1
x[ n] N
X [k ]
N 1 ~
k 0
N 1 ~
kn
x
[
n
]
W
N ( 2), k 0,.. N 1
n 0
WNkn e j 2kn / N
N 1
( k rN ) n ~
~
~
X [ k rN ] x [ n]W
X [k ]
N
n 0
EXAMPLE.
Figure 8.1
~
X [k ]
n 0
kn
W10
N=10
j 4k / 10
sin(k / 2)
sin(k / 10)
k
N
other
FIGURE 8.5
DFS ~
~
~
~
1.linearity : ax1[n] bx2 [n]
aX 1[k ] bX 2 [k ]
both period=12
compositive sequence N=12 12 points DFS
DFS km ~
~
2.shift of a sequence : x [n m]
W
X [k ]
N
nl ~ DFS ~
W
x [ n]
X [k l ]
N
~ DFS ~
3.duality : X [n]
Nx [ k ]
4.symmetry properties :
DFS ~
DFS ~
~
~
x * [n]
X * [ k], x * [ n]
X * [k]
DFS 1 ~
1 ~
~
~
~
~
Re{x[n]} (x[n] x * [n])
(X[k] X * [k]) X e [k]
2
2
DFS 1 ~
1 ~
~
~
~
~
jIm{x[n]} (x[n] x * [n])
(X[k] X *[k]) X o [k]
2
2
DFS 1 ~
1 ~
~
~
~
~
x e [n] (x[n] x * [n])
(X[k] X * [k]) Re{X[ k ]}
2
2
DFS 1 ~
1 ~
~
~
~
~
xo [n] ( x [n] x * [n])
( X [k ] X *[ k ]) j Im{X [ k ]}
2
2
x[n] x* [n]
~
~
X [k ] X * [k ]
~
~
Re{ X [ k ]} Re{ X [ k ]}
~
~
Im{X [ k ]} Im{X [k ]}
~
~
| X [k ] || X [ k ] |
~
~
X [k ] X [k ]
EXAMPLE.
FIGURE 8.2
5. periodic convolution :
N 1
~
x3 [n] ~
x1[n] ~
x2 [ n ] ~
x1[m]~
x2 [ n m ] ~
x2 [ n ] ~
x1[n]
m 0
(1)if : ~
x3 [ n ] ~
x1[n] ~
x 2 [ n]
~
then : X 3 [k ] X 1[k ] X 2 [k ]
~
~
~
(2)if : x3 [n] x1[ n] x2 [ n]
~
~
1
then : X 3 [k ]
X [k ] X 2 [k ]
N 1
graphic method to
calculate periodic
convolution
Figure 8.3
x[n] ~
x [ n ]RN [ n ]
EXAMPLE.
Figure 8.8
duration of sequence is N
X [k ]
N 1
x[n]WNkn , k 0,1,....N 1
n 0
N 1
1
x[n] N
X [k ]WN kn , n 0,1,...N
k 0
N 1
X [0] x[n]
x[0] 1
n 0
N 1
X [k ]
k 0
X [ k ] X ( e j ) |
2
k
N
X ( z) |
z e
2
k
N
EXAMPLE.
Figure 8.10
periodic
extension
with period 5
x[((n))5 ]
DFS{~
x [ n]}
X [k ]R5 [ k ]
EXAMPLE.
Figure 8.5
periodic
extension with
period 10
reflect
frequency
spectrum of
signal more
truly than
figure 8.10
genetic instance
Sequence with length N or infinite length , sample M points
in frequency domain more than or less than or equal to N
then the reconstructed time-domain signal is dominant period of
the periodic extension with period M of original signal maybe
aliasing Viz. if
X ( e j ) |
x[ n]WMkn , k
2
k
M
n
0,1...M 1
X (e ) 1 e
jN
prove :
N 1
X (e ) x[ n]e
j
n0
N 1
N 1
k 0
n0
N 1
X [k ]
/ N
k j
k 0 1 WN e
j n
N 1
1
(
n 0 N
kn jn
X
[
k
](
W
N e )
N 1
X [k ]
k j
1
W
k 0
N e
1 e j N / N
N 1
kn
j n
X
[
k
]
W
)
e
N
k 0
X [k ]
2
M
FT
x1[n] X1 (e
M points IDFT
x2 [ n ]
x[n rM ]
MIDFS
~
~
X
[
k
]
x3[n]
M 3
RM [n]
x[n rM ]
summary 8.5
1. DFT is N-points samples of frequency spectrum of sequence with length N. the
more spectral sampling points, the more genuine to reflect the frequency spectrum
2. get M-points spectral samples of N-points sequence by M-points DFT:
1 M=N do M-points DFT directly
2 M>N extend x[n] to M points by filling 0 then do M-points DFT
3 M<N periodic extension of x[n] with period M and aliasing take the
dominant period with length M, then do M-points DFT
3. whether spectral sampling can reconstruct original time-domain signal
spectral sampling theorem if spectral sampling points is larger than or equal
to the length of signal, the time-domain signal can be reconstructed. Contrarily, it
can not be done.
4. If frequency spectrum is the same its samples are equal contrarily, it does
not come into existence
if frequency spectrum has linear phase, its samples has linear phase, too
contrarily, it does not come into existence
DFT
1.linearity : ax1[ n] bx 2 [ n]
aX 1[ k ] bX 2 [ k ]
DFT km
x[((n m)) N ]RN [n]
W X [k ]
N
ln
DFT
W
x[n]
X [((k l )) N ]RN [k ]
N
EXAMPLE.
circular
shift
Figure 8.12
EXAMPLE. 8.42
| H 1[ k ] |
8points DFT|
H 2 [k ] |
h1[n]
h2 [n]
j
| H 1 (e ) |
| H 2 ( e j ) |
DFT
3.duality : X [n]
Nx[(( k )) N ]RN [ k ] Nx[ N k ]
EXAMPLE.
x[ n] n cos(0.2n)
| X [k ] || DFT {x[n]} |
DFT { X [n]}
X '[k ]
4. Symmetry properties:
*
x [ n]
DFT
DFT *
x [(( n)) N ]RN [n] x [ N n]
X [k ]
DFT 1
1
Re{x[n]} ( x[n] x * [n])
( X [k ] X * [ N k ]) X ep [k ]
2
2
DFT 1
1 ~
~
~
j Im{x [ n]} ( x [n] x * [n])
( X [k ] X * [ N k ]) X op [ k ]
2
2
DFT 1
1
xep [n] ( x[n] x * [ N n])
( X [k ] X * [k ]) Re{ X [k ]}
2
2
DFT 1
1
xop [n] ( x[n] x * [ N n])
( X [k ] X * [k ]) j Im{X [k ]}
2
2
*
definition
X ep [k ] : periodic conjugate - symmetric components
X op [k ] : periodic conjugate - antisymmetric components
sequence with length N can be decomposed
X [k ] X ep [k ] X op [k ]
while
1
X ep [k ] ( X [k ] X * [ N k ]) X ep *[ N k ]length N
2
real part is even symmetric , imaginary part is odd symmetric
1
X op [k ] ( X [k ] X * [ N k ]) X op * [ N k ]lenth N
2
real part is odd symmetric , imaginary part is even symmetric
x[n] x * [n]
X [k ] X * [ N k ]
Re{ X [k ]} Re{ X [ N k ]}
Im{X [k ]} Im{X [ N k ]}
| X [k ] || X [ N k ] |
X [k ] X [ N k ]
Real{X[k]}
|X[k]|
N=10
DFT of real
sequence
Imag{X[k]}
arg{X[k]}
N=9
|X[k]|
Arg{X[k]}
5.circular convolution
x [((m))
1
N ] x2 [((n m)) N ]) RN [ n]
m 0
DFT
(1) x1[n]( N ) x2 [n]
X 1[ k ] X 2 [ k ]
DFT 1
(2) x1[n]x2 [n]
X 1[k ]( N ) X 2 [k ]
N
EXAMPLE.
graphic
method to
calculate
circular
convolution
x[n]( N ) [n 1]
x[((n 1)) N ]RN [n]
Figure 8.14
convolution
result
depends on
N
EXAMPLE.
Figure 8.15
N=L
EXAMPLE.
6.paswals theory
N 1
x[n] y * [n]
n 0
N 1
n 0
| x[ n] |2
1
N
1
N
N 1
k 0
N 1
X [k ]Y * [k ]
k 0
| X [k ] |2
then : x[n
] Nh[n]
y[n rN ]R
N [ n]
PROVE
x[n] h[n]
y[n]
FT
X (e
according to
properties
of circular
convolution
) H (e
Y (e
N point sample
X [ k ]H [ k ]
IDFT
x[n]( N )h[n]
Y [k ]
according to
spectral
sampling
r
If N>=N1+N2-1,then x[n]*h[n]=x[n](N)h[n]
EXAMPLE.
Figure 8.18
linear convolution
Conclusion:
(1)calculate N point circular convolution by linear convolution
(a) y[n] x[n] * h[n]
y[n rN ]R
N [ n]
Figure 8.22
consideration
1 deal with data when input them
2 operation speed
3 the counts of input and output data are
equal
overlap-add method
length of h[n] is P
(1)segment x(n) into sections of length L;
(2)fill 0 into h(n) and some section of x(n) ,
then do L+P-1 points FFT ;
(3) calculate
y (n) IFFT {H (k ) X (k )}n 0,...L P 2
P-1
points
overlap-save method
the length of h[n] is P
(1)segment x(n) into sections of length L,
overlap P-1 points;
(2)fill 0 into h(n) and some section of x(n) ,
then do L points FFT ;
(3) calculate
Figure 8.24
P-1 points
linear convolution
result
DCT 1X [k ]
x[n]
DCT 2X [k ]
x[n]
2
ck
N 1
2
cn
N 1
2
ck
N
2
N
N 1
cn x[ n] cos(
n 0
N 1
ck X [ k ] cos(
n 0
N 1
n 0
N 1
n 0
kn
),0 k N 1
N 1
x[n] cos(
kn
),0 n N 1
N 1
k (2n 1)
),0 k N 1
2N
ck X [ k ] cos(
k (2n 1)
),0 n N 1
2N
DCT 3X [k ]
x[ n]
DCT 4X [k ]
x[ n]
1 / 2, k 0
1,1 k N 1
ck
2
N
N 1
cn x[ n] cos(
n 0
N 1
2
ck
N
X [ k ] cos(
n 0
N 1
x[n] cos(
2
N
X [k ] cos(
n 0
n(2k 1)
),0 n N 1
2N
(2n 1)(2k 1)
),0 k N 1
4N
2
N
n0
N 1
n( 2k 1)
),0 k N 1
2N
( 2n 1)(2k 1)
),0 n N 1
4N
DCT-1
-2
-21.
1.0.0.1.
1.2.
2.3.
3.4.
4.55
DCT-2
-4
-4
-43
3
3-2-1
-2-1
-2-10.
0.0.1.
1.1.2.
2.2.3.
3.3.4.
4.4.5.
5.5.6.
6.6.777
DCT 2
x[n], n 0,...N 1
y[ n]
X [2 N 1 n], n N ,...2 N 1
2 N 1
Y [k ]
y[n]W2knN
n 0
N 1
x[ n]W2knN
n 0
x[ n]W2knN
n 0
2W2Nk / 2
N 1
N 1
x[ n]W2kN( 2 N 1 n )
n 0
N 1
x[ n] cos(
n 0
2W2Nk / 2 X [k ] /(
(2n 1)k
)
2N
2
ck )
N
2 N 1
n N
x[2 N 1 n]W2knN
012 345 67
DFT
012 345 67
summary
8.1 representation of periodic sequences: the discrete fourier series
8.2 the fourier transform of periodic signals
8.3 properties of the discrete fourier series
8.4 fourier representation of finite-duration sequences:
Definition of the discrete fourier transform
8.5 sampling the fourier transform (point of sampling)
8.6 properties of the fourier transform
8.7 linear convolution using the discrete fourier transform
8.8 the discrete cosine transform (DCT)
requirements
definition, calculation and properties of DFS;
derivation of definition of DFT DFS or spectral sampling;
concepts of spectral sampling, time-domain periodic
extension;
properties of DFT linearity circular shift , symmetry,
circular convolution paswals theory;
relationship between linear and circular convolution;
derivation of definition DCT and comparison with DFT.
exercises
8.26
8.29
8.39
8.45
8.49