6.

003: Signals and Systems

DT Fourier Representations

April 15, 2010

Mid-term Examination #3

Wednesday, April 28, 7:30-9:30pm.

No recitations on the day of the exam.

Coverage: Lectures 1–20

Recitations 1–20
Homeworks 1–11

Homework 11 will not collected or graded. Solutions will be posted.

Closed book: 3 pages of notes (8

12
× 11 inches; front and back).

Designed as 1-hour exam; two hours to complete.

Review sessions during open office hours.

Review: DT Frequency Response

The frequency response of a DT LTI system is the value of the

system function evaluated on the unit circle.
 
cos(Ωn) H(z) |H(ejΩ )| cos Ωn + ∠H(ejΩ )

H(e jΩ ) = H(z)|z=e jΩ
Comparision of CT and DT Frequency Responses

CT frequency response: H(s) on the imaginary axis, i.e., s = jω.

DT frequency response: H(z) on the unit circle, i.e., z = e jΩ .

ω s-plane z -plane

 
 
|H(jω)|

H(ejΩ )


0
ω
−π 0 πΩ
Check Yourself

1 − az
A system H(z) = has the following pole-zero diagram.
z−a
z -plane

Classify this system as one of the following filter types.

1. high pass 2. low pass

3. band pass 4. all pass
5. band stop 0. none of the above
Check Yourself

Classify the system ...

1 − az
H(z) =
z−a

Find the frequency response:

1 − ae jΩ e−jΩ − a ← complex
H(e jΩ ) = jΩ = e jΩ jΩ
e −a e −a ← conjugates
 
Because complex conjugates have equal magnitudes,

H(
e )
= 1.
jΩ

→ all-pass filter

Check Yourself

1 − az
A system H(z) = has the following pole-zero diagram.
z−a
z -plane

Classify this system as one of the following filter types. 4

1. high pass 2. low pass

3. band pass 4. all pass
5. band stop 0. none of the above
Effects of Phase

Effects of Phase

Effects of Phase

http://public.research.att.com/~ttsweb/tts/demo.php

Effects of Phase

artificial speech synthesized by Robert Donovan

Effects of Phase

x[n] ??? y[n] = x[−n]

artificial speech synthesized by Robert Donovan

Effects of Phase

x[n] ??? y[n] = x[−n]

How are the phases of X and Y related?

Effects of Phase

How are the phases of X and Y related?


ak = x[n]e−jkΩ0 n
n
 
−jkΩ0 n
bk = x[−n]e = x[m]e jkΩ0 m = a−k
n m

Flipping x[n] about n = 0 flips ak about k = 0.

Because x[n] is real-valued, ak is conjugate symmetric: a−k = a∗k .
bk = a−k = a∗k = |ak |e−j∠ak
The angles are negated at all frequencies.
Review: Periodicity

DT frequency responses are periodic functions of Ω, with period 2π.

If Ω2 = Ω1 + 2πk where k is an integer then

H(e jΩ2 ) = H(e j(Ω1 +2πk) ) = H(e jΩ1 e j2πk ) = H(e jΩ1 )

The periodicity of H(e jΩ ) results because H(e jΩ ) is a function of e jΩ ,

which is itself periodic in Ω. Thus DT complex exponentials have
many “aliases.”

e jΩ2 = e j(Ω1 +2πk) = e jΩ1 e j2πk = e jΩ1

Because of this aliasing, there is a “highest” DT frequency: Ω = π.

Review: Periodic Sinusoids

There are N distinct DT complex exponentials with period N .

If e jΩn is periodic in N then

e jΩn = e jΩ(n+N ) = e jΩn e jΩN

and e jΩN must be 1, and Ω must be one of the N th roots of 1.

Example: N = 8
z -plane
Review: DT Fourier Series

DT Fourier series represent DT signals in terms of the amplitudes

and phases of harmonic components.

DT Fourier Series
1  2π
ak = ak+N = x[n]e−jΩ0 n ; Ω0 = (“analysis” equation)
N N
n=<N >

x[n]= x[n + N ] = ak ejkΩ0 n (“synthesis” equation)
k=<N >
DT Fourier Series

DT Fourier series have simple matrix interpretations.



jk
2π

x[n] = x[n + 4] = jkΩ0 n

ak e
=
ak e 4 n
=
ak j kn

k=<4> k=<4> k=<4>

⎡
⎤
⎡
⎤ ⎡
⎤

x[0]
1 1 1 1 a0

⎢ x[1]
⎥ ⎢ 1 −1 −j ⎥⎢a ⎥
⎢ ⎥ ⎢ j ⎥ ⎢ 1
⎥
⎢ =

⎥ ⎢ ⎥⎢ ⎥
⎣
x[2]
⎦
⎣
1 −1 1 −1 ⎦
⎣
a2
⎦

x[3] 1 −j −1 j a3

1

−jkΩ0 n
1

−jk
2π n
1


ak
= ak+4 = x[n]e
=
e
N
=
x[n]j −kn

4
4
4

n=<4>
n=<4>
n=<4>

⎡
⎤
⎡
⎤⎡ ⎤

a0
1 1 1 1 x[0]

⎢a ⎥ ⎢1 −j −1
j
⎥ ⎢ x[1]
⎥
⎢ 1
⎥ ⎢ ⎥⎢ ⎥
⎢ ⎥ =
⎢ ⎥⎢ ⎥
⎣
a2
⎦
⎣
1 −1 1 −1
⎦
⎣
x[2]
⎦

a3 1 j −1 −j x[3]

These matrices are inverses of each other.

Scaling

DT Fourier series are important computational tools.

However, the DT Fourier series do not scale well with the length N.

1

−jkΩ0 n
1

−jk
2π n
1


ak
= ak+2 = x[n]e
=
e
2 =
x[n](−1)−kn

2
2
2

n=<2>
n=<2>
n=<2>

a0 1
1 1 x[0]

=
a1 2 1 −1 x[1]

1

−jkΩ0 n
1

−jk
2π n
1


ak
= ak+4 = x[n]e
=
e
4 =
x[n]j −kn

4
4
4

n=<4>
n=<4>
n=<4>

⎡
⎤
⎡
⎤⎡ ⎤

a0
1 1 1 1 x[0]

⎢ a ⎥ 1
⎢ 1 −j −1
j
⎥ ⎢ ⎥
⎢ 1
⎥ ⎢ ⎥ ⎢ x[1]
⎥
⎢ ⎥ = ⎢ ⎥⎢ ⎥
⎣
a2
⎦
4 ⎣
1 −1 1 −1
⎦
⎣
x[2]
⎦

a3 1 j −1 −j x[3]
Number of multiples increases as N 2 .
Fast Fourier “Transform”

Exploit structure of Fourier series to simplify its calculation.

Divide FS of length 2N into two of length N (divide and conquer).

Matrix formulation of 8-point FS:

⎡
⎤
⎡
0 ⎤⎡ ⎤

c0
W
8
W
8
0 W
8
0 W80 W
8
0 W
8
0 W80 W80 x[0]
⎢ c1
⎥ ⎢ W
0 W
8
1 W
8
2 W8
3 W
8
4 W
8
5 W86 W87 ⎥ ⎢ ⎥
⎢ ⎥ ⎢ 8
⎥ ⎢ x[1]
⎥
⎢ c ⎥ ⎢ W
0 W
8
2 W
8
4 W86 W
8
0 W
8
2 W84 W86 ⎥ ⎢ ⎥
⎢ 2
⎥ ⎢ 8
⎥ ⎢ x[2]
⎥
⎢ ⎥ ⎢ 0 ⎥⎢ ⎥
⎢ c3
⎥ ⎢ W
8

⎢ ⎥ =
⎢ W
8
3 W
8
6 W81 W
8
4 W
8
7 W82 5
W8 ⎥ ⎢
⎥ ⎢ x[3]
⎥
⎥
⎢ c ⎥ ⎢ W
0 W
8
4 W
8
0 W84 W
8
0 W
8
4 W80 W84 ⎥ ⎢ ⎥
⎢ 4
⎥ ⎢ 8
⎥ ⎢ x[4]
⎥
⎢ ⎥ ⎢ 0 ⎥⎢ ⎥
⎢ c5
⎥ ⎢ W
8
W
8
5 W
8
2 W87 W
8
4 W
8
1 W86 W83 ⎥ ⎢ x[5]
⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣
c6
⎦
⎣
W
0 W
8
6 W
8
4 W
82 W
8
0 W
8
6 W84 W8 ⎦
⎣
x[6]
⎦

2
8

c7
W
8
0 W
8
7 W
8
6 W
85 W
8
4 W
8
3 W82 W81 x[7]

−j 2π
where WN = e N

8 × 8 = 64 multiplications
FFT

Divide into two 4-point series (divide and conquer).

Even-numbered entries in x[n]:

⎡
⎤
⎡
0 ⎤⎡ ⎤

a0
W
4
W
4
0 W
4
0 0
W
4
x[0]
⎢ a ⎥ ⎢ W
0 W
1 W
4
2 W
4
3 ⎥ ⎢ ⎥
⎢ 1
⎥ ⎢ 4
4
⎥ ⎢ x[2]
⎥
⎢ ⎥ =
⎢ 0 ⎥⎢ ⎥
⎣
a2
⎦
⎣
W
4
W
4
2 W
4
0 2
W
4
⎦
⎣
x[4]
⎦

a3
W
4
0 W
4
3 W
4
2 W
4
1 x[6]

Odd-numbered entries in x[n]:

⎡
⎤
⎡
0 ⎤⎡ ⎤

b0 W
4
W
40 W
4
0 0
W
4
x[1]
⎢ b ⎥ ⎢ W
0 W
1 W
4
2 W
4
3 ⎥ ⎢ ⎥
⎢ 1
⎥ ⎢ 4
4 ⎥ ⎢ x[3]
⎥
⎢ ⎥ =
⎢ 0 ⎥⎢ ⎥
⎣
b2
⎦
⎣
W
4
W
4
2 W
4
0 2
W
4
⎦
⎣
x[5]
⎦

b3 W
4
0 W
43 W
4
2 W
4
1 x[7]

Sum of multiplications = 2 × (4 × 4) = 32: fewer than the previous 64.

FFT

Break the original 8-point DTFS coefficients ck into two parts:

ck = dk + ek

where dk comes from the even-numbered x[n] (e.g., ak ) and ek comes

from the odd-numbered x[n] (e.g., bk )
FFT

The 4-point DTFS coefficients ak of the even-numbered x[n]

⎡
⎤ ⎡ ⎤
⎡
⎤ ⎡ 0 ⎤
⎡
⎤

a0
W
4
0 W
4
0 W
4
0 0
W
4
x[0]
W
8
W
8
0 W
8
0 W
80 x[0]

⎢ a ⎥ ⎢ W
0 W
4
1 W
4
2 W
4
3 ⎥⎢ ⎥ ⎢ 0 W
8
2 W
8
4 W
86 ⎥ ⎢ ⎥
⎢ 1
⎥ ⎢ 4
⎥⎢ x[2]
⎥ ⎢ W
8
⎥⎢ x[2]
⎥
⎢ ⎥=
⎢ 0 ⎥⎢ ⎥=
⎢ 0 ⎥⎢ ⎥
⎣
a2
⎦
⎣
W
4
W
4
2 W
4
0 2
W
4
⎦
⎣
x[4]
⎦
⎣
W
8
W
8
4 W
8
0 W
8
4 ⎦
⎣
x[4]
⎦

a3
W
4
0 W
4
3 W
4
2 W
4
1 x[6]
W
8
0 W8
6 W
8
4 W
8
2 x[6]

contribute to the 8-point DTFS coefficients dk :

⎡
⎤
⎡
0 ⎤⎡ ⎤

d0
W
8
W
8
0 W
8
0 W
8
0 W
8
0 W
8
0 W80 W
80 x[0]
⎢ d1
⎥ ⎢ W
0 W
8
1 W
8
2 W
8
3 W
8
4 W
8
5 W86 W
87 ⎥ ⎢ x[1]
⎥
⎢ ⎥ ⎢ 8
⎥⎢ ⎥
⎢ d ⎥ ⎢ W
0 W
8
2 W
8
4 W
8
6 W
8
0 W
8
2 W84 W
8 ⎥ ⎢ x[2]
⎥
6 ⎥ ⎢
⎢ 2
⎥ ⎢ 8
⎥
⎢ ⎥ ⎢ 0 ⎥ ⎢ ⎥
⎢ d3
⎥ ⎢ W
8

⎢ ⎥ =
⎢ W
8
3 W
8
6 W
8
1 W
8
4 W
8
7 W82 5
W
8 ⎥ ⎢ ⎥
⎥ ⎢ x[3]
⎥
⎢ d ⎥ ⎢ W
0 W
8
4 W
8
0 W
84 W
8
0 W
8
4 W80 W
84 ⎥ ⎢ ⎥
⎢ 4
⎥ ⎢ 8
⎥ ⎢ x[4]
⎥
⎢ ⎥ ⎢ 0 ⎥⎢ ⎥
⎢ d5
⎥ ⎢ W
8
W
8
5 W
8
2 W
8
7 W
8
4 W
8
1 W86 W
8
3 ⎥ ⎢ x[5]
⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣
d6
⎦
⎣
W
0 W
8
6 W
8
4 W
82 W
8
0 W
8
6 W
8
4 W
8 ⎦
⎣
x[6]
⎦

2
8
d7
W
80 W
8
7 W
8
6 W
8
5 W
8
4 W
8
3 W
8
2 W
8
1 x[7]

FFT

The 4-point DTFS coefficients ak of the even-numbered x[n]

⎡
⎤ ⎡ ⎤
⎡
⎤ ⎡ 0 ⎤
⎡
⎤

a0
W
4
0 W
4
0 W
4
0 0
W
4
x[0]
W
8
W
8
0 W
8
0 W
80 x[0]

⎢ a ⎥ ⎢ W
0 W
4
1 W
4
2 W
4
3 ⎥⎢ ⎥ ⎢ 0 W
8
2 W
8
4 W
86 ⎥ ⎢ ⎥
⎢ 1
⎥ ⎢ 4
⎥⎢ x[2]
⎥ ⎢ W
8
⎥⎢ x[2]
⎥
⎢ ⎥=
⎢ 0 ⎥⎢ ⎥=
⎢ 0 ⎥⎢ ⎥
⎣
a2
⎦
⎣
W
4
W
4
2 W
4
0 2
W
4
⎦
⎣
x[4]
⎦
⎣
W
8
W
8
4 W
8
0 W
8
4 ⎦
⎣
x[4]
⎦

a3
W
4
0 W
4
3 W
4
2 W
4
1 x[6]
W
8
0 W
8
6 W
8
4 W
8
2 x[6]

contribute to the 8-point DTFS coefficients dk :

⎡
⎤
⎡
0 ⎤⎡ ⎤

d0
W
8
W
8
0 W
8
0 W
8
0 x[0]
⎢ d1
⎥ ⎢ W
0 W
8
2 W
8
4 W
8
6 ⎥⎢ ⎥
⎢ ⎥ ⎢ 8
⎥⎢ ⎥
⎢ d ⎥ ⎢ W
0 W
8
4 W
8
0 W
8
4 ⎥ ⎢ x[2]
⎥
⎢ 2
⎥ ⎢ 8
⎥⎢ ⎥
⎢ ⎥ ⎢ 0 ⎥⎢ ⎥
⎢ d3
⎥ ⎢ W
8

⎢ ⎥ =
⎢ W
8
6 W
8
4 W
8
2 ⎥⎢
⎥⎢
⎥
⎥
⎢ d ⎥ ⎢ W
0 W
8
0 W
8
0 W
8
0 ⎥ ⎢ x[4]
⎥
⎢ 4
⎥ ⎢ 8
⎥⎢ ⎥
⎢ ⎥ ⎢ 0 ⎥⎢ ⎥
⎢ d5
⎥ ⎢ W
8
W
8
2 W
8
4 W
8
6 ⎥⎢ ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣
d6
⎦
⎣
W
0 W
8
4 W
8
0 W
8
4 ⎦
⎣
x[6]
⎦

d7 W
80 W
8
6 W
8
4 W
8
2
FFT

The 4-point DTFS coefficients ak of the even-numbered x[n]

⎡
⎤ ⎡ ⎤
⎡
⎤ ⎡ 0 ⎤
⎡
⎤

a0
W
4
0 W
4
0 W
4
0 0
W
4
x[0]
W
8
W
8
0 W
8
0 0
W
8
x[0]

⎢ a ⎥ ⎢ W
0 W
4
1 W
4
2 W
4
3 ⎥⎢ ⎥ ⎢ 0 W
8
2 W
8
4 W
86 ⎥ ⎢ ⎥
⎢ 1
⎥ ⎢ 4
⎥⎢ x[2]
⎥ ⎢ W
8
⎥⎢ x[2]
⎥
⎢ ⎥=
⎢ 0 ⎥⎢ ⎥=
⎢ 0 ⎥⎢ ⎥
⎣
a2
⎦
⎣
W
4
W
4
2 W
4
0 2
W
4
⎦
⎣
x[4]
⎦
⎣
W
8
W
8
4 W
8
0 W
84 ⎦
⎣
x[4]
⎦

a3
W
4
0 W
43 W
4
2 W
4
1 x[6]
W
8
0 W
8
6 W
8
4 W
8
2 x[6]

contribute to the 8-point DTFS coefficients dk :

⎡
⎤
⎡
⎤
⎡
0 ⎤⎡ ⎤

d0
a0
W
8
W
8
0 W
8
0 W
8
0 x[0]
⎢ d1
⎥ ⎢ a1
⎥ ⎢ W
0 W
8
2 W
8
4 W
8
6 ⎥⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ 8
⎥⎢ ⎥
⎢ d ⎥ ⎢ a ⎥ ⎢ W
0 W
8
4 W
8
0 W
8
4 ⎥ ⎢ x[2]
⎥
⎢ 2
⎥ ⎢ 2
⎥ ⎢ 8
⎥⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ 0 ⎥⎢ ⎥
⎢ d3
⎥ ⎢ a3
⎥ ⎢ W
8

⎢ ⎥ =
⎢ ⎥ =
⎢ W
8
6 W
8
4 W
8
2 ⎥⎢
⎥⎢
⎥
⎥
⎢ d ⎥ ⎢ a ⎥ ⎢ W
0 W
8
0 W
8
0 W
8
0 ⎥ ⎢ x[4]
⎥
⎢ 4
⎥ ⎢ 0
⎥ ⎢ 8
⎥⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ 0 ⎥⎢ ⎥
⎢ d5
⎥ ⎢ a1
⎥ ⎢ W
8
W
8
2 W
8
4 W
8
6 ⎥⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣
d6
⎦
⎣
a2
⎦
⎣
W
0 W
8
4 W
8
0 W
8
4 ⎦
⎣
x[6]
⎦

d7 a3 W
8
0 W
8
6 W
8
4 W
8
2
FFT

The 4-point DTFS coefficients ak of the even-numbered x[n]

⎡
⎤ ⎡ ⎤
⎡
⎤ ⎡ 0 ⎤
⎡
⎤

a0
W
4
0 W
4
0 W
4
0 0
W
4
x[0]
W
8
W
8
0 W
8
0 0
W
8
x[0]

⎢ a ⎥ ⎢ W
0 W
4
1 W
4
2 W
4
3 ⎥⎢ ⎥ ⎢ 0 W
8
2 W
8
4 W
86 ⎥ ⎢ ⎥
⎢ 1
⎥ ⎢ 4
⎥⎢ x[2]
⎥ ⎢ W
8
⎥⎢ x[2]
⎥
⎢ ⎥=
⎢ 0 ⎥⎢ ⎥=
⎢ 0 ⎥⎢ ⎥
⎣
a2
⎦
⎣
W
4
W
4
2 W
4
0 2
W
4
⎦
⎣
x[4]
⎦
⎣
W
8
W
8
4 W
8
0 W
84 ⎦
⎣
x[4]
⎦

a3
W
4
0 W
43 W
4
2 W
4
1 x[6]
W
8
0 W
8
6 W
8
4 W
8
2 x[6]

contribute to the 8-point DTFS coefficients dk :

⎡ ⎤ ⎡ ⎤ ⎡
0 ⎤⎡ ⎤

d0 a0 W
8
W
8
0 W
8
0 W
8
0 x[0]
⎢ d1 ⎥ ⎢ a1 ⎥ ⎢ W
0 W
8
2 W
8
4 W
8
6 ⎥⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ 8
⎥⎢ ⎥
⎢ d ⎥ ⎢ a ⎥ ⎢ W
0 W
8
4 W
8
0 W
8
4 ⎥ ⎢ x[2]
⎥
⎢ 2 ⎥ ⎢ 2 ⎥ ⎢ 8
⎥⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ 0 ⎥⎢ ⎥
⎢ d3 ⎥ ⎢ a3 ⎥ ⎢ W
8

⎢ ⎥ =
⎢ ⎥ =
⎢ W
8
6 W
8
4 W
8
2 ⎥⎢
⎥⎢
⎥
⎥
⎢ d ⎥ ⎢ a ⎥ ⎢ W
0 W
8
0 W
8
0 W
8
0 ⎥ ⎢ x[4]
⎥
⎢ 4 ⎥ ⎢ 0 ⎥ ⎢ 8
⎥⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ 0 ⎥⎢ ⎥
⎢ d5 ⎥ ⎢ a1 ⎥ ⎢ W
8
W
8
2 W
8
4 W
8
6 ⎥⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣ d6 ⎦ ⎣ a2 ⎦ ⎣
W
0 W
8
4 W
8
0 W
8
4 ⎦
⎣
x[6]
⎦

d7 a3 W
8
0 W
8
6 W
8
4 W
8
2
FFT

The ek components result from the odd-number entries in x[n].

⎡
⎤
⎡
⎤
⎡
⎤ ⎡ 0 ⎤
⎡
⎤

b0 W
4
0 W
4
0 W
4
0 0
W
4
x[1]
W
8
W
8
0 W
8
0 0
W
8
x[1]

⎢ b ⎥ ⎢ W
0 W
4
1 W
4
2 W
4
3 ⎥⎢ ⎥ ⎢ 0 W
8
2 W
8
4 W
8
6 ⎥⎢ ⎥
⎢ 1
⎥ ⎢ 4
⎥⎢ x[3]
⎥ ⎢ W
8
⎥⎢ x[3]
⎥
⎢ ⎥=
⎢ 0 ⎥⎢ ⎥=
⎢ 0 ⎥⎢ ⎥
⎣
b2
⎦
⎣
W
4
W
4
2 W
4
0 2
W
4
⎦
⎣
x[5]
⎦
⎣
W
8
W
8
4 W
8
0 W
8
4 ⎦
⎣
x[5]
⎦

b3 W4
0 W
4
3 W
4
2 W
4
1 x[7]
W
8
0 W8
6 W8
4 W
8
2 x[7]

⎡
⎤
⎡
0 ⎤⎡ ⎤

e0
W
8
W
8
0 W
8
0 W80 W
8
0 W
8
0 W80 W
80 x[0]
⎢ e1
⎥ ⎢ W
0 W
8
1 W
8
2 W8
3 W
8
4 W
8
5 W86 W
87 ⎥ ⎢ x[1]
⎥
⎢ ⎥ ⎢ 8
⎥⎢ ⎥
⎢ e ⎥ ⎢ W
0 W
8
2 W
8
4 W86 W
8
0 W
8
2 W84 W
8 ⎥ ⎢ x[2]
⎥
6 ⎥ ⎢
⎢ 2
⎥ ⎢ 8
⎥
⎢ ⎥ ⎢ 0 ⎥ ⎢ ⎥
⎢ e3
⎥ ⎢ W
8

⎢ ⎥ =
⎢ W
8
3 W
8
6 W81 W
8
4 W
8
7 W82 5
W
8
⎥ ⎢ ⎥
⎥ ⎢ x[3]
⎥
⎢ e ⎥ ⎢ W
0 W
8
4 W
8
0 W84 W
8
0 W
8
4 W80 W
84 ⎥ ⎢ ⎥
⎢ 4
⎥ ⎢ 8
⎥ ⎢ x[4]
⎥
⎢ ⎥ ⎢ 0 ⎥⎢ ⎥
⎢ e5
⎥ ⎢ W
8
W
8
5 W
8
2 W
8
7 W
8
4 W
8
1 W86 W
8
3 ⎥ ⎢ x[5]
⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣
e6
⎦
⎣
W
0 W
8
6 W
8
4 W
82 W
8
0 W
8
6 W84 W
8 ⎦
⎣
x[6]
⎦

2
8

e7
W
80 W
8
7 W
8
6 W
8
5 W
8
4 W
8
3 W
8
2 W
8
1 x[7]

FFT

The ek components result from the odd-number entries in x[n].

⎡
⎤
⎡
⎤
⎡
⎤ ⎡ 0 ⎤
⎡
⎤

b0 W
4
0 W
4
0 W
4
0 0
W
4
x[1]
W
8
W
8
0 W
8
0 0
W
8
x[1]

⎢ b ⎥ ⎢ W
0 W
4
1 W
4
2 W
4
3 ⎥⎢ ⎥ ⎢ 0 W
8
2 W
8
4 W
8
6 ⎥⎢ ⎥
⎢ 1
⎥ ⎢ 4
⎥⎢ x[3]
⎥ ⎢ W
8
⎥⎢ x[3]
⎥
⎢ ⎥=
⎢ 0 ⎥⎢ ⎥=
⎢ 0 ⎥⎢ ⎥
⎣
b2
⎦
⎣
W
4
W
4
2 W
4
0 2
W
4
⎦
⎣
x[5]
⎦
⎣
W
8
W
8
4 W
8
0 W
8
4 ⎦
⎣
x[5]
⎦

b3 W4
0 W
4
3 W
4
2 W
4
1 x[7]
W
8
0 W
8
6 W
8
4 W
8
2 x[7]

⎡
⎤
⎡
⎤ ⎡
⎤

e0
W
8
0 W
8
0 W
8
0 W
8
0
⎢ e1
⎥ ⎢ W
8
1 W
8
3 W
8
5 W
8
7 ⎥ ⎢ x[1]
⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢e ⎥ ⎢ W
8
2 W
8
6 W
8
2 6 ⎥ ⎢ ⎥
⎢ 2
⎥ ⎢ W
8
⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ e3
⎥ ⎢
⎢ ⎥ =
⎢ W
8
3 W
8
1 W
8
7 5
W
8
⎥ ⎢
⎥ ⎢ x[3]
⎥
⎥
⎢e ⎥ ⎢ W
8
4 W
8
4 W
8
4 W
8
4 ⎥ ⎢ ⎥
⎢ 4
⎥ ⎢ ⎥⎢ ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ e5
⎥ ⎢ W
8
5 W
8
7 W
8
1 W
8
3 ⎥ ⎢ x[5]
⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣
e6
⎦
⎣
W
8
6 W
8
2 W
8
6 W
8
⎦
⎣

2 ⎦

e7
W
8
7 W
8
5 W
8
3 W
81 x[7]

FFT

The ek components result from the odd-number entries in x[n].

⎡
⎤
⎡
⎤
⎡
⎤ ⎡ 0 ⎤
⎡
⎤

b0 W
4
0 W
4
0 W
4
0 0
W
4
x[1]
W
8
W
8
0 W
8
0 0
W
8
x[1]

⎢ b ⎥ ⎢ W
0 W
4
1 W
4
2 W
4
3 ⎥⎢ ⎥ ⎢ 0 W
8
2 W
8
4 W
8
6 ⎥⎢ ⎥
⎢ 1
⎥ ⎢ 4
⎥⎢ x[3]
⎥ ⎢ W
8
⎥⎢ x[3]
⎥
⎢ ⎥=
⎢ 0 ⎥⎢ ⎥=
⎢ 0 ⎥⎢ ⎥
⎣
b2
⎦
⎣
W
4
W
4
2 W
4
0 2
W
4
⎦
⎣
x[5]
⎦
⎣
W
8
W
8
4 W
8
0 W
8
4 ⎦
⎣
x[5]
⎦

b3 W4
0 W
4
3 W
4
2 W
4
1 x[7]
W
8
0 W
8
6 W
8
4 W
8
2 x[7]

⎡
⎤
⎡
0 ⎤
⎡
⎤ ⎡
⎤

e0
W
8
b0
W
8
0 W
8
0 W
8
0 W
8
0
⎢ e1
⎥ ⎢ W
1 b1
⎥ ⎢
⎢ ⎥ ⎢ 8
⎥ ⎢ W
8
1 W
8
3 W
8
5 W
8
7 ⎥
⎥⎢
⎢ x[1]
⎥
⎥
⎢ e ⎥ ⎢ W
2 b ⎥ ⎢ W
8
2 W
8
6 W
8
2 6 ⎥
W
8
⎥ ⎢⎢ ⎥
⎢ 2
⎥ ⎢ 8
2
⎥ ⎢ ⎥
⎢ ⎥ ⎢ 3 ⎥ ⎢ ⎥ ⎢ ⎥
⎢ e3
⎥ ⎢ W
8
b3
⎥ ⎢
⎢ ⎥ =
⎢ ⎥ ⎢ W
8
3 W
8
1 W
8
7 5
W
8
⎥ ⎢
⎥ ⎢ x[3]
⎥
⎥
⎢ e ⎥ ⎢ W
4 b ⎥ =
⎢ W
8
4 W
8
4 W
8
4 W
8
4 ⎥ ⎢ ⎥
⎢ 4
⎥ ⎢ 8
0
⎥ ⎢ ⎥⎢ ⎥
⎢ ⎥ ⎢ 5 ⎥ ⎢ ⎥⎢ ⎥
⎢ e5
⎥ ⎢ W
8
b1
⎥ ⎢ W
8
5 W
8
7 W
8
1 W
8
3 ⎥ ⎢ x[5]
⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣
e6
⎦
⎣
W
b2
⎦
⎣

6
8
W
8
6 W
8
2 W
8
6 W
8
⎦
⎣

2 ⎦

e7
W
87 b3 W
8
7 W
8
5 W
8
3 W
81 x[7]

FFT

The ek components result from the odd-number entries in x[n].

⎡
⎤ ⎡ ⎤
⎡
⎤ ⎡ 0 ⎤
⎡
⎤

b
0
W
4
0 W
4
0 W
4
0 0
W
4
x
[1]
W
8
W
8
0 W
8
0 0
W
8
x
[1]

⎢ b
⎥ ⎢ W
0 W
4
1 W
4
2 W
4
3 ⎥⎢ ⎥ ⎢ 0 W
8
2 W
8
4 W
8
6 ⎥⎢ ⎥
⎢ 1
⎥ ⎢ 4
⎥⎢ x
[3]
⎥ ⎢ W
8
⎥⎢ x
[3]
⎥
⎢ ⎥=
⎢ 0 ⎥⎢ ⎥=
⎢ 0 ⎥⎢ ⎥
⎣
b
2
⎦
⎣
W
4
W
4
2 W
4
0 2
W
4
⎦
⎣
x
[5]
⎦
⎣
W
8
W
8
4 W
8
0 W
8
4 ⎦
⎣
x
[5]
⎦

b
3
W
4
0 W
43 W
4
2 W
4
1 x[7]
W
8
0 W
8
6 W
8
4 W
8
2 x
[7]

⎡ ⎤ ⎡ 0 ⎤ ⎡
⎤ ⎡
⎤

e0 W 8 b0 W
8
0 W
8
0 W
8
0 W
8
0
⎢ e1 ⎥ ⎢ W 1 b1 ⎥ ⎢ W
8
1 W
8
3 W
8
5 W
8
7 ⎥ ⎢ x
[1]
⎥
⎢ ⎥ ⎢ 8 ⎥ ⎢ ⎥⎢ ⎥
⎢e ⎥ ⎢W2 b ⎥ ⎢ W
8
2 W
8
6 W
8
2 6 ⎥ ⎢ ⎥
⎢ 2⎥ ⎢ 8 2⎥ ⎢ W
8
⎥ ⎢ ⎥
⎢ ⎥ ⎢ 3 ⎥ ⎢ ⎥ ⎢ ⎥
⎢ e3 ⎥ ⎢ W8 b3 ⎥ ⎢
⎢ ⎥=⎢ ⎥ ⎢ W
8
3 W
8
1 W
8
7 5
W
8
⎥ ⎢
⎥ ⎢ x
[3]
⎥
⎥
⎢ e ⎥ ⎢ W 4 b ⎥ =
⎢ W
8
4 W
8
4 W
8
4 W
8
4 ⎥ ⎢ ⎥
⎢ 4⎥ ⎢ 8 0⎥ ⎢ ⎥⎢ ⎥
⎢ ⎥ ⎢ 5 ⎥ ⎢ ⎥⎢ ⎥
⎢ e5 ⎥ ⎢ W8 b1 ⎥ ⎢ W
8
5 W
8
7 W
8
1 W
8
3 ⎥ ⎢ x
[5]
⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣ e6 ⎦ ⎣ W b2 ⎦ ⎣

6 W
8
6 W
8
2 W
8
6 W
8
⎦
⎣

2 ⎦

8
e7 W87 b3 W
8
7 W
8
5 W
8
3 W
81 x
[7]

FFT

Combine ak and bk to get ck .

⎡
⎤
⎡
⎤
⎡
⎤
⎡
0 ⎤

c0
d0 + e0
a0
W 8 b0
⎢ c1
⎥ ⎢ d1 + e1
⎥ ⎢ a1
⎥ ⎢ W 1 b1 ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 8 ⎥
⎢c ⎥ ⎢d + e ⎥ ⎢a ⎥ ⎢W2 b ⎥
⎢ 2
⎥ ⎢ 2 2
⎥ ⎢ 2
⎥ ⎢ 2⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 83 ⎥
⎢ c3
⎥ ⎢ d3 + e3
⎥ ⎢ a3
⎥ ⎢ W8 b3 ⎥
⎢ ⎥ =
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ c ⎥ ⎢ d + e ⎥ =
⎢ a ⎥ +
⎢ W 4 b ⎥
⎢ 4
⎥ ⎢ 4 4
⎥ ⎢ 0
⎥ ⎢ 8 0 ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 5 ⎥
⎢ c5
⎥ ⎢ d5 + e5
⎥ ⎢ a1
⎥ ⎢ W8 b1 ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣
c6
⎦
⎣
d6 + e6
⎦
⎣
a2
⎦
⎣
W 6 b2 ⎦

8
c7 d7 + e7 a3 W87 b3

FFT procedure:
• compute ak and bk : 2 × (4 × 4) = 32 multiplies
• combine ck = ak + W8k bk : 8 multiples
• total 40 multiplies: fewer than the orginal 8 × 8 = 64 multiplies

Scaling of FFT algorithm

How does the new algorithm scale?

Let M (N ) = number of multiplies to perform an N point FFT.

M (1) = 0

M (2) = 2M (1) + 2 = 2

M (4) = 2M (2) + 4 = 2 × 4

M (8) = 2M (4) + 8 = 3 × 8

M (16) = 2M (8) + 16 = 4 × 16

M (32) = 2M (16) + 32 = 5 × 32

M (64) = 2M (32) + 64 = 6 × 64

M (128) = 2M (64) + 128 = 7 × 128

...

M (N ) = (log2 N ) × N

Significantly smaller than N 2 for N large.

Fourier Transform: Generalize to Aperiodic Signals

An aperiodic signal can be thought of as periodic with infinite period.

Fourier Transform: Generalize to Aperiodic Signals

An aperiodic signal can be thought of as periodic with infinite period.

Let x[n] represent an aperiodic signal DT signal.

x[n]
1
n
0
∞

“Periodic extension”: xN [n] = x[n + kN ]
k=−∞

xN [n]
1
n
N
Then x[n] = lim xN [n].
N →∞
Fourier Transform

Represent xN [n] by its Fourier series.

xN [n]
1
n
−N1 N1 N
N1 
1


1 
−j
2
N
π
kn 1 
−j 2
N
π
kn
1 sin
N1 + 2
Ω

ak
= xN [n]e = e
=

N
N
n=−N1

N
sin 12 Ω

N ak sin 32 Ω
sin 12 Ω
k
Ω
2π 2π
Ω0 = Ω = kΩ0 = k
N N

Fourier Transform

Doubling period doubles # of harmonics in given frequency interval.

xN [n]
1
n
−N1 N1 N
N1 
1


1 
−j
2
N
π
kn 1 
−j 2
N
π
kn
1 sin
N1 + 2
Ω

ak
= xN [n]e = e
=

N
N
n=−N1

N
sin 12 Ω

N ak sin 32 Ω
sin 12 Ω
k
Ω
2π 2π
Ω0 = Ω = kΩ0 = k
N N

Fourier Transform

As N → ∞, discrete harmonic amplitudes → a continuum E(Ω).

xN [n]
1
n
−N1 N1 N
N 
1


1

−j
2
N
π kn
1

−j
2
π kn

1
1
sin
N1 + 2
Ω

ak
=
xN [n]e

=
N
=

e
1 Ω

N
N
n=−N
N
sin 2
1

N ak sin 32 Ω
sin 12 Ω
k
Ω
2π 2π
Ω0 = Ω = kΩ 0 = k
N N
 −j
2
N
π kn


N ak
=
=
x[n]e
−jΩn = E(Ω)

x[n]e

n=<N > n=<N >

Fourier Transform

As N → ∞, synthesis sum → integral.

xN [n]
1
n
−N1 N1 N
N ak sin 32 Ω

sin 12 Ω
k
Ω
2π 2π
Ω0 = Ω = kΩ0 = k
N
N
 −j 2π

N ak = x[n]e N kn = x[n]e−jΩn = E(Ω)
n=<N > n=<N > 
 1 j 2π
 Ω0 1
x[n] = E(Ω) e N kn = E(Ω)e jΩn → E(Ω)e jΩn dΩ
N 2π 2π 2π

k=<N >    k=<N >

ak

Fourier Transform

Replacing E(Ω) by X(e jΩ ) yields the DT Fourier transform relations.

∞

X(e jΩ )= x[n]e−jΩn (“analysis” equation)
n=−∞

1
x[n]= X(e jΩ )e jΩn dΩ (“synthesis” equation)
2π 2π
Relation between Fourier and Z Transforms

If the Z transform of a signal exists and if the ROC includes the

unit circle, then the Fourier transform is equal to the Z transform
evaluated on the unit circle.

Z transform:
∞

X(z) = x[n]z −n
n=−∞

DT Fourier transform:
∞
 
jΩ
X(e ) = x[n]e −jΩn
= H(z)z =e jΩ
n=−∞

Relation between Fourier and Z Transforms

Fourier transform “inherits” properties of Z transform.

Property x[n] X(z) X(e jΩ )

Linearity ax1 [n] + bx2 [n] aX1 (s) + bX2 (s) aX1 (e jΩ ) + bX2 (e jΩ )

Time shift x[n − n0 ] z −n0 X(z) e−jΩn0 X(e jΩ )

d d
Multiply by n nx[n] −z X(z) j X(e jΩ )
dz dΩ
Convolution (x1 ∗ x2 )[n] X1 (z) × X2 (z) X1 (e jΩ ) × X2 (e jΩ )
Fourier Representations: Summary

Thinking about signals by their frequency content and systems as

filters has a large number of practical applications.
MIT OpenCourseWare
http://ocw.mit.edu

6.003 Signals and Systems

Spring 2010

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.