A very powerful tool: Fourier Analysis

Jean Baptiste Joseph Fourier in 1807 stated

that:
..any periodic function can be written as a
weighted sum of Sines and Cosines..
…and nobody (Lagrange, Laplace, Poisson)
believed the guy !

Instead, he was right: under certain

conditions (Dirichlet),
a (rather complicate) function can be
decomposed into the sum of simple
sinusoids (harmonics),
each with its proper frequency, amplitude
and phase

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 harmonics
seismic Earth Sciences Department – University of Pisa
trace f(t) A. Mazzotti – E. Stucchi From: Yilmaz (2001)
 Fourier series

a0 ∞ 2πn
f (t )= + ∑ a n cos(ωn t )+b n sin (ωn t ) with ω n=
2 n=1 T

where T is the period of the function

(in practice it is the time length of our observation)

The coefficients  an and bn are determined through the Cauchy integrals:

T /2 T /2
2 2
a n= ∫ f (t )cos(ω n t )dt b n= ∫ f (t )sin(ωn t )dt
T −T /2 T −T /2

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 Dirichlet’s conditions
1) The function f(t) must be periodic, that is : f(t) = f (t + nT), where T
denotes the period.
2) f(t) must be a continuous function with a finite number of maxima and
minima in one period
3) f(t) must have a finite number of discontinuities in one period, and the
discontinuities cannot be infinite.

4) The function must be integrable over a period

+∞
and the integral ∫ |f (t )|dt must have a finite value.
−∞

In theory we can think of signals that violate these conditions, (e.g. sin (log t)) but
it is not possible to have experimental signals that violate these conditions.

In practice, any physical signal will have a Fourier representation.

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 Notes:

The series converges, that is: an and b n ⇒ 0 for n ⇒ ∞

Given a function f(t) : f (t ) ⇔ an and bn

T T
One cycle only is enough to determine f(t): ( )
− , , (0 , T ), (−T , 0)...
2 2

2πn
 is determined by the period T : ω n=
T
2π
1 is the fundamental angular frequency ω1 =
T
Earth Sciences Department – University of Pisa
A. Mazzotti – E. Stucchi
 Fourier series

Another convenient way to represent the Fourier series is:

a0 ∞
f (t )= + ∑ C n cos(ωn t +ϕ n )
2 n=1

with C n = √ a2n +b2n Amplitude

bn
and ϕ n = arctan (− ) Phase
an

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 harmonics as rotating vectors

2 2
b1 bn
C 1 = √ a +b ϕ1 = arctan (− )
1 1
a1
w3 w2
2 2
b2 b3
C 2 = √ a +b ϕ 2 = arctan (− ) w1 b2 b C3
2 2
a2 C2
C1 1

2 2
b3 a1 a3 a2 an
C 3 = √ a +b ϕ 3 = arctan (− )
3 3
a3

AMPLITUDE PHASE
SPECTRUM SPECTRUM

f  t  = C1 cos ω1t + φ1  + C2 cos ω2t + φ2  + C3 cos ω3t + φ3 

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 amplitude and phase spectra
Cn
2 2
b1 C2
C 1 = √ a +b ϕ1 = arctan (− )
1 1
a1 C3

2 2
b2 C1
C 2 = √ a +b ϕ 2 = arctan (− )
2 2
a2 w1 w2 w3 wn
b3
2
C 3 = √ a +b 2
ϕ 3 = arctan (− )
π .
3 3
a3 f1

..
f3
f2
AMPLITUDE PHASE fn 0
SPECTRUM SPECTRUM w1 w2 w3 wn
-π
f  t  = C1 cos ω1t + φ1  + C2 cos ω2t + φ2  + C3 cos ω3t + φ3 
Earth Sciences Department – University of Pisa
A. Mazzotti – E. Stucchi
 amplitude and phase spectra

1) 12.5 Hz , zero phase harmonic

2) 25 Hz , zero phase harmonic
3) 12.5 Hz , -90° phase harmonic From: Yilmaz (2001)

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 Amplitude and Phase spectra
-0.2

0.0
Time [s]

0.2

0.4
0 16 32 48 60
Frequency [Hz] From: Yilmaz (2001)
Earth Sciences Department – University of Pisa
A. Mazzotti – E. Stucchi
seismic trace in time and its frequency spectrum

AMPLITUDE
SPECTRUM

PHASE
SPECTRUM

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 A typical amplitude spectrum of seismic traces

Frequency (Hz)

Amplitude (dB)

Realized by means of
the Landmark Promax
3D software

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 the dB scale

reference amplitude
amplitude (dB) = - 20 log10
measured amplitude

Example:
Recorded highest amplitude: 400
Take it as the reference.

An amplitude of 400 corresponds to - 0 dB

An amplitude of 200 corresponds to - 6 dB
An amplitude of 40 corresponds to - 20 dB
An amplitude of 4 corresponds to - 40 dB

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 the dB scale
It makes it easier to detect small differences in the amplitudes of the harmonics

linear amplitude scale SAME SPECTRA dB scale

Realized by means of the Landmark Promax 3D software
Earth Sciences Department – University of Pisa
A. Mazzotti – E. Stucchi
 Another way to look at Fourier series: the exponential form
Making use of the Eulero’s equations:


cos ωnt  = e
jωnt
+e
 jωn t
/ 2 sin ωnt  =  j e  jωn t
e
 jωn t
/ 2
the Fourier series becomes:
a0  1 1
j n t
f (t )    an e  e
2 n 1 2
 j n t

- jbn e jnt  e  jnt 
2
  
a0  1 j  n t 1  j n t
   e  an  jbn  + e  an  jbn 
2 n 1 2 2

*
and introducing: Fn  an  jbn  / 2 F n  an  jbn  / 2
where Fn* is the complex conjugate of Fn, the Fourier series of f(t) can be written as:
Earth Sciences Department – University of Pisa
A. Mazzotti – E. Stucchi
 Exponential form of the Fourier series

j n t   j n t
f (t ) F0   Fn e +F e n
n 1
 
j n t   j n t
Being:
F e n =F e n
n  1 n 1
to time from frequency

jωnt INVERSE
the Fourier series can be written as: f  t  =  Fn e FOURIER
TRANSFORM


to frequency from time

T /2
1  jωnt DIRECT
and the Cauchy integrals as: Fn =  f t  e dt FOURIER
T  T /2 TRANSFORM

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 Im {f(t)}
A single 1 Hz harmonic with unitary
modulus in complex exponential notation
positive
time
j 2πf 0 t 1 rotation
f t  = e with f0=1Hz sin(2f0t)
{2f0t}

note the “wrapping” t = 0.0 s cos(2f0t) Re {f(t)}

of the phase t = 0.25 s
at p and –p t = 0.50 s

modulus phase
 1

amplitude
0
-1 -0.5 0.5 1 t (s)

-1
- -1 - 0.5 0.0 0.5 1
time (s)
Earth Sciences Department – University of Pisa
A. Mazzotti – E. Stucchi
 Even and Odd time functions

IF f (t) is an even function, that is f (-t) = f (t), e.g. f(t) = t2, or f(t) = cos2t,
THEN its Fourier series is made of cosines only (real components)

IF f (t) is an odd function, that is f (-t) = -f (t), e.g. f(t) = t3, or f(t) = sin2t,
THEN its Fourier series is made of sine terms only (imaginary components)

f (t) = t 2
f(t) = t 3

t t
Earth Sciences Department – University of Pisa
A. Mazzotti – E. Stucchi
 Some time – frequency pairs

“time” domain “frequency” domain “time” domain “frequency” domain

T

0 0 0 0 1/T
cos wave

0 0 0 0
0

0 0

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
Note: due to the Duality Property of the Fourier Transform, which relates to the fact that the direct and
inverse equations look almost identical except for a factor of 1/T and for a minus sign in the exponential
in the integral, the Fourier transform of the Fourier transform is proportional to the original signal reversed
in time. That is whenever we have a transform pair, there is a dual pair with the time and frequency
variables interchanged.
E.g. a boxcar function in time yields a sinc in frequency and vice-versa

s(t ) S (f )
AT
A


-3/T -2/T 2/T 3/T
-T/2 T/2 t -1/T 1/T f

s(t ) S(f )
A
AT=A/F
 F=1/ T

-3T -2T 2T 3T
-T T t -F/2 F/2 f
Earth Sciences Department – University of Pisa
A. Mazzotti – E. Stucchi
The significance of the Phase Spectrum: Temporal shift Phase rotation
A A
f f
 
 

Earth Sciences Department – University of Pisa From: Yilmaz (2001)

A. Mazzotti – E. Stucchi
The significance of the Phase Spectrum: Temporal shift Phase rotation

Earth Sciences Department – University of Pisa From: Yilmaz (2001)

A. Mazzotti – E. Stucchi
The significance of the Phase Spectrum: Phase shift Shape change
A f A f p/2
 
 

Earth Sciences Department – University of Pisa From: Yilmaz (2001)

A. Mazzotti – E. Stucchi
The significance of the Phase Spectrum: Phase shift Shape change

Earth Sciences Department – University of Pisa From: Yilmaz (2001)

A. Mazzotti – E. Stucchi
 Rotation and Shift of the Phase Spectrum
f f
A A p/2
 
 

Earth Sciences Department – University of Pisa From: Yilmaz (2001)

A. Mazzotti – E. Stucchi
Rotation and Shift of the Phase Spectrum

Phase spectrum
clockwise rotated
and shifted of -/2.
Phase spectrum anti-
clockwise rotated and
shifted of +/2

Earth Sciences Department – University of Pisa From: Yilmaz (2001)

A. Mazzotti – E. Stucchi
 Wavelets and Phase

amplitude
time time

time time
Maximum

Phase (un-wrapped)
phase

time time

Minimum
phase
time time frequency

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
Qualitative relations between time signals and their frequency spectra

Short time signals have a wider frequency band than long wavelets


time frequency


time frequency

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
Shorter wavelets are composed by more harmonics (here all the harmonics
are cosines, that is zero-phase harmonics)

Frequency (Hz)
Earth Sciences Department – University of Pisa From: Yilmaz (2001)
A. Mazzotti – E. Stucchi
 Qualitative relations between time signals and their frequency spectra

Given the same phase

spectrum, the wider is the
amplitude spectrum the shorter
is the time duration of the signal frequency frequency

MINIMUM
PHASE time
time

Given the same amplitude

spectrum, a zero-phase
spectrum yields the shortest
ZERO
possible wavelet. PHASE
time time
Earth Sciences Department – UniversityEarth
of Pisa
Sciences Department – University of Pisa
A. Mazzotti – E. Stucchi A. Mazzotti – E. Stucchi
 Wavelets and Phase

Given the same amplitude spectrum, the shortest possible wavelet among all
the physically plausible wavelets (causal wavelets) is the one with a
minimum phase spectrum.
Only a zero-phase wavelet, which is non physical because anti-causal, is
shorter than a minimum phase wavelet. It can be obtained through digital
filtering (deconvolution).

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 Some other properties of the Fourier Transform (1)

Superposition: the spectrum of the sum of N temporal signals is equal

to the sum of their respective complex spectra.

Temporal shift: a shift in the time domain corresponds to

multiplication by a complex exponential in the frequency domain

If f (t)  F() then f  t a   e j a  F  

If a time signal is shifted by a constant value a, its amplitude spectrum
does not change while its phase spectrum is rotated by a quantity
determined by the time shift.

The amplitude and phase spectra of a function f (t  a) are:

|F()| and   a

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 Properties of the Fourier Transforms (2)

Frequency shift: multiplying a signal in the time domain by a complex

exponential corresponds to a shift in the frequency domain. 

If f (t)  F() then F   0   e j0 t f  t 

j t
The amplitude and phase spectra of a function  
f t e 0

are F   F0 e0  e 0

and   0 
It’s the dual theorem of the previous one (Time shift): a time shift
corresponds to multiplying the frequency spectrum by a complex exponential
and vice-versa, a frequency shift corresponds to multiplying the time signal by
a complex exponential.

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 Properties of the Fourier Transforms (3)
Convolution
If f1 (t)  F1() and f2 (t)  F2() then f1 (t) * f2 (t)  F1()  F2()

Convolution between two time signals corresponds to the multiplication of

their respective complex spectra.
Since the spectra are complex numbers, this corresponds to multiplying
their respective amplitude spectra (moduli) and to summing their respective
phase spectra (arguments). Important property (filtering)

|X(  )| = |F1 ()|  |F2 ()| and x(  ) = f1() + f2()

Cross-correlation
Since cross-correlation between two time signals is equal to their convolution
on condition that one is ordered in a reverse time fashion, Xxy(t) = f1(t) * f2(-t),
then cross-correlation between two time signals corresponds to multiplying
their amplitude spectra and to subtracting their phase spectra:
|Xxy(  )| = |F1 ()|  |F2 ()| and Xxy(  ) = f1()  f2()
Earth Sciences Department – University of Pisa
A. Mazzotti – E. Stucchi
 Properties of the Fourier Transforms (4)
Multiplication
If f1 (t)  F1() and f2 (t)  F2() then f1 (t) x f2 (t)  F1() * F2()

Multiplication between two time signals corresponds to the convolution

between their respective complex spectra.
Important property (sampling)

Derivation in time
df (t )
If f (t)  F() then  j ω F (ω)
dt
Note that differentiating in the time domain has the effect of emphasizing
high frequencies in the Fourier spectrum.
t
1 δ (ω)
Integration in time ∫ f (t ) dt  F (ω)+ F (0)
−∞ jω 4π

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 Convolution and Phase
Recall:
x(t) = w1 (t) * w2 (t)  X () = W1()  W2()
| X () | = |W1() |  |W2() |
F x() = F W1() + F W2()
IF w1(t) and w2(t) are minimum phase wavelets
THEN x(t)   is a minimum phase signal
IF w1(t) is minimum phase and w2(t) is zero phase
THEN x(t)   may  or  may not   be a minimum phase signal
x(t) will be minimum phase IF the zero-phase operator w2(t) does not change
the amplitude spectrum of w1(t).
That is, w2(t) is an all-pass filter and thus x(t) = w1(t)
Instead, IF the zero-phase operator w2(t) modifies the
                 amplitude spectrum of w1(t), THEN x(t) will be mixed phase  
Earth Sciences Department – University of Pisa
A. Mazzotti – E. Stucchi
 Convolution and Phase

A |W1() | F W1()
minimum–phase
signal f


x +
A
1 |W2() | F W2()
zero–phase filter f
reject 
pass 
= =
A F x()
| X () |
mixed–phase output f 

The zero-phase operator w2 does not alter the phase of the input signal w1, but it high-
cut filters its amplitude spectrum. The resulting amplitude and phase spectra do not
maintain the minimum phase property for the new signal x(t). BEWARE when filtering..
Earth Sciences Department – University of Pisa
A. Mazzotti – E. Stucchi
 Continuous (analytical) functions and discrete (digital) functions
The previous description refers to continuous analytical functions.

However, in practice we have to deal with successions of numbers which

represent the sampled values of the measured analogical function (say pressure
variations or displacement velocity variations recorded by a sensor).

Therefore, an important step is the sampling operation that converts a

continuous analogical function into a digital function with a limited number of
samples.
In this case the Fourier series are developed through numerical formulations
which go under the name of
Discrete Fourier Transform (DFT)
and of
Inverse Discrete Fourier Transform (IDFT)

Most of the properties (superposition, symmetry, shifting, derivation, integration,

convolution and correlation) of the analytical transform hold true for the discrete transform.

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi
 Discrete Fourier Transforms

For a sequence of N samples, taken with a constant sampling interval, and

with t = 1, 2, ... N,
(note: t is no more a continuous variable: now it is a discrete variable)

N
1 − jω t
F ( ωn )= ∑ f (t )e n DFT Discrete Fourier Transform
N t=1
N
j ωn t
f (t )=∑ F ( ωn )e IDFT Inverse Discrete Fourier Transform
t =1
n 2
with n  in radians
N
more on this topic after we have discussed the sampling operation

Earth Sciences Department – University of Pisa

A. Mazzotti – E. Stucchi