Properties of the DFT

 The DFT as a set of N samples {X(k)} of the Fourier transform X(ω) for a finite-duration
sequence {x(n)} of length L < N.
 The sampling of X(ω) occurs at the N equally spaced frequencies ωk = 2πk/N, k = 0, 1,
2, ......., N-1.
 The N samples {X(k)} uniquely represent the sequence {x(n)} in the frequency domain.
 The DFT and inverse DFT (IDFT) for an N-point sequence {x(n)} are given as
N 1
DFT : X (k )   x(n) WNkn k  0, 1,  , N  1 (1)
n0
N 1
1 (2)
IDFT: x (n) 
N
 X (k ) W
k 0
N
 kn
n  0, 1,  , N  1

where WN is defined as
WN  e  j 2 / N (3)

 The notation used below to denote the N-point DFT pair x(n) and X(k) is
DFT
x(n) 
N  X (k )

December 15, 2022 1

1. Periodicity:
 If x(n) and X(k) are an N-point DFT pair, then

x (n  N )  x ( n) for all n (4)

X (k  N )  X ( k ) for all k (5)

 These periodicities in x(n) and X(k) follow immediately from formulas Eq. (1)
and Eq. (2) for the DFT and IDFT, respectively.
 In some applications it is advantageous to do this.

2. Linearity:
 If
DFT DFT
x1 (n) 
N  X 1 (k ) and x2 (n) 
N  X 2 (k )
then for any real-valued or complex-valued constants a1 and a2,
DFT
a1 x1 (n)  a2 x2 (n) 
N  a1 X 1 (k )  a2 X 2 (k ) (6)

 This property follows immediately from the definition of the DFT given by Eq. (1).

December 15, 2022 2

3. Symmetry properties of the DFT:
 The symmetry properties for the DFT can be obtained by applying the methodology
previously used for the Fourier transform.
 Let us assume that the N-point sequence x(n) and its DFT are both complex valued.
 Then the sequences can be expressed as
x (n)  xR ( n)  jxI (n ) 0  n  N 1 (7)
X (k )  X R ( k )  jX I (k ) 0  k  N 1 (8)
 By substituting Eq.(7) into the expression for the DFT given by Eq.(1), we obtain
N 1
 2 kn 2 kn 
X R ( k )    xR (n) cos  xI ( n) sin (9)
n0  N N 
N 1
 2 kn 2 kn  (10)
X I (k )     xR (n ) sin  xI (n) cos
n0  N N 
 Similarly, by substituting Eq.(8) into the expression for the IDFT given by Eq.(2), we
obtain
1 N 1
 2 kn 2 kn 
xR ( n ) 
N
 X
 R
n0 
( k ) cos
N
 X I ( k ) sin
N  (11)
1 N 1
 2 kn 2 kn 
xI ( n ) 
N
 X
 R
n0 
( k ) sin
N
 X I ( k ) cos
N 
(12)
December 15, 2022 3
4. Real-valued sequences:
 If the sequence x(n) is real, it follows directly from Eq.(1) that

X ( N  k )  X  ( k )  X ( k ) (13)
 Consequently,
|X(N-k)| = |X(k)| and  X(N - k) = -  X(k).
 Furthermore, xl(n) = 0 and therefore x(n) can be determined from Eq.(11), which is
another form for the IDFT.

December 15, 2022 4

5. Real and even sequences:
 If x(n) is real and even, that is,

x( n)  x( N  n) 0  n  N 1

then Eq. (10) yields Xl (k) = 0.

 Hence the DFT reduces to
N 1
2 kn
X (k )  
n0
x (n) cos
N
0  k  N 1 (14)

which is itself real-valued and even.

 Furthermore, since Xl(k) = 0, the IDFT reduces to

1 N 1
2 kn
x ( n) 
N
 X (k ) cos
k 0 N
0  n  N 1 (15)

December 15, 2022 5

6. Real and odd sequences:
 If x(n) is real and odd, that is,

x( n)   x( N  n) 0  n  N 1

then Eq. (9) yields XR(k) = 0.

 Hence,
N 1
2 kn
X ( k )   j  x ( n) sin 0  k  N 1 (16)
n0 N

which is purely imaginary and odd.

 Since XR(k) = 0, the IDFT reduces to

1 N 1
2 kn
x (n)  j
N
 X (k ) sin
k 0 N
0  n  N 1 (17)

December 15, 2022 6

7. Purely imaginary sequences:
 In this case, x(n) = jxl(n).
 Consequently, Eq.(9) and Eq.(10) reduce to
N 1
2 kn
X R ( k )   xl ( n) sin (18)
n0 N
N 1
2 kn (19)
X l (k )   xl (n) cos
n0 N
 We observe that XR(k) is odd and Xl(k) is even.
 If x/(n) is odd, then X/(k) = 0 and hence X(k) is purely real.
 On the other hand, if xl(n ) is even, then XR(k) = 0 and hence X(k) is purely imaginary.
 The symmetry properties given above may be summarized as follows:

(20)

 For example, the DFT of the sequence

x pe ( n)  12 [ xR ( n)  xR ( N  n)]
(21)
is X R (k )  X (k )  X (k )
e
R
0
R
December 15, 2022 7
9. Circular Symmetries of a Sequence:
 The N-point DFT of a finite duration sequence, x(n) of length L ≤ N is equivalent to the
N-point DFT of a periodic sequence xp(n), of period N, which is obtained by periodically
extending x(n), that is,

x p (n)   x( n  lN ) (22)
l 0

 Now suppose that we shift the periodic sequence xp(n) by k units to the right.
 Thus, we obtain another periodic sequence

xp ( n)  x p ( n  k )  
l  
x( n  k  lN ) (23)

 The finite-duration sequence

 x ( n), 0  n  N 1
x( n )   p (24)
 0, otherwise

is related to the original sequence x(n) by a circular shift.

 This relationship is illustrated in Fig. 1 for N = 4.
December 15, 2022 8
 Figure 1: Circular shift o f a sequence.
December 15, 2022 9
9. Circular Symmetries of a Sequence:
 An N-point sequence is called circularly even if it is symmetric about the point zero on
the circle.
 This implies that
x( N  n)  x( n) 1  n  N 1 (25)
 An N-point sequence is called circularly odd if it is antisymmetric about the point zero
on the circle.
 This implies that

x ( N  n )   x ( n) 1  n  N 1 (26)
 The time reversal of an N-point sequence is attained by reversing its samples about the
point zero on the circle.
 Thus the sequence x((-n))N is simply given as

x((n)) N  x( N  n) 0  n  N 1 (27)

 This time reversal is equivalent to plotting x(n) in a clockwise direction on a circle.

December 15, 2022 10

9. Circular Symmetries of a Sequence:
 An equivalent definition of even and odd sequences for the associated periodic
sequence xp(n) is given as follows
Even: x p ( n)  x p ( n )  x p ( N  n)
(28)
Odd: x p ( n)   x p (  n)   x p ( N  n)

 If the periodic sequence is com plex-valued, we have

Conjugate even: x p ( n)  xp ( N  n)

(29)

Conjugate odd: x p ( n)   x ( N  n)
p

 These relationships suggest that we decom pose the sequence xp(n) as

x p (n)  x pe (n)  x po (n) (30)

where

x pe ( n)  12  x p ( n)  xp ( N  n) 
(31)
x po ( n)  12  x p (n )  x p ( N  n) 

December 15, 2022 11

 The symmetry properties of the DFT are summarized in Table 1.

December 15, 2022 12

10. Multiplication of Two DFTs and Circular Convolution

 Let us consider that two finite-duration sequences of length N, x1(n) and x2(n).
 Their respective N-point DFTs are
N 1
X 1 (k )   x1 (n)e  j 2 k / N k  0, 1, , N  1 (32)
n 0
N 1
X 2 ( k )   x2 ( n)e  j 2 k / N k  0, 1, , N  1 (33)
n 0

 Then, multiply the two DFTs together and the result is a DFT, say of a sequence x3(n)
of length N.
 Let us determine the relationship between x3(n) and the sequences x1(n) and x2(n).
 We have
X 3 ( k )  X 1 ( k ) X 2 (k ) k  0, 1,  , N  1 (34)

 The IDFT of {X3(k)} is

N 1 N 1
1 1
x3 ( m) 
N
X
k 0
3 (k ) e 2 km / N

N
 X (k )X
k 0
1 2 ( k ) e 2 km / N (35)

December 15, 2022 13

10. Multiplication of Two DFTs and Circular Convolution

 Put, X1(k) and X2(k) in Eq.(35), then we get

1 N 1 N 1   N 1  j 2 k m / N

x3 ( m)  1
N
k 0 n 0
x ( n ) e  j 2 k n / N
  2 x ( l ) e  j 2 k l / N
e
  l 0 
1 N 1 N 1  N 1 j 2 k ( m  n l ) / N 
  x1 (n)  x2 (l )   e  (36)
N n 0 l 0  k  0 
 The inner sum in the brackets in Eq.(36) has the form
N 1
N a 1

 a k
 1  a N
a 1
k 0  (37)
 1 a
where a is defined as a  e j 2 ( m  n l ) / N
 We observe that a = 1 when m - n - I is a multiple of N.
 On the other hand, aN = 1 for any value of a ≠ 0.
 Consequently, Eq.(37) reduces to
N 1
N l  m  n  pN  ((m  n)) N , p an integer
 a k
  (38)
k 0 0 otherwise
December 15, 2022 14
10. Multiplication of Two DFTs and Circular Convolution

 Using Eq.(38) and Eq.(39), we obtain the desired expression for x3(m) in the form
N 1
x3 (m)   x1 (n) x2 ((m  n)) N , m  0, 1, , N 1 (39)
n 0

 The expression in Eq.(39) has the form of a convolution sum.

 It is not the ordinary linear convolution.
 The convolution sum in Eq.(39) involves the index ((m - n))N and is called circular
convolution.
 We conclude that multiplication of the DFTs of two sequences is equivalent to the
circular convolution of the two sequences in the time domain.
 If
x1 (n) 
DFT
N
 X 1 (k ) and x2 (n) 
DFT
N
 X 2 (k )
 Then

(40)

where x1(n) x2(n) denotes the circular convolution of the sequence x1(n) and x2(n).

December 15, 2022 15

Additional DFT Properties:

1. Time reversal of a sequence:

 If x (n) 
DFT
N
 X (k )
 Then
x(( n)) N  x ( N  n) 
DFT
N
 X ((  k )) N  X ( N  k ) (41)

 Hence, reversing the N-point sequence in time is equivalent to reversing the DFT
values.
 Time reversal of a sequence x(n) is illustrated in Fig.3.

Figure 3: Tim e reversal of a sequence.

December 15, 2022 16
1. Time reversal of a sequence:

Proof: From the definition of the DFT in Eq. (1) we have

N 1
DFT{x( N  n)}   x
n0
( N  n ) e  j 2 kn / N

 If we change the index from n to m = N - n, then

N 1
DFT{x( N  n)}  
m0
x ( m ) e  j 2 k ( N  m ) / N

N 1
 
m0
x(m) e j 2 km / N

N 1
 
m 0
x ( m ) e  j 2 m ( N  k ) / N

 X (N  k)
 We note that

X(N - k) = X (( - k ))N, 0≤ k≤N-l.

December 15, 2022 17
2. Circular time shift of a sequence:

If x(n) 
DFT
N
 X (k )
 j 2 kl / N
Then x((n  l )) N 
DFT
N
 X ( k ) e (42)
Proof: From the definition of the DFT we have
N 1 l 1 N 1
DFT{x ((n  l )) N }   x((n  l ))
n0
N e  j 2 kn / N
  x((n  l ))
n0
N e  j 2 kn / N
  x( n  l ) e  j kn / N
nl

But, x((n  l )) N  x( N  l  n).

Consequently,
l 1 l 1 N 1

 x((n  l ))
n0
N e  j 2 kn / N
  x ( N  l  n) e
n0
 j 2 kn / N
 
m  N l
x (m) e  j k ( m  l ) / N

Furthermore,
N 1 N 1l

 x(n  l ) e
nl
 j kn / N
 
m0
x ( m ) e  j 2 k ( m  l ) / N

Therefore,
N 1
DFT{x ((n  l ))}   x ( m) e
m0
 j 2 k ( m  l ) / N
 X ( k ) e  j 2 k l / N
December 15, 2022 18
3. Circular frequency shift:

If
x ( n) 
DFT
N
 X (k )

Then
x ( n) e j 2 l n / N 
DFT
N
 X ((k  l )) N (43)

 Hence, the multiplication of the sequence x(n) with the complex exponential
sequence e j 2 kn / N is equivalent to the circular shift of the DFT by I units in frequency.
 This is the dual to the circular time-shifting property and its proof is similar to the
latter.

December 15, 2022 19

4. Complex-conjugate properties:

If
x ( n) 
DFT
N
 X (k )
Then
x (n) 
DFT
 X 
((  k ))  X 
(N  k) (44)
N N

Proof:
The IDFT of X*(k) is

1 N 1 1 N 1 
N
X 
( k ) e j kn / N   X (k ) e j 2 k ( N  n ) / N

k 0 N k 0 
Therefore,
N 1
1
N
 x
k 0

((  n )) N  x 
( N  n ) DFTN

N
X 
(k ) (45)

December 15, 2022 20

5. Circular correlation:
 For complex-valued sequences x(n) and y(n), if
DFT DFT
x( n) 
N
 X (k ) and y ( n) 
N
 Y (k )
 Then
rxy (l ) 
DFT
  ( k )  X ( k )Y * ( k )
R (46)
N xy

where rxy (l ) is the (unnormalized) circular cross-correlation sequence, defined as

N 1
rxy (l )   x(n) y ((n  l ))
n0
*
N

Proof: We can write rxy (l ) as the circular convolution of x(n) with y*(–n), that is,

rxy (l )  x(l ) y * ( l )
 Then, with the aid of the properties in Eq.(40) and Eq.(45), the N-point DFT of rxy (l ) is
R ( k )  X (k )Y * ( k )
xy

 In the special case where y(n) = x(n), we have the corresponding expression for the
circular autocorrelation of x(n),
rxx (l ) 
DFT
  (k )  X (k ) 2
R (47)
N Xx

December 15, 2022 21

6. Multiplication of two sequences:
DFT DFT
 If x1 ( n) 
N
 X 1 (k ) and x2 (n) 
N
 X 2 (k )

DFT 1 (48)
 Then x1 (n) x2 ( n) 
N
 X 1 (k ) X 2 (k )
N
 This property is the dual of Eq.(40).
 Its proof follows simply by interchanging the roles of time and frequency in the
expression for the circular convolution of two sequences.

7. Parseval’s theorem:

 For complex-valued sequences x(n) and y(n), in general, If

x ( n) 
DFT
N
 X (k ) and y ( n) 
DFT
N
 Y (k )
 Then,
N 1 N 1
1

n0
x ( n ) y *
( n ) 
N
 X (k )Y
k 0
*
(k ) (49)

December 15, 2022 22

Proof:
 The property follows immediately from the circular correlation property in Eq.(46).
We have
N 1

 x(n) y (n)  r
n0
*
xy (0)

and
N 1 N 1
1 1
rxy (l ) 
N
 R xy (k ) e
k 0
j 2 k l / N

N

k 0
X ( k ) Y  ( k ) e j 2 k l / N

 Hence (49) follows by evaluating the IDFT at I = 0.

 The expression in (49) is the general form of Parseval’s theorem.
 In the special case where y(n) = x(n), Eq. (49) reduces to
N 1 N 1
1
  (50)
2 2
x( n)  X (k )
n0 N k 0

which expresses the energy in the finite-duration sequence x(n) in terms of the
frequency com ponents {X(k)}.

December 15, 2022 23

 The properties of the DFT given above are summarized in the following Table.

December 15, 2022 24

Example-1: Determine the response of the FIR filter with impulse response
h(n)  {1, 2, 3} to the input sequence x(n)  {1, 2, 2, 1}
 
Also determine the 8-point DFT and 8-point IDFT.
Solution: The input sequence has length L = 4 and the impulse response has length M =3.
Linear convolution of these two sequence produces a sequence of length N = 6.
We should also mention that the efficient computation of the DFT via the first Fourier
Transform (FFT) algorithm is usually performed for length N that is a power of 2.
Hence the eight-point DFT of x(n) is
7
X ( k )   x ( n) e  j 2 k n / 8  1  2e  j k / 4  2e  j k / 2  e  j 3 k / 4 , k  0, 1,...., 7
n 0

This computation yields

X (0)  6, X (4)  0
2 2  43 2  2 2  43 2 
X (1)   j    X (5)   j   
2 2 2  2 
 
X (2)  1  j X (6)  1  j

 43 2  2 2  43 2 
2 2 X (7)   j   
X (3)   j    2 2
2  2   

December 15, 2022 25

The eight-point DFT of h(n) is
7
H ( k )   h(n) e  j 2 k n / 8  1  2e  j k / 4  3e  j k / 2
n 0

This computation yields

H (0)  6, H (4)  2
H (1)  1  2  j 3  2   H (5)  1  2  j (3  2) 
H (2)  2  j 2 H (6)  2  j 2
H (3)  1  2  j 3  2   H (7)  1  2  j (3  2) 
The product of these two DFTs yields Y(k), which is
Y (0)  36, Y (1)  14.07  j17.48, Y (2)  j 4, Y (3)  0.07  j 0.515
Y (4)  0, Y (5)  0.07  j 0.515, Y (6)   j 4, Y (7)  14.07  j17.48
The eight-point IDFT is
7
Y ( n )   Y ( k ) e j 2 k n / 8  ....., 7
n  0, 1,
k 0

This computation yields the result

y ( n)  {1, 4, 9, 11, 8, 3, 0, 0}

December 15, 2022 26