Slide 4
Slide 4
Reference Book:
1. S K Mitra “Digital Signal Processing:A Computer Aided Approach”, Prentice Hall, 4th
Edition
3. Proakis J.G. and Manolakis D.G., “Digital Signal Processing”, Macmillan Publishing
Company.
Book Reference:
Digital Signal Processing- S Salivahanan
Discrete-Time Signal Processing-Alan V. Oppenheim, Ronald W. Schafer-3rd Edition
Motivating Examples
Sending image data in the usual format vs. sending only low-frequency data. a)
25% of data is in the usual format. b) 6.25% of data in the usual format. c) 25%
of lowest frequency data. d) 6.25% of lowest frequency data. 5
2-d sinusoid surface plot (left) and image (right)
6
2-D sine and its Fourier Transform
A natural image and its Fourier Transform
7
8
9
10
11
Fourier Series
Fourier Transform
Discrete-Time
Fourier Transform
Uniform sampling of
DTFT spectrum is DFT:
12
The DFT of discrete-time sample signal 𝒙 𝒏
$%&
)*
%(̇ +!
𝑋(𝑘) = & 𝑥 𝑛 𝑒 $
!"#
N = number of samples
n = current sample
k = current frequency, where, k∈ [0, N−1]
x(n) = the value of the signal x(n) nth sample
X(k)= the DFT that includes information of both amplitude and phase
13
1. Periodicity
2. Linearity
3. Shifting Property
4. Time Reversal
5. Multiplication property
6. Complex Conjugate
7. Convolution Theorem
8. Parseval’s Theorem
14
Periodicity:
If X is periodic with period N
𝑥 𝑛 + 𝑁 = 𝑥(𝑛) for all n
𝑋 𝑘 + 𝑁 = 𝑋(𝑘) for all k
Proof:
$%& $%&
)* )*
%(̇ (+-$)! %(̇ /! % ()*!
𝑋[𝑘 + 𝑁] = . 𝑥 𝑛 𝑒 $ = .𝑥 𝑛 𝑒 $ 𝑒 ̇ = 𝑋[𝐾]
!"# !"#
Time reversal:
If X(k) is an N-point DFT of x(n), then
DFT{𝑥 −𝑛 } → 𝑋 𝑁 − 𝑛
DFT{x 𝑁 − 𝑛 } = 𝑥(−𝑘)! = 𝑋(𝑁 − 𝑘)
Linearity:
If 𝑋" 𝑘 and 𝑋# 𝑘 are N-point DFTs of 𝑥" 𝑛 and (n) 𝑥# 𝑛 respectively, and a
and b are arbitrary constants either real or complex-valued. Then,
𝑎𝑥" 𝑛 +𝑏𝑥# 𝑛 = 𝑎𝑋" (k)+𝑏𝑋# (k)
15
Time shifting:
If X(k) is an N-point DFT of x(n), then
!"#$%
%
𝑥(𝑛 − 𝑙) → 𝑋(𝑘)𝑒 &
'!"#
Or, 𝑥(𝑛 − 𝑙) → 𝑋(𝑘)𝑊$+0 [where, WN=𝑒 $ ]
𝑥(𝑛 + 𝑙) → 𝑋(𝑘)𝑊$%+0
Frequency shifting :
If X(k) is an N-point DFT of x(n), then
$#%&'
𝑥(𝑛)𝑒 & → 𝑋(𝑘 − 𝑙)
𝑥(𝑛)𝑊!(&' → 𝑋(𝑘 − 𝑙)
𝑥(𝑛)𝑊!&' → 𝑋(𝑘 + 𝑙)
16
Complex conjugate property:
If X(k) is an N-point DFT of x(n), then
x * (n) → 𝑋∗(−𝑘) = 𝑋∗(𝑁 − 𝑘)
Multiplication property:
If X(k) is an N-point DFT of x(n), then
𝑥& 𝑛 . 𝑥) 𝑛 = 𝑋& 𝑘 ∗ 𝑋) 𝑘
Correlation Property:
If X(k) is an N-point DFT of x(n), then
𝐷𝐹𝑇 𝑌𝑥𝑦 𝑙 → 𝑋& 𝑘 ∗ 𝑋) 𝑘
Parseval’s energy theorem: If X(k) is an N-point DFT of
x(n), then Parseval’s energy theory will be,
!(" !("
1
5 |𝑥(𝑛)|# = 5 |𝑋(𝑘)|#
𝑁
&)* +)*
17
DFT of a finite sequence 𝒙 (n); 0≤ 𝒏 ≤ 𝑵 − 𝟏
Where 𝑋 𝑘 : sequencethe in frequency domain
𝑥 𝑛 : sequence in time domain
𝑵(𝟏 𝟐𝜫 𝑵(𝟏
(𝑱̇ 𝑵 𝒌𝒏
𝑫𝑭𝑻 𝒙 𝒏 =𝑿 𝒌 =5 𝒙 𝒏 𝒆 =5 𝒙 𝒏 . 𝑾𝑵 𝒌𝒏
𝒏)𝟎 𝒏)𝟎
𝟐𝜫
(𝑱̇
𝑾𝑵 = 𝒆 𝑵 where 𝑊! is Twiddle factor or phase factor
We use the twiddle factor to reduce the computational complexity of calculating
DFT and IDFT. Alternatively, we can also say that the twiddle factor has
periodicity/a cyclic property.
&
/-$/) %2)3 +- .$
𝑊$ =𝑒 "
Periodicity Property:
= 𝑒 %2)3+/$ . 𝑒 %23
𝑊!34! = 𝑊!3 /
$#%
= 𝑒 %2)3/$ . −1
(
𝑊! = 𝑒 ! = −𝑊$/
$#%(34!) $#%3 /-$/)
𝑊!34! =𝑒 ( ! = 𝑒 ( !
. 𝑒 ($#% 𝑊$ = −𝑊$/
$#%3 $#% 3
( ! ( !
= 𝑒 .1 = 𝑒 = 𝑊!3
𝑊!34! = 𝑊!3
19
Periodicity Property: Symmetry Property:
34!/#
𝑊!34! = 𝑊!3 𝑊! = −𝑊!3
012 012
( (
Proof: 𝑊! = 𝑒 3
Proof: 𝑊! = 𝑒 3
$#%(34!) $#%3 3
( ( ! 34!/#
𝑊!34! =𝑒 ! = 𝑒 . 𝑒 ($#% 𝑊! =𝑒
($#% +4
1
.!
$#%3
( !
$#% 3
( !
= 𝑒 ($#%+/! . 𝑒 ($%
= 𝑒 .1 = 𝑒 = 𝑊!3 3
= 𝑒 ($#%/! . −1
𝑊!34! = 𝑊!3
34!/#
𝑊! = −𝑊!3
20
21
Twiddle factor matrix:
4-point twiddle factor matrix (4×4) matrix
&
𝑤5& = 𝑒 %2)3/5 = 𝑒 %23/) = −𝑗
)
𝑤5) = 𝑒 %2)3/5 = 𝑒 %23 = −1
5 3𝜋 3𝜋 3𝜋 3𝜋
𝑤45 = 𝑒 %6)*/4 = 𝑒 %65*/) = 𝑐𝑜𝑠 − + 𝑗𝑠𝑖𝑛 − = 𝑐𝑜𝑠 − 𝑗𝑠𝑖𝑛 = 0 − 𝑗 −1 = 𝑗
2 2 2 2
5
𝑤55 = 𝑒 %2)3/5 = 𝑒 %263/5 = 𝑒 %2.)3 = 𝑐𝑜𝑠 2𝜋 − 𝑗𝑠𝑖𝑛 2𝜋 = 1
𝑤57 = −1
𝑤58 = −𝑗
1 1 1 1
1 −𝑗 −1 𝑗
𝑤5 =
1 −1 1 −1
1 𝑗 −1 −𝑗 22
4-point twiddle factor matrix:
w9< = -j
4-Point Twiddle factor 8-Point Twiddle factor
23
Example : Compute N point DFT of the signal given by 𝑥 𝑛 = {1, 0, 0, 1} and N=4
!(" 18
(= ̇ +&
We know , 𝑋 𝑘 = 5 𝑥 𝑛 𝑒 3
&)*
!(" 18
(= ̇ +&
𝑋 𝑘=0 =5 𝑥 𝑛 𝑒 3 = ∑:&)* 𝑥 𝑛 𝑒 *
&)*
=𝑥 0 +𝑥 1 +𝑥 2 +𝑥 3 =1+0+0+1=2
!(" !("
#> #>
(=̇ ! +& (=̇ 9 .".&
𝑋 𝑘 = 1 = L𝑥 𝑛 𝑒 = L𝑥 𝑛 𝑒 .
&)* &)*
012
(
=𝑥 0 + 𝑥 1 𝑒 + 𝑥 2 𝑒 ($% + 𝑥(3)𝑒 ($:%/#
9
= 1 + 0 + 0 + 1. 𝑒 ($.:%/# = 1 + 𝑒 ($.:%/# = 1 + 𝑗
!"
$ %'̇ # .).!
𝑋 𝑘=2 = ∑!"# 𝑥(𝑛)𝑒 = 1 + 0 + 0 + 𝑒 %*.$+ = 1−1=0
!" !"
$ %'̇ # .$.! %'̇ # .$.$
𝑋 𝑘=3 = ∑!"# 𝑥(𝑛)𝑒 =1+𝑒 =1−𝑗
So, 𝑋 𝑘 = {2, 1 + 𝑗, 0, 1 − 𝑗}
24
Example: Compute 4-point DFT of the signal 𝑥 𝑛 = {1, 2, 3, 4}
1
2
Solution: Let, 𝑥 4 = [1 2 3 4] 𝑥59 =
3
4
1 1 1 1
1 −𝑗 −1 𝑗
4-point twiddle factor matrix, 𝑤5 =
1 −1 1 −1
1 𝑗 −1 −𝑗
X 4 = 𝑤5 * 𝑥59
1 1 1 1 1 10
1 −𝑗 −1 𝑗 2 −2 + 𝑗2
= . =
1 −1 1 −1 3 −2
1 𝑗 −1 −𝑗 4 −2 − 𝑗2
25
DFT is the most straightforward mathematical procedure for
determining the frequency content of a time-domain sequence, it’s
terribly inefficient. As the number of points in the DFT is
increased to hundreds, or thousands, the amount of necessary
number crunching becomes excessive.
Example:
!("
27
If N=10! ,
Calculation for DFT, 𝑁 " =10#$ ,
Calculation for FFT, N 𝑙𝑜𝑔"% = 30 ×10!
Let, time per multiplication is 1ns
For DFT : 10#$ ×1 ×10! =10! 𝑠=31.2 years
For FFT : (30×10! ) ×10&! =30s
FFT is computational very efficient than
DFT
28
Radix: The radix (r) of FFT algorithm is a very efficient process for
performing DFTs under the constraint that the DFT size be an integral power
of r.
r: is the radix
29
The N-point DFT of sequence x(n) is given by
$%&
𝑋 𝑘 = ) 𝑥 𝑛 𝑊$!+ , 0 ≤ 𝑘 ≤ 𝑁 − 1
!"#
Splitting x(n) into even and odd points, we obtain
$%& $%&
𝑋 𝑘 = ) 𝑥 𝑛 𝑊$!+ + ) 𝑥 𝑛 𝑊$!+
!"#,! <=<! !"#,! >??
Substituting n=2r for n even and n=2r+1 for n odd, we have
($/)%&) ($/)%&)
()@A&)+
𝑋 𝑘 = ) 𝑥 2𝑟 𝑊$)@+ + ) 𝑥 2𝑟 + 1 𝑊$
@"# @"#
($/)%&) ($/)%&)
"#∗" "#
" &'(%/")
∗ 𝑊%" = 𝑒 &'(")/%) = 𝑒 &'( %
)
=𝑒 = 𝑊%/"
For 4 point DFT Method from 8 point DFT (Stage -2):
0
1
%&
G(k) =B g®W2DE
1
@"#
0 0
3
%& 1 3
%&
=B g 2l W +B
4
2 g 2l + 1 W2)GA& E
1 1
F"# F"#
0 0
3
%& 3
%&
=B g 2l W2GE +W2H B g 2l + 1 W2GE ……………………..(i)
3 1 3
F"# F"#
Where A(k) is the (N/4)-point DFT of even numbers of the above sequence and
B(k) is the (N/4)-point DFT of odd numbers of the above sequence.
Similarly,
0 0
%& %&
3 3
H(k) = B h 2l W0FE +W2E B h 2l + 1 W2GE
3 1 3
F"# F"#
ts X(k) in the frequency domain into sets of smaller and smaller subsequences.