Chapter 7: The Discrete Fourier Transform

Discrete-Time Signals and Systems

The Discrete Fourier Transform Reference:

Dr. Deepa Kundur Sections 7.1-7.2 of

University of Toronto John G. Proakis and Dimitris G. Manolakis, Digital Signal Processing:
Principles, Algorithms, and Applications, 4th edition, 2007.

Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 1 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 2 / 28

Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT

Discrete Fourier Transform Frequency Domain Sampling

I Frequency analysis of discrete-time signals is conveniently

I Recall,
performed on a computer. 1
Z π
I Recall: x(n) = X (ω)e jωn dω
2π −π
∞
F
aperiodic in time ←→ continuous in frequency
X
X (ω) = x(n)e −jωn
F
x(n) ←→ X (ω) n=−∞

I X (ω) must, therefore, be stored in samples on a computer.

I Suppose we sample X (ω).
I Since X (ω) is periodic with period 2π, only a finite (say, N)
consecutive samples are needed.
I For convenience, we consider the N equidistant samples in the
I What happens when we sample in the frequency domain? interval 0 ≤ ω ≤ 2π with spacing δω = 2π N .

Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 3 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 4 / 28
 Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT

Frequency Domain Sampling Frequency Domain Sampling

∞
For k = 0, 1, 2, . . . , N − 1,
X
X (ω) = x(n)e −jωn   ∞ N−1
2π X X n
n=−∞
∞
X k = x(n + lN)e −j2πk N

2π
 X 2π
N l=−∞ n=0
X k = x(n)e −j N kn ,
N n=−∞
N−1
X ∞
X n

∞ lN+N−1
= x(n + lN)e −j2πk N
n n=0 l=−∞
X X
= x(n)e −j2πk N Let n0 = n − lN " ∞
N−1
#
l=−∞ n=lN X X n

∞ N−1
= x(n + lN) e −j2πk N
n0 +lN n=0
X X l=−∞
= x(n0 + lN) −j2πk N
|e {z } | {z }
l=−∞ n0 =0 n0 −j2πk lN equivalent signal xp (n)
=e −j2πk N e N
| {z }
=1

Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 5 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 6 / 28

Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT

Frequency Domain Sampling Frequency Domain Sampling

For k = 0, 1, 2, . . . , N − 1,
DTFS Pair:
N−1
" ∞ #
N−1
 
2π X X n
x(n + lN) e −j2πk N
n
X
X k = xp (n) = ck e j2πk N
N n=0 l=−∞ k=0
N−1 N−1
X n 1 n
xp (n)e −j2πk N
X
= ck = xp (n)e −j2πk N
n=0
N n=0

Comparing to:
I Looks like a DTFS of xp (n)!
N−1
Characteristics of xp (n):
 
I 2π X n

periodic
X k = xp (n)e −j2πk N
I
N n=0
I period = N
I can be expanded via a DTFS

Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 7 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 8 / 28
 Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT

Frequency Domain Sampling and Reconstruction Frequency Domain Sampling and Reconstruction
Therefore,
Therefore,
N−1  
  1 X 2π n
1 2π xp (n) = X k e j2πk N n = 0, 1, . . . , N − 1
ck = X k k = 0, 1, . . . , N − 1 N k=0 N
N N
I Implication: The samples of X (ω) can be used to reconstruct
Since, xp (n).
N−1
n
X
xp (n) = ck e j2πk N then and since,
k=0
N−1
  N−1
1

2π
 2π X n
n
xp (n)e −j2πk N
X
xp (n) = X k e j2πk N n = 0, 1, . . . , N − 1 X k = k = 0, 1, . . . , N − 1
N N N n=0
k=0
I Implication: The signal xp (n) can be used to reconstruct
samples of X (ω).
Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 9 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 10 / 28

Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT

Frequency Domain Sampling and Reconstruction Frequency Domain Sampling and Reconstruction
F
x(n) ←→ X (ω) I x(n) can be recovered from xp (n) if there is no overlap when
 
F 2π taking the periodic extension.
xp (n) ←→ X k
N I If x(n) is finite duration and non-zero in the interval
I FACT: We can reconstrct x(n) from X (ω). 0 ≤ n ≤ L − 1, then
I FACT: We can reconstrct xp (n) from samples of X (ω).
x(n) = xp (n), 0≤n ≤N −1 when N ≥ L
(. . . and vice versa)

I Q: Can we reconstrct x(n) from the samples of X (ω)? I If N < L then, x(n) cannot be recovered from xp (n).
I x(n) Can we reconstrct x(n) from xp (n)? I or equivalently X (ω) cannot be recovered from its samples
2π


I A: Maybe. X N k due to time-domain aliasing
See Figure 7.1.2 of text .

Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 11 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 12 / 28
 Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT

Reconstruction, N ≥ L Reconstruction, N ≥ L
2π

2π


I One way to reconstruct X (ω) from its samples X N
k : I Another way to reconstruct X (ω) from its samples X N
k :
2π


1. Compute xp (n) from X N k :
∞
X
1
N−1
X 
2π

n
X (ω) = x(n)e −jωn
xp (n) = X k e −j2πk N n = 0, 1, . . . , N − 1 n=−∞
N N
k=0 N−1
X
= xp (n)e −jωn
2. Compute x(n) from xp (n):
n=0
 N−1N−1
xp (n) 0 ≤ n ≤ N − 1
 
x(n) =
X1 X 2π n
0 elsewhere = X k e j2πk N e −jωn
n=0
N k=0 N
N−1
" #
3. Compute X (ω) from x(n): X  2π  1 N−1 X n
= X k e −j(ω−2πk) N
∞
X k=0
N N n=0
X (ω) = x(n)e −jωn
n=−∞

Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 13 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 14 / 28

Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT

Reconstruction, N ≥ L Reconstruction

N−1
N−1  # "
1 X −j(ω−2πk) n
X 2π
X (ω) = X e Nk
N N
k=0
| n=0 {z }
interpolation function
N−1 See Figure 7.1.4 of text .
sin ω N2 −jω( N−1 )


1 X −jωn
Let P(ω) = e =
e 2
N n=0 N sin ω2
N−1
X  2π   
2π
Then X (ω) = X k P ω− k N ≥L
k=0
N N

See Figure 7.1.3 of text .

Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 15 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 16 / 28
 Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT

The Discrete Fourier Transform The Discrete Fourier Transform

Summary: Summary:
I If x(n) is infinite duration or has length L>N, the samples I If x(n) is infinite duration or has length L≤N, the samples
X 2πk X 2πk



N
, k = 0, 1, . . . , N − 1 do not uniquely represent the N
, k = 0, 1, . . . , N − 1 uniquely represent the original
original sequence x(n). sequence x(n).
I Instead the frequency samples correspond to a periodic sequence I When x(n) is finite duration of length L ≤ N, then xp (n) is a
xp (n) of period N where xp (n) is a time-aliased version of x(n): periodic repetition of x(n) that can be recovered from a single
period of xp (n) using:
∞
X
xp (n) = x(n − lN) 
x(n) 0 ≤ n ≤ L − 1
l=−∞ xp (n) =
0 L≤n ≤N −1

2πk


I Let X (k) ≡ X N .

Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 17 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 18 / 28

Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT

The Discrete Fourier Transform Pair DFT Example

Q: Determine the N-point DFT of the following sequence for N ≥ L:

1 0≤n ≤L−1
I DFT and inverse-DFT (IDFT): x(n) =
0 otherwise
N−1
n
X
X (k) = x(n)e −j2πk N , k = 0, 1, . . . , N − 1 A: The DTFT of x(n) is given by:
n=0
N−1 L−1
1
X
j2πk Nn X (ω) = x(n)e −jωn
X
x(n) = X (k)e , n = 0, 1, . . . , N − 1
N k=0
n=0
L−1
X sin(ωL/2) −jω(L−1)/2
= e −jωn = e
n=0
sin(ω/2)

See Figure 7.1.5 of text .

Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 19 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 20 / 28
 Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT

DFT Example The DFT as a Linear Transform

N−1
Thus, X n
X (k) = x(n)e −j2πk N , k = 0, 1, . . . , N − 1
sin(ωL/2) −jω(L−1)/2 n=0
X (ω) = e N−1
sin(ω/2) 1X n
x(n) = X (k)e j2πk N , n = 0, 1, . . . , N − 1
sin( 2πk
N
L/2) −j 2πk (L−1)/2 N
X (k) = 2πk
e N k=0
sin( N /2)
sin(πkL/N) −jπk(L−1)/N
= e Want to convert to matrix-vector representation.
sin(πk/N)

See Figure 7.1.6 of text . See Figure 7.1.6b of text .

Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 21 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 22 / 28

Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT

The DFT as a Linear Transform The DFT as a Linear Transform

N−1 N−1
X
−j2πk Nn X (k) = x(n)WNkn , k = 0, 1, . . . , N − 1
X
X (k) = x(n)e , k = 0, 1, . . . , N − 1
n=0 n=0
N−1 N−1
1 1 X
n
x(n) = X (k)WN−kn , n = 0, 1, . . . , N − 1
X
x(n) = X (k)e j2πk N , n = 0, 1, . . . , N − 1 N
N k=0 k=0

Let WN = e −j2π/N
xN = [x(0) x(1) . . . x(N − 1)]T
N−1
X XN = [X (0) X (1) . . . X (N − 1)]T
X (k) = x(n)WNkn , k = 0, 1, . . . , N − 1 
1 1 1 ··· 1


n=0
 1 WN WN2 ··· WNN−1 
2(N−1)
 
N−1 WN =
 1
 WN2 WN4 ··· WN 

1 X  . .. .. .. 
X (k)WN−kn ,
 .
x(n) = n = 0, 1, . . . , N − 1

 . . . ··· . 
N k=0 1 WNN−1 WN
2(N−1)
··· WN
(N−1)(N−1)

Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 23 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 24 / 28
 Chapter 7: The Discrete Fourier Transform 7.1 Frequency Domain Sampling: The DFT Chapter 7: The Discrete Fourier Transform 7.2 Properties of the DFT

The DFT as a Linear Transform Periodicity and Linearity

N−DFT
I In matrix-vector notation: Notation: x(n) ←→ X (k)
I Periodicity:
XN = WN xN
1 ∗ x(n + N) = x(n) for all n
xN = WN−1 XN = W XN
N N
X (k + N) = X (k) for all k
where it can be shown that
1 ∗
WN−1 = W I Linearity: If
N N
N−DFT
x1 (n) ←→ X1 (k)
I Complexity: N complex multiplications and N − 1 complex N−DFT
x2 (n) ←→ X2 (k)
additions;
I For an N-point DFT, a total of N 2 complex multiplications and N−DFT
Then a1 x1 (n) + a2 x2 (n) ←→ a1 X1 (k) + a2 X2 (k)
N(N − 1) complex additions are required.
Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 25 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 26 / 28

Chapter 7: The Discrete Fourier Transform 7.2 Properties of the DFT Chapter 7: The Discrete Fourier Transform 7.2 Properties of the DFT

Circular Symmetry and Convolution Important DFT Properties

I Circular operations: apply the transformation on the periodic

repetition of x(n) and then obtain the final result by taking Property Time Domain Frequency Domain
Notation: x(n) X (k)
points for n = 0, 1, . . . , N − 1 Periodicity: x(n) = x(n + N) X (k) = X (k + N)
I Circular Symmetry: Linearity: a1 x1 (n) + a2 x2 (n) a1 X1 (k) + a2 X2 (k)
Time reversal x(N − n) X (N − k)
I circular time reversal: x((−n))N = x(N − n) Circular time shift: x((n − l))N X (k)e −j2πkl/N
I circularly even: x(N − n) = x(n) Circular frequency shift: x(n)e j2πln/N X ((k − l))N
Complex conjugate: x ∗ (n) X ∗ (N − k)
I circularly odd: x(N − n) = −x(n) Circular convolution: x1 (n) ⊗ x2 (n) X1 (k)X2 (k)
1
I Circular Convolution: Multiplication: x1 (n)x2 (n) N P
X1 (k) ⊗ X2 (k)
P N−1 ∗ 1 N−1 ∗
Parseval’s theorem: n=0 x(n)y (n) N k=0 X (k)Y (k)
N−1
X 
x3 (m) = x1 (n)x2 ((m − n))N , m = 0, 1, . . . , N − 1
n=0

Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 27 / 28 Dr. Deepa Kundur (University of Toronto) The Discrete Fourier Transform 28 / 28