Eck FFT

A Tutorial on Fourier Analysis

Douglas Eck
University of Montreal
NYU March 2006
A fundamental and three odd harmonics (3,5,7)

1
fund (freq 100)
3rd harm
5th harm
7th harmm
0.5
0.5
20
40
60
80
100
120
140
160
180
200
Sum of odd harmonics approximate square wave

1
fund (freq 100)
fund+3rd harm
fund+3rd+5th
fund+3rd+5th+7th
0.5
0.5
20
40
60
80
100
120
140
160
180
200
Sum of odd harmonics from 1 to 127

1
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
20
40
60
80
100
120
140
160
180
200
Linear Combination
In the interval [u1 , u2 ] a function (u) can be written as a linear

combination:
(u) =
i i (u)
i=0
where functions i (u) make up a set of simple elementary

functions. If functions are orthogonal (roughly, perpindicular; inner
product is zero)then coefficients i are independant from one
another.
Continuous Fourier Transform
The most commonly used set of orthogonal functions is the Fourier

series. Here is the analog version of the Fourier and Inverse Fourier:
Z +
X (w ) =
x(t)e (2jwt) dt
x(t) =
X (w )e (2jwt) dw
Discrete Fourier and Inverse Fourier Transform
X (n) =
N1
X
x(k)e 2jnk/N
k=0
x(k) =
N1
1 X
X (n)e 2jnk/N
N
n=0
Taylor Series Expansion
f (x) = f (x0 )+
f 0 (x0 )
f 00 (x0 )
f 000 (x0 )
(xx0 )+
(xx0 )2 +
(xx0 )3 +...
1
12
123
or more compactly as
f (x) =
X
f (n) (x0 )
n=0
n!
(x x0 )n
Taylor series expansion of e j

Since e x is its own derivative, the Taylor series expansion for
f (x) = e x is very simple:
ex =
X
xn
n=0
n!
=1+x +
x2 x3
+
+ ...
2
3!
We can define:
e j ==
X
(j)n
n=0
n!
= 1 + j
2
3
j
+ ...
2
3!
Splitting out real and imaginary parts
All even order terms are real; all odd-order terms are imaginary:
ree j = 1 2 /2 + 4 /4! ...
ime j = 3 /3! + 5 /5! ...
Fourier Transform as sum of sines and cosines

Observe that:
cos() =
n>=0;n is even
sin() =
X
n>=0;n is odd
(1)n/2 n
n!
(1)(n1)/2 n
n!
Thus yielding Eulers formula:

e j = cos() + j sin()
Fourier transform as kernel matrix
Example
Sum of cosines with frequencies 12 and 9, sampling rate = 120
0.02
0.01
0
0.01
0.02
20
40
60
signal
80
100
120
20
40
60
real part
two cosines (freqs=9, 12)
80
100
120
20
40
60
imag part
80
100
120
1
0.5
0
0.5
0.2
0.1
0
0.1
0.2
Example
FFT coefficients mapped onto unit circle
FFT projected onto unit circle
1
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
1
1
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
Impulse response
impulse response
1
0.5
10
20
30
40
50
signal
60
70
80
90
100
10
20
30
40
50
magnitude
60
70
80
90
100
10
20
30
40
50
phase
60
70
80
90
100
0.01
0.01
0.01
1
0.5
0
0.5
1
Impulse response
delayed impulse response
1
0.5
10
20
30
40
50
signal
60
70
80
90
100
10
20
30
40
50
magnitude
60
70
80
90
100
10
20
30
40
50
phase
60
70
80
90
100
0.01
0.01
0.01
4
2
0
2
4
A look at phase shifting

Sinusoid frequency=5 phase shifted multiple times.
sinusoid freq=5 phase shifted repeatedly
49.5361
49.5361
49.5361
10
20
30
40
50
60
number of points shifted
70
80
90
100
10
20
30
40
50
60
number of points shifted
70
80
90
100
3
2
angle freq=6
magnitude
49.5361
1
0
1
2
3
Sin period 10 + period 30

200
100
0
0.5
0.4
0.3
0.2
0.1
0
freq (mag)
0.1
0.2
0.3
0.4
0.5
0.4
0.3
0.2
0.1
0
phase
0.1
0.2
0.3
0.4
0.5
5
0
5
0.5
1
0
1
100
100
100
200
300
component sinusoids
400
500
600
200
300
400
500
reconstructed signal using component sinusoids vs original
600
5
0
5
2
0
2
200
300
400
reconstructed signal using ifft vs original
500
600
Aliasing
The useful range is the Nyquist frequency (fs/2)
cos(21)
1
0
1
0.1
0.2
0.3
0.1
0.2
0.3
0.1
0.2
0.3
0.1
0.2
0.3
0.1
0.2
0.3
0.1
0.2
0.3
0.4
0.5
0.6
cos(21) sampled at 24 Hz
0.7
0.8
0.9
0.7
0.8
0.9
0.7
0.8
0.9
0.7
0.8
0.9
0.7
0.8
0.9
0.7
0.8
0.9
1
0
1
0.4
0.5
cos(45)
0.6
1
0
1
0.4
0.5
0.6
1
0
1
0.4
0.5
cos(3)
0.6
1
0
1
0.4
0.5
0.6
1
0
1
0.4
0.5
0.6
Leakage
Even below Nyquist, when frequencies in the signal do not align
well with sampling rate of signal, there can be leakage. First
consider a well-aligned exampl (freq = .25 sampling rate)
Sinusoid at 1/4 the Sampling Rate
a)
Amplitude
1
0.5
0
0.5
1
Magnitude (Linear)
b)
10
20
30
40
Time (samples)
50
60
70
40
30
20
10
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Normalized Frequency (cycles per sample))
0.8
0.9
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Magnitude (dB)
c)
0
100
200
300
Leakage
Now consider a poorly-aligned example (freq = (.25 + .5/N) *
sampling rate)
Sinusoid NEAR 1/4 the Sampling Rate
a)
Amplitude
1
0.5
0
0.5
1
10
20
30
40
Time (samples)
50
60
70
30
20
10
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
30
Magnitude (dB)
c)
Magnitude (Linear)
b)
20
10
0
10
Leakage
Comparison:

a)
0
0.5
1
Magnitude (Linear)
b)
10
20
30
40
Time (samples)
50
60
70
30
20
10
c)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.5
0
0.5
1
b)
40
1
Amplitude
Amplitude
1
0.5
Magnitude (Linear)
a)
c)
10
20
30
40
Time (samples)
50
60
70
20
10
0
30
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Magnitude (dB)
Magnitude (dB)
30
0
100
200
300
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
20
10
0
10
Windowing can help

Can minimize effects by multiplying time series by a window that
diminishes magnitude of points near signal edge:
Blackman Window
a)
Amplitude
0.5
10
20
30
40
Time (samples)
50
60
70
Magnitude (dB)
b)
0
50
100
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
Magnitude (dB)
c)
0
50
100
0.5
Leakage Reduced
Comparison:
b)
a)
Amplitude
0.5
1
Magnitude (Linear)
10
20
30
40
Time (samples)
50
60
10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Magnitude (dB)
10
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
10
20
0.8
0.9
30
40
Time (samples)
50
60
70
10
5
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
c)
20
15
30
Magnitude (dB)
0.5
1
20
10
70
30
1
0.5
Magnitude (Linear)
Amplitude
b)
c)

a)
1
0.5
20
0
20
40
60
80
Convolution theorem
This can be understood in terms of the Convolution Theorem.

Convolution in the time domain is multiplication in the frequency
domain via the Fourier transform (F).
F(f g ) = F(f ) F(g )
Computing impulse response
The impulse response h[n] is the response of a system to the unit

impulse function.
Using the impulse response
Once computed, the impulse response can be used to filter any

signal x[n] yielding y [n].
Examples
Filtering using DFT
Goal is to choose good impulse response h[n]
Filtering using DFT

Transform signal into frequency domain
Filtering using DFT

Modify frequency properties of signal via multiplication
Filtering using DFT

Modify frequency properties of signal via multiplication
Transform back into time domain
Difficulties (Why not a perfect filter?)
You can have a perfect filter(!)

Need long impulse response function in both directions

Very non causal

Very non causal
In generating causal version, challenges arise
Gibbs Phenomenon
1
0.5
0
500
400
300
200
100
0
100
200
ideal lopass filter in frequency domain
300
400
500
400
300
200
100
0
100
ideal filter coeffs in time domain
200
300
400
500
100
200
300
400
500
600
truncated causal filter
700
800
900
1000
0
500
400
300
200
100
0
100
Gibbs phenomenon
200
300
400
500
0.1
0
0.1
500
0.1
0
0.1
1.5
1
0.5
Spectral Analysis
Often we want to see spectral energy as a signal evolves over

time
Spectral Analysis

time
Compute Fourier Transform over evenly-spaced frames of data
Spectral Analysis

time
Compute Fourier Transform over evenly-spaced frames of data
Apply window to minimize edge effects
Short-Timescale Fourier Transform (STFT)
X (m, n) =
N1
X
x(k)w (k m)e 2jnk/N
k=0
Where w is some windowing function such as Hanning or gaussian

centered around zero.
The spectrogram is simply the squared magnitude of these STFT
values
Trumpet (G4)
5000
Frequency
4000
3000
2000
1000
3
Time
[Listen]
Violin (G4)
5000
Frequency
4000
3000
2000
1000
0.5
1.5
2
Time
2.5
[Listen]
Flute (G4)
5000
Frequency
4000
3000
2000
1000
0.5
1.5
Time
2.5
[Listen]
Piano (G4)
5000
Frequency
4000
3000
2000
1000
0.1
0.2
0.3
0.4
Time
0.5
0.6
0.7
[Listen]
Voice
8000
7000
6000
Frequency
5000
4000
3000
2000
1000
0.2
0.4
0.6
0.8
1
Time
1.2
1.4
1.6
1.8
[Listen]
C Major Scale (Piano)

16000
14000
12000
Frequency
10000
8000
6000
4000
2000
4
Time
[Listen]
C Major Scale (Piano)

Log Spectrogram (Constant-Q Transform) reveals low-frequency
structure
13669
7671
4305
Frequency
2416
1356
761
427
240
135
4
Time
Time-Space Tradeoff
spoken "Steven Usma"
0.6
0.4
Amp.
0.2
0.2
0.4
0.2
0.4
0.6
0.8
1.2
1.4
1.6
Time-Space Tradeoff
Narrowband Spectrogram overlap=152 timepts=1633
4000
Frequency
3000
2000
1000
0.2
0.4
0.6
0.8
Time
1.2
1.4
1.6
Wideband Spectrogram overlap=30 timepts=6593

4000
Frequency
3000
2000
1000
0.2
0.4
0.6
0.8
Time
1.2
1.4
1.6
Auto-correlation and meter

Autocorrelation long used to find meter in music (Brown 1993)
Lag k auto-correlation a(k) is a special case of cross-correlation
where a signal x is correlated with itself:
N1
1 X
a(k) =
x(n) x(n k)
N
n=k
Auto-correlation and meter

Autocorrelation long used to find meter in music (Brown 1993)
Lag k auto-correlation a(k) is a special case of cross-correlation
where a signal x is correlated with itself:
N1
1 X
a(k) =
x(n) x(n k)
N
n=k
Autocorrelation can also be defined in terms of Fourier analysis

a = F 1 (|F(x)|)
where F is the Fourier transform, F 1 is the inverse Fourier
transform and || indicates taking magnitude from a complex value.
10
15
time (seconds)
20
25
30
10
15
time (seconds)
20
25
30
500
1000
1500
2000
2500 3000
lags (ms)
3500
4000
4500
5000
Time series (top), envelope (middle) and autocorrelation (bottom) of

excerpt from ISMIR 2004 Tempo Induction contest
(Albums-Cafe Paradiso-08.wav). A vertical line marks the actual tempo
(484 msec, 124bpm). Stars mark the tempo and its integer multiples.
Triangles mark levels in the metrical hierarchy.
Fast Fourier Transform
Fourier Transform O(N 2 )

Fast version possible O(NlogN)
Size must be a power of two
Strategy is decimation in time or frequency
Divide and conquer
Rearrange the inputs in bit reversed order
Output transformation (Decimation in Time)

Eck FFT

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Eck FFT

Uploaded by

Copyright:

Available Formats

A Tutorial on Fourier Analysis

A Tutorial on Fourier Analysis

NYU March 2006

A Tutorial on Fourier Analysis

A fundamental and three odd harmonics (3,5,7)

Sum of odd harmonics approximate square wave

A Tutorial on Fourier Analysis

Sum of odd harmonics from 1 to 127

A Tutorial on Fourier Analysis

In the interval [u1 , u2 ] a function (u) can be written as a linear

where functions i (u) make up a set of simple elementary

A Tutorial on Fourier Analysis

Continuous Fourier Transform

The most commonly used set of orthogonal functions is the Fourier

A Tutorial on Fourier Analysis

Discrete Fourier and Inverse Fourier Transform

A Tutorial on Fourier Analysis

Taylor Series Expansion

A Tutorial on Fourier Analysis

Taylor series expansion of e j

A Tutorial on Fourier Analysis

Splitting out real and imaginary parts

A Tutorial on Fourier Analysis

Fourier Transform as sum of sines and cosines

Thus yielding Eulers formula:

A Tutorial on Fourier Analysis

A Tutorial on Fourier Analysis

Fourier transform as kernel matrix

A Tutorial on Fourier Analysis

A Tutorial on Fourier Analysis

A Tutorial on Fourier Analysis

A Tutorial on Fourier Analysis

A Tutorial on Fourier Analysis

A look at phase shifting

A Tutorial on Fourier Analysis

Sin period 10 + period 30

A Tutorial on Fourier Analysis

A Tutorial on Fourier Analysis

A Tutorial on Fourier Analysis

A Tutorial on Fourier Analysis

Sinusoid NEAR 1/4 the Sampling Rate

A Tutorial on Fourier Analysis

Windowing can help

A Tutorial on Fourier Analysis

Sinusoid at 1/4 the Sampling Rate

A Tutorial on Fourier Analysis

This can be understood in terms of the Convolution Theorem.

A Tutorial on Fourier Analysis

Computing impulse response

The impulse response h[n] is the response of a system to the unit

A Tutorial on Fourier Analysis

Using the impulse response

Once computed, the impulse response can be used to filter any

A Tutorial on Fourier Analysis

A Tutorial on Fourier Analysis

Filtering using DFT

Goal is to choose good impulse response h[n]

A Tutorial on Fourier Analysis

Filtering using DFT

Goal is to choose good impulse response h[n]

A Tutorial on Fourier Analysis

Filtering using DFT

Goal is to choose good impulse response h[n]