Lec 7

Nguyễn Công Phương
SIGNAL PROCESSING
The Discrete Fourier Transform

Contents
I. Introduction
II. Discrete – Time Signals and Systems
III. The z – Transform
IV. Fourier Representation of Signals
V. Transform Analysis of LTI Systems
VI. Sampling of Continuous – Time Signals
VII.The Discrete Fourier Transform
VIII.Structures for Discrete – Time Systems
IX. Design of FIR Filters
X. Design of IIR Filters
XI. Random Signal Processing
s i tes.google.com/site/ncpdhbkhn 2
1. Computational Fourier Analysis
2. The Discrete Fourier Transform
3. Sampling the Discrete – Time Fourier Transform
4. Properties of the Discrete Fourier Transform
5. Linear Convolution Using the DFT
6. Fourier Analysis of Signals Using the DFT
7. Direct Computation of the DFT
8. The FFT Idea Using a Matrix Approach
9. Decimation – in – Time FFT Algorithms
10. Decimation – in – Frequency FFT Algorithms
11. Generalizations and Additional FFT Algorithms
12. Computation of DFT for Special Applications
Computational Fourier Analysis
(1), Summary
(2), CTFT
∞ ∞
X c ( jΩ) =  xc (t )e− jΩt dt ≈
−∞

n= −∞
xc (nT )e− jΩnT (T ) = Xˆ c ( jΩ)
Xˆ c ( jΩ) = TXɶ ( e jω ) , x[n] = xc (nT ) 

DTFT
→ Xɶ (e jω )
ω =Ω T
1 ∞
 2π 
Xɶ ( e jΩ T
)=
T

l =−∞
X c  jΩ − j

l
T 
TXɶ (e jΩT ), Ω <π /T

X c ( jΩ) = 
0, otherwise
(3), CTFS
xɶ [ n ]= xɶ c ( nT )
xɶc (t + T0 ) = xɶc (t ) 
→ xɶ[n + N ] = xɶ[n]
T =T0 / N
N −1 N −1 2π
1 1 −j
ck ≈  xɶc (nT )e  xɶ[n]e
− jk Ω0nT kn
(T ) = N
= cɶk
T0 n=0 N n=0
2π 2π
Ω0 = = T
T0 N
(4), DTFT
∞
Xɶ (e ) =
jω
 x[n]e
n =−∞
− jωn
N −1
Xɶ (e ) ≈  x[n]e− jωn = Xɶ N (e jω )
jω
n= 0
1, 0 ≤ n ≤ N −1
Xɶ N (e jω ) ←
DTFT
 xN [n ] = x[n ] pN [n ], pN [ n] = 
0, otherwise
X [k ] = Xɶ N (e jω ), k = 0, 1, ..., K − 1
 X [0]   e jω0 0 e jω0 1 ⋯ e jω0 ( N −1)

  x[0] 
 X [1]   jω1 0  
 = e e jω11 ⋯ e jω1 ( N −1)
 x[1]

 ⋮   ⋮ ⋮ ⋱ ⋮  ⋮ 
   jωK −1 0  
 X [ K − 1] e e jωK−11 ⋯ e jωK −1 ( N −1)   x[ N − 1]
↔ X = Wx → x = W−1X
(5), DTFT
N −1 x = W − 1X
Xɶ ( e jω ) ≈  x[n ]e− jωn = Xɶ N (e jω )
n =0
2π
ωk = k, k = 0,1,..., N − 1
N
N −1 2π
−j
→ X [ k ] =  x[ n]e
kn
N
, k = 0, 1,..., N − 1
n= 0
2π 1
1 N −1 − j kn → cɶk =
cɶk =  xɶ[n ]e N N
X [k ]
N n =0 N −1 2π
x[n] =  cɶk e
j kn
xɶ[n] = x[ n] N
k =0
N −1 2π
1
 X [k ]e
j kn
→ x[ n] = N
, n = 0, 1,..., N − 1
N k =0
Ex. (6), DTFT 2π
N −1 −j
x[n] = {0,1, 2,3} X [ k ] =  x[ n]e
kn
N , k = 0,1,..., N − 1
↑
n =0
3 2π
−j
=  x[ n]e
kn
N , k = 0, 1, 2, 3
n =0
2π 2π 2π 2π
−j 0×0 −j 0×1 −j 0×2 −j 0×3
X [0] = x[0]e 4 + x[1]e 4 + x[ 2]e 4 + x[3]e 4
= 0e0 + 1e0 + 2e0 + 3e0 = 6

2π 2π 2π 2π
−j 1×0 −j 1×1 −j 1× 2 −j 1×3
X [1] = x[ 0]e 4 + x[1]e 4 + x[ 2]e 4 + x[3]e 4
= 0e 0 + 1e − j2 π / 4 + 2e − j 4 π / 4 + 3e − j 6π / 4 = e − jπ /2 + 2e − jπ + 3e − j 3π / 2
= 1[cos(−π / 2) + j sin( −π / 2)] + 2[cos(−π ) + j sin( −π )]
+ 3[cos( −3π / 2) + j sin(−3π / 2)]
= 1( 0 − j ) + 2( −1 + j 0) + 3(0 + j ) = −2 + j 2
Ex. (7), DTFT
x[n] = {0,1, 2,3} X [k ] = {6, −2 + j 2, −2, −2 − j 2}
↑ ↑
N −1 2π
1 j kn
x[ n] =
N
 X [k ]e
k =0
N
, n = 0,1,..., N − 1
2π 2π 2π 2π
1 j 0×0 j 1×0 j 2×0 j 3×0 
x[ 0] =  X [0]e 4 + X [1]e 4 + X [ 2]e 4 + X [3]e 4

4 
=
1
4
{
6e0 + ( −2 + j 2)e0 + (−2)e0 + ( −2 − j 2)e0 = 0 }
2π 2π 2π 2π
1 j 0×1 j 1×1 j 2×1 j 3×1 
x[1] =  X [0]e 4 + X [1]e 4 + X [ 2]e 4 + X [3]e 4 
4 
=
1
4
{
6e0 + ( −2 + j 2)e jπ / 2 + ( −2)e jπ + ( −2 − j 2)e j3π / 2 }
1
= {6 + ( −2 + j 2) j + (−2)(−1) + ( −2 − j 2)(− j )} = 1
4
a) Algebraic Formulation of DFT
b) Matrix Formulation of DFT
c) Inherent Periodicity of DFT and IDFT
Algebraic Formulation of DFT
(1)
N −1 2π
−j
X [ k ] =  x[n ]e
kn
N
, 0 ≤ k ≤ N −1
n =0
N −1 2π
1
 X [k ]e
j kn
x[n] = N
, 0 ≤ n ≤ N −1
N k =0
N −1 N −1
1
x[n] =
N
 X [k ]W
k =0
− kn
N
←
−j
2π → X
N , WN = e
DFT
[ k ] =  x[
N
n ]WN
kn
n= 0
Synthesis equation Analysis equation
(2)
N
2π
 
(W )
−j k
−k N − j 2π k
N = e N
 = e =1
 
Ex. 1
W32 W64 W65
N =3
1 1
0.8
0.6
0.8
0.6
N =6
0.4 0.4
Imaginary Par t
Imaginary Par t
0.2 0.2
0
W30 0
W63 W60
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
W61
-1 -1
1
W
-1
3
-0.5 0
Real Part
0.5 1 -1 W62 -0.5 0
Real Par t
0.5 1
N −1
0, if m = 1
W N
km
=
if m = 0
k =0 N,
(3)
N
2π
 
(W )
N −j k
−k − j 2π k
N = e N
 = e =1
 
Ex. 1 2 −j
2π
0
W W =e0 3
= e0 = 1 = 1 0
N =3
1
3 3
0.8
2π
−j 1  2π   2π 
W =e = cos  − + − 3 
0.6
1 3
0.4 3  j sin cos
 3   
Imaginary Par t
0.2
0
W30 = −0.5 − j0.866 = 1 − 2π / 3
-0.2
2π
-0.4 −j 2  4π   4π 
-0.6
W =e
3
2 3
= cos  −  + j sin cos − 3 
-0.8
 3   
-1
W31
-1 -0.5 0 0.5 1
= −0.5 + j 0.866 = 1 2π / 3
Real Part
N −1
0, if m = 1
W N
km
=
if m = 0
k =0 N,
Ex. 2 (4)
Find DFT of unit sample pulse?
N −1 N −1
X [ k ] =  x[ n]W N
kn
=  δ [ n]WNkn
n =0 n =0
N −1
X [0] =  δ [ n]WN0 n
n= 0
= δ [0]WN0×0 + δ [1]WN0×1 + δ [ 2]WN0×2 + ⋯ + δ [ N − 1]WN0( N −1)

= 1 × 1 + 0 ×1 + 0 × 1 + ⋯ + 0 × 1 = 1
N −1
X [1] =  δ [n]WN1n
n =0
= δ [0]WN1×0 + δ [1]WN1×1 + δ [2]WN1×2 + ⋯ + δ [ N − 1]WN1( N −1)

= 1× 1 + 0 × WN1×1 + 0 × WN1×2 + ⋯ + 0 × WN1( N −1) = 1
Ex. 2 (5)
Find DFT of unit sample pulse? X [k ] = {1, 1,1,..., 1}
N −1 2π 2π 2π
1 1 N −1 j N kn 1 N −1
 X [k ]e = 1e e
j kn j kn
x[n] = N
= N
N k =0 N k =0 N k =0
1  j 2Nπ 0×0 2π
j 1× 0
2π
j ( N −1)× 0  1 0 0
x[0] =  e + e N
+ ... + e N
 = 
 e + e + ... + e 0
 = 1
N   N
1  j 2Nπ 0×1 2π
j 1×1
2π
j ( N −1)×1 
x[1] = e + e N + ... + e N 
N  
→ x[1] = 0
N −1
0, if m = 1 2π
 N
j k
W km
= , WNk = e N
k= 0 N , if m = 0
Matrix Formulation of DFT (1)
 X [0]  1 1 ⋯ 1   x[0] 
 X [1]  1 W 1 ⋯ WNN −1   x[1] 
 = N  
 ⋮  ⋮ ⋮ ⋱ ⋮  ⋮ 
   N −1  
 X [ K − 1]  1 W N ⋯ WN( N −1)( N −1)   x[ N − 1]
↔ X N = WN x N → x N = WN−1X N
T
w k = 1 W k
N
⋯ W ( N −1) k
N

w = 1*
H
k  (W ) k *
N ⋯ (W ( N −1) k *
N ) 
N −1
0, if k ≠ m
w w m = W
H − kn
W mn
=
if k = m
k N N
n =0  N,
X N = WN x N → x N = WN−1X N
N −1
0, if k ≠ m
w w m = W
H − kn
W mn
=
if k = m
k N N
n =0  N,
→ WNH WN = NI N
1 H 1 *
→ WN−1 = WN = WN
N N
1 H 1 *
→ x N = WN X N = WN X N
N N
 X [0]  1 1 ⋯ 1   x[0] 
 X [1]  1 W 1 ⋯ WNN −1   x[1] 
 = N  
 ⋮  ⋮ ⋮ ⋱ ⋮  ⋮ 
   N −1  
 X [ K − 1]  1 W N ⋯ WN( N −1)( N −1)   x[ N − 1]
X N = WN x N
N −1
1 H 1 * 1
x N = WN X N = WN X N =
N N N
 X
k =0
[ k ] w *
k
Ex.
Given x[0] = 0, x[1] = 1, x[2] = 2, x[3] = 3. Find their DFT coefficients?
0× 0
 X [0]   WN WN1×0 WN2×0 ⋯ WN( N −1)0   x[0] 
 X [1]   W 0×1 W 1×1
W 2×1
⋯ WN ( N −1)1  
x[ 1] 
   N N N  
 X [ 2]  =  WN0×2 WN1×2 WN2×2 ⋯ WN ( N −1) 2
  x[2] 
    
 ⋮   ⋮ ⋮ ⋮ ⋱ ⋮  ⋮ 
 X [ N − 1] WN0( N −1) WN1( N −1) WN2 ( N −1) ⋯ ( N −1)( N −1) 
WN   x[n − 1]
 X [0] W40×0 W41×0 W42×0 W43×0   0

 X [1]  W 0×1 
W41×1 W42×1 W43×1  1 
→ = 4  
 X [2] W40×2 W41× 2
W42× 2
W4   2
3×2
   0×3 1×3 2× 3 3×3   

 X [3]  W4 W4 W4 W4   3
Ex.
 X [0] W40×0 W41×0 W42×0 W43×0   0

 X [1]  W 0×1 W41×1
W42×1 3×1   
W4  1
 = 4  
 X [2] W40×2 W41×2
W42× 2
W4   2
3×2
   0×3 1×3 2×3 3× 3   

 X [3] W4 W4 W4 W4   3
W40 W40 W40 W40   0 1 1 1 1   0

 0   3
W41 W42 W43  1  1 W4 W4 W4   1
1 2
W4
=   =  
W40 W4 W4 W4   2 1 W4 W4 W4   2
2 4 6 2 4 6
 0 3 6 9   3 6 9 
W4 W4 W4 W4   3  1 W4
W 4
W 4   3
Ex.
 X [0] 1 1 1 1  0  W32 = W35

N =3
1
 X [1]  1 W 1 W 2 W 3  1  0.8
= 4 
0.6
 4 4
  0.4
 X [2] 1 W4 W4 W4  2 
2 4 6
Imagi nary Part

0.2
W30 = W33
   9 
0
3 6 -0.2
 X [ 3]  1 W4
W 4
W 4   3 -0.4
-0.6
2π
−j k -0.8
W 4
k +4
=W = e
4
k 4 -1
W31 = W34
-1 - 0.5 0 0.5 1
Real Pa rt
 X [0] 1 1 1 1   0
 X [1]  1 W 1 W 2 W 3  1 
→ = 4 4 4 
 
 X [2] 1 W4 2
1 W4  2 
2
   3 2 1 
 X [ 3]  1 W4
W 4
W 4   3
Ex.
 X [0] 1 1 1 1   0 1 1 1 1   0   6 
 X [1]  1 W 1 W 2 W 3   1 1 − j −1 j   1   −2 + j2 
 = 4 4 4 
  =   =  
 X [2] 1 W42 1 W42   2 1 −1 1 −1   2   −2 
   1      
 X [ 3]  1 W4
3
W 4
2
W 4  
3 1 j − 1 − j   
3 − 2 − j 2 
2π π
k+ 4
−j k −j k π  π 
W 4 =W = e
4
k 4
=e 2
= cos  k  − j sin  k 
2  2 
W41 = cos(π / 2) − j sin(π / 2) = − j
W42 = cos(2π / 2) − j sin(2π / 2) = −1
W43 = cos(3π / 2) − j sin(3π / 2) = j

Ex.
 X [0]  6 
 X [1]  −2 + j 2
 = 
 X [2]  −2 
   
 X [ 3]   −2 − j 2 
1
x N = WN* X N
N
1 1 1 1  1 1 1 1 
1 − j −1 j  1 j −1 − j 
WN =  →W =
* 
1 −1 1 −1 N
1 −1 1 −1
   
1 j −1 − j  1 − j −1 j 
 x[0] 1 1 1 1   6  0 
 x[1]  1 j −1 − j  −2 + j 2  1 
1
→ =   = 
 x[2] 4 1 −1 1 −1  −2  2 
      
 x [ 3]  1 − j −1 j  − 2 − j 2   3
Inherent Periodicity of
DFT and IDFT
WN( k +N )n = WNkn = WNk ( N +n )
0 n
N −1
x[n]
N −1
Xɶ [ k + N ] = Xɶ [k ] =  x[n]WNkn , for all k DFT
n =0
IDFT
xɶ[n + N ] = xɶ[n], for all n xɶ[n ]
⋯ ⋯
0 N −1 n
a) Frequency – Domain Sampling
b) Time – Domain Aliasing
c) Reconstruction of DTFT
d) Relationship between CTFT, DTFT, and DFT
Frequency – Domain Sampling
(1)
X(z)
Unit circle
2π
X[k]
1
x[n] =  Xɶ (e )e
jω − jω n
dω
2π 0
2π ∞ 2π
−j
 x[ n]e
j k kn
X [ k ] = Xɶ ( e N
)= N
n =−∞
2π
j k 2π
X [ k ] = X ( z ) z=e j (2 π / N )k = Xɶ (e N
) 0 N
N −1
1
0
jω
(2) X (e )
x[ n]
IDTFT
ω
0 n 0 π 2π
X ( e jω )
∞
 x[n − 12l]
l =−∞
IDFT
N = 12
ω
0 n 0 π 2π
∞
X ( e jω )
 x[n − 6l ]
l =−∞
IDFT
N=6
ω
0 n 0 π 2π
(3)
∞ 2π ∞ lN + N −1 2π
−j −j
 
kn
 x[ n]e
kn
X [k] = N = x[n]e N
, 0 ≤ k ≤ N −1
n= −∞ l =−∞ n= lN
N −1 2π
 ∞  − j N kn
=    x[n − lN ]  e
n = 0  l =−∞ 
∞ N −1 2π
 x[n − lN ] = xɶ[n ] =  cɶ e
j kn
N
k
l =−∞ k =0
N −1 2π
1 −j
 xɶ[n]e
kn
cɶk = N
N k =0
N −1 2π 2π N −1
1 1
 Xɶ ( e 
j kn j kn
− kn
→ xɶ[n ] = N
)e N
= X [ k ]WN
N k =0 N k =0
Time – Domain Aliasing
Ex. 1
1
x[n] = a u[n ], 0 < a < 1 ←→ Xɶ ( e ) =
DTFT
n jω
1 − ae− jω
2
N= 8
N = 16
Origin
1 .5
Amplitude
N −1 2π 2π
1 j k j kn 1
xɶ[n ] =
N
 Xɶ (e
k =0
N
)e N
0 .5
0
2 4 6 8 10 12 14 16
x[n] → Xɶ → Xɶ N → x[ n] → Xɶ
n
10
Origin
9
N = 16
8 N= 8
7
M agnitude
2π N −1 2π 6
) =  xɶ[ n]e
j k j kn
Xɶ N (e N N 5
4
k =0
3
1
100 200 300 400 500 600
ω
Reconstruction of DTFT (1)
2π
j k
x[ n] 
DTFT
→ X [k ] = Xɶ ( e N ), 0 ≤ k ≤ N − 1 
IDTFT
→ x[ n], 0 ≤ n ≤ N − 1
2π N −1 2π
−j
) =  x[n]e
j k kn
Xɶ ( e N K
, 0 ≤ k ≤ K −1
n =0
 x[n], 0 ≤ n ≤ N − 1
xzero − padded [n ] = 
0, N ≤ n ≤ K −1
2π N −1 2π N −1
−j
) =  x[ n]e =  xzp [ n]WKkn = X zp [k ]
j k kn
→ Xɶ ( e N K
n =0 n =0
Ex. 2
1, 0 ≤ n ≤ N −1
x[n] = u[n] − u[n − N ] = 
0, otherwise
N −1
1 − e− jω N sin(ω N / 2) − jω ( N −1) / 2
Xɶ ( e jω ) =  x[n]e − jω n = − jωn
= e
n =0 1− e sin(ω / 2)
2π
j k sin(π Nk / K ) − jπ ( N −1) k / K
X [ k ] = Xɶ (e K
)= e
sin(π k / K )
N −1
X ( z ) =  x[n ]z −n
n= 0
2π 2π
1
N −1 N −1
 −n
=   Xɶ (e
j k j kn
N N
)e z
n =0  N k =0 
n
2π 2π
1 N −1 j k N −1  
 Xɶ (e )  e
j kn
−1
= N N
z 
N k =0 k =0  
2π
j k
1 − z−N N −1
Xɶ (e N
)
=
N

k =0
j
2π
k
1−e N
z −1
Relationships between CTFT,
x (t)
c
DTFT, and DFT X c ( jΩ)
CTFT
Ω
0 t 0
Sampling Periodization
1 ∞  2π 
X ( e ) =  X c  jΩ − jm
ɶ j ΩT
x[ n] = xc ( nT ) 
T m =−∞  T 
DTFT
0 nT 0 Ω
Periodization Sampling
 2π k 2π 
∞
1 ∞
xɶ[ n] =  x (nT − lNT )
c X [k ] =  X c  j
T m= −∞  NT
− jm
T 

l =−∞
DFT
n N
0 s i tes.google.com/site/ncpdhbkhn 0 N 39
N k
a) Linearity
b) Periodic, Circular, and Modulo – N Operations
c) Symmetry
d) Circular Shift of a Sequence
e) Circular Convolution
f) Circular Correlation
g) DFT of Stretched and Sampled Sequences
Linearity
y[n ] = a1 x1[n ] + a2 x2 [n ]
N1 −1 N 2 −2
→ Y [k ] = a1  x1[n ]W + a2  x2 [n ]WNkn , 0 ≤ k ≤ N − 1
kn
N
n =0 n =0
= a1 X 1[k ] + a2 X 2 [k ], 0 ≤ k ≤ N −1
a) Linearity
c) Symmetry
Periodic, Circular, and
Modulo – N Operations (1)
x[n]
n = lN + r, 0 ≤ r ≤ N − 1
 n N
= n modulo N = r n
0 ( N − 1)
xɶ[n]
xɶ[n] = x[ n N ], for all n
n
0 N
Xɶ [k ] = X [ k ], for all k x[4] x  n 

N
x[3] N
x[5] x[2]
n=0
2 1
0
n
x[n]
x[ n N
]
n
0 ( N − 1)
N −1
xɶ[n]
1 0
n
0 N
zɶ[n ] = xɶ[− n ]
x[ −n N ]
N −1 n
0 N
1, 0 ≤ n ≤ N − 1
pN [ n] = 
z[n] = zɶ[n] pN [ n] 0, otherwise
1 0
0 n
s i tes.google.com/site/ncpdhbkhn N 44
x[n] = x[ −n N ], (circular even symmetry)

x[n] = − x[ −n N ], (circular odd symmetry)
 x[0], n=0
x[n] =  (circular even symmetry)
 x[ N − n], 1 ≤ n ≤ N −1
0, n=0
x[n] =  (circular odd symmetry)
− x[ N − n], 1 ≤ n ≤ N −1
a) Linearity
c) Symmetry
Symmetry (1)
1 1
0.8 0.8
0.6 0.6
x1 [n]
x2 [n]
0.4 0.4
0.2 0.2
0 0
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7
n n
3 3
2.5
2
2
X 1R [k]
X 2R [k]
1.5 1
1
0
0.5
0 -1
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7
k k
2 2
1 1
X 1I[k]
X 2I[k]
0 0
-1 -1
-2 -2
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7
k k
Symmetry (2)
x[ n] = xR [n] + jxI [n ], 0 ≤ n ≤ N −1
X [ k ] = X R [k ] + jX I [k ], 0 ≤ k ≤ N − 1
N −1 2π N −1 2π
−j 1
X [ k ] =  x[n ]e  X [ k ]e
kn j kn
N
, x[n ] = N
n =0 N k =0
 N −1
  2π   2π  
X
 R [k ] =  x
 R
n =0 
[ n ]cos 
 N
kn +
 I

x [ n ]sin 
 N
kn  ,

0≤k < N

 N −1
  2π   2π  
 X I [k ] =   xR [n]sin  kn  − xI [n ]cos  kn  , 0≤k < N
 n =0   N   N 
→
 x [ n] =  X [ k ]cos  2π kn  − X [k ]sin  2π kn   ,
N −1
 R   R
k =0 
 N



I  N



0≤n< N


 N −1
  2π   2π  
x
 I [ k ] =  R X [k ]sin  N

kn 

+ X I [ k ]cos  N kn  ,
 
0≤k< N
 k =0 
Symmetry (3)
 N −1
  2π   2π 
 X R [k ] =   xR [n]cos  kn  + xI [n ]sin  kn , 0 ≤ k < N
 n= 0   N   N 

 X [k ] =  x [n ]sin  2π kn  − x [n]cos  2π kn  , 0 ≤ k < N
N −1
 I   R 

 I





 n = 0  N N 
If x[n] is real,then xI [ n] = 0
 N −1
 2π  N −1  2π 
X
 R

[ k ] = 
n =0
x R [ n ]cos 
 N
kn  
=
 n =0
x [ n ]cos 
 N
kn 

→
 X [ k ] = x [n]sin  2π kn  = x[n]sin  2π kn 
N −1 N −1
 I 
n= 0
R  N

 
 n =0
 N



  X R [0], k =0
 X R [k ] = X R [ − k N ] = 
  X R [ N − k ], 1 ≤ k ≤ N − 1
 0, k =0

→  X I [ k ] = − X I [ −k N ] = 
  − X I [ N − k ], 1 ≤ k ≤ N − 1
  X [0], k =0
 X *[k ] = X [ −k ] = 
  X [ N − k ], 1 ≤ k ≤ N − 1
N
Symmetry (4)
x[n] = xce [ n] + x co [n], 0 ≤ n ≤ N −1
x[n] + x[ − n  x[ n] + x[ N − n ]
]  , 1 ≤ n ≤ N −1
x ce[ n] = N
= x ce[ −n N ] =  2
2  x[0], n=0
x[ n] − x[ −n  x[ n] − x[ N − n]
]  , 1 ≤ n ≤ N −1
x co [n ] = N
= − x [ −n N ] = 
ce
2
2 0, n=0
a) Linearity
c) Symmetry
Circular Shift of a Sequence
x[ n N ] x[n]
N −1 n
0 N −1
1 0 xɶ[n ]
x[ n − m N ] ⋯ ⋯ n
0 N −1
zɶ[n ] = x[n − 2]
ɶ
1 ⋯ ⋯ n
N −1 0
0 N −1
z[n] = zɶ[n ] pN [n] = x[ n − 2 N ]
x[ n − m N
] ←
DFT
N
→ WN
km
X [k ]
n
0
s i tes.google.com/site/ncpdhbkhn N −1 52
a) Linearity
c) Symmetry
Circular Convolution (1)
∞
y[n] =  h[m]x[n − m]
m =−∞
←
DTFT
→ Yɶ (e jω ) = Hɶ (e jω ) Xɶ (e jω )
Y [k ] = H [k ] X [k ], 0 ≤ k ≤ N − 1
N −1
1
y[n] =
N
 Y [k ]W
k =0
− kn
N
N −1 N −1
1 1  N −1 km  
N −1
kl 
=
N
 H [k ] X [k ]W
k=0
− kn
N
=
N
 
k = 0 m = 0
h[ m ]WN  
  l =0
x [l ]WN  N

W − kn
N −1 N −1 N −1
1 1 k ( m +l − n ) 
=
N
 
m =0
h[ m ]
l =0
x[l ] N


k =0
WN 

1 N −1
1, m + l − n = rN
W k ( m +l −n )
N =
N k =0 0, otherwise
m + l − n = rN → l = n − m + rN = (n − m) modulo N = n − m N
N −1 N −1
y[n] =  h[m ]x[n − m] =  h[m]x[ n − m N
]
m =0 m= 0
N −1 N −1
y[ n] =  h[m ]x[n − m ] =  h[m] x[ n − m N
]
m= 0 m =0
y[n ] = h[n] N x[n]
y[n ] = h[n] N x[ n] ←

DFT
N
→ Y [ k ] = H [k ] X [k ]
x [n ] x[ n − 2 8 ] x[ n − 8 8 ]
x[4] x[3] x[2] x[1] x[4] x[3]
x[5] x[2] x[3] x[0] x[5] x[2]
n=0 n=0 n=0
Circular mapping Circular addressing

Ex. 3
Given h[ n] = {1 2 3 4}, x[n ] = {0 1 2 3}. Find y[ n] = h[n] N x[n].

↑ ↑
N −1
y[n ] =  h[m]x[ n − m N ] Method 1
m =0
y[0] = h[0] x[0] + h[1]x[3] + h[2]x[2] + h[3]x[1] x[3]

x[2]
x[0]
x[1] x[0]
x[3]
x[1]
x[2]
= 1 × 0 + 2 × 3 + 3 × 2 + 4 × 1 = 16 h[ 2] h[1]
y[1] = h[0]x[1] + h[1]x[0] + h[2]x[3] + h[3]x[2]
nn = 1
0
23
= 1 × 1 + 2 × 0 + 3 × 3 + 4 × 2 = 18
y[2] = h[0]x[2] + h[1]x[1] + h[2]x[0] + h[3]x[3] h[ 3] h[ 0]

= 1 × 2 + 2 × 1 + 3 × 0 + 4 × 3 = 16 x[1]
x[0]
x[2]
x[3] x[0]
x[3]
x[1]
x[2]
y[ 3] = h[ 0] x[ 3] + h[1] x[ 2] + h[ 2] x[1] + h[ 3] x[ 0]
= 1 × 3 + 2 × 2 + 3 × 1 + 4 × 0 = 10
Ex. 3

↑ ↑
y[0] = h[0]x[0] + h[1]x[3] + h[2]x[2] + h[3]x[1] = 1 × 0 + 2 × 3 + 3 × 2 + 4 × 1 = 16

y[1] = h[0]x[1] + h[1]x[0] + h[2]x[3] + h[3]x[2] = 1 ×1 + 2 × 0 + 3 × 3 + 4 × 2 = 18
y[2] = h[0]x[2] + h[1]x[1] + h[2]x[0] + h[3]x[3] = 1 × 2 + 2 × 1 + 3 × 0 + 4 × 3 = 16
y[3] = h[0]x[3] + h[1] x[2] + h[2]x[1] + h[3]x[0] = 1 × 3 + 2 × 2 + 3 × 1 + 4 × 0 = 10
 y[0]  x[ 0] x[3] x[ 2] x[1]   h[0]

 y[1]   x[1] x[ 0] x[ 3] x[2]  h[1] 
→ =  
 y[ 2]  x[ 2] x[1] x[ 0] x[ 3]  h[ 2]
    
 y[ 3]   x[3] x[ 2] x[1] x[0]  h[ 3] 
Ex. 3

↑ ↑
y[ n] = h[n] N x[n] ←

DFT
N
→ Y [k ] = H [k ] X [k ] Method 2
 X [ 0] 1 1 1 1   0 1 1 1 1   0  6 
 X [1]  1 W 1 W 2 W 3   1 1 − j −1 j   1  −2 + j 2 
 = 4 4 4  
=   = 
 X [ 2] 1 W4 2
1 W4   2 1 −1 1 −1  2  −2 
2
   1       
 X [ 3]  1 W4
3
W 4
2
W 3
4   1 j − 1 − j 3
   −2 − j 2 
 H [ 0]  1 1 1 1   0 1 1 1 1   1  10 
 H [1]  1 W 1 W 2 W 3   1 1 − j −1 j   2  −2 + j 2 
 = 4 4 4 
 =  =  
 H [ 2]  1 W4 2
1 W4   2 1 −1 1 −1  3  −2 
2
   1       
 H [ 3]  1 W 4
3
W 4
2
W 3
4   1 j − 1 − j 4
   −2 − j 2 
Ex. 3

↑ ↑
y[ n] = h[n] N x[n] ←

DFT
N
→ Y [k ] = H [k ] X [k ] Method 2
X = [6; − 2 + j 2; − 2; − 2 − j 2] ; H = [10; − 2 + j 2; − 2; − 2 − j 2]
→ Y [0] = H [ 0] X [ 0] = 6 × 10 = 60; Y [1] = − j 8; Y [ 2] = 4; Y [ 3] = j8
 y[0] 1 1 1 1  Y [0] 1 1 1 1   60  16

 y[1]  1 − j −1 j   Y [1]  1 − j −1 j   − j8 10
 1
=   1
=   =  
 y[2] 4 1 −1 1 −1 Y [2] 4 1 −1 1 −1  4  16
         
 y[3] 1 j −1 − j  Y [3] 1 j −1 − j   j 8  18
a) Linearity
c) Symmetry
Circular Correlation
N −1
rxy [l ] =  x[ n] y[ n − l N
], 0 ≤ l ≤ N −1
n= 0
rxy [l ] = x[n] N y[ −n N
] ←
DFT
N
→ R xy [ k ] = X [ k ]Y *
[k ]
a) Linearity
c) Symmetry
DFT of Stretched Sequences
x[n]
n
0 N −1
x ( L ) [n ]
x( L ) [n ] ←
DFT
NL
→ Xɶ [k ] = X [ k N
]
n
0 L −1
1 ( L) 1
x [n ] = xɶ[n ] ←
DFT
→ X [k ]
L L NL  x[ n / L], n = 0, L, 2 L, ..., ( N − 1) L
x [ n] = 
( L)
0, otherwise
DFT of Sampled Sequences
x[n]
n
M −1 N −1
1  N 0
x( M ) [n ] ←→
DFT
N /M
M

m =0
X k + m 
 M x( M ) [n ]
n
0
M −1
1  N
M
  x n + m 
M
←
DFT
N /M
→ X (M )[k ] x( M )[ n] = x[ nM ], 0≤ n ≤
N
−1
m= 0 M
Properties of the DFT
a) Linear Convolution Using Circular Convolution
b) Implementation of FIR Filters Using the DFT
Linear Convolution Using
∞
y[n ] =  h[k ]x[n − k ], 0≤n≤L+M −2
k =−∞
Y ( e jω ) = H (e jω ) X (e jω )
Y [ k ] = Y (e j 2π k / N ), 0 ≤ k ≤ N − 1
H [k ] = H (e j 2π k / N ), X [k ] = X (e j 2π k / N )
→ yzero − padded [n] = hzp [n] N xzp [ n] ←

DFT
N
→ Y [ k ] = H [k ] X [k ]
x[n] x zp [n ] X [k ]
Pad with N – point
Length (M – 1) zeros DFT Y [ k ] = H [k ] X [k ]
L y [n]
N – point
N = L + M −1
IDFT Length
h[n] Pad with N – point N
Length (L – 1) zeros h [n ] DFT H [k ]
zp
M s i tes.google.com/site/ncpdhbkhn 67
Linear Convolution Using
∞
y[n ] =  h[k ]x[n − k ], 0≤n≤L+M −2
k =−∞
 y[0]  x[0] 0 0 
 y[1]   x[1] x[0] 0 
   
 y[2]  x[2] x[1] x[0]  h[0]
    
→  y[3] =  x[3] x[2] x[1]   h[1] 
 y[4]  x[4] x[3] x[2]  h[2]
   
 y [5 ]   0 x[4] x[3]
 y[6]  0 0 x[4]
 x[0] 0 0 x[4] x[3] x[1]  h[0]
x[2]
 x[1] x[0] 0 0 x[4] x[3] x[2]  h[1] 
  
 x[2] x[1] x[0] 0 0 x[4] x[3] h[ 2]
  
=  x[3] x[2] x[1] x[0] 0 0 x[4]  0 
 x[4] x[3] x[ 2] x[1] x[0] 0 0  0 
  
 0 x[4] x[3] x[2] x[1] x[0] 0   0 
 0 0 x[ 4] x[3] x[2] x[1] x[0]  0 
a) Linear Convolution Using Circular Convolution
b) Implementation of FIR Filters Using the DFT
Implementation of FIR Filters
Using the DFT (1)
x[ n ] x zp [n ] X [k ]
Pad with N – point
Length (M – 1) zeros DFT Y [ k ] = H [k ] X [k ]
L y [n]
N – point
N = L + M −1
IDFT Length
h[n ] Pad with N – point N
Length (L – 1) zeros h [n ] DFT H [k ]
zp
M
• The input may have indefinite length

• The input’s length may be too large for storage &
computation
• The computation can not be started untill all input
signal samples have been collected  delays
Using the DFT (2)
The overlap – add method
 y[0]  h[0] 0 0 0 0 0 0 0 
 y[1]   h[1] h[0] 0 0 0 0 0 0   x[0]
   
 y[2]  h[2] h[1] h[0] 0 0 0 0 0   x[1] 
    
 y[3]  0 h[2] h[1] h[0] 0 0 0 0   x[2]
 y[4]  0  
0 h[2] h[1] h[0] 0 0 0   x[3] 
 = 
 y[5]  0 0 0 h[ 2] h[1] h[0] 0 0   x[4]
 y[6]  0  
0 0 0 h[2] h[1] h[0] 0   x[5] 
   
 y[7]  0 0 0 0 0 h[2] h[1] h[0]  x[6]
 y[8]  0   
0 0 0 0 0 h[2] h[1]  x[7]
   
 y[9]  0 0 0 0 0 0 0 h[ 2]
Using the DFT (3)
The overlap – add method
 y[0]  h[0] 0 0 0 0 x[1] 0

0 0 
 y[1]   h[1]  x[2]
y[h4[]0]= [h[00] 0 0 h[2] 0 h[1]]0 0   x[0]
0
   
 y[2]  h[2] h[1] h[0] 0 0 0 x[3]0 0   x[1] 
      
 y[ 3]   0 h[2] h[1] h[0] 0 0 x[ 4 ] 0
0   x[2]
 y[4]  0  
0 h[2] h[1] h[0] 0 0 0   x[3] 
 = 
 y[5]  0 0 0 h[ 2] h[1] h[0] 0 0   x[4]
 y[6]  0  
0 0 0 h[2] h[1] h[0] 0   x[5] 
    x[3] 
 y[7]  0 0 0 0 0 h[2] h[1] h[0]  x[6]
x[4]  
 y[8]  0 y[50] = [ h[01] h[0
0] 0 0 h[2]]0  
h[2] h[1]  x[7]
    x[5] 
 y[9]  0 0 0 0 0 0 0 h[ 2]
 x[ 6] 
Using the DFT (4)
The overlap – save method
 y[0]  h[0] 0 0 0 0 0 0 0 
 y [1]   h[1] h[0] 0 0 0 0 0 0   x[0]
   
 y [2]  h[2] h[1] h[0] 0 0 0 0 0   x[1] 
    
 y[3]  0 h[2] h[1] h[0] 0 0 0 0   x[2]
 y [4]  0  
0 h[2] h[1] h[0] 0 0 0   x[3] 
 = 
 y[5]  0 0 0 h[ 2] h[1] h[0] 0 0   x[4]
 y[6]  0  
0 0 0 h[2] h[1] h[0] 0   x[5] 
   
 y[7]  0 0 0 0 0 h[2] h[1] h[0]  x[6]
 y[8]  0   
0 0 0 0 0 h[2] h[1]  x[7]
   
 y[9]  0 0 0 0 0 0 0 h[ 2]
a) Effects of Time – Windowing on Sinusoidal Signals
b) Effects of Time – Windowing on Signals with Continuous Spectra
c) “Good” Window and the Uncertainty Principle
d) The Spectrogram
Effects of Time – Windowing on
Sinusoidal Signals (1)
1
xc (t )
0
-1
-6 -4 -2 0 2 4 6
t
1
1, 0 ≤ t ≤ T0 / 2
wc (t ) = 
wc(t)
0.5
0, otherwise 0
-6 -4 -2 0 2 4 6
t
1
wc(t)xc(t)
xˆc (t ) = wc (t ) xc (t ) 0
-1
-6 -4 -2 0 2 4 6
t
1, 0 ≤ n ≤ L − 1
w[n] =  → xˆ[n] = xˆc (nT ) = wc (nT ) x c (nT ) = w[ n]x[ n]
0, otherwise
x c (t ) x[ n ] xˆ[ n ] Xˆ [k ]
N – point
Ideal A/D
DFT
1
T= w[n ]
Fs
x c (t ) xˆc (t ) xˆ[n ] Xˆ [k ]
N – point
Ideal A/D
DFT
1
w c (t ) T=
Fs
x c ( t ) = A1 cos(Ω1t + φ1 ), −∞<t <∞
1 1
= A1e − jφ1 e − jΩ1t + A1e jφ1 e jΩ1t
2 2
1 − jφ1 1
xˆc (t ) = wc (t ) x c (t ) = A1e wc (t )e − jΩ1t
+ A1e jφ1 wc (t )e jΩ1t
2 2
1 1
→ Xˆ c ( jΩ) = A1e − jφ1Wc [ j(Ω + Ω1 )] + A1e jφ1Wc [ j(Ω − Ω1 )]
2 2
1, 0 ≤ t ≤ T0
wc (t ) = wR (t) = 
0, otherwise
sin(Ω T0 / 2) − jΩT0 / 2
Wc ( jΩ) = WR ( jΩ ) = e
Ω/2
Sinusoidal Signals (4) 1
1
0.5 xc ( t ) X c ( jΩ)
0 0.5
-0.5 −Ω1 Ω1
-1
t 0 Ω
-6 -4 -2 0 2 4 6 -6 -4 -2 0 2 4 6
1 3
wc (t ) 2 Wc ( jΩ)
0.5 1
Ω
0
0 t -1
-6 -4 -2 0 2 4 6 -6 -4 -2 0 2 4 6
T0
3
0
Ω
xˆc (t ) = wc (t ) xc (t )
-1
-6 -4 -2 0 2 4 6
1 3
0.5 2 Xˆ c ( jΩ)
0 1
Ω
-0.5 0
-1 t -1
-6 -4 -2 0 2 4 6 -6 -4 -2 0 2 4 6
3
W c(j(Ω - Ω2))
2.5 W c(j(Ω - Ω1))
1.5
0.5
-0.5
0 2 4 6 8 10 12
Ω
2π 1
∆Ω = Ω 2 − Ω1 ≥ Ω 0 = → T0 ≥
T0 F2 − F1
W c(j(Ω - Ω2)) + Wc(j(Ω - Ω1))
3.5
2.5
1.5
0.5
0 2 4 6 8 10 12
Ω
d) The Spectrogram
Signals with Continuous Spectra
1
Spectrum of infinite duration signal
Magnitude
0.5
0
-8 -6 -4 -2 0 2 4 6 8
F (Hz)
1
Spectrum of rectangular window
Magnitude
0.5
-0.5
-8 -6 -4 -2 0 2 4 6 8
F (Hz)
1
Shifted copies of window spectrum
M agnitude
0.5
200 400 600 800 1000 1200 1400 1600

F (Hz)
Spectrum of windowed si gnal

Magnitude
0
-8 -6 -4 -2 0 2 4 6 8
F (Hz) ∞
1 ∞ 1
Xɶ c ( jΩ ) =
2π
θ )W c
( j ( θ )) d θ
−∞ c s i tes.google.com/site/ncpdhbkhn
X ( j Ω − ≈
2π
X
k = −∞
c
( jθ k )Wc ( j(Ω − θk )) ∆θ 81
d) The Spectrogram
“Good” Windows and
the Uncertainty Principle (1)
Smearing & leakage
1  jΩ 
x c ( at ) ←
CTFT
→ Xc  
a  a 
xc (at ) = cos[Ω0 (at )] = cos[(aΩ 0 )t ]
Time and frequency scaling property
Uncertainty principle
∞
σ =  t xc (t ) dt
2 2 2
t −∞
1 ∞
σ =  Ω 2 X c ( jΩ) d Ω
2 2
Ω
2π −∞
1
σ tσ Ω ≥
2
12
WR(jΩ ) wRectangular (t)
0.8
10
0.6
8
0.4
6
0.2
4
0 2
-0.2
-1.5 -1 -0.5 0 0.5 1 1.5 - 1.5 -1 -0.5 0 0.5 1 1.5
Ω t
2
12
WB (jΩ) = [WR(jΩ) ] wBartlett (t)
0.9
0.8 10
4 sin (Ω T0 / 4)
2
WBartlett ( jΩ ) =
0.7
8
0.6
Ω 2T0 / 2
0.5 6
0.4
0.3 4
0.2
2
0.1
-1.5 -1 -0.5 0 0.5 1 1.5 - 1.5 -1 -0.5 0 0.5 1 1.5

Ω s i tes.google.com/site/ncpdhbkhn t 85
WHann ( j Ω) = 0.5WRectangular ( j Ω)
+ 0.25WRectangular ( j (Ω − 2π / T0 ))
+ 0.25WRectangular ( j (Ω + 2π / T0 ))
  2π t  
wHann (t ) =  0.50 + 0.50 cos    wRect (t )
  T0  
WHamming ( jΩ) = 0.54WRectangular ( jΩ)
+ 0.23WRectangular ( j (Ω − 2π / T0 ))
+ 0.23WRectangular ( j (Ω + 2π / T0 ))
  2π t  
wHamm (t ) =  0.54 + 0.46 cos    wRect (t )
  T0  
1.2
Rectangular 0.5 Bartlett
0.45
1
0.4
0.8 0.35
Amplitude
Amplitude
0.3
0.6 0.25
0.2
0.4
0.15
0.2 0.1
0.05
100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000
F (Hz) F (Hz)
Ex. x(t ) = cos(2π Ft ) + cos(2π 3Ft ) + cos(2π 4Ft )

0.5 0.55
Hann Hamming
0.45 0.5
0.4 0.45
0.35 0.4
0.35
Amplitude
Amplitude
0.3
0.3
0.25
0.25
0.2
0.2
0.15 0.15
0.1 0.1
0.05 0.05
100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000
F (Hz) s i tes.google.com/site/ncpdhbkhn F (Hz) 89
d) The Spectrogram
The Spectrogram (1)
L−1
X [k , n] =  w[m]x[n + m]e− j ( 2π k / N )m
m= 0
The Spectrogram (2)
Direct Computation of
the Discrete – Fourier Transform
N −1
X [ k ] =  x[n]WNkn , k = 0, 1,..., N − 1
n =0
N −1
1
x[n] =
N
 X
k =0
[ k ]WN
− kn
, n = 0,1,..., N − 1
N −1
  2π   2π  
X R[ k ] =   xR [n]cos  kn  + xI [n ]sin  kn  
k =0   N  N 
N −1
  2π   2π 
X I [k ] =   xR[ n]sin  kn  − xI [n]cos  kn 
k =0   N   N 
The total cost of computing all X[k] coefficients is approximately N2 operations!!!
Fast Fourier Transform

8. The Fast Fourier Transform Idea Using a Matrix
Approach
The Fast Fourier Transform Idea
Using a Matrix Approach (1)
 X [0] 1 1 1 1 1 1 1 1   x[0]
 X [1]  1 W8 W82 W83 W84 W85 W86 W87   x[1] 
     a b   e   ae + bf 
 X [2] 1 2 4 6
1 W8 W8 W8   x[ 2]
2 4 6
c =
d   f   ce + df 
W8 W8 W8
   3   
 X [ 3]  = 1 W8 W86 W8 W84 W87 W82 W85   x[3]
 X [4] 1 W8
4
1 W84 1 W84 1 W84   x[ 4]
   5   b a   f   bf + ae 
 X [ 5]  1 W W82 W87 W84 W8 W86 W83   x[5] =
8
d c   e   df + ce 
  
 X [6] 1 W 6
W84 W82 1 W86 W84 W82   x[6] 
  
8
 7 6 5 4 3 2 
 X [7] 1 W8 W8 W8 W8 W8 W8 W8   x[7]
1 1 1 1 1 1 1 1   x[0]
1 W82 W84 W86 W8 W83 W85 W87   x[2]
  
1 W84 1 W84 W82 W8 W8 W8   x[4]
6 2 6
  
1 W86 W84 W82 W83 W8 W87 W85   x[6]
=
1 1 1 1 W84 W84 W84 W84   x[1] 
  
1 W82 W84 W86 W85 W87 W8 W83   x[3]
1 W84 1 W84 W86 W82 W86 W82   x[5]
  
1 W86 W84 W82 W87 W85 W83 W8   x[7]
 X [0] 1 1 1 1 1 1 1 1   x[0] W82
 X [1]  1 W82 W84 W86 W8 W83 W85 W87   x[ 2]
1
     0.8
W83 W81
 X [2] 1 W84 1 W84 W82 W8 W8 W8   x[ 4]
6 2 6 0.6
    
0.4
X [ 3]  = 1 W86 W84 W82 W83 W8 W87 W85   x[6]
Imagi nary Part


0.2
0
W84 8
W80
 X [4] 1 1 1 1 W84 W84 W84 W84   x[1]  -0.2
     -0.4
 X [ 5]  1 W82 W84 W86 W85 W87 W8 W83   x[3] -0.6
 X [6] 1 W 4
1 W 4
W 6
W82 W86 W82   x[5] -0.8
W85 W87
  
8 8 8
 6  -1
 X [7] 1 W W84 W 2

W 7 5 3
W8 W8 W8   x[7] -1 -0.5 0 0.5 1
8 8 8
W86
Rea l Part
1 1 1 1 1 1 1 1   x[0]
1 W82 W84 W86 W83 W85 W87   x[2]

W8
  W84 = −1
1 W84 1 W84 W82 W86 W82 W8   x[4]
6
  W85 = −W81

1 W86 W84 W82 W83 W8 W87 W85   x[6]
= W86 = −W82
1 1 1 1 −1 −1 −1 −1   x[1] 
   W87 = −W83
1 W82 W84 W86 −W8 −W83 −W85 −W87   x[3] 
1 W84 1 W84 −W82 −W86 −W82 −W86   x[5] 
  
1 W86 W84 W82 −W83 −W8 −W87 −W85   x[7]
 X [0] 1 1 1 1 1 1 1 1   x[0]
 X [1]  1 W82 W84 W86 W8 W83 W85 W87   x[2]
    
 X [2] 1 W84 1 W84 W82 W86 W82 W8   x[4]
6
    
 X [ 3]  = 1 W86 W84 W82 W83 W8 W87 W85   x[6]
 X [4] 1 1 1 1 −1 −1 −1 −1   x[1] 
    
 X [ 5]  1 W82 W84 W86 −W8 −W83 −W85 −W87   x[3]
 X [6] 1 W84 1 W84 −W82 −W86 −W82 −W86   x[5]
   5  
 X [7] 1 W86 W84 W82 −W83 −W8 −W87 −W8   x[7]
1 1 1 1  1 0 0 0   x[0]  x[1] 
1 W4 W42 W43  0 W 0 0   x [2 ]  x[3]
W4 =  ; D =  8 ; X
Even =
  ; XOdd =  
1 W42 1 W42  8 0 0 W82 0   x [4 ]  x[5]
       
1 W43 W42 W4  0 0 0 W83   x[6]  x[7]
XTop = [ X [0] X [1] X [2] X [3]] ; X Bottom = [ X [4] X [5] X [6] X [7]]
T T
 XTop   W4 D8 W4   x Even 
→ =
 X Bottom   W4 −D8 W4   xOdd 
 XTop   W4 D8 W4  x Even 
 = X T = X E + DN X O
 Bottom   W4
X −D8 W4   xOdd  →
N X B = X E − DN X O
− point DFT : X E = WN / 2 x E & XO = WN / 2x O
2
N – point Input Sequence
n = Even n = Odd
N/2 – point N/2 – point
Sequence Sequence
xE [n ] xO [ n]
N – point FFT
N/2 – point N/2 – point
DFT DFT
X E [k ] X O [k ]
Merge N/2 – point DFT
N – point DFT Coefficients X [k ]

 X [0] 1 1 1 1 1 1 1 1   x[0]
 X [1]  1 W82 W84 W86 W8 W83 W85 W87   x[2]
    
 X [2] 1 W84 1 W84 W82 W86 W82 W8   x[4]
6
    
 X [ 3]  = 1 W86 W84 W82 W83 W8 W87 W85   x[6]
 X [4] 1 1 1 1 −1 −1 −1 −1   x[1] 
    
 X [ 5]  1 W82 W84 W86 −W8 −W83 −W85 −W87   x[3]
 X [6] 1 W84 1 W84 −W82 −W86 −W82 −W86   x[5]
   5  
 X [7] 1 W86 W84 W82 −W83 −W8 −W87 −W8   x[7]
 X [0] 1 1 1 1 1 1 1 1   x[0]
 X [2] 1 W82 W84 W86 1 W82 W84 W86 
x [1] 
    
 X [4] 1 W84 1 W84 1 W84 1 W84   x [2 ]
  1 W86 W84 W82 1 W86 W84 W82
 
 X [ 6 ]=  x [3]  →  X E  =  W4 W4   x T 
 X [1]  1     
W8 W82 W83 −1 −W8 −W82 −W83   x[4]  X O   W4D 8 − W4D 8   x B 
    
 X [ 3]  1 W83 W86 W8 −1 −W83 −W86 −W8   x[5] 
 X [5] 1 W85 W82 W87 −1 −W85 −W82 −W87   x[6]
    
 X [ 7 ] 1 W87 W86 W85 −1 −W87 −W86 −W85   x[7]
8. The Fast Fourier Transform Idea Using a Matrix Approach
a) Algebraic Derivation
b) Practical Programming Considerations
c) Alternative Forms
Algebraic Derivation (1)
N −1
X [k ] =  x[n]WNkn , k = 0,1,..., N − 1
n= 0
N N
−1 −1
2 2
=  x[2m ]WNk ( 2m ) +  x[2m + 1]WNk (2 m+1)
m =0 m= 0
N N
−1 −1
2 2
=  x[2m ]WNk (2 m ) + WNk  x[2m + 1]WNk (2 m )
m =0 m= 0
a[n] = x[2n], n = 0,1, 2,..., N / 2 − 1

b[n] = x[2n + 1], n = 0, 1, 2,..., N / 2 − 1
WN2 = WN / 2
→ X [k ] = A[k ] + WNk B[k ], k = 0,1,..., N − 1

N / 2 −1
A[k ] = 
m =0
a[m ]WNkm/ 2 , k = 0,1,..., N / 2 − 1
N / 2 −1
B[ k ] = 
m =0
b[m]WNkm/ 2 , k = 0, 1,..., N / 2 − 1
X [ k ] = A[k ] + WNk B[ k ], k = 0, 1, ..., N / 2 − 1
N
 N  N k+  N
→ X  k +  = A  k +  + WN 2 B  k + 
 2  2  2
 N
 k+ 2 
W 
N

= −WNk
 N
→ X k +  = A[k ] − WNk B[k ]
 2
N
X [k ] = A[k ] + WNk B[k ], k = 0, 1, ..., −1
2
 N N
X  k +  = A[k ] − WNk B[k ], k = 0, 1, ..., − 1
 2 2
Ex.
X [k ] = DFT8{x[0], x[1], x[2], x[3], x[4], x[5], x[6], x[7]}, 0 ≤ k ≤ 7
ր
C [k ] = DFT2{x[0], x[4]}, k = 0,1
A[k ] = DFT4{x[0], x[2], x[4], x[6]}, 0 ≤ k ≤ 3
ց
D[k ] = DFT2{x[2], x[6]}, k = 0,1
ր
X [k ]
ց
ր
E[ k ] = DFT2{x[1], x [5]}, k = 0,1
B[k ] = DFT4{x[1], x[3], x[5], x[7]}, 0≤k≤3
ց
F [ k ] = DFT2 {x [3], x[ 7]}, k = 0,1
A[k ] = C[k ] + W82k D[k ], k = 0,1

A[k + 2] = C[k ] − W82 k D[k ], k = 0,1  X [k ] = A[ k ] + W8k B[k ], k = 0, 1, 2, 3
→
 X [ k + 2] = A[ k ] − W8 B[ k ], k = 0,1, 2, 3
k
B[ k ] = E [k ] + W82k F [ k ], k = 0,1
B[ k + 2] = E [k ] − W82 k F [ k ], k = 0, 1
Ex.
X [ k ] = DFT8{x[ 0], x[1], x[ 2], x[3], x[ 4], x[ 5], x[ 6], x[ 7]}, 0 ≤ k ≤ 7
A[ k ] = C[ k ] + W82 k D[ k ], k = 0,1
A[k + 2] = C[ k ] − W82k D[ k ], k = 0,1
 X m [ p] = X m−1[ p] + WNr X m −1[ q]

→
 X m [ q] = X m −1[ p ] − WN X m −1[ q]
r
X m −1[ p] X m[ p ]
WNr
X m −1[ q] X m [ q]
−1
Ex.
X [ k ] = DFT8{x[ 0], x[1], x[ 2], x[3], x[ 4], x[ 5], x[ 6], x[ 7]}, 0 ≤ k ≤ 7
x[ 0] x[ 0] x[ 0] x[0] Merge two X [ 0]
1 – point
x[1] x[ 2] x[ 4] x[4] DFTs Merge two X [1]
2 – point
x[ 2] DFTs
x[ 2] x[ 4] x[ 2] Merge two X [ 2]
1 – point
x[ 3] x[ 6] x[ 6] x[ 6] DFTs Merge two X [ 3]
4 – point
x[ 4] x[1] x[1] x[1] Merge two DFTs X [ 4]
1 – point
x[ 5] x[ 3] x[ 5] x[ 5] DFTs Merge two X [ 5]
4 – point
x[ 6] x[ 5] x[ 3] x[ 3] DFTs X [ 6]
Merge two
1 – point
x[ 7] x[ 7] x[ 7] x[ 7] DFTs X [ 7]
Algebraic Merging
Practical Programming
Considerations
Stage 1 Stage 1 Stage 1
C[ 0] A[0]
x[0] X [0]
W80 C[1] A[1]
x[ 4] X [1]
−1
D[ 0] A[2]
x[ 2] X [2]
−1
0 0
W8
D[1] W 8
A[3]
x[ 6] X [3]
−1 2
−1
E [0] W 8
B[0] W80
x[1] X [ 4]
−1
0 1
W8
E[1] B[1] W 8
x[5] X [5]
−1 2
−1
0
F[ 0 ] W 8
B[ 2] W 8
x[ 3] X [6]
−1 −1
0 3
W8
F [1] W 8
2
B[3] W 8
x[ 7] X [7]
−1 −1 −1108
s i tes.google.com/site/ncpdhbkhn
Alternative Form (1)
Alternative Form (2)
Decimation – in – Frequency
FFT Algorithms (1)
N −1
X [k ] =  x[ n]WNkn , k = 0,1,..., N − 1
n= 0
N −1 N −1
N
X [2k ] =  x[ n]W N
(2 k )n
=  x[ n]WNkn/ 2 , k = 0,1,..., −1
n= 0 n= 0 2
N N
−1 −1  N
2 2
 N  k  k + 2 
=  x[ n]W kn
N /2 +  x  n +  WN / 2
n =0 n =0  2
N N
−1 −1
2
 N
2
=  x[ n]WNkn/ 2 +  x  n +  WNkn/ 2
n =0 n =0  2
N
−1
2
  N  N
=   x[n] + x n +  WNkn/ 2 , k = 0, 1,..., − 1
n =0   2  2
N
−1
2   N   n  kn N
X [2k + 1] =  
n =0  
x[ n ] − x 

n +  
2  
WN
WN /2
, k = 0, 1,...,
2
−1
FFT Algorithms (2)
N
−1
2   N  N
X [k ] =   x [n ] +
n =0 
x  n +  WNkn/ 2 , k = 0, 1,..., − 1
 2  2
N
−1
2   N   N
X [2k + 1] =    x[n] − x n +   WNn WNkn/ 2 , k = 0, 1,..., − 1
n =0    2   2
a[n ] = x[n ] + x[n + N / 2], n = 0, 1,..., N / 2 − 1
b[n ] = ( x[n ] − x[ n + N / 2])WNn , n = 0,1,..., N / 2 − 1
N
−1
2
N
A[k ] =  a[n]W
n= 0
kn
N/2 , k = 0,1,...,
2
−1
N
−1
2 N
B [k ] =  b[n ]W
n =0
kn
N /2 , k = 0, 1,...,
2
−1
 N
 X [ 2 k ] = A[ k ], k = 0, 1,..., −1
→ 2
 X [2k + 1] = B[ k ], k = 0, 1,..., N − 1
 2
Ex. FFT Algorithms (3)
X [k ] = DFT8{x[0], x[1], x[2], x[3], x[4], x[5], x[6], x[7]}, 0 ≤ k ≤ 7
a[n] = x[n] + x[n + 4]
b[n] = ( x[n] − x[n + 4])W8n
A[0] = X [0] B[0] = X [1]

A[1] = X [2] B[1] = X [3]
A[2] = X [4] B[2] = X [5]
A[3] = X [6] B[3] = X [7]
c[ n] = a[ n] + a[ n + 2]
d [n ] = ( a[n ] − a[ n + 2])W82n
e[ n] = b[ n] + b[ n + 2]
f [n ] = ( b[ n] − b[n + 2])W82n
X [k ] = DFT8{x[0], x[1], x[2], x[3], x[4], x[5], x[6], x[7]}, 0 ≤ k ≤ 7
N
−1
2
N C[ 0] = c[ 0] + c[1] = A[ 0] = X [ 0]
A[ k ] =  a[n]WNkn/ 2 , k = 0,1,...,
2
−1  C[1] = c[ 0] − c[1] = A[ 2] = X [ 4]
n =0 
N
−1  D[0] = d [ 0] + d [1] = A[1] = X [ 2]

2
N
B[k ] =  b[ n]WNkn/ 2 , k = 0,1,..., −1  D[1] = d [0] − d [1] = A[ 3] = X [6]
n= 0 2 →
N  E[ 0] = e[ 0] + e[1] = B[ 0] = X [1]
X [2k ] = A[ k ], k = 0, 1, ..., − 1  E[1] = e[ 0] − e[1] = B[ 2] = X [ 5]
2 
N  F [ 0] = f [ 0] + f [1] = B[1] = X [ 3]
X [2k + 1] = B[ k ], k = 0,1,..., − 1  F [1] = f [ 0] − f [1] = B[3] = X [ 7]
2
X m−1[ p] X m[ p]
 X m [ p ] = X m−1[ p] + X m −1[ q]

 X m [ q] = ( X m −1[ p ] − X m−1[ q])WN
r
X m−1[q] WNr X m[q]
s i tes.google.com/site/ncpdhbkhn
−1 116
X [ k ] = DFT8{x[ 0], x[1], x[ 2], x[3], x[ 4], x[ 5], x[ 6], x[ 7]}, 0 ≤ k ≤ 7
x[0] X [0]
W80
x[1] X [ 4]
−1
W80
x[ 2] X [2]
−1
W82 W80
x[ 3] X [6]
0
−1 −1
W 8
x[4] X [1]
−1
W81 W80
x[5] X [5]
−1 −1
W82 W8
0
x[ 6] X [3]
−1 −1
W83 W82 W80
x[ 7] X [7]
−1 −1 −1
Generalizations and Additional
FFT Algorithms (1)
n = N 2n1 + n2 , n1 = 0, 1,..., N 1 − 1, n2 = 0,1,..., N 2 − 1
k = k1 + N1k2 , n1 = 0, 1,..., N1 − 1, n2 = 0, 1,..., N 2 − 1
→ nk = ( N 2 n1 + n 2 )( k1 + N1k 2 ) = Nn1k 2 + N 1n2 k 2 + N 2n1k1 + n2 k1
WNN = 1, WNN1 = WN 2 , WNN 2 = WN1
N −1
X [ k ] =  x[n]WNkn , k = 0, 1,..., N − 1
n =0
N 2 −1 N1 −1
→ X [k1 + N 1k2 ] =   21 2 N
x
n2 =0 n1 =0
[ N n + n ]W ( N 2 n1 +n2 )( k1 + N1 k2 )
  N1 −1
N 2 −1
 k1n2  k2n2
=     x[ N 2 n1 + n2 ]WN1  WN  WN 2
k1n1
n2 =0 
  n1=0  
FFT Algorithms (2)
  N 1−1
N 2 −1
 k1n2  k2n2
X [ k1 + N1k2 ] =     x[ N 2 n1 + n2 ]WN1  WN  WN 2
k1n1
n2 =0 
  n1 =0  
x n2 [n1 ] = x[ N 2n1 + n2 ]
N −1
→ X n2 [k1 ] =  n2 1 N1
x [n ]W k1n1
n1 =0
N2 −1
→ X [k1 + N 1k2 ] =  N n2 1 N2
W k1 n2
n2 =0
X [k ]W k 2 n2
x k1 [n 2 ] = W Nk1n2 X n 2 [ k1 ]
N 2 −1
→ X [k1 + N 1k 2 ] = 
n2 = 0
x k1 [n2 ]WNk22 n2
FFT Algorithms (3)
1. Form the N2 decimated sequences
xn2 [n1 ] = x[ N 2 n1 + n2 ],
compute the N1 – point DFT of each sequences.
2. Multiply each N1 – point DFT by the twiddle

factor:
xk1 [n2 ] = W k1n2
N X n2 [k1 ].
3. Compute the N2 – point DFTs of the N1

sequences determined from step 2.
a) Goertzel’s Algorithm
b) Chirp Transform Algorithm
c) Sliding DFT
Goertzel’s Algorithm (1)
N −1
X [k ] =  x[n ]WNkn , k = 0,1,..., N − 1
n= 0
W N− kN = 1
N −1 N −1
→ X [k ] = W − kN
N  x[n ]W
n= 0
N
kn
=  x[n ]WN− k ( N −n )
n= 0
3
N = 4 → X [k ] =  x[n]W4− k ( 4− n)
n= 0
= x[3]W4−k + x[2]W4−2 k + x[1]W4−3k + x[0]W4−4k
{ {
= W4− k x[3] + W4− k x[2] + W4− k {x[1] + W4− k x[0]} }}
Goertzel’s Algorithm (2)
{ {
X [ k ] = W4− k x[3] + W4− k x[2] + W4−k {x[1] + W4− k x[0]} }}
yk [−1] = 0, yk [0] = x[0] + W4− k yk [ −1],

yk [1] = x[1] + W4−k yk [0], yk [2] = x[2] + W4−k yk [1]
yk [3] = x[3] + W4− k yk [2], yk [ 4] = x[ 4] + W4−k yk [3] = W4− k yk [3]
yk [n] = WN−k yk [n − 1] + x[n], 0 ≤ n ≤ N

X [ k ] = yk [ N ]
c) Sliding DFT
Chirp Transform Algorithm
N −1
X [ k ] =  x[n]WNkn , k = 0, 1,..., N − 1
n =0
1 2
kn = [k − ( n − k )2 + n 2 ]
2
N −1 k2 −( n −k )2 n2
→ X [k ] =  x[ n]W W N
2
N
2 W N
2
n =0
 N −1  k2  −( n −k ) 2  n2
=    x[n]WN2  WN 2  WN2
 n =0  
 
( 2
)
=  x[ n]WNn / 2 * WN− n / 2  WNn / 2 ,
2

2
0 ≤ n ≤ N −1
h[ n] = WN− n = e j (π n / N ) n
2
/2
 x[n]  −n 2 / 2
 h[n]  * W N
→ X [k ] =  
h[ n]
c) Sliding DFT
Sliding DFT
n − N + 1 x [m ] n
xN −1[ m] n
xn −1[m] n
0 N −1 n−N n −1
 x[n − N + 1 + m], 0 ≤ m ≤ N − 1
xn [ m] =  , n ≥ N −1
0, otherwise
N −1 N −1
X n [k ] =  xn [ m]W mk
N =  x[ n − N + 1 + m]WNmk , n ≥ N − 1, 0 ≤ k ≤ N − 1
m =0 m= 0
= {X n −1[ k ] + x[ n] − x[n − N ]}WN− k , n ≥ N − 1, 0 ≤ k ≤ N − 1

N −1
X N −1[k ] =  x[ n]WNmk , 0 ≤ k ≤ N −1
m= 0

Lec 7

Uploaded by

Lec 7

Uploaded by

Nguyễn Công Phương

The Discrete Fourier Transform

Xˆ c ( jΩ) = TXɶ ( e jω ) , x[n] = xc (nT ) 

TXɶ (e jΩT ), Ω <π /T

 X [0]   e jω0 0 e jω0 1 ⋯ e jω0 ( N −1)

= 0e0 + 1e0 + 2e0 + 3e0 = 6

Synthesis equation Analysis equation

= δ [0]WN0×0 + δ [1]WN0×1 + δ [ 2]WN0×2 + ⋯ + δ [ N − 1]WN0( N −1)

= δ [0]WN1×0 + δ [1]WN1×1 + δ [2]WN1×2 + ⋯ + δ [ N − 1]WN1( N −1)

 X [0] W40×0 W41×0 W42×0 W43×0   0

   0×3 1×3 2× 3 3×3   

 X [0] W40×0 W41×0 W42×0 W43×0   0

   0×3 1×3 2×3 3× 3   

W40 W40 W40 W40   0 1 1 1 1   0

 X [0] 1 1 1 1  0  W32 = W35

Imagi nary Part

W42 = cos(2π / 2) − j sin(2π / 2) = −1

W43 = cos(3π / 2) − j sin(3π / 2) = j

x[n] = x[ −n N ], (circular even symmetry)

x[n] = xce [ n] + x co [n], 0 ≤ n ≤ N −1

y[n ] = h[n] N x[n]

y[n ] = h[n] N x[ n] ←

x[5] x[2] x[3] x[0] x[5] x[2]

n=0 n=0 n=0

Circular mapping Circular addressing

Given h[ n] = {1 2 3 4}, x[n ] = {0 1 2 3}. Find y[ n] = h[n] N x[n].

y[0] = h[0] x[0] + h[1]x[3] + h[2]x[2] + h[3]x[1] x[3]

y[2] = h[0]x[2] + h[1]x[1] + h[2]x[0] + h[3]x[3] h[ 3] h[ 0]

Given h[ n] = {1 2 3 4}, x[n ] = {0 1 2 3}. Find y[ n] = h[n] N x[n].

y[0] = h[0]x[0] + h[1]x[3] + h[2]x[2] + h[3]x[1] = 1 × 0 + 2 × 3 + 3 × 2 + 4 × 1 = 16

 y[0]  x[ 0] x[3] x[ 2] x[1]   h[0]

Given h[ n] = {1 2 3 4}, x[n ] = {0 1 2 3}. Find y[ n] = h[n] N x[n].

y[ n] = h[n] N x[n] ←

Given h[ n] = {1 2 3 4}, x[n ] = {0 1 2 3}. Find y[ n] = h[n] N x[n].

y[ n] = h[n] N x[n] ←

→ Y [0] = H [ 0] X [ 0] = 6 × 10 = 60; Y [1] = − j 8; Y [ 2] = 4; Y [ 3] = j8

 y[0] 1 1 1 1  Y [0] 1 1 1 1   60  16

→ yzero − padded [n] = hzp [n] N xzp [ n] ←

• The input may have indefinite length

 y[0]  h[0] 0 0 0 0 x[1] 0

200 400 600 800 1000 1200 1400 1600

Spectrum of windowed si gnal

Smearing & leakage

-1.5 -1 -0.5 0 0.5 1 1.5 - 1.5 -1 -0.5 0 0.5 1 1.5

Ex. x(t ) = cos(2π Ft ) + cos(2π 3Ft ) + cos(2π 4Ft )

The total cost of computing all X[k] coefficients is approximately N2 operations!!!

Fast Fourier Transform

X [ 3]  = 1 W86 W84 W82 W83 W8 W87 W85   x[6]

Imagi nary Part

 X [ 5]  1 W82 W84 W86 W85 W87 W8 W83   x[3] -0.6

 X [7] 1 W W84 W 2

N – point DFT Coefficients X [k ]

a[n] = x[2n], n = 0,1, 2,..., N / 2 − 1

→ X [k ] = A[k ] + WNk B[k ], k = 0,1,..., N − 1

A[k ] = C[k ] + W82k D[k ], k = 0,1

 X m [ p] = X m−1[ p] + WNr X m −1[ q]

A[0] = X [0] B[0] = X [1]

X m−1[q] WNr X m[q]

→ nk = ( N 2 n1 + n 2 )( k1 + N1k 2 ) = Nn1k 2 + N 1n2 k 2 + N 2n1k1 + n2 k1

WNN = 1, WNN1 = WN 2 , WNN 2 = WN1

2. Multiply each N1 – point DFT by the twiddle

3. Compute the N2 – point DFTs of the N1

= x[3]W4−k + x[2]W4−2 k + x[1]W4−3k + x[0]W4−4k

yk [−1] = 0, yk [0] = x[0] + W4− k yk [ −1],

yk [n] = WN−k yk [n − 1] + x[n], 0 ≤ n ≤ N

= {X n −1[ k ] + x[ n] − x[n − N ]}WN− k , n ≥ N − 1, 0 ≤ k ≤ N − 1

You might also like