Digital Signal Processing

The Discrete Fourier

Transform (DFT)
Overview
• Introduction
• Definition of the DFT
• Properties of the DFT
Introduction
• In previous chapters, we have seen how to represent a
sequence in terms of a linear combination of complex
exponentials using the discrete-time Fourier transform (DTFT)
• Also, we used the sequence values as the coefficients in a
power series expansion of a complex-valued function of z.
• Practically, the DTFT and the z-transform are
– Defined for infinite-length sequences
– Functions of continuous variable (𝜔 or z)
– Not numerically computable transforms because we have to
evaluate infinite sums at uncountably infinite frequencies
• For finite-length sequences there is another representation,
called the discrete Fourier transform (DFT).
• DFT is very useful for digital computations and for digital
hardware implementations.
Definition of the DFT
• Unlike the DTFT, which is a continuous function of a continuous
variable, 𝜔, the DFT is a sequence that corresponds to samples
of the DTFT. It is defined as
𝑋[𝑘] ≝ DFT 𝑥[𝑛] = 𝑋(𝜔)ቚ
𝜔=2𝜋𝑘/𝑁

• The DFT of a signal 𝑥[𝑛] is defined as

𝑁−1 𝑁−1 where
2𝜋𝑘𝑛 2𝜋𝐾𝑛
−𝑗 𝑁
𝑋[𝑘] = ෍ 𝑥[𝑛]𝑒 = ෍ 𝑥[𝑛]𝑊𝑁𝑘𝑛 𝑊𝑁𝐾𝑛 = 𝑒 −𝑗 𝑁

𝑛=0 𝑛=0 𝑘 = 0,1, ⋯ , 𝑁 − 1

n= 0,1,⋯, 𝑁−1
• The inverse DFT (IDFT) is given by
𝑁−1 𝑁−1
1 2𝜋𝑘𝑛
𝑗 𝑁 1
𝑥[𝑛] = IDFT 𝑋[𝑘] = ෍ 𝑋[𝑘]𝑒 = ෍ 𝑋[𝑘]𝑊𝑁−𝑘𝑛
𝑁 𝑁
𝑘=0 𝑘=0
Definition of the DFT
• If we represent the sequence 𝑥[𝑛] as an N-point vector x and its
DFT by the N-point vector 𝐗, then we can express the DFT and
IDFT in a matrix form as follows
𝑋[0] 𝑥[0]
𝑋[1] 𝑥[1] 1 ∗
𝐗= = 𝐖𝑁 𝐱 𝐱= = 𝐖𝑁 𝐗
⋮ ⋮ 𝑁
𝑋[𝑁 − 1] 𝑥[𝑁 − 1]

Where the transform matrix 𝐖𝑁 is defined as

1 1 1 ⋯ 1 Note that:
1 𝑊𝑁 𝑊𝑁2 ⋯ 𝑊𝑁𝑁−1
2(𝑁−1) 1 𝐻 1 ∗
𝐖𝑁 = 1 𝑊𝑁2 𝑊𝑁4 ⋯ 𝑊𝑁 𝐖𝑁−1 = 𝐖𝑁 = 𝐖𝑁
⋮ ⋮ ⋮ ⋱ ⋮ 𝑁 𝑁
2(𝑁−1) (𝑁−1)(𝑁−1)
1 𝑊𝑁𝑁−1 𝑊𝑁 ⋯ 𝑊𝑁
Definition of the DFT
Example:
Compute the 4-point (𝑁 = 4) DFT of the sequence
𝑥[𝑛] = 1, 2, 3
Solution:
We have to pad the sequence with one zero to become
𝑥[𝑛] = 1, 2, 3, 0
1 1 1 1 1
1 −𝑗 −1 𝑗 2
𝐖4 = and 𝐱=
1 −1 1 −1 3
1 𝑗 −1 −𝑗 0
6
−2 − 2𝑗
therefore 𝐗 = 𝐖4 𝐱 =
2
−2 + 2𝑗
1
Note (IDFT): Given X[K]. Find x[n] --------x[n]=𝑁 𝐖𝑁∗ X
Definition of the DFT
Computational Complexity of the DFT
• According to its definition, the DFT requires:
▪ 𝑁 2 complex multiplications
▪ 𝑁(𝑁 − 1) complex additions
• Computing DFT directly from the definition is inefficient due to
the high computational complexity of 𝑂(𝑁 2 ) arithmetical
operations.
• A more simplified and efficient method to compute the same
DFT is called the Fast Fourier Transform (FFT). It drastically
reduces the computational complexity to 𝑂(𝑁 log 2 𝑁)
operations.
• Matlab uses the FFT method to compute the DFT.
Definition of the DFT
• Mathematically, the sampling of the DTFT results in a periodic
sequence 𝑥[𝑛] of a period 𝑁 too.
• Therefore, both 𝑥[𝑛] and 𝑋[𝑘] are now 𝑁-point periodic
sequences of period 𝑁.
• If 𝑥[𝑛] has a length 𝑁1 < 𝑁, we should append it with (𝑁 − 𝑁1 )
zero samples to calculate the 𝑁-point DFT. This is called zero
padding. So, the variables 𝑛 and 𝑘 are restricted to the range:
0 ≤ 𝑛, 𝑘 ≤ 𝑁 − 1
• The zero padding gives a high-density spectrum and provides a
better displayed version for plotting. However, zero padding
does not give a higher resolution spectrum because no new
information is added to the signal.
• To get higher resolution spectrum, one has to obtain more data
from the experiment or observations (increase 𝑁).
Properties of the DFT
Circular Shifting
Since both 𝑥[𝑛] and 𝑋[𝑘] are periodic of period 𝑁, their shifts are circular.
They are denoted as 𝑥 𝑛 − 𝑛0 𝑁 and 𝑋 𝑘 − 𝑘0 𝑁 , respectively. The
modulo-𝑵 notation 𝑚 𝑁 means that (remainder)
m modulo N , 𝑚 𝑁 = 𝑚 + 𝑖𝑁
for some integer 𝑖 such that
0≤ 𝑚 𝑁 ≤ 𝑁 − 1.
𝑭𝒐𝒓 𝒆𝒙𝒂𝒎𝒑𝒍𝒆: 12 8 =rem(12,8)=4 , −12 8 = −12 + 8 8 = ‫ۦ‬−12 +
2𝑥8ۧ8 = 4 8
Note : Make sure the # inside is positive, if not, multiply N by 2,3….
𝑭𝒐𝒓 𝒆𝒙𝒂𝒎𝒑𝒍𝒆: −20 + 3𝑥8 8 = 4 8
Therefore : 12 8 = 4 8 = −12 8 = −4 8 = −20 8

and 𝑅𝑁 [𝑛] is an 𝑁-point rectangular window defined as

1, 0≤𝑛 ≤𝑁−1
𝑅𝑁 [𝑛] = ቊ
0, otherwise
 Circular Shifting
𝑬𝒙𝒂𝒎𝒑𝒍𝒆: given x[n]={1,2,3,4} and N=4
Circular shift by 1 sample is:
y[n] = 𝑥 𝑛 − 𝑛0 4

y[0] =𝑥 0−1 4 =𝑥 −1 + 4 4 = 𝑥 3 4 = 𝑥 3 =4
y[1] =𝑥 1−1 4 =𝑥 0 4 = 𝑥 0 =1
y[2] =𝑥 2−1 4 =𝑥 1 4 = 𝑥 1 =2
y[3] =𝑥 3−1 4 =𝑥 2 4 = 𝑥 2 =3
Properties of the DFT
Illustration of circular shift:

8-points sequence Circular shift by 2 samples

𝒏=𝟎 𝒏=𝟎
7 1 7 1

6 2 6 2

5 3 5 3
4 4
Properties of the DFT
Example of circular shift:

4-points sequence Circular shift by 1

𝑥[𝑛] 𝑥[ 𝑛 − 1 4 ]

Circular shift by 2 Circular shift by 3

𝑥[ 𝑛 − 2 4 ] 𝑥[ 𝑛 − 3 4 ]
Properties of the DFT
Circular Folding
Similarly, circular folding is different than linear folding in using the
module-𝑁 notation operation. If an 𝑁-point sequence, 𝑥[𝑛] is
circularly folded, then
𝑥[0] 𝑛=0
𝑥 −𝑛 𝑁 = ቊ
𝑥[𝑁 − 𝑛] 1 ≤ 𝑛 ≤ 𝑁 − 1

Example of circular folding: X(n)=[1,-1,1,0], X(-n)=[1,0,1,-1] see next (circle)

𝑥[𝑛]
𝑥 −𝑛 4
Properties of the DFT
Illustration of circular fold:

8-points sequence Circular fold

𝑛=0 𝑛=0
7 1 1 7

6 2 2 6

5 3 3 5
4 4
Properties of the DFT
1. Linearity:
𝐷𝐹𝑇
𝑎1 𝑥1 [𝑛] + 𝑎2 𝑥2 [𝑛] 𝑎1 𝑋1 [𝑘] + 𝑎2 𝑋[𝑘]
2. Time Shifting
𝐷𝐹𝑇 𝑘𝑛 2𝜋
𝑥 𝑛 − 𝑛0 𝑁 𝑊𝑁 0 𝑋[𝑘] 𝑊𝑁 = 𝑒 −𝑗 𝑁

3. Frequency Shifting
−𝑛𝑘0 𝐷𝐹𝑇
𝑊𝑁 𝑥[𝑛] 𝑋 𝑘 − 𝑘0 𝑁

4. Circular Convolution
𝑁−1
𝐷𝐹𝑇
𝑥1 𝑛 ⊛ 𝑥2 𝑛 = ෍ 𝑥1 𝑚 𝑥2 𝑛 − 𝑚 𝑁 𝑋1 [𝑘]𝑋2 [𝑘]
𝑚=0
 Properties of the DFT
Example of Circular Convolution
Perform the 4-point circular convolution of the following sequences
ℎ[𝑛] =[1,-1,1,0] 𝑥[𝑛] =[1,2,3,3]

ℎ −𝑚 4 =[1,0,1,-1]
3

𝑦 𝑛 = ෍ 𝑥 𝑚 ℎ[ 𝑛 − 𝑚 4 ]
𝑚=0

𝑦[0] = σ3𝑘=0 𝑥 𝑚 ℎ[ −𝑚 4 ] = 1x1+2x0+3x1+3x-1=1

Properties of the DFT
Example of Circular Convolution (cont.) 𝑥[𝑛] =[1,2,3,3]
Similarly
3

𝑦 1 = ෍𝑥 𝑚 ℎ 1−𝑚 4 =
𝑘=0

= 1𝑥(−1) + 2𝑥1 + 3𝑥0 + 3𝑥1 = 4

ℎ[ 1 − 𝑚 4 ] =[-1,1,0,1]

𝑦[2] = ෍ 𝑥 𝑚 ℎ[ 2 − 𝑚 4 ] = 2
𝑘=0
3

𝑦[3] = ෍ 𝑥 𝑚 ℎ[ 3 − 𝑚 4 ] = 2 Note:CW
𝑘=0
Thus,
𝑦[𝑛] = ℎ[𝑛] ⊛ 𝑥[𝑛] = 𝛿[𝑛] + 4𝛿[𝑛 − 1] + 2𝛿[𝑛 − 2] + 2𝛿[𝑛 − 3]
 Properties of the DFT
Circular Convolution Using Tabular Method 𝑥[𝑛]
𝑥[𝑛]
1 2 3 3
1 1 2 3 3 1 2 3 3
-1 -1 -2 -3 -3 -1 -2 -3 -3
ℎ[𝑛]

1 1 2 3 3 1 2 3 3 ℎ[𝑛]
0 0 0 0 0 0 0 0 0

𝑦[𝑛] 1 4 2 2

𝑦[𝑛] = ℎ[𝑛] ⊛ 𝑥[𝑛] = 𝛿[𝑛] + 4𝛿[𝑛 − 1] + 2𝛿[𝑛 − 2] + 2𝛿[𝑛 − 3]

Matlab Commands
circshift Shift array circularly
fft Discrete Fourier transform using the FFT technique
fftshift Shift zero-frequency component to center of spectrum
ifft Inverse Discrete Fourier transform using the FFT technique