[ ],h n k− [ ]h k [ ]h k−
;k n=
0 1 -2 -1 0 0 1 2 3 4 5
h k h k− h k6−
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
[ ] [ ] [ ]
[ ] [ ][ ]
1, 0 10, otherwise
0; 0
n
h n u n u n N
n N
x n a u n
n y n
= − −
≤ ≤ −⎧= ⎨⎩
=
< =
Ex
[ ]1
0
10 1; 1
nnk
k
an N y n aa
+
=
−≤ ≤ − = =
−∑
n N− −1b g n 0 k
[ ]1 1
1
1
1; 1
11
n N nnk
k n N
Nn N
a an N y n aa
aaa
− + +
= − +
− +
−> − = =
−
⎛ ⎞−= ⎜ ⎟−⎝ ⎠
∑
n0
y n
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
• Properties of LTI system
– Commutative
– Cascaded connection
Letting we have
[ ] [ ] [ ]( ) [ ] [ ] [ ] [ ]1 2 1 2x n h n h n x n h n x n h n∗ + = ∗ + ∗
[ ]x n [ ]v n [ ]y nh1 h2
[ ] [ ] [ ] [ ]
[ ] [ ] [ ]
[ ] [ ]k
k
x n h n h n x n
y n x k h n k
x n k h k
∞
=−∞
∞
=−∞
∗ = ∗
⇒ = −
= −
∑
∑
[ ] [ ] [ ]1 2 ,h n h n h nΔ ∗
[ ] [ ] [ ][ ] [ ] [ ]
[ ] [ ] [ ][ ] [ ]
1
2
1 2
v n x n h n
y n v n h n
x n h n h n
x n h n
= ∗
= ∗
= ∗ ∗
= ∗
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
– A Linear system is stable iff the impulse response is absolutely summable, i,e.,
Proof
Sufficient condition:
If is bounded, i.e.,
then,
Necessary condition:
Show that if , a bounded input can cause an unbounded output.
Consider a bounded sequence given by
( ) k
S h k∞
=−∞
Δ < ∞∑
[ ] [ ] [ ] [ ] [ ]k k
y n h k x n k h k x n k∞ ∞
=−∞ =−∞
= − ≤ −∑ ∑[ ] Xx n B≤[ ]x n
[ ] [ ]Xk
y n B h k∞
=−∞
≤ ∑
[ ] [ ] k
h k y n∞
=−∞
< ∞ ⇒ <∞∑
S = ∞
[ ][ ][ ] [ ]
[ ][ ] [ ] [ ] [ ]
[ ]
2*, 0
00 , 0
k k
h nh n h kh nx n y x k h k S
h kh n
∞ ∞
=−∞ =−∞
⎧ −≠⎪ −= ⇒ = − = = = ∞⎨
⎪ =⎩
∑ ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
Ex
Note: The property of convolution can be used to analyze the time-invariant system
[ ] [ ]
0
11
1
1
n
n
n
h n a u n
S aa
a S
a S
∞
=
=
⇒ = =−
< ⇒ <∞
≥ ⇒ =∞⇒
∑
unstable
Accumulator Backwarddifference
x n y n x n
[ ] [ ] [ ] [ ]( )[ ] [ ][ ]
1
1
h n u n n n
u n u n
n
δ δ
δ
= ∗ − −
= − −
=
y none sampledelay
Forwarddifference
x n
δ n−1 δ δn n+ −1
Backwarddifference
[ ] [ ] [ ] [ ]( )[ ] [ ]
1 1
1
h n n n n
n n
δ δ δ
δ δ
= − ∗ + −
= − −
⇒
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
• Linear constant coefficients difference equations
– An Nth order linear const. difference eq. is represented by
When the right side term is equal to zero, it is called the homogeneous difference eqn.
Ex: Accumulator
Ex: Moving average
[ ] [ ]0 0
N M
k kk k
a y n k b x n k= =
− = −∑ ∑
[ ] [ ] [ ] [ ] [ ] [ ]
[ ] [ ] [ ]
1
1
1
n n
k k
y n x n x n x k x n y n
y n y n x n
−
=−∞ =−∞
= = + = + −
⇒ − − =
∑ ∑
D
[ ]x n [ ]y n+
[ ]1y n −
[ ] [ ] [ ]( )
[ ] [ ]
[ ] [ ] [ ]( ) [ ]
1
0
1
1
1
M
k
h n u n u n MM
y n x n kM
h n n n M u nM
δ δ
−
=
= − −
⇒ = −
= − − ∗
∑
M-sampledelay
x n y n∑1
M _
[ ] [ ] [ ] [ ]( )11y n y n x n x n MM
− − = − −
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
Ex[ ] [ ] [ ]
[ ] [ ] [ ]
[ ][ ] [ ]
[ ] [ ] [ ]
[ ] [ ] [ ][ ]
[ ]
2
1 1
1 1
1 1 2
1
1
y -1 .0;
0
1 0 0:
, 01;
2 1 1
3:
, 1
n n n n
n
y n ay n x n
x n K n cn
y ac K
y ay a c aK
y n a c a K n y n a c Ka u nn
y a y x a c
y a a c a c
y n a c n
δ
+ +
− −
− − −
+
= − +
= = ←
≥
⇒ = +
= + = +
= ⋅ + ≥ ⇒ = ⋅ +
< −
⎡ ⎤− = − − − = ⋅⎣ ⎦− = ⋅ =
= ⋅ ≤ −
Consider and auxiliary condFor
For
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
• Frequency-domain representation of DTS
Let
eigen-value eigen-function of the system
[ ]
[ ] [ ] [ ]
[ ]
j n
j n k
k
j n j k
k
x n e
y n h k e
e h k e
ω
ω
ω ω
∞−
=−∞
∞−
=−∞
=
⇒ =
=
∑
∑
( ) [ ]
( ) ( )( ) ( )
j
H
j j k
k
j jR
j ej
H e h k e
H e jH e
H e eω
ω ω
ω ω
θω
∞−
=−∞
Δ
= +
=
∑
[ ] ( )j j ny n H e eω ω=
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
Ex: Ideal delay
Since for an ideal delay system,
– For
the output of the system is
← principle of superposition
Ex:
[ ] [ ][ ]
[ ] [ ] ( )0 0 0
0
,j n
j n n j n j nj n j
y n x n n
x n e
y n e e e H e e
ω
ω ω ωω ω− − −
= −
=
= = ⇒ =
For
[ ] [ ]0h n n nδ= −
( ) [ ] ( )
( ){ }
00
0
1j nj j n j
n
j
H e n n e e H e
Arg H e n
ωω ω ω
ω
δ
ω
∞−−
=−∞
= − = ⇒ =
= −
∑
[ ] ,kj nk
k
x n e ωα=∑
[ ] ( )k kj j nk
ky n H e eω ωα=∑
[ ] ( ) ( )
[ ] ( ) ( )[ ] ( ) ( ) ( )
[ ] ( ) ( ) ( )
0 0
0 0 0
0
0
cos2
2
cos , =arg
j n j nj j
j j j nj j n j
j j j j
j j
Ax n A n e e e e
Ay n H e e e H e e e
h n H e H e H e e
y n A H e n H e
ω ωφ φ
ω ω ωφ ω φ
ω ω ω θ
ω ω
ω φ
ω φ θ θ
−−
− −−
− ∗ − −
= + = +
⎡ ⎤⇒ = +⎣ ⎦
= =
⎡ ⎤⇒ = + + ⎣ ⎦
If is real,
where
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
Ex: Delay
– Discrete-time LTI → periodicity of
Ex: Ideal frequency selective filters (LPF)
( )[ ] ( )( )
00 0
0 0
1;
cos
jH e n
y n A n n
ω θ ω
ω φ
= = −
⇒ = − +
2π
( )( ) [ ] ( )
[ ] ( )
( )
2 2
2
j k j k n
k
jj n
k
j
H e h n e
h n e H e
H e
ω π ω π
ωω
ω π
∞+ − +
=−∞
∞−
=−∞
⇒ =
⎡ ⎤= = ⎣ ⎦
⇒
∑
∑is periodic with a period of
− −π ω 0 −π ω 0 π 2 0π ωr 2π 2 0π ω+ω
H e jωc h
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
Ex: [ ] 1 21 2
11
0 otherwise
M n MM Mh n
⎧ − ≤ ≤⎪ + += ⎨⎪⎩
−π−π
−π
πθ
( )( )
( ) ( ) ( )
( )
( )
2
1
21
2 1 1 2 1 2
2 1
1 2
1
1 2
1 / 2 1 / 2 1 / 2
1 2
1 2
1 2 _
11
11 1
1 11
sin1 21 sin
2
Mj j n
n M
j Mj M
j
j M M j M M j M M
j
j M M
H e eM M
e eM M e
e e eM M e
M M
eM M
ω ω
ωω
ω
ω ω ω
ω
ω
ω
ω
−
=−
− ++
−
− + − + + − + +
−
− −
⇒ =+ +
−=
+ + −
−=
+ + −
+ +
=+
∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
• Representation of sequences by Fourier transform
– A stable sequence can be represented by a Fourier integral of the form
where
Since is periodic with period , it is of the form of Fourier series for the
continuous-variable periodic function .
– Note that Eq (A) is the inverse of Eq. (B)
[ ] ( )2
1 - - - ( )2
j j nx n X e e dπ ω ω
πω
π= ∫ A
( ) [ ]
( ) ( )
( ) ( )1
- - - ( )
j
X
j j n
n
j ej
j jR
X e x n e
X e e
X e jX e
ω
ω ω
θω
ω ω
∞−
=−∞
=
= ←
= + ←
∑ B
polar cord. form
rectangular cord. form
( )jX e ω 2π
( )jX e ω
[ ] ( ) [ ]
[ ] ( ) [ ]
( ) ( )( ) [ ]
2
2
1 12 2
12
sin1 2
j j n j n j n
n
j n m
n
j n m
x n X e e x m e e d
x m e d x n
n me d n m
n m
π πω ω ω ω
π π
π ω
π
π ω
π
ωπ π
ωπ
πω δ
π π
∞−
− −=−∞
∞−
=−∞
−
⎛ ⎞= = ⎜ ⎟
⎝ ⎠
= =
⎛ ⎞−= = −⎜ ⎟
−⎝ ⎠
∑∫ ∫
∑ ∫
∫∵
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
– If is absolutely summable (i.e., stable sequence), exists.
– Furthermore, can be showen to converge uniformly to a continuous function of
⇒ Any stable sequence (or system) have a finite and continuous frequency response:
(Sufficient condition for existence of )
– A finite-length sequence is absolutely summable
Ex:
– If a sequence is not absolutely summable, but square summable ; i.e.,
it can be represented by a Fourier transform with mean-square convergence, i.e.,
( ) [ ]
[ ] [ ]
j j n
n
j
n n
X e x n e
x n e x n
ω ω
ω
∞−
=−∞
∞ ∞−
=−∞ =−∞
=
≤ ≤ < ∞
∑
∑ ∑
[ ]x n ( )jX e ω
[ ]x n .ω
( )jH e ω
[ ] [ ]
( ) [ ]
1 , 11
n
j n j n n j n
n n
jj
x n a u n
X e a u n e a e
ae aae
ω ω ω
ωω
∞ ∞− −
=−∞ =−∞
−−
=
= =
= = <−
∑ ∑
if
[ ] 2
n
x n∞
=−∞
< ∞∑
( ) [ ]2
lim 0M
j j n
M n MX e x n e d
π ω ω
πω−
−→∞=−
− =∑∫
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
Ex: Ideal LPF
non causal and not absolutely summable
is discontinuous at
But is mean-square summable
Ex: : neither absolutely nor square summable
Since
When
is not finite for all
( )
( )
1
0
1 sin , 2
c
c
cjLP
c
j n cLP
H e
nh n e d nn
ω
ω ω
ω
ω ω
ω ω π
πωωπ π−
⎧ <⎪= ⎨< <⎪⎩
⇒ = = ∀
⇒
∫
( )jLpH e ω
cω ω=
[ ] 1, x n n= ∀
( ) ( )2 2j
k
X e kω πδ ω π∞
=−∞
⇒ = +∑( ) ( )
[ ] ( )0
0
0
2 2
1 22
j
k
j n
j n
X e k
x n e d
e
ω
π ω
π
ω
πδ ω ω π
πδ ω ω ωπ
∞
=−∞
−
= − +
⇒ = −
= ⇒
∑
∫[ ]0 0, 1x nω = =
( )jX e ωϖ
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
– Theory of generalized functions
– Conjugate symmetric sequence:
– Conjugate anti-symmetric sequence:
where
( ) ( )
[ ] ( ) ( )
002 2
2 2m
j n
n k
j n jm m m
m k m
e k
x n a e X e a k
ω ω
ω ω
πδ ω ω π
π δ ω ω π
∞ ∞−
=−∞ =−∞
∞
=−∞
= − +
⇒ = ⇒ = − +
∑ ∑
∑ ∑∑
[ ] [ ]*e ex n x n= −
[ ] [ ]*o ox n x n= − −
[ ] [ ] [ ]
[ ] [ ] [ ]( )
[ ] [ ] [ ]( )
1212
e o
e
o
x n x n x n
x n x n x n
x n x n x n
∗
∗
⇒ = +
= + −
= − −
[ ] [ ][ ] [ ]
:
: e e
o o
x n x n
x n x n
= −
= − −
even sequence
odd sequence
( ) ( ) ( ) ( ) ( )j j j j je o R IX e X e X e X e jX eω ω ω ω ω⇒ = + = +
( ) ( ) ( )( )( ) ( ) ( )( )
1212
j j je
j j jo
X e X e X e
X e X e X e
ω ω ω
ω ω ω
∗
∗ −
= +
= −
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
– Note that
Ex:
( ) ( ) ( ) ( ) j j j je oX e X e X e X eω ω ω ω∗ − ∗ −= = −and
[ ]{ } ( ) ( )j jx n X e X eω ω∗↔ =Re[ ]{ } ( )[ ]{ } ( )[ ] ( )[ ] ( )
Re
Im
je
jo
je R
jo I
x n X e
j x n X e
x n X e
x n jX e
ω
ω
ω
ω
↔
↔
↔
↔
( )
( ) ( )
( ) ( ) ( )( ) ( )( ) ( )
2
1
1 11
11
1 1 cos1 2 cos
sintan1 cos
jj
j jj
j jR R
j jI I
j jX X
X e aae
X e X eae
X e a X ea a
X e X e
ae ea
ωω
ω ωω
ω ω
ω ω
ω ω
ωω
ωθ θω
−
∗−
−
− −
= <−
= =−
= − =+ −
= −
−= = −
−
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
• Fourier transform theorems
– Linearity:
– Delay:
– Time reversal:
If is real,
– Differentiation:
– Parseval’s theorem:
[ ]{ } ( )[ ] ( ){ }
[ ] ( )
1
j
j
F j
F x n X e
x n F x e
X n X e
ω
ω
ω
−
=
=
⇒ ←⎯→
[ ] [ ] ( ) ( )1 2 1 2F j jax n bx n aX e bX eω ω+ ←⎯→ +
[ ] ( )( )
0
0 0
0
( )
j nF j
j n j nF
x n n X e e
e X e
ωω
ω ω ω
−
−
− ←⎯→
←⎯→
[ ] ( )F jx n X e ω−− ←⎯→
[ ] ( )F jx n X e ω∗ +− ←⎯→[ ]x n
[ ] ( )F jdnx n j X ed
ω
ω←⎯→
[ ] ( )2
21 2
j
n
E x n X e dπ ω
πω
π
∞
−=−∞
Δ =∑ ∫
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
– Convolution:
periodic convolution
⇒ Multiplication of two sequences is equivalent to periodic convolution of
corresponding Fourier Transforms.
Ex:
[ ] ( ) [ ] ( )[ ] [ ] [ ] [ ] [ ] ( ) ( ) ( )
, F Fj j
F j j j
k
x n X e h n H e
y n x k h n k x n h n Y e X e H e
ω ω
ω ω ω∞
=−∞
←⎯→ ←⎯→
⇒ = − = ∗ ←⎯→ =∑
[ ] [ ] [ ] ( ) ( ) ( )( )12
jF j jy n x n h n Y e X e H e dπ ω ξω ξ
πξ
π−
−= ←⎯→ = ∫
[ ] [ ] [ ] [ ]
[ ] [ ] [ ] [ ] ( )
1 11 12 4
1 111 1 14 41 1 1 1 14 2 1 1 12 2 2
j j
j
j j j
y n y n x n x n
e en n h n h n H e
e e e
ω ω
ω
ω ω ωδ δ
− −
− − −
− − = − −
−⇒ − − = − − ⇒ = = −
− − −
[ ] [ ] [ ] [ ]11 1 1 1; 1 1
1 2 4 2
n nFn
ja u n h n u n u nae ω α
−
−⎛ ⎞ ⎛ ⎞←⎯→ < ⇒ = − −⎜ ⎟ ⎜ ⎟− ⎝ ⎠ ⎝ ⎠
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
• Discrete-time Random Signals
– Consider an LTI system with impulse response
When is a WSS discrete-time random process,
– If stationary,
[ ]h n
[ ] [ ] [ ]k
y n h n k x k∞
=−∞
= −∑[ ]x n
[ ] [ ]{ } [ ] [ ]{ } , X Yn E X n n E y nμ μΔ Δ
[ ]X Xnμ μ=
[ ] [ ] [ ]
[ ] [ ]{ } [ ] [ ]{ }
[ ]
( )0
Yk
k k
Xk
jX
n E h n k x k
h n k E x k h k E x n k
h k
H e
μ
μ
μ
∞
=−∞
∞ ∞
=−∞ =−∞
∞
=−∞
⎧ ⎫= −⎨ ⎬
⎩ ⎭
= − = −
=
=
∑
∑ ∑
∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
1. Discrete time signals and systems
– Autocorrelation of [ ]y n[ ] [ ] [ ]{ }
[ ] [ ] [ ] [ ]
[ ] [ ] [ ] [ ]{ }
[ ] [ ] [ ] [ ]
,yy
k l
k l
xx yyk l
n n m E y n y n m
E h k h l x n k x n l m
h k h l E x n k x m m l
h k h l m k l m
φ
φ φ
∞ ∞
=−∞ =−∞
+ = +
⎧ ⎫= − − +⎨ ⎬
⎩ ⎭= − + −
= + − =
∑ ∑
∑∑
∑∑
[ ] [ ] [ ] [ ]
[ ] [ ] [ ] [ ]
[ ] [ ] [ ]
.
:
yy xxi k
hk
yy h xxi
l k i
m m i h k h i k
h n i h k h i k
m i m i
φ φ
φ
φ φ φ
∞ ∞
=−∞ =−∞
∞
=−∞
∞
=−∞
− Δ
= − +
Δ +
⇒ = −
∑ ∑
∑
∑
Let Then,
Autocorrelation sequence of
( ) ( ) ( ) ( ) ( )( ) ( ) ( ) ( )[ ]{ } ( ) ( ) ( ) ( )
2
2
22
1 1: 0 02 2
j j j j jyy h xx xx
j j j jh
j j jyy yy xx
e e e H e e
e H e H e H e
E y n e d H e e d
ω ω ω ω ω
ω ω ω ω
π πω ω ω
π πφ ω ω
π π
∗
− −
⇒Φ =Φ Φ = Φ
Φ = =
= = Φ = Φ ≥∫ ∫
where
Note
( )jH e ω
ω a ω b
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
2. Periodic Sampling of Continuous-time Signals• Impulse sampling: ideal continuous-to–discrete (C/D) converter
where is the sampling period and is called the sampling frequency
( )cx t ( )sx t
( ) ( )s t t nTδ= −∑
[ ] ( ),cx n x nT n= −∞ < < ∞
( ) ( ) ( )
( ) ( )
s cn
cn
x t x t t nT
x nT t nT
δ
δ
∞
=−∞∞
=−∞
⇐= −
= −
∑
∑
shifting property of the impulse function
1
s
Tf
⎛ ⎞=⎜ ⎟⎝ ⎠
sf
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
– Fourier transform of
( ) ( ) ( )
( )1
1
s cn
cn
cn
X f X f F t nT
nX f fT T
nX fT T
δ
δ
∞
=−∞
∞
=−∞
∞
=∞
⎧ ⎫= ∗ −⎨ ⎬
⎩ ⎭⎛ ⎞= ∗ −⎜ ⎟⎝ ⎠
⎛ ⎞= − ⇐⎜ ⎟⎝ ⎠
∑
∑
∑ periodic function
( )sx t
( ) ( ){ } ( ) ( )
( )
2
2
j fts s c
n
j fnTc
n
X f F x t x nT t nT e dt
x nT e
π
π
δ∞
−
=∞
∞−
=∞
= = −
=
∑∫
∑
− f s W− 0 W sf W− f s
( )cX f
W− W
( )sX f
2
s sf W W f W− ≤ ⇔ ≤⇒ Aliasing
= 2 / 2s sT fπ πΩ =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• Natural Sampling
( )cx t ( )sx t
( )p t0 τ T T +τ
2 1( ) ; sinsj f ntn n
n
np t p e p cT T
π τ∞
=−∞
= =∑
( )sX f
sf− sf0
( ) ( ) ( )
( ) 2
( )
s
s c
j f nts c n
n
n cn
x t x t p t
X f F x t p e
np X fT
π∞
=−∞
∞
=−∞
=
⎧ ⎫⇒ = ⎨ ⎬
⎩ ⎭⎛ ⎞= −⎜ ⎟⎝ ⎠
∑
∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• Sample and hold (S/H)
( )cx t ( )sx t
0 τ
( )s t( )p t
( ) ( ) ( ) ( )
( ) ( ) ( )
( )
1
1
s cn
s cn
n
x t x t t nT p t
nX f X f f p fT T
np f X fT T
δ
δ
∞
=−∞
∞
=−∞
∞
=−∞
⎡ ⎤= − ∗⎢ ⎥⎣ ⎦
⎡ ⎤⎛ ⎞⇒ = ∗ −⎜ ⎟⎢ ⎥⎝ ⎠⎣ ⎦⎛ ⎞= −⎜ ⎟⎝ ⎠
∑
∑
∑ ( )sX f
sf− sf0
1( ) sincP f fT T
τ⎛ ⎞= ⎜ ⎟⎝ ⎠
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• Nyquist Sampling Theorem
– If is strictly band limited, i.e.,
and the sampling frequency is chosen such that
then
Since
if the sample values are specified for all time, is uniquely determined
by using the Fourier series.
⇒ is uniquely determined by
( )cx t ( ) 0 X f f W= >for1 2 ,sf WT
= =
( ) ( ) 2j fnTs c
n
X f x nT e π∞
−
=−∞
= ∑
( ) ( )1 , ,2c sX f X f f WW
= ≤
( )cx nT ( )cX f
( )cx t ( ), 0, 1, 2, .cx nT n = ± ± ⋅ ⋅ ⋅
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• Reconstruction of a band limited signal from its samples: Interpolation formula
⇒ A band-limited signal of finite energy can be completely recovered by its samples
taken at a rate of 2W/sec
– Ideal reconstruction filter
( ) ( )
[ ]
[ ] ( )
[ ] ( )( )
[ ]
2
2
2
12
12sin /
/
sin /( ); ( )/
j ftc c
fjW j ftWW
n
W j f t nT
Wn
n
n
x t X f e df
x n e e dfW
x n e dfW
t nT Tx n
t nT T
t Tx n h t nT h tt T
π
ππ
π
ππ
ππ
∞
−∞
∞ −
−=−∞
∞−
−=−∞
∞
=−∞
∞
=−∞
=
=
=
−=
−
= − =
∫
∑∫
∑ ∫
∑
∑
( )H f
f−
12T
12T
sin /( )/t Th t
t Tπ
π=
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• Poisson sum formula
A periodic signal with period T is represented by the Fourier series,( )sx t
( )
( )
( ) ( )
22
2
2 22
2
2
1
1
1 1
1
;s
s
s s
Tj f nt
Tn s
Tj f nt j f nt
Tnn
j f nts n s
n
n
x x t e dtT
t nT e x e dtT T
t x e fT
T
x
π
π π
π
δ δ τ
−
−
∞ ∞−
−=−∞ =−∞
∞
=−∞
=
⎛ ⎞− = = =⎜
⎝ ⎠
=
⎟
=
∫
∫
∑
∑ ∑ ∵
where
Ex :
( ) ( )x t t nTδ= −∑ ( )y th tb g
( ) ( ) ( ) ( )
( ) ( )
( ) ( ) ( )
( ) ( )
2 2
2
21
1 1
1
s s
s
s
n n
j f nt
n
j f n t j f nt
j f nts
n n
sn n
y t h t t nT h t nT
y t h t eT
h e d H
h t nT H nf
nT
e
e
T
fT
π
π τ π
π
δ
τ τ
∞ ∞
=−∞ =−∞
∞
=−∞
∞ ∞−
=−
∞ ∞
=−∞ =−
∞ =−∞
∞
= ∗ − = −
⎡ ⎤= ∗ ⎢ ⎥⎣ ⎦
−
= =
=⇒
∑ ∑
∑
∑
∑
∑∫
∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• Discrete-time processing of continuous-time signals
low passreconstruction filter
– The freq-domain representation is easier than the time-domain representation
x tcb g x n y n y tr b g
T T
C D/ h n D C/
discrete - timesystem
[ ] ( ) ( )
( ) [ ]( )
( )
( ) ( ) ( )
( )
1 2 ;
sin;
,
0 , otherwise
jc c
k
rn
j Tr r
j T
kx n x nT X e XT T T
t nTTy t y n
t nTT
Y H Y e
TY eT
ω ω π
π
π
π
∞
=−∞
∞
=−∞
Ω
Ω
⎛ ⎞= ⇒ = −⎜ ⎟⎝ ⎠
−
=−
⇒ Ω = Ω
⎧ Ω <⎪= ⎨⎪⎩
∑
∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• LTI discrete-time systems
– If is band limited and the sampling rate is larger than the Nyquist rate,
the effective frequency response is
( ) ( ) ( )j j jY e H e X eω ω ω=
( ) ( ) ( )r eff cY H XΩ = Ω Ω
( ) ( ),0 , otherwise
j T
eff
H eH T
πΩ⎧ Ω <⎪Ω = ⎨⎪⎩
( )cX Ω
( ) ( ) ( ) ( )( ) ( )
( ) ( )
;
1 2
,
0 , otherwise
j T j Tr r
j Tr c
k
j Tc
Y j H H e X e T
kH H e XT T
H e XT
ω
π
π
Ω Ω
∞Ω
=−∞
Ω
⇒ Ω = Ω =Ω
⎛ ⎞= Ω Ω −⎜ ⎟⎝ ⎠
⎧ Ω Ω <⎪= ⎨⎪⎩
∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• Example of signal reconstruction
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
Ex: Ideal LPF
Ex: Ideal band-limited differentiator
( ) ( ) ( )
( )
( )
[ ] 2
,
0 , otherwise
,
cos , 0cos sin
0 , 0
c c c
eff
j
dy t x t H jdt
jH T
jH eT
n nn n nh n nTn T n
ω
π
ω ω π
ππ π π
π
= ⇒ Ω = Ω
⎧ Ω Ω <⎪⇒ Ω = ⎨⎪⎩
= <
⎧ ≠− ⎪⇒ = = ⎨⎪ =⎩
( )1, 0, otherwise
ceff
TH
ω⎧ Ω <Ω = ⎨
⎩
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
– Impulse invariance
⇒ When , the discrete-time system is said to be an impulse-invariance
version of the continuous-time system.
Ex:
The resulting discrete-time frequency response is aliased
( ) ( ); jcH e H H
Tω ω
ω π= = Ω <⎛ ⎞⎜ ⎟⎝ ⎠
[ ] ( )ch n Th nT=
( ) ( ) ( )
( ) ,
( )1 2
1
j j Ts T
jc
k
jc
X X ekH e HT T T
H eT
e
HT
Xω
ω
ω
ω
ω π
ω ω π
∞
=−
Ω
Ω∞
=
⎛ ⎞= − ⇐⎜ ⎟⎝ ⎠⎛ ⎞⇒ = <⎜ ⎟⎝ ⎠
Ω = =∑∵
[ ] ( ) [ ]
( )
0
0
0 0
0
1
( ) ( ) ( )
( ) 1 1
( )
s tc c
s nTc
jcs T s T j
c
Ah t Ae u t H ss s
h n Th nT Ae u n
AT ATH z H e He z e e Th t
ωω
ω− −
= ⇒ =−
= =
⎛ ⎞= = ≠ ⎜ ⎟− − ⎝ ⎠∵
or
is not strictly band - limited
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• Continuous-time processing of discrete-time signals
( ) [ ]( )
( ) ( )
( ) [ ]( )
( ) ( ) ( ) ( )
sin; ( ) ,
sin; ;
j Tc c
n
c c c cn
t nTTx t x n X T X e
t nT TTt nTTy t y n Y H X
t nT TT
ππ
π
ππ
π
∞Ω
=−∞
∞
=−∞
−
= Ω = Ω ≤−
−
= Ω = Ω Ω Ω <−
∑
∑
x tc b gx n y n( )cy t
T T
C D/( )ch tD C/
( ) ( ) ( ); /j Tc effH e H H TπΩ = Ω = Ω Ω ≤⇒
( )( ) ( ) ( ) ( )
1 ,jc
j jc c
Y e YT T
H X H e X e
ω
ω ω
ω ω π⎛ ⎞= <⎜ ⎟⎝ ⎠
= Ω Ω =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
Ex: Non-integer delay ( ) ,j jH e eω ω ω π− Δ= <
[ ] [ ] [ ] [ ]( )
( ) ( )[ ] ( )
[ ]( )
( )
[ ] ( )( ) [ ]
2
,
,
sin
sin s
j f Tc
c c
c
k
t nT
k
y n x n h n n
H f e
y t x t T
y n x nT T
t T kTTx k
t T kTT
n kx k h n
n k
π
δ
π
π
ππ
− Δ
∞
=−∞
=
∞
=−∞
Δ = − Δ ⇒ = − Δ
Δ =
= − Δ
⇒ = − Δ
− Δ −
=− Δ −
− − Δ= ⇒ =
− − Δ
∑
∑
If is an integer
If is not an integer
( )( )
in
1 . ., ;2
nn
e g
ππ
− Δ− Δ
Δ =
( )2c cTy t x t⎛ ⎞= −⎜ ⎟
⎝ ⎠
0 T 2T 3TT−
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
– Ex: Moving average with non-integer delay
When M=5,
for ( )( )
[ ] cos 0.25
[ ] 0.308cos 0.25 2.5
x n n
y n n
π
π
=
⇒ = −⎡ ⎤⎣ ⎦
[ ]1 , 0
10, otherwise
n Mh n M
⎧ ≤ ≤⎪= +⎨⎪⎩
( ) 2
( 1)sin1 2 , 1 sin
2
Mjj
M
H e eM
ωω
ω
ω πω−
+
⇒ = <+
[ ] 2My n w n⎡ ⎤= −⎢ ⎥⎣ ⎦
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• Changing the sampling rate
– Decimation by M (sampling rate compressor)
Let Then,
[ ] [ ] [ ] ( )
( )
( )
1 2
c d c
jc
k
x n x nT x n x nM x nMT
kX e XT T T
ω ω π∞
=−∞
=
⎛
⇒ Δ =
⎞= −⎜ ⎟⎝ ⎠
∑
( )
1
0
'
21
'
0
1 1 2 2
1
1 2'1 2
M
ci k
iM jM
i
jd c
c
i
kX e XT T T
X i kMMT
kXM T M
MT M
M
T
T T
X e
ω
ω π
ω
ω π
π π
ω π
∞
=−∞
∞
=−
− ∞
= =−∞
∞
−−
=
⎡ − ⎤⎛ ⎞= −⎜ ⎟⎢ ⎥⎝ ⎠⎣ ⎦⎛ ⎞
=
⎛ ⎞= −⎜ ⎟⎝ ⎠⎛ ⎞= − ⇐ = +⎜ ⎟
⎜ ⎟⎝ ⎠
⎝ ⎠
∑ ∑
∑
∑
∑
.T MT′ =
x nM↓
[ ] [ ]dx n x nM=
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
– Decimation by M (cont.)
To avoid aliasing, a LPF is
required before down-sampling
Example: When M=3:
[ ]x nx nT
LPF
[ ]dh n'T MT=
[ ]dx nM↓
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
– Interpolation by L (sampling rate expander)
[ ] ( );
i cTx n x nT TL
nxL
′ ′= =
⎡ ⎤= ⎢ ⎥⎣ ⎦
[ ]ex nx nT ' /T T L=L↑ LPF
[ ]ix n
[ ]ih n
[ ]
[ ] [ ]
( ) [ ] [ ]
[ ] ( )
, 0, , 2 ,...
0 , otherwise
e
k
j j ne
k k
j Lk j L
k
nx n L Lx n L
x k n kL
X e x k n kL e
x k e X e
ω ω
ω ω
δ
δ
∞
=−∞
∞ ∞−
=−∞ =−∞
∞−
=−∞
⎧ ⎡ ⎤ = ± ±⎪ ⎢ ⎥= ⎣ ⎦⎨⎪⎩
= −
⎛ ⎞⇒ = −⎜ ⎟
⎝ ⎠
= =
∑
∑ ∑
∑
W− WΩ
/ Lπ0
2πω = ′ΩT
π−π2π− 2πTω =Ω
( )jX e ω
( )cX Ω
( )jeX e ω
/ Lπ−3 / Lπ− 3 / Lπω = ′ΩT
/ Lπ/ Lπ−2π−
( )jiX e ω
L
π π
H eijωc h
L
π−
L
π− 2π2π−
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
– D/C conversion
In practice, we use an approximate LPF.
– Linear interpolation
[ ]
[ ] [ ] [ ] '
sin 1 00 , 2 ...
( )
i
i i c ck
nnLh n n n L L
Ln nTx n x k h n k x x nT xL L
π
π=⎧
= ⇒ ⎨ = ± ±⎩
⎡ ⎤ ⎛ ⎞= − = = = ⎜ ⎟⎢ ⎥⎣ ⎦ ⎝ ⎠∑
[ ]
[ ] [ ] [ ]
[ ] [ ]
( )
2
1 ,
0, otherwise
sin1 2sin
2
ek
m
j
nn Lh n L
x n x k h n k
x m h n mL
L
H eL
ω
ω
ω
⎧− <⎪= ⎨
⎪⎩= −
= −
⎡ ⎤⎢ ⎥
= ⎢ ⎥⎢ ⎥⎣ ⎦
∑
∑
π5
25π 4
5π
H eijωc h
H e Ljωc h: = 5
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
– Changing the sampling rate by a non-integer ratio
[ ]ex nx n [ ]ix n
TL
LPFLPFL↑[ ]dx n
M↓MTL
TLT
L
T
[ ]ex nx n [ ]dx n[ ]ix nM↓LPFL↑
LRM
=
min ,
G L
c L Mπ πω
=
⎛ ⎞= ⎜ ⎟⎝ ⎠
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
– Changing the sampling rate by a non-integer ratio (cont.)LRM
=
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• Multirate signal processing
– Interchange the filtering and the down sampling process
[ ]x n [ ]ax n [ ]y nM↓ ( )H z
[ ]x n( )MH z
[ ]bx n [ ]y nM↓
( ) ( ) ( )j j M jbX e H e X eω ω ω=
( )
( )( )
( )
( ) ( )
21
0
212
0
21
0
1
1
1
iM jj Mb
i
iM j j iM
i
iM jj M
i
j ja
Y e X eM
X e H eM
H e X eM
H e X e
ω πω
ω πω π
ω πω
ω ω
−−
=
−−−
=
−−
=
⎛ ⎞= ⎜ ⎟
⎝ ⎠⎛ ⎞
= ⎜ ⎟⎝ ⎠
⎛ ⎞= ⎜ ⎟
⎝ ⎠
=
∑
∑
∑
( )21
0
1 ( )iM jj M
ai
X e X eM
ω πω
−−
=
= ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• Multirate signal processing
– Interchange the filtering and the down sampling process
– Interchange the filtering and the interpolation process
( ) ( ) ( )j j M jbX e H e X eω ω ω=
( ) ( )( )
( ) ( ) ( )
2 21 12
0 0
21
0
1 1
1
i iM Mj j j ij M Mb
i i
iM jj j jMa
i
Y e X e X e H eM M
H e X e H e X eM
ω π ω πω πω
ω πω ω ω
− −− −−
= =
−−
=
⎛ ⎞ ⎛ ⎞= =⎜ ⎟ ⎜ ⎟
⎝ ⎠ ⎝ ⎠⎛ ⎞
= =⎜ ⎟⎝ ⎠
∑ ∑
∑
( )21
0
1 ( )iM jj M
ai
X e X eM
ω πω
−−
=
= ∑
[ ]x n ( )H z [ ]ax n [ ]y nL↑
[ ]x nL↑
[ ]bx n [ ]y n( )LH z
( ) ( ) ( ) ( )( ) ( ) ( ) ( ) ( ) ( ) ( )
j j L j L j La
j j L j j L j j L j Lb b
Y e X e X e H e
X e X e Y e H e X e X e H e
ω ω ω ω
ω ω ω ω ω ω ω
= =
= ⇒ = =
[ ]x nM↓
[ ]ax n [ ]y n( )H z
[ ]x n ( )MH z [ ]bx n [ ]y nM↓
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
– Poly phase implementation of interpolation filters
Note that only every Lth sample of is nonzero
Consider decomposition of an impulse response by
By successively delaying these subsequences, we can reconstruct by
[ ]x nL↑
[ ]u n [ ]y n( )H z
[ ]u n
z
...
[ ]h n [ ]0e n[ ]h n [ ]0h n
[ ]1h n +
[ ]2h n +
[ ]1h n M+ −
[ ]1e n
[ ]2e n
[ ]1Me n−
L↓ L↑
z
z
L↑
L↑
L↓
L↓
1z−
2z−
+
L↓ L↑ ( )1Mz− −
...
...
[ ] [ ] [ ]k ke n h nL k h nL= + =
[ ] [ ], , 0, 1, 2,0 , otherwisek
h n k n mL mh n
+ = = ± ± ⋅ ⋅ ⋅⎧= ⎨⎩
[ ] [ ]1
0
L
kk
h n h n k−
=
= −∑
[ ]h n
[ ]h n
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
z
[ ]h n [ ]0e n
[ ]1h n +
[ ]1h n M+ −
[ ]1e n
[ ]1Me n−
L↓ L↑
z
L↑L↓
L↓ L↑
...
...
1z−
1z−
+
+
+
:
1z−
[ ]x n
[ ]y n
( )0LE z
1z−...
( )1L
ME z−
+( )1LE z
:
( )0E z +[ ]x n L↑
+L↑
L↑
( )1E z
( )1LE z−
1z −
1z −
[ ]y n
⇒
⇒
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
2. Periodic sampling of continuous-time signals
• Digital processing of analog signals
– Quantization error:
– Assuming that the error sequence is uncorrelated with and it is uniformly
distributed over
– For a (K+1)- bit quantizer with full-scale value
SQNR is increases approximately 6dB for each bit added to the quantizer.
Anti-aliasing
filter
( )cx t ( )xa tS&H
Compensatedreconstructure
filterD/ADiscrete-time
systemA/Dy n⎡ ⎤⎣ ⎦x n⎡ ⎤
⎣ ⎦x n⎡ ⎤⎣ ⎦ ( )y t
[ ]x nQ
[ ]x n
[ ]q n
⇔[ ]x n [ ]x n
↑+
[ ] [ ] [ ]ˆ2 2
q n x n x nΔ Δ− < = − ≤
[ ]q n [ ]x n
, ,2 2Δ Δ⎡ ⎤−⎢ ⎥⎣ ⎦ 2
2 22
2
1 12q q dqσ
Δ
Δ−
Δ= ⋅ =
Δ∫
,mX2 2
2 212
km
qXσ
−
=
2
210 log 6.02 10.8 20logX m
q X
XKσσ σ
= = + −SQNR
⇒
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
Transform analysis of L.T.I. systems• Frequency response of LTI systems
– Ideal frequency-selective filters
x n y nh n
[ ] [ ] [ ] ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( )( )j j
k
j ArgH e ArgX ej j j j j
y n x k h n k Y z H z X z
Y e H e X e H e X e eω ω
ω ω ω ω ω +
= − ⇔ =
⇒ = =
∑
[ ] j nx n e ω=
( )
( ) ( )[ ] [ ] [ ]
[ ]
c
1,
0, otherwise
1 :
sin
cjlp
j jhp lp
hp lp
c
H e
H e H e
h n n h n
nnn
ω
ω ω
ω ω
ω ω π
δ
ωδπ
⎧ ≤⎪= ⎨< ≤⎪⎩
⇒ = −
⇒ = −
= − ⇐
ideal high - pass filter
not computationally realizable
↑complex gain (or eigenvalue)
eigenfunction
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– Ideal delay
– Low-pass filter with linear phase (or delay)
– Group delay: A measure of the nonlinearity of the phase
Example
⇒ The time delay of the envelope of narrowband signal centered at is
given by the negative of the slope of the phase at
( )
[ ] ( )( )
c
,
0 ,
sin,
dj ncj
p
c dhp d
d
eH e
n nh n n
n n
ωω ω ω
ω ω π
ωπ
−⎧ ≤⎪= ⎨≤ ≤⎪⎩
−⇒ = ←
−Regardless of it is still noncausal
( ) ( )( )Arg jdI H ed
ωωω
= −
( )( ) 0Arg jdH e nω φ ω≈ − −
[ ] [ ][ ] [ ] ( )( )
0
0 0
cos ,
cosd d
x n s n n
y n s n n n n
ω
ω φ
=
≈ − − −
For an input
[ ]s n 0ω
0 .ω
[ ] [ ]( ) 1 ,d d
id d
j n j njid
h n n n
H e e eω ωω
δ
ω π− −
= −
⇒ = = <
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
• Linear constant coefficient difference equation
Example
[ ] [ ]
( ) ( )
( ) ( )( )
( )
( )
0 0
0 0
0
0
10
1
10
1
1 zeros: ; poles:
1
N M
k kk k
N Mk k
k kk k
Mk
kkN
kk
k
M
kk
k kN
kk
a y n k b x n k
a z Y z b z X z
b zY zH z
X z a z
b c zz c z d
a d z
= =
− −
= =
−
=
−
=
−
=
−
=
− = −
⇒ =
⇒ Δ = ⇐
−= ⇒ = =
−
∑ ∑
∑ ∑
∑
∑
∏
∏
It does specifythe impulse response of an
not uniquel
LTI s
y
ystem
( ) ( ) ( )( )
[ ] [ ] [ ] [ ] [ ] [ ]
21 1 2
1 21 1
1 1 2 1 3 zeros: 1, 1; poles: ,1 31 3 2 411 14 82 4
1 31 2 2 1 24 8
z Y zz zH z z zX zz zz z
y n y n y n x n x n x n
− − −
− −− −
+ + += = = ⇒ = − − = −⎛ ⎞⎛ ⎞ + −− +⎜ ⎟⎜ ⎟⎝ ⎠⎝ ⎠
⇒ + − − − = + − + −
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– Causal system
is a right-sided sequence
⇒ The region of convergence of should be outside the outermost pole
– Stable system: absolutely summable
– Example
To be causal, ROC ⇒
To be stable, ROC ⇒
– In order for an LTI system to be both stable and causal, the ROC must be outside the outermost pole and include the unit circle, i.e., all poles inside the unit circle
– Depending upon the choice of ROC, the same difference equation results in a different
impulse response
[ ]h n
[ ] [ ] 1n
n nh n h n z z
∞−
=−∞
< ∞⇔ <∞ =
⇒
∑ ∑The ROC of H(z) includes the unit ci
for
rcle
[ ] [ ] [ ] [ ]
( )( )1 2 1 1
5 1 22
1 15 11 1 1 22 2
y n y n y n x n
H zz z z z− − − −
− − + − =
⇒ = =⎛ ⎞− + − −⎜ ⎟⎝ ⎠
12
2
2z >
1 2 2
z< <
( )cH z
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– Inverse system
Let be
To hold this equation, the ROC of Hi(z) and H(z) must overlap
If H(z) is causal, the ROC is
⇒ The ROC of should overlap with
Example 1:
Example 2:
( )H zi ( ) ( )1H zi H z
=
( ) ( ) ( ) [ ] ( ) [ ] [ ] 1 i iG z H z H z g n h n h n nδ= ⇒ = ∗ =
max kkz d>
( )iH z max kkz d>
( )
( )
[ ] [ ] [ ]
1
1
1
1
1
1 0.5 : z 0.91 0.9
1 0.9 : z 0.5 ( z 0.9)1 0.50.5 0.9 0.5 1
i
n ni
zH zz
zH zz
h n u n u n
−
−
−
−
−
−= >
−−
⇒ = ⇒ > >−
⇒ = − ⋅ −
with ROC
ROC to be overlapped with
( )
( )
[ ] [ ] ( ) [ ][ ] [ ] [ ]2 2; 2 2 1 8 2 1iz h n u n u n> = − ⋅ + ⋅ ⋅ − ⇒For an ROC unstable & causal
1
1
1
1
11
1
0.5 z 0.91 0.9
2 1.82 1 2
2; 2 2 1 1 8 2
i
n ni
n n
zH zz
zH zz
z h n u n u n
−
−
−
−
−
−
−= >
−− +
⇒ =−
< = ⋅ − − − ⋅ ⋅ − ⇒
with ROC:
For an ROC stable & noncausal
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
If H(z) is causal with zeros at , will be causal iff the Roc of is
If is stable, the RoC of must include the unit circle,
H(z) and its inverse are stable and causal iff both poles and zeros of H(z) are
inside the limit circle. ⇒ minimum-phase system
– Impulse response of for rational system functions
If causal,
Example: A first-order IIR filter
( )1
0 1
,1
M N Nk
k k
AH z B z M N
d z
−−
−= =
= + ≥−
∑ ∑
[ ] [ ] [ ]
[ ] [ ] [ ] [ ]0 1
0 1
, M N N
nk k
k
M N Nn
k kk
h n B n A d u n M N
y n B x n A d u n x n
δ−
= =
−
= =
= − + ≥
⇒ = − + ∗
∑ ∑
∑ ∑
[ ] [ ] [ ] ( )
[ ] [ ]1
11 ; : 1
1 n
y n ay n x n H z z aaz
a h n a u n
−− − = ⇒ = >−
< ⇒ =
ROC
For stability,
max 1kkc <
max kkz c>
( )H zi( )H zi
( )H zi
IIR termFIR term
kc
( )H zi
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
FIR system
Example
0
0
0
( )
, 00, otherwise
Mk
kk
Mn
kkM
kk
H z b z
b n Mh n b n k
y n b x n k
δ
−
=
=
=
⎧⎪⎡ ⎤⎡ ⎤ ⎨⎣ ⎦ ⎣ ⎦ ⎪⎩
⎡ ⎤⎡ ⎤⎣ ⎦ ⎣ ⎦
=
≤ ≤⇒ = − =
⇒ = −
∑
∑
∑
[ ]
( )( )
[ ] [ ]
[ ] [ ] [ ] [ ]
11
10
21
0
1
, 00 , otherwise
11
, 0,1,2,...,
1 1
n
MMMn n
n
kjM
k
Mk
k
M
a n Mh n
a zH z a zaz
z ae k M
y n a x n k
y n ay n x n a x n M
π
− ++−
−=
+
=
+
⎧ ≤ ≤= ⎨⎩
−⇒ = =
−
= =
⇒ = −
− − = − − −
∑
∑
Zeros :
or
7M =
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
• Frequency response for rational systems
– Gain (or attenuation)
– Phase response
– Group delay
( ) ( )( )
( )
( )
00 1
0
0 1
0 1
0
1
1
1
1
1
j
MMjj k
kkj k k
N Nz ej k j
k kk k
Mj
kj k
Nj
kk
c eb ebH e H zaa e d e
c ebH ea d e
ω
ωω
ω
ω ω
ω
ω
ω
−−
= ==
− −
= =
−
=
−
=
−= = =
−
−⇒ =
−
∑ ∏
∑ ∏
∏
∏
( )10
010 10 10
1 10
20log
20log 20 log 1 20 log 1 ( )
j
M Mj j
k kk k
H e
b c e d e dBa
ω
ω ω− −
= =
=
= + − − −∑ ∑
Gain
( )( )( )
( )( )
*2
20 1
*0
1
1 1
1 1
Mj j
k kj k
Nj j
k kk
c e c ebH ea d e d e
ω ω
ω
ω ω
−
=
−
=
− −⎛ ⎞
= ⎜ ⎟⎝ ⎠ − −
∏
∏
( ) ( ) ( ) ( )0
1 10
Arg ( ) Arg Arg 1 Arg 1 ; g x ArM M
j j jk k
k k
bH e c e d ea
ω ω ω π π− −
= =
−⎛ ⎞
= + − − −⎜ <⎝
≤⎟⎠∑ ∑
( ) ( ) ( )1 1
grd ( ) Arg 1 Arg 1M M
j j jk k
k k
d dH e c e d ed d
ω ω ω
ω ω− −
= =
= − − + −∑ ∑
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
• Frequency response of a single pole or zero
– Single zero
( ) ( ) ( )( ) ( )( )
( )
( )( ) ( )( )
( )( ) ( )( )( )( )
1
2
2
1
2
2
1 1 ;
1 1 1
1 2 cos
sinrg tan
1 cos
grd rg
cos
1 2 cos
j j j
j j j j j j j
j
j j
H z az H e ae a re
H e re e re e re e
r r
rA H e
rdH e A H e
dr rr r
ω ω θ
ω θ ω θ ω θ ω
ω
ω ω
ω θ
θ ωθ ω
ωθ ωθ ω
− −
− − −
−
= − ⇒ = − =
⇒ = − = − −
= + − −
−⇒ =
− −
⎡ ⎤⇒ = ⎣ ⎦
− −=
+ − −2v
φ 3
3vω 1vθ
`
( )
( )
( )( )
33
1
3 1 3 rg
j jj
j
j
z aH zz
ve reH e ve v
A H e
ω θω
ω
ω φ φ φ ω
−=
−⇒ = = =
= − = −
0
0.5
θ
θ π
θ π
=
=
=
0.9r =
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– Example: Real single zero
( )( ) ( )
( )
( )( )( )( ) ( )( )
1
2
2
1
2
2
1
1 ;
= ,
1
1 2 cossinrg tan
1 cos
grd rg
cos 1 2 cos
j j j
j j
j
j j
H z az
H e ae a re
a r
H e re
r rrA H er
dH e A H ed
r rr r
ω ω θ
ω ω
ω
ω ω
θ π
ωωω
ωωω
−
−
−
−
= −
⇒ = − =
= −
⇒ = +
= + −
⇒ =+
⎡ ⎤⇒ = ⎣ ⎦
+=
+ +
If
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– Second-order IIR system
( ) ( )( )[ ] [ ] [ ] [ ]
[ ] ( ) [ ]
1 2 21 1
2
1 11 2 cos1 1
2 cos 1 2
sin 1sin
j j
n
H zr z r zre z re z
y n r y n r y n x n
r nh n u n
θ θ θ
θ
θθ
− −− −= =
− +− −
⇔ − − + − =
+⇒ =
1vω 3v
2vθ−
( )
( )( ) ( )( )
( )( )
23
1 2 1 2
1 1
1
sin sinrg tan tan
1 cos 1 cos
j
j
vH e
v v v v
r rA H e
r r
ω
ω θ ω θ ωθ ω θ ω
− −
= =⋅
− += − −
− − − +
( ) 1 2
11 0.9 2 0.81
H zz z− −
=− +
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– Third-order IIR system
( ) ( )( )( )( )
1 1 2
1 1 2
1.0376
0.6257
0.05634 1 1 1.0166
1 0.683 1 1.4461 0.7957
1, (59.45 )
0.683, 0.892 (35.85 )
jz
jp
z z zH z
z z z
z e
z e
− − −
− − −
±
±
+ − +=
− − +
⇒ = −
⇒ =
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
• Relationship between magnitude and phase
– In general, magnitude information phase information
– In cases of rational system functions, there is some constraint between magnitude and
phase
– The square of the magnitude frequency response is the evaluation of z-transform on
the unit circle
– Question: Can we know of H(z) from C(z) ?
( ) ( ) ( ) ( )
( )( )
( )
( )( )
2
*
1
0 01 1
10 0
1 1
1
1 11;
1 1
j
j j j
z e
M M
k kk kN N
k kk k
H e H e H e H z Hz
c z c zb bH z Ha z ad z d z
ω
ω ω ω∗ ∗
=
∗−
∗= =∗
∗−
= =
⎛ ⎞= = ⎜ ⎟⎝ ⎠
− −⎛ ⎞= =⎜ ⎟⎝ ⎠− −
∏ ∏
∏ ∏
( ) ( ) ( )*
* 1 C z H z Hz
Δ
No
No⎯⎯→←⎯⎯
( )
( ) ( )1 1
: ; 1 : ;
k k
k k
H z d c
H d cz
− −∗ ∗ ∗∗
⎛ ⎞⎜ ⎟⎝ ⎠
pole zero conjugate reciprocal pairs
pole zero
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– If H(z) is assumed to be causal & stable, all its poles are in the unit circle
⇒ Poles of H(z) can be identified uniquely
Example System with the same C(z)
( ) ( )( )
( ) ( )( )
( ) ( )
( ) ( )
( ) ( ) ( ) ( )
1 1
11 14 4
1 1
21 14 4
1 2
*1 1 1 *
*2 2 2 1 2*
2 1 1 0.5
1 0.8 1 0.8
1 1 2
1 0.8 1 0.8
1
1
j j
j j
j j
z zH z
e z e z
z zH z
e z e z
H e H e
C z H z Hz
C z H z H C z C zz
π π
π π
ω ω
− −
−− −
− −
−− −
− +=⎛ ⎞⎛ ⎞− −⎜ ⎟⎜ ⎟
⎝ ⎠⎝ ⎠
− +=⎛ ⎞⎛ ⎞− −⎜ ⎟⎜ ⎟
⎝ ⎠⎝ ⎠
⇒ =
⎛ ⎞= ⎜ ⎟⎝ ⎠⎛ ⎞= ⇒ =⎜ ⎟⎝ ⎠
−05. −2
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
• All pass system
– A stable system function of the form
has a frequency-response magnitude independent of
– In general, the transfer function of an all-pass system is given by
– All-pass system can be used for compensating group delay distortion.
– Causal all-pass system has positive group delay (but not continuous phase) property
( )1
11apz aH z
az
− ∗−=
−
( ) ( )( )( )( )
1 11
1 1 11 1
; 0, 1, 11 1 1
cr MMk kk
ap k kk kk k k
z e z ez dH z A A d ed z e z e z
∗− −−
− − ∗ −= =
− −−= > < <
− − −∏ ∏
( ) ( )
( )( ) ( )( )
1
1
sinrg 2 tan
1 cos
11 1
j jj j j
ap ap
ap
j
j
j
rA H e
e a a eH e e H eae ae
r
ω ωω ω ω
ω ω
ω ω θω
ω θ−
− ∗ ∗−
− −
− −= = ⇒
−−
−
=− −
= −−
.ω
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– Example
( )1
1
0.91 0.9zH z
z
−
=∓
∓
0.9
0.9p
p
z
z
=
= −
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– Example
4π
1 20.5.75−43
−
0.8
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
• Minimum-phase system
– It has its poles and zeros all inside the unit circle.
⇒ Any minimum-phase system is causal and stable.
⇒ is a causal, stable, and min-phase system
– Cascade of both systems will not introduce any delay
– Energy concentration
For an arbitrary stable and causal system , let be the min-phase
function s.t.,
Then, for any sequence and K,
⇒ The signal energy is transported with the smallest possible delay using a
minimum-phase system
Example If
( )1H z−
( ) ( )1 1 joH z H z e− = ⋅
( )jF e ω
( ) ( ) , .j jG e F eω ω ω= ∀
[ ]x n
[ ] [ ]( ) [ ] [ ]( )2 2
0 0
K K
k kk k
x n g n x n f n= =
∗ ≥ ∗∑ ∑
[ ] [ ] [ ] [ ]2 2
0 0
, K K
k k
x n n g k f kδ= =
= ≥∑ ∑
( )jG e ω
( )H z
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– The min-phase function has its energy maximally concentrated near the time origin.
proof
Let
The energy of
Assume vanishes for
But is not necessarily zero for
– cannot be uniquely determined from , because cascading all-pass filters
does not affect the magnitude of the frequency response.
[ ]y n
( ) ( )( ) ,
jj
j
F eH e
G e
ωω
ω ωΔ ∀ be stable, causal and all pass
x n z nG e jωc h H e jωc h
y n
min-phase all-passphase shifter
( ) 1jH e ω =
[ ]{ } ( )
[ ]{ } ( ) ( )
0
0
0
0
22 20
0 2
22 2
0 2
1 2,
1
jy
j jz y
E E y n Y e dT
E E z n H e Y e d E
ωω
ω
ωω ω
ω
πω ωω
ωω
−
−
Δ = =
Δ = =
∫
∫[ ]y n [ ]2
00 , . .,
K
yn
n n K i e E y n=
< < = ∑a n d[ ]z n n K>
[ ]2
0
K
z yn
E E z n=
= ≥ ∑( )H z ( )G z
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– Any rational rational transfer function can be expressed as
Assume that has all zeros and poles inside the unit circle except one
zero outside the unit circle at
where is minimum phase
has poles and zeros of inside the unit circle plus zeros that are
conjugate reciprocals of the zeros of outside the unit circle
contains all the zeros of outside the unit circle
Example
( ) ( ) ( )m in apH z H z H z=
*
1 , 1z cc
= <
( )H z
( ) 1 *1
1 *1
1 1
( )( )
( ) (1 )1
H z H z z c
z cH z czcz
−
−−
−=
= −
−⋅ − ⋅
−minimum phase
all pass
( )
m in (
1 11
(
1 1
)
1
)
1 111 3 3 331 1 11 1 12 2 3
apH z H z
z zzH zz z z
− −−
− − −
+ ++= = ⋅ ⋅
+ + +
( )1H z
( )minH z
( )apH z ( )H z
( )H z
( )H z
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– The amplitude characteristics of a minimum-phase filter is fully determined by the
phase characteristics and vice versa
Proof
Since has its poles and zeros inside the unit circle,
has a causal and well behaved inverse Fourier transform, i.e.,
We can see that and form a Hilbert transform pair.
Since is the logarithm of a real-valued symmetric function of frequency, it
is real and symmetric
Thus we can compute
( ) ( ) j jG e G eω ω⇒ We can compute from
( )jX e ω ( )jG e ωφ
( )jX e ω
[ ] [ ]
( ) [ ] [ ]
[ ] [ ] [ ] [ ]0
0 2 cos
0 0 2 , 1
,
j
nX e x x n n
g x g n x n n
x n x n n
ω ω∞
=
⇒
⇒ +
⇒ =
= − ∀
= ≥
∑ =
and
( ) jG e ωφ
( )G z ( )ln 1jG e zω > is analytic for
( )ln jG e ω⇒
( ) [ ] ( ) ( )
( ) ( ) [ ] ( ) [ ]0
0 0 cos ;
si
ln
l nn
j n j jG
n
j jG
n n
j
j
g n e X e j e
X e g n n e g n
G e
G e n
ω ω ω
ω
ω
ω ω
φ
ω φ ω
∞−
=
∞ ∞
= =
=
= = −
= +∑
∑ ∑where =
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
– Frequency-response compensation
If is a min-phase system, perfect compensation is generally possible
The frequency response magnitude is exactly compensated with some delay
Example
s n s nc
H zdb g H zcb g( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
min
min
1
d ap
c d c ap
H z H z H z
H z G z H z H z H zH z
=
= ⇒ = =Let
( )dH z
( )( )( )( )( )( )
( )( )( )( )
0.6 1 0.6 1 0.8 1 0.8 1
2 0.6 1 0.6 1min
0.8 1 0.8 1
0.8 1 0.8 1
( ) 1 0.9 1 0.9 1 1.25 1 1.25
( ) 1.25 1 0.9 1 0.9
1 0.8 1 0.8
1 1.25 1 1.25( )
1 0
j j j j
j j
j j
j j
ap
H z e z e z e z e z
H z e z e z
e z e z
e z e zH z
π π π π
π π
π π
π π
− − − − − −
− − −
− − −
− − −
= − − − −
⇒ = − −
⋅ − −
− −⇒ =
−( )( )0.8 1 0.8 1.8 1 0.8j je z e zπ π− − −−
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
( )H z ( )apH zmin ( )H z
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
In general, min-energy delay property
– Maximum phase system is a stable system whose poles and zeros are all outside the
unit circle
stable smallest
⇒ all poles contribute to with terms of the form
⇒ anti causal!
[ ] [ ]m in0 0h h≤
( ) ( )
( ) ( )
[ ] [ ] [ ] [ ]
[ ] [ ]
1
m in 1
1
m in 1
m in m in
m in
, 11
1lim lim
10 0 0 0 1
0 0
z z
z aH z H z aaz
a zH z H zz a
h h h ah aa
h h
−
−
−
−→ ∞ → ∞
−= <
−−
=−
⇒ = − ⇒ = − <
⇒ ≤
( )( )
( )1
1
1, 1
M
kk
k kN
kk
z cH z c d
z d
=
=
−= > >
−
∏
∏
R o c : kz d⇒ <
[ ]h n ( ) [ ]1nkd u n− − −
Digital Signal Processing Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
Note that FIR max-phase sequence can be made causal by introducing a finite
delay
– Property of min-phase systems
minimum phase-lag
negative for
minimum group-delay
minimum energy delay property
( ) [ ] ( )( ) ( )
1
m ax m in
m in
0 11
kk
k
ap
z cH z h cc z
H z H z
−∗
∗
−= −
−
=
∏ ∏
( ) ( ) ( )minarg arg argj j japH e H e H eω ω ω⎡ ⎤ ⎡ ⎤ ⎡ ⎤= +⎣ ⎦ ⎣ ⎦ ⎣ ⎦
0 ω π≤ ≤
[ ] [ ]m in0 0h h≤
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
Structure for discrete-time systems• Characterization of an LTI system can be represented by difference equation, impulse
response or system function
Example:
• For description of systems for implementation, use the block diagram and/or signal flow
graph method
( )
[ ] [ ] [ ]
10 1
1
10 1
,1
1
n n
b b zH z z aaz
h n b a u n b a u n
−
−
−
+= >
−⇒ = + −
⇒⇒⇒
Impulse response :Infinite duration impulse response Not possible to implement the system by discrete time convolutionThe
[ ] [ ] [ ] [ ]0 1 1 1y n ay n b x n b x n= − + + −
output can be calculated by a recursive computation algorithm
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
• Basic block diagram symbols
Example
y n
x n
z n x n y n= +
x n a y n ax n=
x n y n x n= −1z−1
[ ] [ ] [ ] [ ]1 21 2y n a y n a y n bx n= − + − +
z−1
z−1
x n b y n
y n −1a1
a2
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
–
< direct form I>
z−1
z−1
::
z−1
z−1
:
z−1 z−1
x n b0
b1
b2
bM −1
bM
x n −1
x n − 2
x n M− +1
x n M−
H z1b g H z2 b gy n
a1
a2
aN −1
aN
y n −1
y n N− +1
y n N−
[ ] [ ] [ ]1 0
N M
k kk k
y n a y n k b x n k= =
= − + −∑ ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
– By interchanging the cascaded blocks,
< direct form II or canonic direct form >
( )( ) ( )
( ) ( )( ) ( )
1 2
2 1
Y zH z
X z
H z H z
H z H z
=
=
=
z − 1
:
:z − 1
:
b0
1a
2a
Na
[ ]1nω −
[ ]2nω −
[ ]n Nω −
y n
1Nb −
z − 1z − 1
z − 1
x n
1Na −
Nb
[ ]nω
:
1b
z − 1
z −1
z −1
:
:
z −1
b0
b1
b2
bN −1
bN
a N
a N −1
a 2
a1
[ ]x n [ ]y n
⇒
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
• Signal flow graph representation
Example
node jj nω ⎡ ⎤⎣ ⎦
branch j k,
node kk nω ⎡ ⎤⎣ ⎦
x n ω 1 n c
b
d
ey nω 2 n
a
[ ] [ ] [ ] [ ][ ] [ ][ ] [ ] [ ]
1 2 2
2 1
2
n x n a n b n
n c n
y n dx n e n
ω ω ω
ω ω
ω
= + +
=
= +
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
Example
z−1
x n ω nb0
ω n −1
y n
b1a
x n ω n y nb0
b1
z−1
a
[ ] [ ] [ ][ ] [ ] [ ]
( ) ( ) ( ) ( ) ( ) ( )( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( )
( ) ( )( )
0 1
1 1
1 10 1 0 1
1 10 1
10 1
1
1
1
1
1
1
n x n a n
y n b n b n
W z X z aW z z X z az W z
Y z b W z bW z z Y z b b z W z
Y z X zW z
b b z azY z b b zH zX z az
ω ω
ω ω− −
− −
− −
−
−
= + −
= + −
⇒ = + ⇒ = +
= + ⇒ = +
⇒ = =+ +
+⇒ = =
+
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
• Basic structures for IIR systems
– Direct form I
– Direct form II
( )1 2
1 2
1 21 0.75 0.125
z zH zz z
− −
− −
+ +=
− +x n
z−1
z−1
1
2
1
075.
z−1
z−1
−0125.
y n
x nz−1
2
z−1075.
y n
−0125.
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
– Cascade form
( )( ) ( )( )
( ) ( )( )
1 2
1 2
1 1 * 1
1 1
1 1 * 1
1 1
1 1 1
1 1 1
M M
k k kk kN N
k k kk k
g z h z h zH z A
c z d z d z
− − −
= =
− − −
= =
− − −=
− − −
∏ ∏
∏ ∏
z−1
z−1
− d12 h1
2
d d1 1+ *c h − +h h1 1*c h
d d2 2+ *c h − d22 h2
2
( )*2 2h h− + c1 −g1
z−1
z−1
z−1
( ) 2* 1 21 k k kd d z d z− −− + +
0 25. 05.
z−1 z−1
x n y n
( ) ( )( )( )( )
1 1
1 1
1 1
1 0.5 1 0.25z z
H zz z
− −
− −
+ +=
− −Example
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
– Parallel form
( ) ( )( )( )
1 21
1 1 * -10 1 1
1 2
11 1 1
2,
pN N Nk kk k
kk k kk k k
p
B e zAH z g zc z d z d z
N N NM N N M N
−−
− −= = =
−= + +
− − −
= +≥ = −
∑ ∑ ∑
21
0 11 2
1 1 21
Nk k
k k k
e e za z a z
−
− −=
+− −
∑
y nx n e01
e11
a11z −1
z −1
a21
g0
: :
z −1
z −1
g1
g2
g N p
z −1
A1
c1
:
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
Example
x n y n
8
−7
z−1
z−180 75.
−01125.
z−1
z−1
x n y n8 18
−0 25.05.
0 25.
( )1 2
1 1
1
1 1
1 1
1 21 0.75 0.125
7 881 0.75 0.125
18 2581 0.5 1 0.25
z zH zz z
zz z
z z
− −
− −
−
− −
− −
+ +=
− +− +
= +− +
= + −− −
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
• Transposed forms
– Signal flow graph methods provide procedures for transforming graphs into different forms
– Transposition of a flow graph (flow graph reversal) is obtained by reversing the directions
of all branches, while keeping the branch transnittances and reversing the roles of the
input and output
Examplex n y n
z −1
a
y n x n
z −1
a
x n y n
z −1
a
( ) 1
11
H zaz−
=−
The difference is the change in ordering
Transposed form
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
Example Transposed form of a second-order structure
Transposed direct form II
z−1
y nx n ω n
a1 b1
a2 b2
z−1
0b
[ ] [ ] [ ] [ ][ ] [ ] [ ] [ ]
1 2
0 1 2
1 2
1 2
n a n a n x n
y n b n b n b n
ω ω ω
ω ω ω
= − + − +
= + − + −
b2a2
b0
b1
z −1
a1
z −1
y n x n
x n y nb0 v n0
z −1
z −1
b1 a1
b2 a2
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
• Basic network structure for FIR systems
– Direct form
Causal FIR systems
In the transposed form,
[ ] [ ] [ ]0
, 00 , otherwise
Mn
kk
b n My n b x n k h n
=
≤ ≤⎧= − ⇒ = ⎨
⎩∑
x n z−1 z−1 z−1
h 0 h 1 h 2
y n
[ ]h M
x n
z−1y n z−1
h 1 h 2 h Mh 0
z−1 z−1 z−1
h M h M −1
y n
h 0x n
z−1
⋅ ⋅ ⋅
[ ]2h M −
Transversal filter or tapped delay line
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
– Cascade form
( ) [ ] ( ) ( )1 20 1 2
0 1; 1 / 2
sMMn
k k k sk k
H z h n z b b z b z M M− − −
= =
= = + + = +⎢ ⎥⎣ ⎦∑ ∏
z−1
z−1
z−1
z−1
b02b01
b11 b12
b21 b22
z−1
z−1
0 sMb
1 sMb
2 sMb
[ ]x n [ ]y n
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
– Structure for linear phase FIR systems
Causal linear phase FIR systems
[ ] [ ] [ ] [ ]
[ ] [ ] [ ]
[ ] [ ] [ ] [ ]
[ ] [ ] [ ] [ ]
0
12
0 / 2 1
1 12 2
0 0
for 0
2 2
2 2
M
k
MM
k k M
M M
k k
h M n h n h M n h n n M
y n h k x n k
M Mh k x n k h x n h k x n k
M Mh k x n k h x n h M k x n M k
=
−
= = +
− −
= =
− = − = − ≤ ≤
⇒ = −
⎡ ⎤ ⎡ ⎤= − + − + −⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦
⎡ ⎤ ⎡ ⎤= − + − + − − +⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦
∑
∑ ∑
∑ ∑
or
[ ] [ ]
[ ] [ ] [ ] [ ]( )1
2
0
2 2
M
k
h M n h n
M My n h k x n k x n M k h x n−
=
− =
⎡ ⎤ ⎡ ⎤= − + − + + −⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦∑
When (type - I system),
z−1 z−1 z−1
z−1 z−1z−1
h 0 h 1 h 2 h M2
1−LNMOQP h M
2LNMOQP
y n
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
4. Structure for discrete-time systems
Symmetry condition
⇒ zeros of occur mirror image pairs
If is real, zeros of occur in complex conjugate pairs.
z3
z4
z3*
z1*
z1
z2 1
2z
1
1z
1
1z*
( )H z
[ ]h n ( )H z
( ) [ ]( )( )( )( )
{ }
1 1 2 1 2 1 2 3 4
2
2 3 1 12 1 1
0 1 1 1 1
1 1 1, 2Re , 2Re , 2
H z h z az z bz z cz dz cz z
a z b z c z d zz z z
− − − − − − − − −= + + + + + + + + +
⎧ ⎫= + = = − + = + +⎨ ⎬
⎩ ⎭where
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Filter Design Techniques• Design procedure
– Specify the desired frequency Response
– Approximate the specifications
– Realize the filter ( ) ,jeffH e H
Tω ω ω π⎛ ⎞= <⎜ ⎟
⎝ ⎠
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Design of discrete-time IIR filters from continuous-time filters
– Filter design by impulse invariance
The discrete-time filter specifications are transformed to continuous-time filter
specifications by
Design a suitable continuous-time filter
Transform to the desired discrete-time filter
[ ] ( )d c dh n T h nT=
C D/ h n D C/xa τb g x n y n ya τb g
TTdT
( ) 2jc
k d d
H e H kT T
ω ω π∞
=−∞
⎛ ⎞= +⎜ ⎟
⎝ ⎠∑
( ) ( ) 1, , 0,2
jc c
d d
H e H H f fT T
ω ω ω π⎛ ⎞
= ≤ = ≥⎜ ⎟⎝ ⎠
if
( )2d
fTω π⇔ =Ω
( )cH s
( )H z
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
For
By sampling
A pole at in the s-plane is transformed to a pole at in the z-plane.
If is stable, the real part of
⇒
⇒ pole is inside the unit circle
⇒ is also stable
Although the poles in the s-plane are mapped in the z-plane in a simple manner,
the zeros are not.
( ),ch t
[ ] [ ] [ ] ( ) [ ]
( )1 1
11 1
k d k d
k d
N N ns nT s Td c d d k d k
k k
Nd ks T
k
h n T h nT T A e u n T A e u n
T AH ze z
= =
−=
= = =
⇒ =−
∑ ∑
∑
ks s= k ds Tz e=
( )cH s 0ks <
1k ds Te <
( )H z
( ) ( ) 11
, 0,
0 , 0
kN
s tNkk
kc ck k
A e tAH s h ts s t
==
⎧ ≥⎪= = ⎨− ⎪ >⎩
∑∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Butterworth (BW) filter
The magnitude response of a BW filter is
where is the 3dB frequency
( ) ( ) ( )2 2
2 1 1
1
12
Nc
c
N
c
H s H ssf
H fff π
= ⇐ − =⎛ ⎞
+ ⎜ ⎟⎝
⎞⎜ ⎟⎝ ⎠
⎛+
⎠
cf
( )2 12 ; 0,1,2, ,2 1
j k NN
k cs e k Nπ
+ −⎧ ⎫⇒ =Ω = −⎨ ⎬⎩ ⎭
poles are
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Example Design a continuous-time Butterworth (BW) filter s.t.
The magnitude response of a BW filter is
where is the frequency
By solving
1
1 1− δ
δ 2
ω p ω s
1
2
0.108750.177830.20.3
p
s
δδω π
ω π
===
=
( )
( )
0.89125 1, 0 0.1
10.17783 0.152
c
c
H f f
H f f
≤ ≤ ≤ ≤
≤ ≤ ≤
( ) ( ) ( )2 2
2 1 1
1
12
Nc
c
N
c
H s H ssf
H fff π
= ⇐ − =⎛ ⎞
+ ⎜ ⎟⎝
⎞⎜ ⎟⎝ ⎠
⎛+
⎠
2 2
2 2
0.1 110.89125
0.15 110.17783
N
c
N
c
f
f
⎧ ⎛ ⎞ ⎛ ⎞+ =⎪ ⎜ ⎟ ⎜ ⎟⎝ ⎠⎪ ⎝ ⎠
⎨⎛ ⎞⎪ ⎛ ⎞+ =⎜ ⎟ ⎜ ⎟⎪ ⎝ ⎠⎝ ⎠⎩
5.8858 60.11216c
N Nf
⇒ = ⇒ ==
cf 3dB
1 dT ω= = ΩLet so that
( )2 12
0.182 0.6792 0.497 0.497
0.679 0.182
j k NN
k c
js f e j
j
π
π+ −
− ±⎛ ⎞⎜ ⎟= ⇒ − ±⎜ ⎟⎜ ⎟− ±⎝ ⎠
poles are12
0.11 1.258928
0.111918c
c
ff
⎛ ⎞+ =⎜ ⎟⎝ ⎠
⇒ =
0.7032
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
This method is useful only for band-limited filters
( ) ( )( ) ( )
( ) ( )
( ) ( )
( ) ( )
2 2 2
1 2
3 4
5 6
0.120930.3645 0.4945 0.99455 0.4945 1.35855 0.4945
0.182 0.679 0.182 0.679
0.497 0.497 0.497 0.497
0.679 0.182 0.679 0.182
cH ss s s
A As j s j
A As j s j
A As j s j
=+ + + + ⋅ + +
= ++ + + −
+ ++ + + −
+ ++ + + −
( )0.182 0.6791
2
1
0.8336 0.7782 0.6280
0.6487 0.5235 0.69489
jz e j
j z
− −= = −
= − ⇒ =
( ) ( ) ( )1 2
0.182 0.679 1 0.182 0.679 11 1j j
A AH ze e z e e z− − − + −
⇒ = + + ⋅ ⋅ ⋅− −
( )1
1 2
1
1 2
1
1 2
0.2871 0.44661 1.2974 0.6949
2.1428 1.14551 1.0691 0.3699
1.8557 0.630321 0.9972 0.2570
zH zz z
zz z
zz z
−
− −
−
− −
−
− −
−=
− +− +
+− +
−+
− +
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Bilinear transformation
To avoid aliasing, we employ the bilinear transform
∞ −∞
2πf
−π πω
( )1
1
2 11c
d
zH z HT z
−
−
⎡ ⎤−⎛ ⎞= ⎢ ⎥⎜ ⎟+⎝ ⎠⎣ ⎦
1
1
12 2
12
12
2 1
1
21, 0
1
d
dd
dd
d
sz s jzs
T zfT s
T j fT
T j fT
z
π
σ π
σ π
σ π
σ
−
−
⎛ ⎞ +⇒ = ⇐ = +
−
+ +=
− −
⇒ <
−= ⎜ ⎟+⎝ ⎠
<if
1
1
2 11d
zsT z
−
−
−⎛ ⎞= ⎜ ⎟+⎝ ⎠
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
The axis of the s-plane maps onto the unit circle of the z-plane
For
or
2j fπ
2 ,1 1 1 1 1
jd d
d d
s j fj fT j fTz z ej fT j fT
ω
ππ ππ π
=+ +
= ⇒ = ⇒ =− −
( )
2
2
2 1 21
2 sin2 2
2 cos2
2 tan2
j
jd
j
jd
d
es j fT e
e j
T e
jT
ω
ω
ω
ω
σ π
ω
ω
ω
−
−
−
−
⎛ ⎞−= = +⎜ ⎟+⎝ ⎠
⎛ ⎞⎜ ⎟⎝ ⎠=⎛ ⎞⎜ ⎟⎝ ⎠
=
-1
1 20 tan tan2 2
=2tan2
d d
d
fT T
T
ω ωσπ
ω
⎛ ⎞⇒ = = Ω =⎜ ⎟
⎝ ⎠Ω⎛ ⎞
⎜ ⎟⎝ ⎠
and or
or
ωπ
−π
Ω 2πfb gor
ω
π
= FHGIKJ
=
−
−
22
2
1
1
tan
tan
ΩT
T f
d
db g
σ
jΩ
−π
Re
Im
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Hc
(Ω)
Ωp
Ωs
Ω
ω p ω s πωω
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Example Bilinear transformation of a BW filter
( )2 2
2
2
2 2
1 0.6498 11 1.25890.89125
1
1.01905 1 1 31.62200.17783
N
c Nc
c
N
c
H⎛ ⎞ ⎛ ⎞Ω = ⇒ + = =⎜ ⎟ ⎜ ⎟Ω ⎝ ⎠⎛ ⎞ ⎝ ⎠Ω
+ ⎜ ⎟Ω⎝ ⎠
⎛ ⎞ ⎛ ⎞+ = =⎜ ⎟ ⎜ ⎟Ω ⎝ ⎠⎝ ⎠
( )( )
0 0.2
0.3
2 0.20 0.2
0.891
0 tan 0.6498; 12
2 0.30.3 tan 1.01905
25 1,
0.17783,
2
dd
j
j
d
T
H e
H e
T
T
ω
ω
ω π
π π
πω π
ππ
ω
ω
≤ ≤
≤ ≤ ⇒ ≤Ω ≤
≤ ≤
≤ ≤ ≤
= =
≤ Ω⇒≤ ⇒ ≤Ω
2 21 11 10.17783 1.89125 2.0729 5.30267
1.01905 0.39092log0.6498
6 0.76622c
N
N
⎧ ⎫⎡ ⎤ ⎡ ⎤⎪ ⎪⎛ ⎞ ⎛ ⎞− −⎢ ⎥ ⎢ ⎥⎨ ⎬⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠⎢ ⎥ ⎢ ⎥⎪ ⎪⎣ ⎦ ⎣ ⎦⎩ ⎭⇒ = ≅ ≅
⎛ ⎞⎜ ⎟⎝ ⎠
⇒ = ⇒Ω =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
( ) ( )( )
( )
2 2
2
0.202380.39965 0.5871 1.08365 0.5871
11.48025 0.5871
cH ss s
s
⇒ =+ + + +
×+ +
( ) ( )( )( )
( )
61
1 2 1 2
1 2
0.0007378 1
1 1.2686 0.7051 1 1.0106 0.3583
11 0.9044 0.2155
zH z
z z z z
z z
−
− − − −
− −
+⇒ =
− + − +
×− +
12
312
512
0.76622 , 0.7401 0.1983
0.76622 0.5418 0.5418
0.76622 0.1983 0.7401
j
j
j
e j
e j
e j
π
π
π
±
±
±
⎛⋅ = ±⎜
⎜⎜⇒ ⋅ = ±⎜⎜ ⋅ = ±⎜⎝
Poles are
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
The magnitude of N-th order Butterworth filter can be represented by
Note that this BW filter is maximally flat, , and periodic with a period
of 2π.
Maximally flat ⇔ The first (2N-1) derivatives of are zero at
It is not easy to directly find the poles and zeros in the z-plane from
since the magnitude-squared function should be factorized into to
determine
( ) 2
21 2; tan tan
2 2 2tan
21tan
2
j c c DN
d
c
TH eT
ω ω ωω
ω
Ω= = ⇐Ω =
⎛ ⎞⎜ ⎟
+ ⎜ ⎟⎜ ⎟⎝ ⎠
( ) 12
cjH e ω =
0ω =
( ) 2jH e ω
( ) ( )1H z H z−
( ).H z
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Design of type-I Chebyshev filters
– is the N-th order Chebyshev polynomial defined by
( ) 2
2 2
1
1 Nc
HVε
Ω =⎛ ⎞Ω+ ⎜ ⎟Ω⎝ ⎠
( )NV x
( ) ( )1cos cosNV x N x−=
( )( ) ( )( ) ( )( ) ( )
( ) ( ) ( )( )
( )
0
11
1 22
2 33
1 1
2
1
cos0 1
cos cos
cos 2cos 2 1
2 1 4 3
2
0 1 0 1
1 cos
N N N
N
N
V x
V x x x
V x x x
V x x x x x x
V x xV x V x
x V x
x x V x
−
−
+ −
−
= =
= =
= = −
= − − = −
= −
⇒ ≤ ≤ ⇒ ≤ ≤
> ⇒ ⇒ ∼is imaginary hyperbolic cosine ( ) NV x⇒ increases monotonically
11 ε−
Ω p Ωs
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Property of
forms an orthogonal set over
For each N,
For even N,
For odd N,
The zeros of lie in
The range of is between –1 and 1 for
( )NV x
( ){ }NV x
( ) ( )1
21
0, 1 / 2, 0
1 , 0N M
M NV x V x dx M N
x M Nππ
−
≠⎧⎪= = ≠⎨
− ⎪ = =⎩∫
( )1 1NV =
( ) ( ) ( )( ) ( )
/ 21 1; 0 1 NN N
N N
V V
V x V x
− = = −
= −
( ) ( )( ) ( )
1 1; 0 0N N
N N
V V
V x V x
− = − =
= − −
1 1x− ≤ ≤( ){ }NV x
( ){ }NV x 1 1x− ≤ ≤
1 1x− ≤ ≤
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Ripple factor:
Since
Rule: choose and
( )
( )
2
2
1 0 1, 11
1,
c
c
H
H
εΩ
⇒ ≤ ≤ ≤ Ω ≤Ω +
Ω> Ω
Ω
For
decreases monotonically
, Nε cΩ
11 ε−
Ω p Ωs2
2
1 =1-1+
ε δε
⇒ ripple amplitude( )1 1NV =
( ) ( ) ( )2 2
1 1 1 1 tan / 2
pj T j T
N
H e H eV k T
ω ω
ε ε ω= ⇐ =
+ + ⎡ ⎤⎣ ⎦
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Consider the first zeros of
Letting
The poles of the Chebyshev filter lie on an ellipse in the s-plane
( )2 21 0NV xε+ =
( )1
1cos cosh 0 & sin sinh
2 1 / 2 , 0,1,2, ,2 11 1 sinh
Nu Nv Nu Nv
u M N M N
vN
επ
ε−
⇒ = = ±
⇒ = + = −
=
2 2
2 2 1; sinh cosh
s jv v
σ σΩ+ = = + Ω
( ) ( )1
1
cos cos( ) cos
cos cos cos
cos cosh sin sinh/
N
r u jv x x r
V x N x Nr
Nu Nv j Nu Nvj ε
−
−
≡ + = ⇒ =
= =
= −= ±
1 1 1 11 21 1, , 1
2 2N N N Na bα α α α α ε ε
− − − −⎛ ⎞ ⎛ ⎞= − = + = + +⎜ ⎟ ⎜ ⎟
⎝ ⎠ ⎝ ⎠
( )( )sinh cos cos
cosh sin sinp c
p c
v a
v b
σ θ θ
θ θ
= = Ω
Ω = = Ω
3N =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Design of Type II Chebyshev Filters
• Elliptic filter
– The error is distributed over the entire frequency band ⇒ Equal ripple in all band
⇒ min order of the filter
: Chebyshev rational function
of degree N
( ) 2
12
1
1c
cN
HVε
−Ω =Ω⎡ ⎤⎛ ⎞+ ⎜ ⎟⎢ ⎥Ω⎝ ⎠⎣ ⎦
( ) 2
2 2
1
1c
Nc
HUε
Ω =⎛ ⎞Ω+ ⎜ ⎟Ω⎝ ⎠
( )NU Ω
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Example
– Filter specifications
Passband
Stopband
This implies that 1 20.01, 0.001, 0.4 0.6p sδ δ ω π ω π= = = =and
( ) 0.001; 0.6jH e ω π ω π≤ ≤ ≤
( )0.99 1.01; 0.4jH e ω ω π≤ ≤ ≤
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Butterworth filter with N=14 Chev-type I filter with N=8
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Chev-type II filter with N=8 Elliptic filter with N=6
1
1
2 11d
zsT z
−
−
−⎛ ⎞= ⎜ ⎟+⎝ ⎠
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Poles and zeros of the designed filters
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Design of FIR filters by windowing
– FIR filter design directly approximates the desired frequency response of the discrete-
time system
– When the desired frequency response is
consider the design of a causal FIR filter that approximates the ideal response
– The truncation of the desired impulse response results in the Gibbs phenomenon
– Rectangular Windowing
where is the windowing function given by
Note that is a smeared version of
( ) [ ] [ ] ( )1 2
j j n jd d d d
hH e h n e h n H e d
πω ω ω
πω
π
∞−
−=−∞
= =∑ ∫;
[ ] [ ] [ ] [ ] [ ], 0
0 , otherwised
d r
h n n Mh n h n h n nω
≤ ≤⎧= ⇒ =⎨⎩
[ ]r nω
[ ] 1, 00, otherwiser
n Mnω
≤ ≤⎧= ⎨⎩
( ) ( ) ( )( )12
jj jd rH e H e e d
π ω θω θ
πω θ
π−
−⇒ = ∫
( )jH e ω ( )jdH e ω
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
0
0
H edjωc h
ω ω
( )jrW e ω
– Effect of windowing
( )( )
( )
0
1
2
11
1sin
2sin
2
Mj j n
rn
j M
j
Mj
W e e
ee
M
e
ω ω
ω
ω
ωω
ω
−
=
− +
−
−
=
−=
−+
=
∑
ω ωr
jec h
21
πM +
ΔWMm =
+4
1π
peak side lobe
design paramters
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Common windowing functions
– Bartlett (triangular)
– Hanning
– Hamming
– Blackman
[ ]
2 , 02
22 ,2
0, otherwise
B
n MnM
n Mw n n MM
⎧ ≤ ≤⎪⎪⎪= − ≤ ≤⎨⎪⎪⎪⎩
[ ]20.5 1 cos , 0
0 , otherwisehan
n n Mw n M
π⎧ ⎛ ⎞− ≤ ≤⎪ ⎜ ⎟= ⎝ ⎠⎨⎪⎩
[ ]20.54 0.46cos , 0
0 , otherwiseham
n n Mw n M
π⎧ − ≤ ≤⎪= ⎨⎪⎩
[ ]2 40.42 0.5cos 0.08cos , 0
0BL
n n n Mw n M M
π π⎧ ⎛ ⎞− + ≤ ≤⎪ ⎜ ⎟= ⎝ ⎠⎨⎪⎩
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Trade off between the main lobe width and the peak of the side lobe
Peak side lobe Main-lobe Peak approx.
width error
13 4 /( 1) 21 25 8 / 25
31
MM
ππ
− + −− −−
RectangularBartlettHanning 8 / 44
41 8 / 5357 12 /
MM
M
πππ
−− −− −
Hamming Blackman 74
Rectangular window
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Hanning window
Bartlett window Hamming window
Blackman window
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Kaiser window
where is the zeroth-order modified Bessel function of the first kind defined by
As β = 0 → rectangular window case
If the window is tapered more (i.e., β is increased), the main lobe width becomes
wider and the peak of the side lobe becomes lower
[ ]( )
2
0
0
1
,0 , 2
0 ,otherwise
nIMw n n M
I
αβα
αβ
⎧ ⎡ ⎤−⎡ ⎤⎪ ⎢ ⎥− ⎢ ⎥⎣ ⎦⎪ ⎢ ⎥⎣ ⎦= ⎨ ≤ ≤ =⎪⎪⎩
( )0I ⋅
( ) ( )sin0
1 2 cos for large 2 4
jxI x e d x xx
π θ
π
πθπ π−
⎛ ⎞= ≈ −⎜ ⎟⎝ ⎠
∫
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Increasing M while holding β constant
results in the main lobe width decreased
but does not affect the peak of the side
lobe.
Given is fixed, is defined to be
the highest frequency such that
and to be the lowest frequency such that
Then β can be empirically determined by
δ pω
( ) 1jH e ω δ≥ −
sω
( )jH e ω δ≤
( )( ) ( )0.4
0.1102 8.7 , 50
0.5842 21 0.07886 21 21 500 , 21
8 ; 2.285
, 20logs p
A A
A A AA
AM
A
β
ωω ω ω δ
− >⎧⎪⎪= − + − ≤ ≤⎨⎪ <⎪⎩
−=
ΔΔ = − = −where
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Comparison of commonly used windows
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Design example
( )
1 2
0.4 ; 0.6 0.52
0.20.01; 0.001 0.001
20log 600.1102 8.7 5.65326
8 60 8 36.219 37; 2.285 2.285 0.2
p sp s c
s p
A dBA
AM M
ω ωω π ω π ω π
ω ω ω π
δ δ δδ
β
ω π
+= = ⇒ = =
Δ = − =
= = ⇒ =⇒ = − =
⇒ = − =
− −= = = ⇒ =
Δ ×Type - II FIR
[ ] [ ] [ ]
( )( ) ( )
12 2
0
0
18.55.65326 118.5sin 0.5 18.5 , 0
18.5 5.653260, otherwise
dh n h n n
nIn n M
n I
ω
ππ
⇒ = ⋅
⎧ ⎡ ⎤⎛ ⎞−⎛ ⎞⎪ ⎢ ⎥−⎜ ⎟⎜ ⎟⎪ ⎢ ⎥⎝ ⎠⎪ ⎝ ⎠− ⎢ ⎥= ≤ ≤⎣ ⎦⎨ ⋅⎪ −⎪⎪⎩
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Linear system with generalized linear phase
– For casual system, zero phase is not achievable
Phase distortion exists
Linear phase results in a simple time shift
– System with linear phase
Ideal delay
From the inverse Fourier transform,
For an input
( )( )
( )
( )
,
1
rg
grd
j jid
jid
jid
jid
H e e
H e
A H e
H e
ω ωα
ω
ω
ω
ω π
αω
αω αω
−= <
⇒ =
⎡ ⎤ = −⎣ ⎦∂⎡ ⎤ = − =⎣ ⎦ ∂
[ ] ( )( )
sin,id
nh n n
nπ α
π α−
= −∞ < < ∞−
[ ],x n
[ ] [ ] ( )( )
[ ] ( )( )
sin
sin
k
ny n x n
n
n kx k
n k
π απ α
π απ α
∞
=−∞
−= ∗
−
− −=
− −∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
In particular, if
Ideal LPF with linear phase
If the impulse response is symmetric about i.e.,
[ ] [ ][ ] [ ] [ ] [ ]
, d id d
id d
n h n n n
y n x n h n x n n
α δ= = −
⇒ = ∗ = −
( ) [ ] ( )( )
, sin
0,
jcj c
p pc
e nH e h n
n
ωαω ω ω ω α
π αω ω π
−⎧ < −⎪= ⇒ =⎨ −< ≤⎪⎩
,dnα = ,n α=
[ ] [ ]2p ph n h nα − =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Generalized linear phase
This is a necessary condition for to have a constant group delay
There are other systems having linear phase characteristics without this
symmetry condition
( ) ( )( )
[ ][ ] [ ] ( )sinsin
tan sin 0,cos cos n
h n nh n n
h n nωβ ωα
β ωα ω α β ωβ ωα ω−
− = = − ⇒ − + = ∀⎡ ⎤⎣ ⎦−∑ ∑∑
[ ]h n
[ ] [ ][ ] ( )
[ ] [ ]
0 or , 2 2
= /2 3 / 2, cos 0
2 2
h n h n
h n n
M h n h n
β π α α
β π π ω α
α α
= − =
− =⎡ ⎤⎣ ⎦⇒ = = − = −
∑This is satisfied when is an integer, and
Alternatively, if or
an integer and
( ) ( ) ( )
( ) ( ) ( ) ( )cos sinj j
jj jH e
A e jA
A e e
e
αω βω ω
ω ωβ ωα β ωα
− −
+
=
= − − ⇒ ( ) ( )
( ){ }grd
rg
j
j
H e
d A H ed
ω
ω
τ ω
ωα
⎡ ⎤= ⎣ ⎦
⎡ ⎤= − ⎣ ⎦
=( ) [ ]
[ ] [ ]cos sin
j j n
n
n n
H e h n e
h n n j h n n
ω ω
ω ω
∞−
=−∞
∞ ∞
=−∞ =−∞
=
= −
∑
∑ ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Causal generalized linear phase systems
If causal,
Causal FIR systems of length (M+1) have generalized linear phase, if
If
If
[ ] ( )0
sin 0,n
h n nω α β ω∞
=
− + = ∀⎡ ⎤⎣ ⎦∑
[ ] [ ]
( ) ( )( )
2
, 0
0 , otherwise
j Mj j
e
je
h M n n Mh n
H e A e e
A e
ωω ω
ω ω
−
− ≤ ≤⎧= ⎨⎩
=then
where is a real, even periodic function of
[ ] [ ]
( ) ( ) ( )( )
2 2
, 0
0 , otherwise
j M j Mj j j
o o
jo
h M n n Mh n
H e jA e e A e e
A e
ω ω πω ω ω
ω ω
+− −
− − ≤ ≤⎧= ⎨⎩
= =then
where is a real, odd periodic function of
[ ] [ ] [ ] [ ] 2 or 2h n h n h n h nα α= − = − −
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Type I FIR linear phase systems
where M is an ever integer
where
where
[ ] [ ] 0h n h M n n M= − ≤ ≤
( ) [ ] [ ] [ ]
[ ] [ ] ( )
[ ] ( )( )
[ ]
12
2
0 0 12
1 12 2
2
0 0
12
2
0
12
0
2
2
2
2 cos2 2
MMM Mjj j n j n j n
Mn n n
M MMj j M nj n
n n
MMjj M nj n
n
M
n
MH e h n e h n e h e h n e
Mh n e h e h M n e
Mh n e e h e
M Mh n n h e
ωω ω ω ω
ωωω
ωωω
ω
−−− − −
= = = +
− −− − −−
= =
−−− −−
=
−
=
⎡ ⎤= = + +⎢ ⎥⎣ ⎦
⎡ ⎤= + + −⎢ ⎥⎣ ⎦
⎡ ⎤= + + ⎢ ⎥⎣ ⎦
⎡ ⎤ ⎡ ⎤= − +⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦
∑ ∑ ∑
∑ ∑
∑
∑ [ ]2
2 2
0cos
MM Mj j
ka k k e
ωωω
− −
=
= ⋅∑
[ ] [ ]0 , 2 , 1,2,.... .2 2 2M M Ma h a k h k k⎡ ⎤ ⎡ ⎤= = − =⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦
( ) ( ) 2Mjj j
eH e A e eω
ω ω −⇒ =
( ) [ ]2
0
cos
M
je
k
A e a k kω ω=
=∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Example
[ ]
( )4
2
0
1, 0 40, otherwise
5sin2
sin2
j j n j
n
nh n
H e e eω ω ω
ω
ω− −
=
≤ ≤⎧= ⎨⎩
⇒ = =∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Type II linear phase systems
where is an odd integer
Example
[ ] [ ], 0h n h M n n M= − ≤ ≤
M
( ) [ ]0
12
2
1
1 12 cos2 2
Mj j n
n
MMj
k
H e h n e
Mh k k e
ω ω
ω
ω
−
=
+
−
=
⇒ =
+ ⎡ ⎤⎡ ⎤ ⎛ ⎞= − − ⋅⎜ ⎟⎢ ⎥⎢ ⎥⎣ ⎦ ⎝ ⎠⎣ ⎦
∑
∑
[ ]
( )52
1, 0 50, otherwise
sin3
sin2
jj
nh n
H e eωω ω
ω−
≤ ≤⎧= ⎨⎩
⇒ =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Type III FIR linear- phase systems
where M is an even integer
Example
[ ] [ ], 0h n h M n n M= − − ≤ ≤
( )2
2
1
2 sin2
MMjj
k
MH e j h k k eω
ω ω−
=
⎡ ⎤⇒ = − ⋅ ⋅⎢ ⎥⎣ ⎦∑
[ ] [ ] [ ]( ) 2
2
1 2sinj j j
h n n n
H e e j eω ω ω
δ δ
ω− −
= − −
= − = ⋅
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Type IV FIR linear- phase systems
where is an odd integer
Example
[ ] [ ], 0h n h M n n M= − − ≤ ≤
( )1
22
1
1 12 sin2 2
MMjj
k
MH e j h k k eω
ω ω
+
−
=
+ ⎡ ⎤⎡ ⎤ ⎛ ⎞⇒ = − ⋅ − ⋅⎜ ⎟⎢ ⎥⎢ ⎥⎣ ⎦ ⎝ ⎠⎣ ⎦∑
[ ] [ ] [ ]( ) ( )
2
1
1 1
2sin2
j j j
j
h n n n
H e e e
j e
ω ω ω
ω
δ δ
ω
−
−
= − −
⇒ = + = −
= ⋅
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Location of zeros for FIR linear-phase systems
If are symmetric,
If is a zero of then
That is, is also a zero of
When is real and is a zero of will also be a zero of
and so will
Therefore, when is real, there will be four conjugate reciprocal zeros of the
form
unless
( ) [ ]0
Mn
n
H z h n z−=
=∑[ ]{ }h n
( ) [ ] [ ] ( )1
0
M on k M M
n k M
H z h M n z h k z z H z− − − −
= =
= − = =∑ ∑
0jz re θ= ( ),H z
( ) ( )10 0 0 0MH z z H z− −= =
1 10
jz r e θ− − −=
[ ]h n( ).H z
0z ( ) 0, jH z z re θ∗ −= ( ),H z
( ) 1
0 .z−∗
( )( )( )( )1 1 1 1 1 11 1 1 1j j j jre z re z r e z r z zθ θ θ θ− − − − − − − −− − − −
[ ]h n
1.r =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
If are anti-symmetric, we have
In particular, when
Thus, must have a zero at
It also must have a zero at when is even.
( ) [ ] [ ] ( )1
0
M on k M M
n k M
H z h M n z h k z z H z− − − −
= =
= − − = − = −∑ ∑
[ ]{ }h n
( ) ( ) ( ) ( ) ( )11, 1 1 1, 1 1 1 .Mz H H z H H+= = − = − − = − −and if
( )H z 1.z =
1z ω π= − ⇔ = M
type - II type - III type - IV
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Relation of FIR linear-phase systems to minimum-phase systems
– Any FIR linear-phase system can be represented by
where
and has only zeros on the unit circle.
⇒ The order of
( ) ( ) ( )1max min min, iM
iH z H z z M H z−−= = no.of zeros of
( )ucH z
( ) ( ) ( ) ( )min max ucH z H z H z H z=
0M
0 2 iM M M+is
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Example: Decomposition of a linear-phase system
The overall system has linear phase( ) ( ) ( )min maxH z H z H z=
( )( )( )( )( ) ( )( )( )( )
2 0.6 1 0.6 1 0.8 1 0.8 1min
2 0.6 1 0.6 1 0.8 1 0.8 1max
( ) 1.25 1 0.9 1 0.9 1 0.8 1 0.8
( ) 0.9 1 1.1111 1 1.1111 1 1.25 1 1.25
j j j j
j j j j
H z e z e z e z e z
H z e z e z e z e z
π π π π
π π π π
− − − − − −
− − − − − −
= − − − −
⇒ = − − − −
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Causal and linear phase FIR filter
The impulse response of a desired filter is also symmetric;
The overall response
The resulting response has generalized linear phase and its magnitude is also an
even real function.
[ ] [ ] ( ) ( ) 20
0
Mjj je
w M n n Mw n W e W e e
ωω ω −− ≤ ≤⎧= ⇒ =⎨⎩
ωeven function of [ ] [ ]d dh n h M n= −
( ) ( ) 2Mjj j
d eH e H e eω
ω ω −=
( ) ( ) ( )( )
( ) ( )( )( )
( ) ( )( )
( )
( ) ( ) ( )( )
2 2
2
2
121
21
2
12
jj jd
MMj jjje e
Mjjje e
Mjje
jj je e e
H e H e W e d
H e e W e e d
H e W e e d
A e e
A e H e W e d
ω θω θ
ω θθω θθ
ωω θθ
ωω
π ω θω θ
π
θπ
θπ
θπ
θπ
−
−− −−
−−
−
−
−
=
=
=
=
=
∫
∫
∫
∫where
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Example Linear-phase lowpass filter
[ ] [ ] [ ]dh n h n w n=[ ] [ ] [ ]dh n h n nω=
( ) [ ]2
21, 20 ,
sin2
2
c
c
Mj Mjj j ncd d
c
c
eH e h n e e d
Mn
Mn
ωω
ωω ω
ω
ω ω ωπω ω π
ω
π
−−
−
⎧⎪ <= ⇒ =⎨< ≤⎪⎩
⎡ ⎤⎛ ⎞−⎜ ⎟⎢ ⎥⎝ ⎠⎣ ⎦=⎛ ⎞−⎜ ⎟⎝ ⎠
∫
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Filter design by Kaiser method
High-pass filter
• Design example:
( )
( ) ( )
[ ]
2
2
0 0
sin sin2 2
2 2
cj
Mhp j
c
Mjj jhp ep
c
hp
H ee
H e e H e
M Mn nh n
M Mn n
ωω
ωω ω
ω ω
ω ω π
π ω
π π
−
−
≤ ≤⎧⎪= ⎨≤ ≤⎪⎩
= −
⎛ ⎞ ⎛ ⎞− −⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠⇒ = −
⎛ ⎞ ⎛ ⎞− −⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠
( ) ( )
1 2
0.4
0.35 , 0.5 0.02133.56
0.5842 21 0.07886 212.5974
33.56 8 23.73 242.285 0.15
s p
A
A A
M M
ω π ω π δ δ δ
β
π
= = = = =
⇒ =
⇒ = − + −
=−
= = ⇒ =×
If and
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
High-pass filter example
Actual error=0.0213>0.021
33.562.597424
A
Mβ=== ⇒ type I FIR
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
High-pass Kaiser high-pass filter example with M=25 (type-II)
• Use of a type-II Kaiser HPF
may not be appropriate
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– FIR filter with generalized response
– Differentiator
Kaiser formulae were developed for frequency response with simple magnitude
discontinuities, but still works on differentiator
G1
G2
G4
G3
0 ω 1 ω 2 ω 3 π
4N =[ ] ( )
[ ] [ ] [ ]
11
1
sin2
20
kN
d k kk
N
d
Mnh n G G
Mn
Gh n h n n
ω
π
ω
+=
+
⎛ ⎞−⎜ ⎟⎝ ⎠⇒ = −
⎛ ⎞−⎜ ⎟⎝ ⎠
=
⇒ =
∑
where
( )
[ ]
2
2
cos sin2 2
2 2
Mjjd
d
H e j e
M Mn nh n M Mn n
ωω ω π ω π
π π
π
−= − < <
⎡ ⎤ ⎡ ⎤⎛ ⎞ ⎛ ⎞− −⎜ ⎟ ⎜ ⎟⎢ ⎥ ⎢ ⎥⎝ ⎠ ⎝ ⎠⎣ ⎦ ⎣ ⎦⇒ = −⎛ ⎞− −⎜ ⎟⎝ ⎠
[ ] [ ] [ ][ ] [ ]
1, dM h n h n n
h M n h n
ω+ = ⋅
− = −
In the case of using a symmetric window of size results in
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Kaiser window design of a differentiator
The amplitude error:
By not imposing constraint H(z) to have a zero at z=-1, it is possible to design a
filter having better approximation to the desired response, while using a less
number of filter taps.
( ) ( )( ) ( )
; 0
1
jo
jo
E A e
A e H z z
ω
ω
ω ω ω π= − ≤ ≤
= ±where is the amplitude of the filter and has zeros at
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
Example: Impulse response of Kaiser-windowed differentiatore
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Optimum approximation of FIR filters
– Design a filter that best approximates the desired response for a given M
– Rectangular window provides the best-mean squared approximation:
minimizes
Do not permit individual control errors in
different bands
[ ] [ ], 0
0 , otherwisedh n n M
h n≤ ≤⎧
= ⎨⎩
[ ][ ] [ ][ ]
; 0
; otherwiswd
d
h n h n n Me n
h n
− ≤ ≤⎧⎪⎨⎪⎩
( ) ( )
[ ]
22
2
12
j jd
n
H e H e d
e n
π ω ω
πε ω
π −
∞
=−∞
= −
=
∫
∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Consider the design of a type-I FIR filter with zero phase, i.e.,
– Corresponding frequency response
is a real, even and periodic function of
A causal system can be obtained by delaying it by L samples
– From a polynomial approximation
can be represented as an L-th degree trigonometric polynomial
– Define an error function by
where is a weighting function
[ ] [ ] [ ]2e eh n h n L h L n− = −
( ) [ ] [ ] [ ] ( )0
0 2 cosL L
j j ne e e e
n L n
A e h n e h h n nω ω ω−
=− =
= = +∑ ∑
( ) ( ) ( )1cos cos ; cos cosn nn T T x n xω ω −= =
( )jA e ω
[ ] [ ]e eh n h n= −
( )W ω
ω
( ) ( ) ( ) ( ) 2j jdE W H e H eω ωω ω= −
( ) ( ) ( )cos
0 0
cos ; L L
j k kk kx
k k
A e a P x P x a xωω
ω=
= =
= = =∑ ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Example
Maximum weighted absolute approximation error
is
( )1, 0
0,
pjd
s
H e ωω ω
ω ω π
≤ ≤⎧= ⎨
≤ ≤⎩
2δ δ=
( ) 1
2
0
1,
1
; ;p
s
W KKω ω
ω ω
δωπ δ
≤ ≤
≤
⎧⎪= =⎨⎪⎩ ≤
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Minimax or Chevyshev criterion
Find a set of impulse response that minimizes
– Alternative theorem:
Let denote a closed subset comprising the disjoint union of closed subsets of real
axis x. Then
is an m-th order polynomial
For desired function and positive function which are continuous
on , and weighted error defined by
a necessary and sufficient condition that be a unique r-th order polynomial
minimizing is that has at least alternations
At least values in such that and
( )PE x
{ }0 , : p sF ω ω πω ω ω= ≤ ≤ ≤ ≤[ ]{ }( );0min max ( )
eh n n L FE
ωω
≤ ≤ ∈
( )PW x( )PD x
2r +
δ
( )maxP
Px FE E x
∈=
( )0
mk
kk
P x a x=
=∑
PF
( ) ( ) ( ) ( )P P PE x W x D x P x= −⎡ ⎤⎣ ⎦
1 2 2rx x x +< < <( ) ( )1 , 1,2, , 2P i P iE x E x E i r+= − = ± = +
ix2r + PF
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Example: Consider 5-th order polynomials that
approximate unity for and zero for
Assume that for these two regions
The optimum 5-th order polynomial has at least 7
alternations of the error in the region in
has 3 alternations in and
2 alternations in
also has 5 alternations; 3 in
and 2 in
has 8 alternations in and
optimum polynomial
( )1P x
( )3P x
( ) 1PW x =
0.1 1x≤ ≤
0.1 1x≤ ≤1 0.1x− ≤ ≤ −
( )2P x
( )jP x
PF
1 0.1x− ≤ ≤ −
0.1 1 x≤ ≤ ⇒
1 0.1x− ≤ ≤ −
1 0.1x− ≤ ≤ −
0.1 1x≤ ≤
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Optimum type-I lowpass filters
– Consider polynomial defined by
– Assume that
– Equi-ripple approximation
A set of coefficients is designed to make have at least
alternations on
{ }ka ( )2L +
PF
( )1/ ; cos cos 1
cos0; 1 cos cos
pP
s
KW
ω ωω
ω ω
≤ ≤⎧= ⎨
− ≤ ≤⎩
( )1; cos cos 1
cos0; 1 cos cos
pP
s
Dω ω
ωω ω
≤ ≤⎧= ⎨
− ≤ ≤⎩
( ) ( )0
cos cosL
kk
kP aω ω
=
=∑ 7L =
( )P x
{ }: 0 , P p sF ω ω ω πω ω= ≤ ≤ ≤ ≤
( )cosPE ω
( ) ( ) ( ) ( )cos cos cos cosP P PE W D Pω ω ω ω= −⎡ ⎤⎣ ⎦
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Optimum type-I lowpass filters (cont.)
– Redraw in terms of
7L =
( )jA e ω cosx ω=
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Optimum type-I lowpass filters (cont.)
– Max. possible number of alternations of the error
is since an L-th degree polynomial can
have at most points with zero slope
Has zeros at and roots of
-th order polynomial
– Alternations always occur at
– The filter will be equi-ripple except possibly at
( )1L −
( )1L −
p sω ω ω ω= =and
( ) ( )
( ) ( )
1
0
1
10
cos sin cos
sin 1 cos
Lk
kk
Lk
kk
P ka
k a
ω ω ωω
ω ω
−
=
−
+=
∂= −
∂
= − +
∑
∑
( )3L +
0 ω ω π= =or
0, ω ω π= =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Parks-McClellan algorithm
– The alternation theorem indicates that the optimum filter will satisfies
– can be rewritten as
– Solution for can be found by means of polynomial interpolation
( ) [ ] [ ] [ ] ( )0 0 0
0 2 cos cosL L L L
j j n k ke e e e k k
n L n k k
A e h n e h h n n a a xω ω ω ω−
=− = = =
= = + = =∑ ∑ ∑ ∑
( )ijdH e ω
ka δand
( )jeA e ω
( )( )
( )
( )
( )
( )( )
1
2
2
21 1 1
02
2 2 2 1
22
2
1
2 2
2
2
11
11
11
L
L
L
LL
L L L
jd
jd
jd
L
WH e
H e W
x x x
ax x x a
x x xH e
W
ω
ω
ω
ω
δ
ω
ω
+ +
+ + ++
⎡ ⎤⎢ ⎥
⎡ ⎤ ⎢ ⎥ ⎡⎢ ⎥ ⎢ ⎥− ⎢⎢ ⎥ ⎢ ⎥ ⎢=⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎣⎢ ⎥ −⎢ ⎥⎣ ⎦
⎢ ⎥⎣ ⎦
⎤⎥⎥
⎢ ⎥⎢ ⎥
⎦
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
1
1
1 ; 1,2, , 2
1 /
i i
i i
ij ji d e
ij jd e i
W H e A e i L
H e A e W
ω ω
ω ω
ω δ
δ ω
+
+
⎡ ⎤− = − = +⎣ ⎦
⇒ = + −
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Parks-McClellan algorithm (cont.)
– For a given set of extremal frequencies, the P-M algorithm determines
– For given , the error function has magnitude at frequencies
– has values for and for
– An L-th order trigonometric polynimial can be obtained by interpolating ( )iE ω
( ) ( )( ) ( )
1
21
11cos , , 1
k
L
k k Li k ii k
kj
k d kk
b xx C xx
ex
H dW
ω δω
ω =≠
+ +
+= −−
−= = − =∏
( ) ( )
1
11
1
cos
Lk k
j k ke L
k
k k
d Cx xA e P
dx x
ω ω
+
=+
=
−= =
−
∑
∑
2
1
1L
ki k ii k
bx x
+
=≠
=−∏
( )jeA e ω
iω
sω ω π≤ ≤δ±
ka δand
( )( )( )
2
112
1
;1
k
Lj
k dk
kLk
k k
b H e
bW
ω
δ
ω
+
=++
=
=−
∑
∑
( )jeA e ω
1 Kδ±
δ± 2L +
0 pω ω≤ ≤
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Parks-McClellan algorithm (cont.)
Example: 7 L =
( )jeA e ω
ka δand
1 Kδ±
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Characteristics of optimum FIR filters
– Size of the filter tap( )( )
1 210log 132.324 s p
Mδ δω ω
− −=
−
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Filter design examples by Parks-McClellan algorithm
26M = 27M =
0.4 ; 0.6p sω π ω π= =
2 31 210 ; 10δ δ −= =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Filter design example (cont.)
– Compensation of zero-order holder
28 14M L= ⇒ =
( ) ( )/ 2 ; 0
sin / 20;
pjd
s
H e ω
ω ω ωω
ω ω π
⎧ ≤ ≤⎪= ⎨⎪ ≤ ≤⎩
2 31 210 ; 10 ; 0.4 ; 0.6p sδ δ ω π ω π−= = = =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Optimum bandpass filter design
– Local extrema may occur in the transition
regions
– The approximation need not be equi-ripple
in the approximation regions
( )
( )
0; 0 0.31; 0.35 0.30; 0.7
1; 0 0.31; 0.35 0.30.2; 0.7
jd
j
H e
W e
ω
ω
ω ππ ω ππ ω π
ω ππ ω ππ ω π
≤ ≤⎧⎪= ≤ ≤⎨⎪ ≤ ≤⎩
≤ ≤⎧⎪= ≤ ≤⎨⎪ ≤ ≤⎩
74 37M L= ⇒ =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Optimum approximation of FIR filters
– Design a filter that best approximates the desired response for a given M
– Rectangular window provides the best-mean squared approximation:
minimizes
Do not permit individual control errors in
different bands
[ ] [ ], 0
0 , otherwisedh n n M
h n≤ ≤⎧
= ⎨⎩
[ ][ ] [ ][ ]
; 0
; otherwiswd
d
h n h n n Me n
h n
− ≤ ≤⎧⎪⎨⎪⎩
( ) ( )
[ ]
22
2
12
j jd
n
H e H e d
e n
π ω ω
πε ω
π −
∞
=−∞
= −
=
∫
∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Consider the design of a type-I FIR filter with zero phase, i.e.,
– Corresponding frequency response
is a real, even and periodic function of
A causal system can be obtained by delaying it by L samples
– From a polynomial approximation
can be represented as an L-th degree trigonometric polynomial
– Define an error function by
where is a weighting function
[ ] [ ] [ ]2e eh n h n L h L n− = −
( ) [ ] [ ] [ ] ( )0
0 2 cosL L
j j ne e e e
n L n
A e h n e h h n nω ω ω−
=− =
= = +∑ ∑
( ) ( ) ( )1cos cos ; cos cosn nn T T x n xω ω −= =
( )jA e ω
[ ] [ ]e eh n h n= −
( )W ω
ω
( ) ( ) ( ) ( ) 2j jdE W H e H eω ωω ω= −
( ) ( ) ( )cos
0 0
cos ; L L
j k kk kx
k k
A e a P x P x a xωω
ω=
= =
= = =∑ ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Example
Maximum weighted absolute approximation error
is
( )1, 0
0,
pjd
s
H e ωω ω
ω ω π
≤ ≤⎧= ⎨
≤ ≤⎩
2δ δ=
( ) 1
2
0
1,
1
; ;p
s
W KKω ω
ω ω
δωπ δ
≤ ≤
≤
⎧⎪= =⎨⎪⎩ ≤
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Minimax or Chevyshev criterion
Find a set of impulse response that minimizes
– Alternative theorem:
Let denote a closed subset comprising the disjoint union of closed subsets of real
axis x. Then
is an m-th order polynomial
For desired function and positive function which are continuous
on , and weighted error defined by
a necessary and sufficient condition that be a unique r-th order polynomial
minimizing is that has at least alternations
At least values in such that and
( )PE x
{ }0 , : p sF ω ω πω ω ω= ≤ ≤ ≤ ≤[ ]{ }( );0min max ( )
eh n n L FE
ωω
≤ ≤ ∈
( )PW x( )PD x
2r +
δ
( )maxP
Px FE E x
∈=
( )0
mk
kk
P x a x=
=∑
PF
( ) ( ) ( ) ( )P P PE x W x D x P x= −⎡ ⎤⎣ ⎦
1 2 2rx x x +< < <( ) ( )1 , 1,2, , 2P i P iE x E x E i r+= − = ± = +
ix2r + PF
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
– Example: Consider 5-th order polynomials that
approximate unity for and zero for
Assume that for these two regions
The optimum 5-th order polynomial has at least 7
alternations of the error in the region in
has 3 alternations in and
2 alternations in
also has 5 alternations; 3 in
and 2 in
has 8 alternations in and
optimum polynomial
( )1P x
( )3P x
( ) 1PW x =
0.1 1x≤ ≤
0.1 1x≤ ≤1 0.1x− ≤ ≤ −
( )2P x
( )jP x
PF
1 0.1x− ≤ ≤ −
0.1 1 x≤ ≤ ⇒
1 0.1x− ≤ ≤ −
1 0.1x− ≤ ≤ −
0.1 1x≤ ≤
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Optimum type-I lowpass filters
– Consider polynomial defined by
– Assume that
– Equi-ripple approximation
A set of coefficients is designed to make have at least
alternations on
{ }ka ( )2L +
PF
( )1/ ; cos cos 1
cos0; 1 cos cos
pP
s
KW
ω ωω
ω ω
≤ ≤⎧= ⎨
− ≤ ≤⎩
( )1; cos cos 1
cos0; 1 cos cos
pP
s
Dω ω
ωω ω
≤ ≤⎧= ⎨
− ≤ ≤⎩
( ) ( )0
cos cosL
kk
kP aω ω
=
=∑ 7L =
( )P x
{ }: 0 , P p sF ω ω ω πω ω= ≤ ≤ ≤ ≤
( )cosPE ω
( ) ( ) ( ) ( )cos cos cos cosP P PE W D Pω ω ω ω= −⎡ ⎤⎣ ⎦
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Optimum type-I lowpass filters (cont.)
– Redraw in terms of
7L =
( )jA e ω cosx ω=
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Optimum type-I lowpass filters (cont.)
– Max. possible number of alternations of the error
is since an L-th degree polynomial can
have at most points with zero slope
Has zeros at and roots of
-th order polynomial
– Alternations always occur at
– The filter will be equi-ripple except possibly at
( )1L −
( )1L −
p sω ω ω ω= =and
( ) ( )
( ) ( )
1
0
1
10
cos sin cos
sin 1 cos
Lk
kk
Lk
kk
P ka
k a
ω ω ωω
ω ω
−
=
−
+=
∂= −
∂
= − +
∑
∑
( )3L +
0 ω ω π= =or
0, ω ω π= =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Parks-McClellan algorithm
– The alternation theorem indicates that the optimum filter will satisfies
– can be rewritten as
– Solution for can be found by means of polynomial interpolation
( ) [ ] [ ] [ ] ( )0 0 0
0 2 cos cosL L L L
j j n k ke e e e k k
n L n k k
A e h n e h h n n a a xω ω ω ω−
=− = = =
= = + = =∑ ∑ ∑ ∑
( )ijdH e ω
ka δand
( )jeA e ω
( )( )
( )
( )
( )
( )( )
1
2
2
21 1 1
02
2 2 2 1
22
2
1
2 2
2
2
11
11
11
L
L
L
LL
L L L
jd
jd
jd
L
WH e
H e W
x x x
ax x x a
x x xH e
W
ω
ω
ω
ω
δ
ω
ω
+ +
+ + ++
⎡ ⎤⎢ ⎥
⎡ ⎤ ⎢ ⎥ ⎡⎢ ⎥ ⎢ ⎥− ⎢⎢ ⎥ ⎢ ⎥ ⎢=⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎣⎢ ⎥ −⎢ ⎥⎣ ⎦
⎢ ⎥⎣ ⎦
⎤⎥⎥
⎢ ⎥⎢ ⎥
⎦
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
1
1
1 ; 1,2, , 2
1 /
i i
i i
ij ji d e
ij jd e i
W H e A e i L
H e A e W
ω ω
ω ω
ω δ
δ ω
+
+
⎡ ⎤− = − = +⎣ ⎦
⇒ = + −
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Parks-McClellan algorithm (cont.)
– For a given set of extremal frequencies, the P-M algorithm determines
– For given , the error function has magnitude at frequencies
– has values for and for
– An L-th order trigonometric polynimial can be obtained by interpolating ( )iE ω
( ) ( )( ) ( )
1
21
11cos , , 1
k
L
k k Li k ii k
kj
k d kk
b xx C xx
ex
H dW
ω δω
ω =≠
+ +
+= −−
−= = − =∏
( ) ( )
1
11
1
cos
Lk k
j k ke L
k
k k
d Cx xA e P
dx x
ω ω
+
=+
=
−= =
−
∑
∑
2
1
1L
ki k ii k
bx x
+
=≠
=−∏
( )jeA e ω
iω
sω ω π≤ ≤δ±
ka δand
( )( )( )
2
112
1
;1
k
Lj
k dk
kLk
k k
b H e
bW
ω
δ
ω
+
=++
=
=−
∑
∑
( )jeA e ω
1 Kδ±
δ± 2L +
0 pω ω≤ ≤
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Parks-McClellan algorithm (cont.)
Example: 7 L =
( )jeA e ω
ka δand
1 Kδ±
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Characteristics of optimum FIR filters
– Size of the filter tap( )( )
1 210log 132.324 s p
Mδ δω ω
− −=
−
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Filter design examples by Parks-McClellan algorithm
26M = 27M =
0.4 ; 0.6p sω π ω π= =
2 31 210 ; 10δ δ −= =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Filter design example (cont.)
– Compensation of zero-order holder
28 14M L= ⇒ =
( ) ( )/ 2 ; 0
sin / 20;
pjd
s
H e ω
ω ω ωω
ω ω π
⎧ ≤ ≤⎪= ⎨⎪ ≤ ≤⎩
2 31 210 ; 10 ; 0.4 ; 0.6p sδ δ ω π ω π−= = = =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
5. Filter design techniques
• Optimum bandpass filter design
– Local extrema may occur in the transition
regions, which may not be acceptable
– The approximation need not be equi-ripple
in the approximation regions
( )
( )
0; 0 0.31; 0.35 0.30; 0.7
1; 0 0.31; 0.35 0.30.2; 0.7
jd
j
H e
W e
ω
ω
ω ππ ω ππ ω π
ω ππ ω ππ ω π
≤ ≤⎧⎪= ≤ ≤⎨⎪ ≤ ≤⎩
≤ ≤⎧⎪= ≤ ≤⎨⎪ ≤ ≤⎩
74 37M L= ⇒ =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Discrete Fourier transform• Discrete Fourier series representation of periodic sequences
– Let be a periodic sequence with period N such that for any
integer m
Example A periodic complex exponential
– A set of N periodic complex exponentials defines all the
frequency components
– Fourier series representation
[ ]x n
[ ]{ }, 0,1,2,... 1,ke n k N= −
[ ] [ ] k k Ne n e n+=Note that for any integer
[ ] [ ]x n x n mN= +
[ ] [ ]2
; knj
Nk ke n e e n mN k
π
= = + is an interger
[ ] [ ]
[ ] [ ]
21
0
21
0
1
knN jN
k
mnN jN
n
x n X k eN
X k x n e
π
π
−
=
− −
=
⇒ =
=
∑
∑
[ ] [ ]21
0
1 knN jN
k
x n X k eN
π−
=
= ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
The Fourier series coefficient can be obtained by
The Fourier series coefficients is a periodic sequence
– Discrete Fourier series (DFS)
Define by
[ ] [ ]
[ ]( )
( )
2 2 21 1 1
0 0 0
21 1
0 0
21
0
1
1
1,10, otherwise
rn kn rnN N Nj j jN N N
n n k
k r nN N jN
k k
k r nN jN
n
x n e X k e eN
X k eN
k r Ne
N
π π π
π
π
− − −− −
= = =
−− −
= =
−−
=
=
=
− =⎧= ⎨⎩
∑ ∑ ∑
∑ ∑
∑Since
[ ] [ ]21
0
rnN jN
nx n e X r
π−
=
=∑
[ ] [ ]DFSx n X k←⎯⎯→
[ ] [ ]
[ ] [ ]
21
0
1
0
;
1
N jkn NN N
n
Nkn
Nk
X k x n w w e
x n X k wN
π− −
=
−−
=
= =
=
∑
∑
Analysis equation :
Synthesis equation :
[ ] [ ]
[ ] [ ] [ ] [ ]
21
0
2 ( ) 21 12
0 0
knN jN
n
k N n knN Nj j j nN N
n n
X k x n e
X k N x n e x n e e X k
π
π ππ
− −
=
+− −− − −
− −
=
+ = = =
∑
∑ ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example Periodic pulse train
Example Duality in DFS
Let the DFS coefficient be a periodic impulse train
[ ] [ ]
[ ] [ ]
[ ] [ ]
[ ] [ ]
1
0
21 1
0 0
1, 0, otherwise
, 0 1
1
1 1 1
r
Nkn
Nn
N N j knkn kn NN N
k k
n mMx n n mN
x n n n N
X k n w
x n X k w w eN N N
π
δ
δ
δ
∞
=−∞
−−
=
− −− −
= =
=⎧= − = ⎨
⎩= ≤ ≤ −
= =
⇒ = = =
∑
∑
∑ ∑ ∑
Since
[ ] [ ] [ ] [ ] Y k Nx n y n X n= ↔ =
[ ] [ ]1
0
01,
Nkn
N Nk
y n N k w w nδ−
− −
=
⇒ = = = ∀∑
[ ] [ ]r
Y k N k rNδ∞
=−∞
= −∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example Periodic rectangular pulse train
Any periodic sequence can be represented as a sum of complex exponential
sequences
[ ]2 454 4
1010 1010
0 0 10
sin1 21 sin
10
kkj kn jnkk
n n
kWX k W e ekW
π ππ
π− −
− =
−= = = =
−∑ ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
• Properties of DFS
– Linearity
– Shift of a sequence
– Duality
– Symmetry properties
[ ] [ ] [ ] [ ]1 2DFSax n bx n aX k bX k+ ←⎯⎯→ +
[ ] [ ] ( ) [ ] [ ]
[ ] [ ][ ] [ ]
2 2 2 21 1
0 0
N Nj kn j n m j j km kmN N N NN
n n
DFS kmN
DFSnN
x n m e x n m e e X k e X k W
x n m X k W
x n W X k
π π π π− −− − − − −
= =
−
− = − = =
⇒ − ←⎯⎯→
←⎯⎯→ −
∑ ∑
[ ] [ ] [ ] [ ]
[ ] [ ]
[ ] [ ] [ ] [ ]
1 1
1
0
0 0
1
,
Nkn
Nk
Nkn
Nk
DFS DF
k
S
Nkn
N Nx n k W
Nx
x n X k
k x n W
x n k X n Nx
XN
k
w
X
−
=
−
−−
=
=
− =
⇒ − =
←⎯⎯→ ←⎯⎯→
= ⇒
−
∑ ∑
∑If
[ ] [ ][ ] [ ]
* *
* *
DFS
DFS
x n X k
x n X k
←⎯⎯→ −
− ←⎯⎯→
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
⇒ If is real,
– Periodic convolution
Duality
Difference between aperiodic convolution and periodic convolution
• The sum is over an interval
•
[ ] [ ][ ] [ ]
*X k X k
X k X k
= −
= −
[ ]x n
[ ] [ ] [ ] [ ] [ ]
[ ] [ ] [ ]
[ ] [ ]
[ ] [ ] [ ] [ ]
1 1
3 1 2 2 10 0
1 1
3 1 20 0
1 1
1 20 0
1
2 1 1 20
N N
m m
N Nkn
Nn m
N Nkn
Nm n
Nkm
Nm
x n x m x n m x m x n m
X k x m x n m W
x m x n m W
X k x m W X k X k
− −
= =
− −
= =
− −
= =
−
=
= − = −
⎛ ⎞= −⎜ ⎟
⎝ ⎠
= −
= =
∑ ∑
∑ ∑
∑ ∑
∑
0 1m N≤ ≤ −
[ ] ( )2 2x n m x m n− = − −⎡ ⎤⎣ ⎦
[ ] [ ] [ ] [ ]1
1 2 1 20
NDFS
mx m x n m X k X k
−
=
⇒ − ←⎯⎯→∑
[ ] [ ] [ ] [ ]1
1 2 1 20
1 NDFS
m
x n x n X m X k mN
−
=
←⎯⎯→ −∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example Periodic convolution
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Properties of DFS
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Properties of DFS (cont.)
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
• Fourier transform (F.T.) of periodic signals
– A sequence should be absolutely summable to guarantee the uniform convergence of
its Fourier transform
– Periodic signals are not absolutely summable
– The F.T. of a periodic signal can be interpreted as an impulse train in the frequency
domain with impulse values proportional to the DFS coefficients for the sequence
– If is periodic with period N, the Fourier transform of is defined by
Although the F.T. of a periodic sequence does not converge in the normal sense,
the introduction of impulses enables to use the F. T.
Note: has a necessary periodicity with period
[ ] ( ) [ ]2 2j
k
kx n X e X kN N
ω π πδ ω∞
=−∞
⎛ ⎞⇔ = −⎜ ⎟⎝ ⎠
∑
[ ]x n[ ]x n
( ) [ ]
[ ]
[ ] [ ]
2 2
0 0
2
0
2
1 1 2 2 ; 0 2 /2 2
1 2
1
j j n j n
k
j n
k
j knN
k
kX e e d X k e d NN N
kX k e dN N
X k e x nN
π ε π εω ω ω
ε ε
π ε ω
ε
π
π πω δ ω ω ε ππ π
πδ ω ω
− −
− −
∞ −
−=−∞
∞
=−∞
⎛ ⎞= − < <⎜ ⎟⎝ ⎠
⎛ ⎞= −⎜ ⎟⎝ ⎠
= =
∑∫ ∫
∑ ∫
∑
( )jX e ω 2π
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example: Fourier transform of a periodic impulse train
– For a finite-length sequence
Thus, is obtained by sampling the Fourier transform of a finite sequence
[ ] [ ] [ ]
[ ] [ ] [ ]
( ) ( ) ( )
( )2
2 2
2 2
j j j
j
k
kjN
k
m m
x n x n p n
X e X e P e
kX e
mN
N N
kX eN
x n x n mN
N
ω ω ω
ω
π
π πδ
ω
δ
ω
π πδ
∞ ∞ ∞
=−∞ =−∞ =−
∞
=−∞
∞
=−
∞
∞
= ∗
=
⎛ ⎞= −⎜ ⎟⎝ ⎠
⎛ ⎞ ⎛ ⎞= −
− −
⎜ ⎟ ⎜ ⎟⎝ ⎠⎝
−
⎠
= =∑ ∑ ∑
∑
∑
[ ] [ ] [ ]
( )
1,
2
2m
j
k
p n n mN P k k
kP e
NNω
δ
πδ ω
π
∞
=−∞
∞
=−∞
= − ⇒ = ∀
⇒ = −⎡ ⎤⎢ ⎥⎣ ⎦
∑
∑[ ] [ ]( ) , 0 for 0 and x n x n n n N= < ≥., i.e
[ ] ( ) [ ]2 21
20
Nj k j kj N Nk nN
X k X e X e x n eπ π
ωπω
− −
= =
⎛ ⎞⇒ = = =⎜ ⎟
⎝ ⎠∑
[ ] x n[ ] X k
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example: Relationship between the DFS coefficients and the FT of one period
[ ] 1; 0 40;
x nn≤ ≤⎧
= ⎨⎩ otherwise
( ) ( )( )
42
0
sin 5 / 2sin / 2
j j n j
nX e e eω ω ωω
ω− −
=
⇒ = =∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
• Sampling the Fourier transform
– Since is periodic with period is also periodic in k with period N.
– Since
The samples of the F.T. of can be thought of as DFS coefficients of a periodic sequence obtained through periodic replication of
( )jX e ω [ ]2 , X kπ
[ ] ( ) [ ]
[ ] [ ]
[ ] ( )
[ ] [ ]
[ ] [ ]
[ ]
2 21
20
2 21
0
21
0
,
1
1
Nj k j kj N Nk nN
N j mk j knN N
k m
N j k m nN
m k
m
X k X e X e x n e
x n x m e eN
x m eN
x m n m N
x n n N
x n N
π πω
πω
π π
π
δ
δ
− −
==
− ∞ −
= =−∞
∞ − −
=−∞ =
∞ ∞
=−∞ =−∞
∞
=−∞
∞
=−∞
⎛ ⎞= = =⎜ ⎟
⎝ ⎠⎡ ⎤
= ⎢ ⎥⎣ ⎦
=
= − −
= ∗ −
= −
∑
∑ ∑
∑ ∑
∑ ∑
∑
∑
[ ]x n[ ]x n[ ]x n
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– A periodic sequence can be interpreted as equally spaced samples of the Fourier transform of one period of
Example
[ ] [ ]
( ) [ ] [ ]
[ ] ( )
1 1
0 0
2
, 0 10 , otherwise
2
N Nj j n j n
n n
jk
N
x n n Nx n
X e x n e x n e
X k X e
ω ω ω
ωπω
πω
− −− −
= =
=
⎧ ≤ ≤ −= ⎨⎩
= =
⇒ = ⇐
∑ ∑
Sampling of the Fourier transform at a rate of
[ ]
( ) [ ] ( )( )
42 4 /10
0
1, 0 40, otherwise
5sin sin / 22sin /10sin
2
j j n j j k
n
nx n
kX e e e X k e
kω ω ω π
ωπ
ω π− − −
=
≤ ≤⎧= ⎨⎩
= = ⇒ =∑
[ ]x n[ ]X k
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
• Fourier representation of finite-duration sequences: Discrete Fourier transform (DFT)
– can be visualized as warping the finite sequence around the cylinder
– Since the Fourier series coefficients of is also a periodic sequence with
a period of N, the DFT of can be obtained by
– Since
[ ] [ ] [ ] [ ] ( )( )
( )( ) ( )
, 0 1
0 , otherwise
modulo
N
N
x n n Nx n x n x n N x n
x n x n N
∞
=−∞
⎧ ≤ ≤ −⎡ ⎤= ⇔ = − =⎨ ⎣ ⎦
⎩⎡ ⎤ ⎡ ⎤⎣ ⎦⎣ ⎦
∑
where
[ ]x n
[ ] [ ]
[ ] ( )( ) [ ] [ ]
, 0 10 , otherwise
Nm
X k k NX k
X k X k X k X k mN∞
=−∞
⎧ ≤ ≤ −⎪= ⎨⎪⎩
⎡ ⎤= ⇔ = −⎣ ⎦ ∑
[ ]x n[ ]X k[ ]x n
[ ] [ ] [ ] [ ]
[ ] [ ] [ ] [ ]
1 1
0 0
1 1
0 0
1 ,
1, 0 1 , 0 1
0 , otherwise 0 , otherwise
N Nnk nk
N Nn k
N Nnk nk
N Nk k
X k x n W x n X k WN
x n W n N X k W n NX k x n N
− −−
= =
− −−
= =
= =
⎧ ⎧≤ ≤ − ≤ ≤ −⎪ ⎪= =⎨ ⎨
⎪ ⎪⎩ ⎩
∑ ∑
∑ ∑
and
and
[ ] [ ]DFTx n X k←⎯⎯→
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Example
N=5;
[ ]2 21
20
, 0, , 2 ,10, otherwise
1
j kN jN
kjn N
N k N NeX k ee
π π
π
−− −
−=
= ± ± ⋅ ⋅ ⋅⎧−= = = ⎨
⎩−∑
[ ]X k [ ]X k
[ ]x n
[ ]x n
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Example: N=10
x n⎡ ⎤⎣ ⎦
x n⎡ ⎤⎣ ⎦
[ ]X k
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
• Properties of the DFS
– Linearity
– Circular shift
Let
Since
[ ] [ ] [ ] [ ] [ ] [ ]1 2 1 2
1 2 1, 2
a
m x
DFT DFTx n x n X k X k
N N N
x n X k
N N
α β α β⇒ + +
⎡ ⎤≥ ⎣ ⎦
←⎯⎯→ ←⎯⎯→
with zero padding
[ ] [ ] ( )1DFT j j mx n x n m X e eω ω−←⎯⎯→= −
[ ] [ ] [ ]2
1 1
j kmFT NDx n X k X k eπ
−←⎯⎯→ =
[ ] [ ][ ]
1 1 0 0 -1, x n n N x n
x n
= ≤ ≤outside cannot be obtained by simply time shifting
[ ] ( )( ) [ ] ( )( )
[ ] ( )( )( )( )
( )( ) [ ]
[ ] [ ] ( )( )
1 1 1 1
2 2 2
1
11
, 0 1
0 , otherwise
N
N N
m k km kmj j jN N N
N
DFT
N
N
x n x n X k X k
X k X k e X k e X k e
x n x n m n Nx n
π π π− − −
⎡ ⎤ ⎡ ⎤⇒ = =⎣ ⎦ ⎣ ⎦
⎡ ⎤ ⎡ ⎤⇒ = = =⎣ ⎦ ⎣ ⎦⎧ ⎡ ⎤= − ≤ ≤ −⎪ ⎣ ⎦⇒ =
←⎯⎯→
⎨⎪⎩
( )( ) [ ]2
, 0 1: kmj
NN
DFTx n m X k e n Nπ
−←⎡ ⎤⇒ − ≤ ≤⎣ ⎯ −⎯→⎦ circular shift
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Example: circular shift
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Duality
Let
– Symmetry
[ ] ( )( ) [ ] ( )( )[ ] [ ]
DFTN
D
N
FT
x n X n X k X k
X k Nx k
⎡ ⎤ ⎡ ⎤= ←⎯ =⎣ ⎦ ⎯→
←⎣
⇒ −⎯⎯→⎦
[ ] [ ] [ ] [ ] [ ]
[ ] [ ]
( )( )
[ ] ( )( )
1 1
1
1
, 0 10, otherwise
, 0 1
0, otherwise
, 0 1
N
DFT
DFN
T
Nx k k NX k
N
x n X n x n
x k k N
X n N x k k
X k Nx
N
k←⎯⎯→= ⇒
←⎯⎯→
=
⎧ − ≤ ≤ −= ⎨⎩⎧ ⎡ ⎤− ≤ ≤ −⎪ ⎣ ⎦= ⎨⎪⎩
⎡ ⎤⇒ − ≤⎣ ⎦
−
≤ −
[ ] [ ] ( )( )[ ] ( )( )( )( ) [ ]
,
0 1DFT
N
DFT
N
N
x n
x n
X k n N
x n X k
x n x n∗ ∗
∗ ∗
⎡ ⎤=
⎡ ⎤⇒ ←⎯⎯→ − ≤ ≤ −⎣ ⎦⎡ ⎤− ←⎯⎯→⎣
= ⎣
⎦
⎦DFT of DFS coefficients of
[ ] [ ] [ ]{ } [ ] ( )( )
[ ] [ ] [ ]{ }0
1 ; 212
e Nx n x n x n x n x n
x n x n x n
∗
∗
⎡ ⎤= + − = ⎣ ⎦
= − −
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Similarly,
– N-point Circular convolution
[ ] [ ]
( )( ) ( )( ){ }[ ] [ ]{ } ( )( ) ( ) ( )( )
[ ] [ ] [ ] [ ]{ } [ ]{ }
, 0 11 21 2
1 Re2
; ; 0 1
DFT
e e
N
e e
N N
N
x n x n n N
x n x n
x n x n N n nN n
x n X k X k
n n
X k
N
k X
∗
∗
∗
= ≤ ≤ −
⎡ ⎤ ⎡ ⎤= +⎣ ⎦ ⎣
⇐ − = −
⎦
=
←⎯⎯
= ≤+ −
= + =
−
→
≤
[ ] [ ][ ] [ ]{ }0
, 0 1
Imo
DF
o
T
x n x n n N
x n j X k
= ≤ ≤ −
⇒ ←⎯⎯→
[ ] [ ] [ ]
( )( ) ( )( )
[ ] ( )( )[ ] [ ]
( )( ) [ ]
[ ] [ ]
1
3 1 20
1
1 20
1
1 20
1 2
1
1 20
2 1
, 0 1
N
m
N
N Nm
N
Nm
N
Nm
x n x m x n m n N
x m x n m
x m x n m
x n x n
x n m x m
x n x n
−
=
−
=
−
=
−
=
= − ≤ ≤ −
⎡ ⎤ ⎡ ⎤= −⎣ ⎦ ⎣ ⎦
⎡ ⎤= −⎣ ⎦
Δ
⎡ ⎤= −⎣ ⎦
=
∑
∑
∑
∑
N
N
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example
[ ] [ ]1 1x n nδ= −
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example: N-point circular convolution
N=L:
N =2 L: Linear convolution ?
[ ] [ ] [ ] [ ]
[ ] [ ] [ ] [ ] ( )( )
[ ] [ ]
1 2 1 2
1
1 2 3 1 20
1 2
1
1
DFS
NDFS
N
x n x n X k X k
x n x n X k X X kN
X k X kN
−
=
←⎯⎯→
⎡ ⎤⇔ ←⎯⎯→ = −⎣ ⎦∑
N
N
[ ] [ ]
[ ] [ ] [ ]
[ ]
1
1 20
2
3 1 2
3
, 01
0, otherwise
, 00, otherwise
, 0 1
Nkn
Nn
N kX k X k W
N kX k X k X k
x n N n N
−
=
=⎧= = ⋅ = ⎨
⎩=⎧
⇒ = = ⎨⎩
⇒ = ≤ ≤ −
∑
[ ] [ ]1 2
1 0 10 otherwise
n Lx n x n
≤ ≤ −⎧= = ⎨
⎩
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example: N(=2L)-point circular convolution
[ ] [ ] [ ] [ ]
[ ] [ ] [ ] [ ] ( )( )
[ ] [ ]
1 2 1 2
1
1 2 3 1 20
1 2
1
1
DFT
NDFT
N
x n x n X k X k
x n x n X k X X kN
X k X kN
−
=
←⎯⎯→
⎡ ⎤←⎯⎯→ = −⎣ ⎦
=
∑
N
N
[ ] [ ]1 2
1 0 10 otherwise
n Lx n x n
≤ ≤ −⎧= = ⎨
⎩
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
• Linear convolution using the DFT
– -point circular convolution with zero padding
0 L
0 p0 L p+ −1
[ ]2x n
[ ] [ ] [ ]3 1 2m
x n x m x n m∞
=−∞
= −∑
( )1N L pΔ + −
[ ] [ ] [ ] [ ], 0 1;
0, otherwise mp
x n n Nx n x n x n mN
≤ ≤ −⎧= = −⎨⎩
∑
[ ]1x n
( ) ( ) ( ) [ ] ( )
( ) ( )
3 1 3
1 2
2 3 2
2
,
0
1
j j
j j j j
k
k
N
N
X
X e X e X e X k X e k N
e X e
ω ω ω
ω
ω
ω
π
π
ω
ω
=
=
= ⇒ = ≤ ≤
=
−
[ ] [ ][ ] [ ] [ ]
1 2
3 1 2
p
X k X k
x n x n x n
=
⇒ = N
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– The circular convolution of two finite-length sequences can be equivalent to linear
convolution of the two sequences.
– Calculation of the output of an LTI using the DFT
– How to resolve the issue when is infinite?
Assume
The sequence can be decomposed in to a sum of shifted finite-length
segments of length L
[ ][ ]3
3
, 0 1
0, otherwisem
p
x n mN n Nx n
− ≤ ≤ −⎧⎪= ⎨⎪⎩
∑
0 p
h n
zero padding
0
L −1
L L
x n [ ] [ ] [ ]y n h n x n= ∗
[ ]x n
[ ] 0 0x n n= <for
[ ]x n
[ ] [ ] [ ] [ ]0
, 0 1;
0 , otherwisemm m
x n mL n Nx n x n mL x n
∞
=
+ ≤ ≤ −⎧= − = ⎨
⎩∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Illustration of warping around due to improper circular convolution
N L= 1N L p= + −
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
• Linear convolution using the discrete Fourier transform
– How to resolve the issue when is infinite?
Assume
The sequence can be decomposed in to a sum of shifted finite-length
segments of length L
[ ]x n
[ ] 0 0x n n= <for
[ ]x n
[ ] [ ]
[ ] [ ]0
;
, 0 1
0 , otherwise
mm
m
x n x n mL
x n mL n Nx n
∞
=
= −
+ ≤ ≤ −⎧= ⎨⎩
∑
where
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Overlap-add method [ ] [ ] [ ]y n h n x n= ∗
[ ] [ ]
[ ] [ ] [ ]0m
m
m m
y n y n mL
y n h n x n
∞
== −
= ∗
∑
[ ] [ ], 0 1mx n x n mL n L= + ≤ ≤ −
-N point circular convolution
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Overlap-save method [ ] [ ] [ ]y n h n x n= ∗
-N point circular convolution
[ ] ( ) 1 1
0 1mx n x n m L p p
n L
= + − + − +⎡ ⎤⎣ ⎦≤ ≤ −
[ ] ( )
[ ] [ ] [ ]0
1 1m
m
m m
y n y n m L p p
y n h n x n
∞
== − − + + −⎡ ⎤⎣ ⎦
= ∗
∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
• The discrete Cosine transform (DCT)
– General class of finite-length transform
where are referred to the basis sequences s.t.
e.g. DFT:
– DCT definition
The basis sequences are cosine functions
Assume that are periodic and even symmetric
– Type-I periodic sequence: even periodic symmetry about n=0, N-1, 2(N-1),
[ ] [ ] [ ]
[ ] [ ] [ ]
1
0
1
0
1
N
kn
N
kk
k x n n
x n k nN
φ
φ
−∗
=
−
=
Α =
= Α
∑
∑[ ]{ }k nφ [ ] [ ]
1
0
1,10,
N
k mn
m kn n
m kNφ φ
−∗
=
=⎧= ⎨ ≠⎩
∑[ ]
2 knjN
k n eπ
φ =
[ ]{ }k nφ cos knNπ
[ ]x n
[ ] ( )( ) ( )( )1 2 2 2 2N Nx n x n x nα α− −
⎡ ⎤ ⎡ ⎤= + −⎣ ⎦ ⎣ ⎦
[ ] [ ] [ ]
[ ]1 0 and -121 1 2
x n n x n
n Nn
n N
α α
α
=
⎧ =⎪= ⎨⎪ ≤ ≤ −⎩
:8 - 29, - 31, - 34, - 36, - 39, - 49, & - 64Due
Homework #
Ju y
7
b ne 9
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
DCT-1 is defined by
– Type-II periodic sequence: periodic with period 2N
DCT-2 is defined by
[ ] [ ] [ ]
[ ] [ ] [ ]
1
0
1
0
1 1
1 1
2 cos , 0 11
1 cos , 0 11 1
N
n
N
k
c
c
knX k n x n k NN
knx n k X k n NN N
πα
πα
−
=
−
=
= ≤ ≤ −−
= ≤ ≤ −− −
∑
∑
[ ] ( )( ) ( )( )2 22 1N N
x n x n x n⎡ ⎤ ⎡ ⎤= + − −⎣ ⎦ ⎣ ⎦
[ ]1 , 021 , 1 1
kk
k Nβ
⎧ =⎪= ⎨⎪ ≤ ≤ −⎩
[ ] [ ] ( )
[ ] [ ] [ ] ( )
1
0
1
0
2 2
2 2
2 12 cos , 0 1
22 11 cos , 0 12
N
n
N
k
c
c
k nX k x n k N
Nk n
x n k X k n NN N
π
πβ
−
=
−
=
+= ≤ ≤ −
+= ≤ ≤ −
∑
∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Relationship between DFT and DCT-1
2(N-1)-point DFT of
The DCT-1 of N-point sequence is identical to 2(N-1)-point DFT of
[ ] ( )( ) ( )( ) [ ]1 12 2 2 2, 0 2 3
N Nx n x n x n x n n Nα α− −
⎡ ⎤ ⎡ ⎤= + − = ≤ ≤ −⎣ ⎦ ⎣ ⎦
[ ]1 .x n
[ ] ( ) [ ] ( )
( ) [ ] ( ) [ ] ( )
( ) [ ] [ ]
( ) [ ] [ ]
22 32 1
10
1 2 31 1
0
1 21 1
0 1
21 1
1
1
1 1
1 1
1 1 1
12 1
1 2( 1)2 1
1 2 1
1 02 1
knN jN
k
kn knN Nj jN N
k k N
kn knN Nj jN N
k k
kn knN j jN N
k
c c
c c
c c c
x n X k eN
X k e X N k eN
X k e X k eN
X k e e X X NN
π
π π
π π
π π
−−
=
− −− −
= =
− −− −
= =
− −− −
=
=−
⎡ ⎤= + − −⎢ ⎥− ⎣ ⎦
⎡ ⎤= +⎢ ⎥− ⎣ ⎦
⎛ ⎞= + + +⎜ ⎟− ⎝ ⎠
∑
∑ ∑
∑ ∑
∑ [ ]( )
[ ] [ ] [ ]
11
1
01
1
1 , 0 and 11 cos , 0 1; 21 1
N nj
N
N
k
c
e
k Nnkk X k n N kN N k N
π
πα α
−−
−
=
⎡ ⎤−⎢ ⎥
⎣ ⎦
⎧
1 , 1 2
= −⎪= ≤ ≤ − = ⎨− − ⎪ ≤ ≤ −∑
⎩
[ ] [ ] [ ] [ ]{ }1 2Re , 0 2 3X k X k X k X k k Nα α α∗= + = ≤ ≤ −
[ ] [ ] [ ] ( )[ ]
1
0
1
22 cos2 1
N
n
c
knX k n x nN
X k
απα
−
=
=−
=
∑
[ ]1 :x n
[ ] ( )
[ ]( )
[ ]
2 31
2 221
1
21
1
1
1
1
2 2
knN jN
k N
N m nN jN
m
nmN jN
m
c
c
c
X N k e
X m e
X m e
π
π
π
−−
=
− −−−
=
−−−
=
− −
=
=
∑
∑
∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Note : DCT-1 involves only real-valued coefficients
⇒ reduce the implementation complexity.
– Relationship between DFT and DCT-2
2N-point DFT of 2N-point sequence
[ ] ( )( ) ( )( ) [ ]2 22 21 , 0 2 1
N Nx n x n x n x n n N⎡ ⎤ ⎡ ⎤= + − − = ≤ ≤ −⎣ ⎦ ⎣ ⎦
[ ] [ ] [ ][ ] [ ]( )
[ ]{ }
* 2 / 2
/ 2 / 2 * / 2
/ 2 / 2
2
2 Re , 0 2 1
j k N
j k N j k N j k N
j k N j k N
X k X k X k e
e X k e X k e
e X k e k N
π
π π π
π π
−
−
= +
= +
= ≤ ≤ −
[ ]{ } [ ]
[ ] [ ]{ }[ ] [ ]
1/ 2
0
/ 2
/ 2
2
2 2
(2 1)Re cos2
2Re ; 0 1
; 0 1
Nj k N
k
j k N
j k N
c
c
k nX k e x nN
X k X k e k N
X k X k e k N
π
π
π
π−−
=
−
−
+=
⇒ = ≤ ≤ −
⇒ = ≤ ≤ −
∑
[ ] [ ] [ ]1
02 2
(2 )(2 1)2 2 cos ; 0 2 12
N
k
c cN k nX N k x n X k k NN
π−
=
− +− = = − ≤ ≤ −∑
[ ]
[ ][ ]
[ ]
/ 2
/ 2
2
22
2
0 , 0
, 1, , 10,
2 , 1, ,2 1
j k N
j k N
c
c
c
X k
X k e k NX k
k NX N k e k N N
π
π
=⎧⎪
= ⋅ ⋅ ⋅ −⎪⇒ = ⎨=⎪
⎪− − = + ⋅ ⋅ ⋅ −⎩
[ ]2x n
[ ] [ ]2 1
2 / 2
02 2
1 , 2
0,1, ,2 1
Nj k N
kx n X k e
Nn N
π−
=
⇔ =
= ⋅ ⋅ ⋅ −
∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Energy compaction property of DCT-2
Since DCT-2 coefficients of a finite-length sequence are concentrated in the low
indices than DFT, DCT-2 is often used for data compression applications in
preference to DFT
Example
[ ] [ ] [ ]1 12 2
0 02
1N N
n k
cx n k X kN
β− −
= =
=∑ ∑
[ ] ( )0cosnx n a nω φ= +
32⇐ −point DCT
32 - point DFT
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Energy concentration can be examined by measuring the truncation error of the
transformations.
[ ] [ ] [ ]
[ ] [ ] [ ]
1 2
0
1 2
0
1
1
N
DFTn
N
DCTn
m
m
c
E m x n x nN
E m x n x nN
−
=
−
=
= −
= −
∑
∑
MS truncation error :
[ ] [ ] [ ]1
02
1 (2 1)cos ; 0 12
N m
k
c cm
k nx n k X k n NN N
πβ− −
=
+= ≤ ≤ −∑
[ ]( )
( ) ( )( )
1, 0 1 / 2
0, 1 / 2 1 / 2
1, 1 / 2 1m
k N m
T k N m k N m
N m k N
≤ ≤ − −⎧⎪
= + − ≤ ≤ − +⎨⎪ + + ≤ ≤ −⎩
[ ] [ ] [ ]1
2 / 2
0
1 , 0,1, ,2 1; 1,3,5, , 1N
j k N
km mx n T k X k e n N m N
Nπ
−
=
= = ⋅ ⋅ ⋅ − = ⋅ ⋅ ⋅ −∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Discrete Fourier transform• Discrete Fourier series representation of periodic sequences
– Let be a periodic sequence with period N such that for any
integer m
Example A periodic complex exponential
– A set of N periodic complex exponentials defines all the
frequency components
– Fourier series representation
[ ]x n
[ ]{ }, 0,1,2,... 1,ke n k N= −
[ ] [ ] k k Ne n e n+=Note that for any integer
[ ] [ ]x n x n mN= +
[ ] [ ]2
; knj
Nk ke n e e n mN k
π
= = + is an interger
[ ] [ ]
[ ] [ ]
21
0
21
0
1
N j knN
k
mnN jN
n
x n X k eN
X k x n e
π
π
−
=
− −
=
⇒ =
=
∑
∑
[ ] [ ]21
0
1 N j knN
k
x n X k eN
π−
=
= ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
The Fourier series coefficient can be obtained by
The Fourier series coefficients is a periodic sequence
– Discrete Fourier series (DFS)
Define by
[ ] [ ]
[ ] ( )
( )
2 2 21 1 1
0 0 0
21 1
0 0
21
0
1
1
1,10, otherwise
knN N Nj rn j j rnN N N
n n k
N N j k r nN
k k
N j k r nN
n
x n e X k e eN
X k eN
k r Ne
N
π π π
π
π
− − −− −
= = =
− − −
= =
− −
=
=
=
− =⎧= ⎨⎩
∑ ∑ ∑
∑ ∑
∑Since
[ ] [ ]21
0
N j rn
N
n
x n e X rπ−
=
=∑
[ ] [ ]DFSx n X k←⎯⎯→
[ ] [ ]
[ ] [ ]
21
0
1
0
;
1
N jkn NN N
n
Nkn
Nk
X k x n w w e
x n X k wN
π− −
=
−−
=
= =
=
∑
∑
Analysis equation :
Synthesis equation :
[ ] [ ]
[ ] [ ] [ ] [ ]
21
0
2 21 1( ) 2
0 0
N j knN
n
N Nj k N n j kn j nN N
n n
X k x n e
X k N x n e x n e e X k
π
π ππ
− −
=
− −− + − −
− −
=
+ = = =
∑
∑ ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example Periodic pulse train
Example Duality in DFS
Let the DFS coefficient be a periodic impulse train
[ ] [ ]
[ ] [ ]
[ ] [ ]
[ ] [ ]
1
0
21 1
0 0
1, 0, otherwise
, 0 1
1
1 1 1
r
Nkn
Nn
N N j knkn kn NN N
k k
n mMx n n mN
x n n n N
X k n w
x n X k w w eN N N
π
δ
δ
δ
∞
=−∞
−−
=
− −− −
= =
=⎧= − = ⎨
⎩= ≤ ≤ −
= =
⇒ = = =
∑
∑
∑ ∑ ∑
Since
[ ] [ ] [ ] [ ] Y k Nx n y n X n= ↔ =
[ ] [ ]1
0
01,
Nkn
N Nk
y n N k w w nδ−
− −
=
⇒ = = = ∀∑
[ ] [ ]r
Y k N k rNδ∞
=−∞
= −∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example Periodic rectangular pulse train
Any periodic sequence can be represented as a sum of complex exponential
sequences
[ ]2 454 4
1010 1010
0 0 10
sin1 21 sin
10
kkj kn jnkk
n n
kWX k W e ekW
π ππ
π− −
− =
−= = = =
−∑ ∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
• Properties of DFS
– Linearity
– Shift of a sequence
– Duality
– Symmetry properties
[ ] [ ] [ ] [ ]1 2DFSax n bx n aX k bX k+ ←⎯⎯→ +
[ ] [ ] ( ) [ ] [ ]
[ ] [ ][ ] [ ]
2 2 2 21 1
0 0
N Nj kn j n m j j km kmN N N NN
n n
DFS kmN
DFSnN
x n m e x n m e e X k e X k W
x n m X k W
x n W X k
π π π π− −− − − − −
= =
−
− = − = =
⇒ − ←⎯⎯→
←⎯⎯→ −
∑ ∑
[ ] [ ] [ ] [ ]
[ ] [ ]
[ ] [ ] [ ] [ ]
1 1
1
0
0 0
1
,
Nkn
Nk
Nkn
Nk
DFS DF
k
S
Nkn
N Nx n k W
Nx
x n X k
k x n W
x n k X n Nx
XN
k
w
X
−
=
−
−−
=
=
− =
⇒ − =
←⎯⎯→ ←⎯⎯→
= ⇒
−
∑ ∑
∑If
[ ] [ ][ ] [ ]
* *
* *
DFS
DFS
x n X k
x n X k
←⎯⎯→ −
− ←⎯⎯→
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
⇒ If is real,
– Periodic convolution
Duality
Difference between aperiodic convolution and periodic convolution
• The sum is over an interval
•
[ ] [ ][ ] [ ]
*X k X k
X k X k
= −
= −
[ ]x n
[ ] [ ] [ ] [ ] [ ]
[ ] [ ] [ ]
[ ] [ ]
[ ] [ ] [ ] [ ]
1 1
3 1 2 2 10 0
1 1
3 1 20 0
1 1
1 20 0
1
2 1 1 20
N N
m m
N Nkn
Nn m
N Nkn
Nm n
Nkm
Nm
x n x m x n m x m x n m
X k x m x n m W
x m x n m W
X k x m W X k X k
− −
= =
− −
= =
− −
= =
−
=
= − = −
⎛ ⎞= −⎜ ⎟
⎝ ⎠
= −
= =
∑ ∑
∑ ∑
∑ ∑
∑
0 1m N≤ ≤ −
[ ] ( )2 2x n m x m n− = − −⎡ ⎤⎣ ⎦
[ ] [ ] [ ] [ ]1
1 2 1 20
NDFS
mx m x n m X k X k
−
=
⇒ − ←⎯⎯→∑
[ ] [ ] [ ] [ ]1
1 2 1 20
1 NDFS
m
x n x n X m X k mN
−
=
←⎯⎯→ −∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example Periodic convolution
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Properties of DFS
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
– Properties of DFS (cont.)
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
• Fourier transform of periodic signals
– A sequence should be absolutely summable to guarantee the uniform convergence of
its Fourier transform
– Periodic signals are not absolutely summable
– The F.T. of a periodic signal can be interpreted to be an impulse train in the frequency
domain with the impulse values proportional to the DFS coefficients for the sequence
– If is periodic with period N, the Fourier transform of is defined as
Although the F.T. of a periodic sequence does not converge in the normal sense,
the introduction of impulses enables to use the F. T.
Note: has the necessary periodicity with period
[ ] ( ) [ ]2 2j
k
kx n X e X kN N
ω π πδ ω∞
=−∞
⎛ ⎞⇔ = −⎜ ⎟⎝ ⎠
∑
[ ]x n[ ]x n
( ) [ ]
[ ]
[ ] [ ]
2 2
0 0
2
0
2
1 1 2 2 ; 0 2 /2 2
1 2
1
j j n j n
k
j n
k
j knN
k
kX e e d X k e d NN N
kX k e dN N
X k e x nN
π ε π εω ω ω
ε ε
π ε ω
ε
π
π πω δ ω ω ε ππ π
πδ ω ω
− −
− −
∞ −
−=−∞
∞
=−∞
⎛ ⎞= − < <⎜ ⎟⎝ ⎠
⎛ ⎞= −⎜ ⎟⎝ ⎠
= =
∑∫ ∫
∑ ∫
∑
( )jX e ω 2π
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example: Fourier transform of a periodic impulse train
– For a finite-length sequence
Thus, is obtained by sampling the Fourier transform of the finite sequence
[ ] [ ] [ ]
[ ] [ ] [ ]
( ) ( ) ( )
( )2
2 2
2 2
j j j
j
k
kjN
k
m m
x n x n p n
X e X e P e
kX e
mN
N N
kX eN
x n x n mN
N
ω ω ω
ω
π
π πδ
ω
δ
ω
π πδ
∞ ∞ ∞
=−∞ =−∞ =−
∞
=−∞
∞
=−
∞
∞
= ∗
=
⎛ ⎞= −⎜ ⎟⎝ ⎠
⎛ ⎞ ⎛ ⎞= −
− −
⎜ ⎟ ⎜ ⎟⎝ ⎠⎝
−
⎠
= =∑ ∑ ∑
∑
∑
[ ] [ ] [ ]
( )
1,
2
2m
j
k
p n n mN P k k
kP e
NNω
δ
πδ ω
π
∞
=−∞
∞
=−∞
= − ⇒ = ∀
⇒ = −⎡ ⎤⎢ ⎥⎣ ⎦
∑
∑[ ] [ ]( ) , 0 for 0 and x n x n n n N= < ≥., i.e
[ ] ( ) [ ]2 21
20
Nj k j kj N Nk nN
X k X e X e x n eπ π
ωπω
− −
= =
⎛ ⎞⇒ = = =⎜ ⎟
⎝ ⎠∑
[ ] x n[ ] X k
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
6. Discrete Fourier transform
Example: Relationship between the DFS coefficients and the FT of one period
[ ] 1; 0 40;
x nn≤ ≤⎧
= ⎨⎩ otherwise
( ) ( )( )
42
0
sin 5 / 2sin / 2
j j n j
nX e e eω ω ωω
ω− −
=
⇒ = =∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
• Discrete Fourier transform (DFT) can analyze the frequency content of continuous-time
signal
Fourier Analysis of Signals Using DFT
( ) ( ) ( )( )12
j j jV e X e W e dπ
π
ω θ ω θ θπ
−
−= ∫
( ) 1 2jc
k
kX e XT T T
ω ω π∞
=−∞
⎛ ⎞= +⎜ ⎟⎝ ⎠
∑
[ ] [ ] [ ]
[ ] [ ] ( )21
20
; knN j jN
kn N
v n x n w n
V k v n V e Teπ
ω πω ω
− −
= =
=
= = = Ω∑
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
– Example
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
– Length of the DFT
Example: T=1/5000 and 10Hz resolution
– Relationship between DFT values
[ ] [ ] [ ] [ ][ ] [ ] [ ] ( )
( ) [ ] ( )
* *
*
11
11
(( ))
512 11 501 11 2000 12 11 5000/512 2 107.4
11 0.4 1
N
c
V k V k V N k V k
V V V j
X T V jπ π
= − ⇒ − =
⇒ − = = = −
Ω = ⋅ ⋅ = ⋅
⇒ Ω ≈ ⋅ = +
[ ] [ ] ( )21
20
;
2 2
knN jN
kn N
jV eV T
NNT T
k v n e ω
ω
π
π ω
π π
=
− −
=
= Ω
ΔΩ ≤ ⇒ ≥
= =
ΔΩ
∑
9
1/ 5000
2 500010 2 500 512 210T
N NNTππ
=
⋅ ≤ ⇒ ≥ = ⇒ = =
[ ] ( )512; 1/5000
If 11 2000 1 ;N T
V j= =
= +
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
• DFT analysis of sinusoidal signals
– Windowing and spectral sampling have an important effect on the analysis of
sinusoidal signals using the DFT
– Effect of windowing
After the windowing
it can be suffered from reduced resolution and leakage of spectrum
[ ] [ ] [ ][ ] [ ]
[ ] [ ]( ) [ ] [ ]( )0 0 0 0 1 1 1 1
0 0 0 1 1 1
( ) (0 1
cos( ) cos( )
2 2j n j n j n j n
v n x n w n
A w n n A w n nA Aw n e w n e w n e w n eω θ ω θ ω θ ω θ
ω θ ω θ
+ − + + − +
=
= + + +
= + + +
[ ]0 0 0 1 1 1
0 0 0 1 1 1
( ) cos( ) cos( )cos( ) cos( )
cs t A t A tx n A n A n
θ θω θ ω θ
= Ω + + Ω +
⇒ = + + +
( ) ( )( ) ( )( ) ( )( ) ( )( )0 0 1 10 0 1 10 1
2 2j j j jj j j jj A AV e W e e W e e W e e W e eω ω ω ω ω ω ω ωθ θ θ θω − + − +− −⎡ ⎤ ⎡ ⎤= + + +⎣ ⎦ ⎣ ⎦
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
Example Leakage due to the use of a rectangular window
0
1
26
23
πω
πω
=
=
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
The resolution is primarily affected by the width of the main lobe of
The degree of leakage depends on the relative amplitude of the main and side
lobe of
These are associated with the window length L and the shape of the window( )jW e ω
( )jW e ω
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
Example: Kaiser window
[ ]( )
2
0
0
11,0 1,
20 ,otherwise
nILw n n L
I
αβα
αβ
⎧ ⎡ ⎤−⎡ ⎤⎪ ⎢ ⎥− ⎢ ⎥⎣ ⎦⎪ ⎢ ⎥ −⎣ ⎦= ⎨ ≤ ≤ − =⎪⎪⎩
sl
A
( )24 121
155sl
ml
AL
π ++
Δ
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
– Effect of spectral sampling
Example
[ ] ( ) ( )2 ; 2 /jk k
N
V k V e k NTωπω
π=
= Ω =
[ ] 2 4cos 0.75cos ;14 15
0 63
n nv n
n
π π= +
≤ ≤
64N =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
– Spectral sampling with frequencies matching DFT frequencies
Example [ ] 2 2cos 0.75cos ; 0 6316 8
n nv n nπ π= + ≤ ≤
64N =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
– DFT analysis by Kaiser window
Example [ ] [ ] [ ]2 4cos 0.75 cos ; 0 6314 15K K
n nv n w n w n nπ π= + ≤ ≤
64L = 32L =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
– DFT analysis with 32-point Kaiser window and zero padding
Example [ ] 2 4cos 0.75cos ; 5.48, 32 Kaiser window14 15
n nv n Lπ π β= + = =
32N = 128N =
1024N =64N =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
– DFT analysis with Kaiser window and zero padding
Example [ ] 2 4cos 0.75cos ; 5.48 Kaiser window14 15
n nv n π π β= + =
1024, 32N L= = 1024, 54N L= =
1024, 64N L= =1024, 42N L= =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
• Fourier analysis of stationary random signals: Periodogram
– The sample mean and sample variance respectively defined by
are unbiased and asymptotically unbiased estimators, respectively
– Periodogram analysis: estimation of the power spectrum
Periodogram is the F. T. of aperiodic correlation of windowed data sequence
Modified periodogram: is not a rectangular window
( ) [ ] [ ]1
0
Lj j n
n
V e x n w n eω ω−
−
=
=∑
[ ] [ ]( )21 1
2
0 0
1 1ˆ ˆ ˆ; L L
X X Xn n
m x n x n mL L
σ− −
= =
= = −∑ ∑
[ ]w n
( ) ( ) [ ]
[ ] [ ] [ ] [ ] [ ]
12
1
1
0
1 1 Lj j m
vvm L
L
vvn
I V e c m eLU LU
c m x n w n x n m w n m
ω ωω−
−
=− +
−
=
= =
= + +
∑
∑
( ) 1XX SSP P
T Tωω ⎛ ⎞= ⎜ ⎟⎝ ⎠
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
– Periodogram can be calculated by means of the DFT
– Properties of the periodogram:
To make the periodogram unbiased,
[ ] [ ] [ ]{ } [ ] [ ] [ ]1
0
; L
xx wwn
m E x n x n m c m w n w n mφ−
=
= + = +∑
( ){ } [ ]{ }1
1
1 Lj m
vvm L
E I E c m eLU
ωω−
−
=− +
= ∑
( ) [ ] 21kI V k
LUω =
[ ]{ } [ ] [ ] [ ] [ ]{ }
[ ] [ ] [ ]
[ ] [ ]
1
0
1
0
L
vvn
L
xxn
ww xx
E c m w n w n m E x n x n m
w n w n m m
c m m
φ
φ
−
=
−
=
= + +
= +
=
∑
∑
( ){ } ( ) ( )( ) ( ) ( ) 21 ; j j jXX WW WWE I P C e d C e W e
LUπ ω θ ω ω
πω θ θ−
−= =∫
( ) [ ] [ ]1 12 2 2
0 0
1 1 112
L Lj
n nW e d w n U w n
LU LU Lπ ω
πω
π
− −
−= =
= = =⇒∑ ∑∫
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
– As the window size increases, the peridogram becomes a consistent estimate
Example: white noise generation with rectangular windowing
256L =
[ ] ( ) [ ] [ ] [ ]212 2 /
1
1 1 Lj kn N
km L
I k I V k w n x n eL L
πω−
−
=− +
= = = ∑
( ){ } ( )2var k XXI Pω ω64L =
16L =
1024L =
256L =
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
– Edge effect due to covolution of finite sequences
As m becomes close to L, the
calculation becomes inconsistent
This problem can be alleviated by
averaging multiple independent
periodogram estimates
[ ] [ ] [ ] [ ] [ ]1
0
L
vvn
c m w n w n m x n x n m−
=
= + +∑
( ) ( )1 1 2
0 0
1 1K Lj
rr n
I X eK LU
ωω− −
= =
= ∑ ∑
[ ] [ ] [ ]; 0 1rx n w n x rR n n Q= + ≤ ≤ −
420.461: Digital Signal Processing copyright@Yong-Hwan Lee
Seoul National UniversitySchool of Electrical Engineering
7. Fourier analysis of signals using DFT
– Example
( ) ( ) ( )2
20 02XX e
AP πω δ ω ω δ ω ω σ⎡ ⎤= − + + +⎣ ⎦
[ ] ( ) [ ] [ ] ( ) 20.5cos 2 / 21 ; 3, 3 1ex n n e n e n Uπ θ σ= + + − ⇒ =∼
1024; 1Q U= =
127, 16K L= =
31, 64K L= =
1, 1024K L= =