Scribd
Upload
97 views

9 Signal Processing

Fourier analysis decomposes signals and waveforms into their constituent frequencies. This allows any periodic function to be expressed as the sum of sines and cosines of different frequencies. The Fourier transform extends this analysis to non-periodic functions by representing them as a sum of infinite sinusoids. Convolution in the time domain equals multiplication in the frequency domain, providing the basis for modern digital signal processing using the fast Fourier transform (FFT) algorithm.

Uploaded by

bidyut_iitkgp
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
97 views

9 Signal Processing

Fourier analysis decomposes signals and waveforms into their constituent frequencies. This allows any periodic function to be expressed as the sum of sines and cosines of different frequencies. The Fourier transform extends this analysis to non-periodic functions by representing them as a sum of infinite sinusoids. Convolution in the time domain equals multiplication in the frequency domain, providing the basis for modern digital signal processing using the fast Fourier transform (FFT) algorithm.

Uploaded by

bidyut_iitkgp
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 22

Theoretical Seismology

Lecture 9 Signal Processing

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Fourier analysis
Fourier studied at the cole Normale in Paris, taught by Lagrange, who Fourier described as the first among European men of science, Laplace, who Fourier rated less highly, and by Monge. He was appointed to the cole Centrale des Travaux Publiques, the school being under the direction of Carnot and Monge, which was soon to be renamed cole Polytechnique. In 1807 he astounded many of his contemporary mathematicians and scientists by asserting that an arbitrary function could be expressed as a linear combination of sine and cosine Jean Baptiste Joseph Fourier functions, called Fourier trignometric Born: 21 March 1768 in Auxerre, Bourgogne, France series, are applied to the analysis of Died: 16 May 1830 in Paris, France periodic phenomena including vibrations and wave motion.
GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Response Spectrum

Arup

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

The Seismometer
Basic principle
a a mass attached to a moveable frame when frame is shaken by seismic waves the inertia of the mass causes its motion to lag behind relative motion recorded on rotating drum, on magnetic tape or digitally

Mass is damped to prevent continued oscillation


a This limits the frequency response of the seismometer

Relative motion amplified up to 100s of thousands of times

Schematic of a horizontal motion mechanical seismometer

(Simon Day)
GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Vibrations: Simple Harmonic Motion


Simplest vibrating system: x

2u ( x ) 2 + u ( x) = 0 2 t
Displacement u

is the angular frequency, f = / There are two solutions: u(x)= A sin (t) and u(x) = B cos (t) A and B are amplitude We can check this by substituting:

u ( x) = A cos( t ) t 2u ( x ) 2 2 = A sin( t ) = u ( x) 2 t

u ( x) = B sin( t ) t 2u ( x) 2 2 = B sin( t ) = u ( x) 2 t

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Vibrations: Simple Harmonic Motion 2


u(x)= A sin (t) and u(x) = B cos (t) These are both solutions and the only difference is the point on the cycle at which motion starts. If we shift the cos plot by a phase shift of we get the sin plot. So this means another solution is u(x) = C cos (t + ) Or the general solution is u(x) = A sin (t) + B cos (t) Another solution is u(x) = C exp[-i t + ] Differentiate this to check that it is a solution

(e i t = cos t + i sin t )
i is sqrt of -1

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Fourier decomposition
Higher orders of r Means higher and higher frequencies

f (t ) =

a0 + {ar cos(rt / T ) + br sin (rt / T )} 2 r =1 harmonic no. d.c. offset


Can make any function such as this square wave (aka box car function) out of the superposition of sine and cosine waves

sum of sin & cos

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Time domain
Any f(t) can be described by a Fourier series of sin and cos terms of different frequencies: Fourier series T < t T

a0 f (t ) = + {ar cos(rt / T ) + br sin (rt / T )} 2 r =1 1 T 1 T a r = f (t ) cos(rt / T )dt br = f (t ) sin (rt / T ) T T T T 1 T a0 = f (t )dt T T


For even fns br = 0 For odd fns ar = 0 [ f(t) = f(-t) ] [ f(t) = -f(-t) ]

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Fourier decomposition
Square wave or box car function f(t)
1

-T/2

T/2

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Frequency domain
Amplitude spectrum ar cosine harmonics br sin harmonics

r Phase r = tan-1 (br / ar) r

=r/T r harmonic
no.

Amplitude Ar = (ar2 + br2) Ar

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Stage post
We have seen that vibrations and waves contain a whole range of frequencies They can be represented as a sum of sines and cosines of different frequencies a Fourier series But given a wave or impulse, how do we find out what frequencies it contains:

We use the Fourier transform


The study of frequency content of signals (seismic, electronic, light, radiowaves etc., is called spectral analysis Only a bit can be done analytically, the rest relies on computers

Matlab tutorial

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Fourier Transform
Extend interest from T < t T to - < t The Fourier Transform of a function f(t) is

Inverse Fourier Transform is

F ( ) = e

i t

f (t ) dt

f (t ) = e F ( ) d
i t

Reminder

ei t = cos t + i sin t e i t = cos t i sin t

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Fourier Transform of a box-car function


f(t) 1 f(t) = 1 T < t T f(t) = 0 elsewhere Duration = 2T t

-T

F ( ) = e

i t

f (t ) dt = e
T

i t

dt

e i t e i T e i T = = i i
GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Fourier Transform of a box-car function


After some algrebraic manipulation we can find

F ( ) =

2 sin T

sinc function 1/x decay

Phase and amplitude


In general F() contain Real and Imaginary components even if f(t) is real Im A Re A2 = Re2 + Im2 = tan-1 (Re/Im)

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Convolution
The measured response of the seismometer is the convolution of the earthquake source-time response, the Earth response u(t) and the response of the seismometer itself z(t). The convolution of 2 functions f(t) and g(t) is defined mathematically as:

h(t ) = f (t ) g (t ) = f ( x) g (t x) dx

* Denotes convolution x is a dummy variable of integration


GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Convolution 2
The Fourier Transform is:

H ( ) = e i t h(t ) dt

f ( x) g (t x) e i t dt dx

= f ( x) g (t x) e i t dt dx

Introduce a factor 1 = eix e-ix

H ( ) =

f ( x) e

i x

g (t x)e i t e i x dt dx

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Convolution 3
Let u = t x du = dt

f ( x )e

i x

g (u ) e i u du dx

f ( x )e

i x

dx g (u ) e i u du

Since u and x are now independent

H ( ) = F ( ) G ( )
Capitals are Fourier Transforms

Now you dont need to remember this proof, but you need remember:
GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Convolution 4
So we have shown that convolution in the time domain

h (t ) = f (t ) g (t )
has become a multiplication in the frequency domain

H ( ) = F ( ) G ( )
Convolution in the Time domain = Multiplication in the Frequency domain

This is the basis for all modern signal processing


GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Fast Fourier Transform


Computer analysis Replace the infinite signal f(t) by finite samples of the signal f(nTS) Replace the integral of the Fourier Transform with the sum of the discrete Fourier transform f(t) f(nTS) Approximation of a signal by sampling: note if you dont have enough samples you miss key features t 0 t = T/N = TS T = NTS N samples The computation process is called the fast Fourier transform

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Comparison of Fourier techniques


Fourier series Fourier transform f(t) continuous F() discrete f(t) continuous F() continuous Analysis of signals Frequency analysis of signals Analysis of sampled signals Algorithm to compute DFT

Discrete Fourier f(t) discrete transform (DFT) F() discrete Fast Fourier transform (FFT) f(t) discrete F() discrete

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Comparison of Fourier techniques


t Periodic function Fourier series f(t) Discrete Spectrum f()

t Continuous f(t) Fourier transform Continuous Spectrum

Discrete periodic

t Discrete f(t) Discrete Fourier transform Spectrum DFT

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

Point of Fast Fourier Transform


Fast Fourier transform introduced by seismologists Rapidly spread to oil industry and more recently to all signal processing, including music Direct calculation of DFT required N2 multiplications Calculation of FFT requires N log2 N multiplications For small sample, e.g. N=4096, DFT requires more than 16 million calculations, while the FFT requires less than 50,000 So this is the basis of modern digital signal processing

GNH7/GG09/GEOL4002 EARTHQUAKE SEISMOLOGY AND EARTHQUAKE HAZARD

You might also like