Lagg - Specttrral Anallysis and Time Seriies

Spectral Analysis and Time Series
Andreas Lagg
Part I: fundamentals Part II: Fourier series Part III: Wavelets

on time series
classification definition why wavelet

prob. density func. method transforms?
autocorrelation properties fundamentals:
FT, STFT and
power spectral density convolution
resolution problems
crosscorrelation correlations
multiresolution
applications leakage / windowing analysis: CWT
preprocessing irregular grid DWT
sampling noise removal
trend removal
Exercises
A. Lagg – Spectral Analysis
Basic description of physical data
deterministic: described by explicit mathematical relation
x t =X cos
 k
t
t
n
non deterministic: no way to predict an exact value at a future instant of time
o
n

d
e
t
e
r
m A. Lagg – Spectral Analysis
Classifications of deterministic data
Deterministic
Periodic Nonperiodic
Sinusoidal Complex Periodic Almost periodic Transient
Sinusoidal data
x t =X sin 2  f 0 t

T =1/ f 0
time history frequency spectrogram
Complex periodic data
x t =x t±nT  n=1,2,3,...

a0
x t  =
2
 ∑ an cos 2  n f 1 t  b n sin 2  n f 1 t 
(T = fundamental period)
Almost periodic data
x t =X 1 sin 2 t1   X 2 sin 3 t2   X 3 sin   50 t3 
no highest common divisor > infinitely long period T
Transient nonperiodic data
all nonperiodic data other than almost periodic data
{
−at
Ae t≥0
x t =
0 t0
{
−at
Ae cos bt t≥0
x t =
0 t0
x t =
{ A
0
c≥t≥0
ct0
Classification of random data
Random Data
Stationary Nonstationary
Ergodic Nonergodic Special classifications

of nonstationarity
stationary / non stationary
collection of sample functions = ensemble
data can be (hypothetically) described by
computing ensemble averages (averaging over
multiple measurements / sample functions)
mean value (first moment):
N
1
 x t 1 = lim ∑ x k t 1 
N  ∞ N k=1
autocorrelation function (joint moment):
N
1
R x t 1, t 1= lim ∑ x k t 1  x k t 1
N  ∞ N k=1
stationary:  x t 1 = x , R x t 1, t 1=R x

weakly stationary:  x t 1 = x , R x t 1, t 1=R x 
ergodic / non ergodic
Ergodic random process:
properties of a stationary random process
described by computing averages over only one
single sample function in the ensemble
mean value of kth sample function:
T
1
 x k =lim
T ∞
∫
T 0
x k t  dt
autocorrelation function (joint moment):
T
1
R x  , k =lim ∫ x k t  x k t dt
T ∞ T 0
ergodic:  x k = x , R x  , k =R x 
Basic descriptive properties of random data
mean square values
propability density function
autocorrelation functions
power spectral density functions
(from now on: assume random data to be stationary and ergodic)
Mean square values (mean values and variances)
T
1
describes general intensity of random data:  2x = lim ∫ x 2 t  dt
T ∞ T 0
rout mean square value:  x
rms
=  2
x
often convenient: T
1
static component described by mean value:  x = lim
T ∞ T
∫ x t  dt
0
T
1
dynamic component described by variance:  2x = lim
T ∞ T
∫ [ x t − 2
x dt
]
0
=  2x − 2
standard deviation:  x =  2
x
x
Probability density functions
describes the probability that the data will assume a value within some defined
range at any instant of time k
Tx
Prob[ xx t ≤x x ] = lim , T x =∑  t i
T ∞ T i=1
 x Prob[ xx t ≤x x ] ≈ p x x
for small :
probability density function
p x = lim
x0
Prob[ xx t ≤x x ]
x
= lim
1
lim
Tx
 x 0  x T ∞ T
[ ]
probability distribution function
P  x = Prob[x t≤x]
x
= ∫ p d 
−∞
Illustration: probability density function
sample time histories:
sine wave (a)
sine wave + random noise
narrowband random noise
wideband random noise
all 4 cases: mean value  x =0
Illustration: probability density function
probability density function
Autocorrelation functions
describes the general dependence of the data values at one time on the
values at another time.
T
1
R x  = lim ∫ x t  x t dt
T ∞ T 0
x =  R x ∞ 2
 x = R x 0 (not for special cases like sine waves)
Illustration: ACF
autocorrelation functions (autocorrelogram)
Illustrations
autocorrelation function of a rectangular pulse
x t  x t−
autocorrelation function
Power spectral density functions
(also called autospectral density functions)
describe the general frequency composition of the data in terms
of the spectral density of its mean square value
mean square value in frequency range :
 f , f  f 
T
1
 2x  f ,  f  = lim
T ∞
∫
T 0 
x t , f ,  f 
2
dt
portion of x(t) in (f,f+∆f)
2
definition of power spectral density function:  x  f ,  f  ≈ G x  f  f
 2x  f ,  f 
[ ]
T
1 1
G x  f  = lim = lim lim ∫ x t , f ,  f 2 dt
 f 0 f  f 0  f T ∞ T 0
important property: spectral density function is related to the
autocorrelation function by a Fourier transform:
∞ ∞
G x  f  = 2 ∫ R x e d  = 4 ∫ R x  cos 2  f  d 
−i 2  f 
−∞ 0
Illustration: PSD
power spectral density functions
Dirac delta function at f=f0 sine wave
sine wave
+ random noise
narrowband
noise
“white” noise:
spectrum is uniform over broadband
all frequencies noise
Joint properties of random data
until now: described properties of an
individual random process
Joint probability density functions
joint properties in the amplitude domain
Crosscorrelation functions
joint properties in the time domain
joint probability measurement
Crossspectral density functions
joint properties in the frequency domain
Crosscorrelation function
describes the general dependence of
one data set to another
T
1
R xy  = lim ∫ x t  y t dt
T ∞ T 0
similar to autocorrelation function
R xy =0 functions are
uncorrelated
crosscorrelation measurement
typical crosscorrelation plot (crosscorrelogram):
sharp peaks indicate the existence of a correlation
between x(t) and y(t) for specific time displacements
Applications
Measurement of time delays
2 signals:
different offset
different S/N
time delay 4s
often used:
'discrete' cross correlation coefficient
lag =l, for l >=0:
N −l
∑  x k − x  y kl −y 
k=1
R xy l =

N N
∑  x k − x 2
∑  y k −y 2
k=1 k=1
Applications
Detection and recovery from
signals in noise
3 signals:
noise free replica of the signal
(e.g. model)
2 noisy signals
cross correlation can be used to
determine if theoretical signal is
present in data
Preprocessing Operations
sampling considerations
trend removal
filtering methods
sampling
cutoff frequency (=Nyquist
frequency or folding frequency)
1
fc =
2h
Trend removal
often desirable before performing a
spectral analysis
Leastsquare method:
time series: ut 
K
desired fit
(e.g. polynomial):
∑
u = bk nh
k
n=1,2,... , N
k=0
N
LsqFit: minimize Qb=∑ u n −u n 

2
n=1
N
set partial ∂Q
=
derivatives to 0: ∂ b n=1 ∑
2u n − u
 n [−nh
l
]
l
K N N
K+1 equations: ∑ bk ∑ nhkl = ∑ un nhl

k=0 n=1 n=1
Digital filtering
end of part I ...
Andreas Lagg

on time series

FT, STFT and
resolution problems
multiresolution
trend removal
Exercises
Fourier Series and Fast Fourier Transforms
Standard Fourier series procedure:
if a transformed sample record x(t) is periodic with a period Tp (fundamental
frequency f1=1/Tp), then x(t) can be represented by the Fourier series:
a0 ∞
x t  = ∑ a q cos 2  q f 1 t  bq sin 2  q f 1 t 
2 q=1
T
2
where aq =
T
∫ x t  cos 2  q f 1 t dt q=0,1,2,. ..
0
T
2
bq =
T
∫ x t sin 2  q f 1 t dt q=1,2,3,...
0
Fourier series procedure method
sample record of finite length, equally spaced sampled:
x n = x nh n=1,2,. .. , N
Fourier series passing through these N data values:
   
N /2 N /2−1
2qt 2qt
x t  = A0  ∑ Aq cos  ∑ Bq sin
q=1 Tp q=1 Tp
Fill in particular points: t=nh , n=1,2,... , N , T p =Nh , x n =x nh = ...
N N
1 1
coefficients Aq and Bq: A0 = ∑
N n=1
x n = x A N /2 = ∑
N n=1
x n cos n 
N
2 2qn N
Aq = ∑
N n=1
x n cos
N
q=1,2,... ,
2
−1
N
2 2qn N
Bq = ∑
N n=1
x n sin
N
q=1,2,... ,
2
−1
inefficient & slow => Fast Fourier Trafos developed
Fourier Transforms Properties
Linearity DFT
{x n } ⇔ {X k }
DFT
{y n } ⇔ {Y k }
DFT
a {x n }b {y n } ⇔ a {X k }b {Y k }
Symmetry {X k } = ∗
{X −k }
ℜ {X k } is even ℑ {X k } is odd
Circular time shift DFT
−i k n 0
{x n−n } 0
⇔ {e Xk}
DFT
i k0 n
{e yn } ⇔ {Y k−k } 0
Using FFT for Convolution
∞
r*s ≡ ∫
−∞
r  st− d 
original data
Convolution Theorem:
FT
r * s ⇔ R f  S  f 
response function
Fourier transform of the convolution
is product of the individual Fourier
transforms
convolved data
discrete case:
N /2
(note how the response function for negative times is wrapped
r * s j ≡ ∑ s j−k r k around and stored at the extreme right end of the array)
k=−N /21
Convolution Theorem: constraints:
N /2
FT duration of r and s are not the same
∑ s j−k r k ⇔ R n S n
signal is not periodic
k=−N /21
Treatment of end effects by zero padding
constraint 1: simply expand response function to length N by padding it with zeros
constraint 2: extend data at one end with a number of zeros equal to the max.
positive / negative duration of r (whichever is larger)
FFT for Convolution
1. zeropad data
2. zeropad response function
(> data and response function have N elements)
3. calculate FFT of data and response function
4. multiply FFT of data with FFT of response function
5. calculate inverse FFT for this product
Deconvolution
> undo smearing caused by a response function
use steps (13), and then:
4. divide FFT of convolved data with FFT of response
function
5. calculate inverse FFT for this product
Correlation / Autocorrelation with FFT
definition of correlation / autocorrelation see first lecture
∞
Corr g , h = g * h = ∫
−∞
gt h d 
Correlation Theorem:
FT *
Corr g , h ⇔ G  f  H  f 
AutoCorrelation:
FT 2
Corr g , g ⇔ ∣G  f ∣
discrete correlation theorem:
N −1
Corr g , h j ≡ ∑ g jk hk
k=0
FT
⇔ G k H *k
Fourier Transform problems
spectral leakage
T =n T p T ≠n T p
reducing leakage by windowing (1)
Applying windowing (apodizing) function to data record:
x t  = x t  wt  (original data record x windowing function)
x n = x n w n
No Window: Bartlett Window: Hamming Window:
∣t−T /2∣
wt =1− wt =0.540.46∗cos t 
T /2
Blackman Window: Hann Window: Gaussian Window:
wt  = 0.420.5∗cos t  t
wt =cos2  
0.08 cos2  t  2 wt =exp −0.5a t 2 
reducing leakage by windowing (2)
without windowing with Gaussian windowing
No constant sampling frequency
Fourier transformation requires constant sampling (data points at
equal distances)
> not the case for most physical data
Solution: Interpolation
linear:
inear interpolation between yk and yk+1
IDL> idata=interpol(data,t,t_reg)
quadratic: important:
quadratic interpolation using yk1, yk and yk+1
IDL> idata=interpol(data,t,t_reg,/quadratic) interpolation changes
sampling rate!
leastsquare quadratic
leastsquare quadratic fit using yk1, yk, yk+1 and yk+2
> careful choice of
IDL> idata=interpol(data,t,t_reg,/lsq) new (regular) time grid
spline necessary!
IDL> idata=interpol(data,t,t_reg,/spline)
IDL> idata=spline(t,data,t_reg[,tension])
Fourier Transorm on irregular gridded data Interpolation
original data: sine wave + noise
FT of original data
irregular sampling of data (measurement)
interpolation: linear, lsq, spline, quadratic
'resampling'
FT of interpolated data
Noise removal
Frequency threshold (lowpass)
make FT of data
set high frequencies to 0
transorm back to time
domain
Noise removal
signal threshold for
weak frequencies (dBthreshold)
make FT of data
set frequencies with amplitudes
below a given threshold to 0
transorm back to time domain
Optimal Filtering with FFT
normal situation with measured data: ∞
underlying, uncorrupted signal u(t) st  = ∫
−∞
r t−u d 
+ response function of measurement r(t) c t  = st nt 
= smeared signal s(t)
+ noise n(t)
= smeared, noisy signal c(t)
estimate true signal u(t) with:
C  f  f 
U  f  =
R f 
 f  , t  = optimal filter
(Wiener filter)
Calculation of optimal filter
reconstructed signal and uncorrupted signal should be close in leastsquare sense:
> minimize ∞ ∞
∫ ∣u t −ut ∣ dt =
−∞
2
∫ ∣U  f −U  f ∣ df
−∞
2
∣ ∣ =0
2
∂ [S  f N  f ] f  S  f 
⇒ −
∂  f  R f  R f 
2
∣S  f ∣
⇒  f  = 2 2
∣S  f ∣ ∣N  f ∣
additional information:
power spectral density can often
be used to disentangle noise
function N(f) from smeared
signal S(f)
Using FFT for Power Spectrum Estimation
discrete Fourier transform of c(t)
> Fourier coefficients: N −1
C k= ∑ c j e
2i j k / N
k=0,... , N −1
j=0
> periodogram estimate of
power spectrum:
Periodogram
1 2
2 ∣ 0∣
P 0 = P  f 0 = C
N
1
P f k =
N
[
2 ∣ k∣
C
2
∣ N −k∣
C
2
]
1 2
P  f c = P  f N / 2 = 2 ∣C N / 2∣
N
end of FT
Andreas Lagg

on time series

FT, STFT and
resolution problems
multiresolution
trend removal
Exercises
Introduction to Wavelets
why wavelet transforms?
fundamentals: FT, short term FT and resolution
problems
multiresolution analysis:
continous wavelet transform
multiresolution analysis:
discrete wavelet transform
Fourier: lost time information
6 Hz, 4 Hz, 2 Hz, 1 Hz 6 Hz + 4 Hz + 2 Hz + 1 Hz
Solution: Short Time Fourier Transform
(STFT)
perform FT on 'windowed' function:
example: rectangular window
move window in small steps over data
perform FT for every time step
STFT  f , t '  = ∫ [ x t t−t ' ]e −i 2  f t

dt
t
Short Time Fourier Transform
STFT
STFTspectrogram shows both time and
frequency information!
Short Time Fourier Transform: Problem
narrow window function > good time resolution
wide window function > good frequency resolution
Gaussfunctions as
windows
Solution: Wavelet Transformation
time vs. frequency resolution is intrinsic problem (Heisenberg Uncertainty Principle)
approach: analyze the signal at different frequencies with different resolutions
> multiresolution analysis (MRA)
Continuous Wavelet Transform
similar to STFT: but:
signal is multiplied with a the FT of the windowed signals are
function (the wavelet) not taken
transform is calculated (no negative frequencies)
separately for different segments The width of the window is changed
of the time domain as the transform is computed for
every single spectral component
Continuous Wavelet Transform
CWT 
x

=  , s =
x
1
∫
∣s∣ t
x t  
* t−
s
dt
 ... translation parameter , s ... scale parameter

t  ... mother wavelet  = small wave
mother wavelet:
finite length (compactly
supported) >'let'
oscillatory >'wave'
scale parameter s replaces frequency in STFT
functions for different
regions are derived from
this function > 'mother'
The Scale
similar to scales used in maps:
high scale = non detailed global view (of the signal)
low scale = detailed view
in practical applications:
low scales (= high frequencies)
appear usually as short bursts or
spikes
high scales (= low frequencies) last
for entire signal
scaling dilates (stretches out) or
compresses a signal:
s > 1 > dilation
s < 1 > compression
Computation of the CWT
signal to be analyzed: x(t), mother wavelet: Morlet or Mexican Hat
start with scale s=1 (lowest scale,
highest frequency)
> most compressed wavelet
shift wavelet in time from t0 to t1
increase s by small value
shift dilated wavelet from t0 to t1
repeat steps for all scales
CWT Example
signal x(t)
axes of CWT: translation and
scale (not time and frequency)
translation > time
scale > 1/frequency
Time and Frequency Resolution
every box corresponds to a value of the wavelet
transform in the time frequency plane
all boxes have constant area
Δf Δt = const.
low frequencies: high resolution
in f, low time resolution
high frequencies: good time
resolution
STFT: time and frequency
resolution is constant (all boxes
are the same)
Wavelets: Mathematical Approach
 
WLtransform: 1 * t−
CWT x =x  , s = ∫ x t  dt
∣s∣ t s
Mexican Hat wavelet: Morlet wavelet:
 
2
−t 2 t2
1 t −
t  = e 2 2
−1 t  = ei a t e 2
s 3
 2
 
inverse WLtransform: 1 1 t−
x t  = 2 ∫∫  x  , s 2 

d  ds
c s  s s
admissibility ∣
 ∞
∣
2 1/ 2 FT
condition: c={2  ∫ d } ∞ with 
 ⇔ t 
−∞ ∣∣
Discretization of CWT: Wavelet Series
> sampling the time – frequency (or scale) plane
advantage:
sampling high for high frequencies (low scales) s1
N2 = N1
scale s1 and rate N1 s2
sampling rate can be decreased for low f2
frequencies (high scales) N2 = N1
f1
scale s2 and rate N2
continuous wavelet discrete wavelet
1
 , s = 
s  
t−
s
  j , k t  = s−0 j /2 s−0 j t−k 0  ,  j , k orthonormal
x j , k = ∫ x t *j , k t  dt WLtransformation

t
reconstruction of
x t  = c ∑ ∑ x j , k  j , k t 
j k
signal
Discrete Wavelet Transform
(DWT)
discretized continuous wavelet transform is only a sampled version of the CWT
The discrete wavelet transform (DWT) has significant advantages for
implementation in computers.
excellent tutorial:
http://users.rowan.edu/~polikar/WAVELETS/WTtutorial.html
IDLWavelet Tools:
IDL> wv_applet
Wavelet expert at MPS:
Rajat Thomas
end of Wavelets
Exercises
Part I: Fourier Analysis
(Andreas Lagg)
Instructions:
http://www.linmpi.mpg.de/~lagg
Part II: Wavelets
(Rajat Thomas)
Seminar room
Time: 15:00
Exercise: Galileo magnetic field
data set from Galileo magnetometer Tips:
(synthesized) restore,'gll_data.sav'
file: gll_data.sav, contains: use IDLFFT
total magnetic field remember basic plasma
radial distance physics formula for the
time in seconds ion cyclotron wave:
qB gyro
gyro= , f gyro=
m 2
your tasks: Background:
If the density of ions is high enough they will excite ion cyclotron waves
Which ions are present? during gyration around the magnetic field lines. This gyration frequency
only depends on mass per charge and on the magnitude of the magnetic
Is the time resolution of
field.
the magnetometer
In a lowbeta plasma the magnetic field dominates over plasma effects.
sufficient to detect The magnetic field shows only very little influence from the plasma and
electrons or protons? can be considered as a magnetic dipole.
http://www.sciencemag.org/cgi/content/full/274/5286/396
Literature
Random Data: Analysis and Measurement Procedures
Bendat and Piersol, Wiley Interscience, 1971
The Analysis of Time Series: An Introduction
Chris Chatfield, Chapman and Hall / CRC, 2004
Time Series Analysis and Its Applications
Shumway and Stoffer, Springer, 2000
Numerical Recipies in C
Cambridge University Press, 19881992
http://www.nr.com/
The Wavelet Tutorial
Robi Polikar, 2001
http://users.rowan.edu/~polikar/WAVELETS/WTtutorial.html

Lagg - Specttrral Anallysis and Time Seriies

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Lagg - Specttrral Anallysis and Time Seriies

Hochgeladen von

Copyright:

Verfügbare Formate

Spectral Analysis and Time Series

Part I: fundamentals Part II: Fourier series Part III: Wavelets

classification definition why wavelet

Sinusoidal Complex Periodic Almost periodic Transient

x t =X sin 2  f 0 t

x t =x t±nT  n=1,2,3,...

x t =X 1 sin 2 t1   X 2 sin 3 t2   X 3 sin   50 t3 

Ergodic Nonergodic Special classifications

stationary:  x t 1 = x , R x t 1, t 1=R x

ergodic:  x k = x , R x  , k =R x 

LsqFit: minimize Qb=∑ u n −u n 

K+1 equations: ∑ bk ∑ nhkl = ∑ un nhl

Part I: fundamentals Part II: Fourier series Part III: Wavelets

classification definition why wavelet

No Window: Bartlett Window: Hamming Window:

Blackman Window: Hann Window: Gaussian Window:

Part I: fundamentals Part II: Fourier series Part III: Wavelets

classification definition why wavelet

STFT  f , t '  = ∫ [ x t t−t ' ]e −i 2  f t

 ... translation parameter , s ... scale parameter

x j , k = ∫ x t *j , k t  dt WLtransformation

Das könnte Ihnen auch gefallen

Lagg - Specttrral Anallysis and Time Seriies

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Lagg - Specttrral Anallysis and Time Seriies

Hochgeladen von

Copyright:

Verfügbare Formate

Spectral Analysis and Time Series

Part I: fundamentals Part II: Fourier series Part III: Wavelets

classification definition why wavelet

Sinusoidal Complex Periodic Almost periodic Transient

x t =X sin 2  f 0 t

x t =x t±nT  n=1,2,3,...

x t =X 1 sin 2 t1   X 2 sin 3 t2   X 3 sin   50 t3 

Ergodic Nonergodic Special classifications

stationary:  x t 1 = x , R x t 1, t 1=R x

ergodic:  x k = x , R x  , k =R x 

Lsq­Fit: minimize Qb=∑ u n −u n 

K+1 equations: ∑ bk ∑ nhkl = ∑ un nhl

Part I: fundamentals Part II: Fourier series Part III: Wavelets

classification definition why wavelet

No Window: Bartlett Window: Hamming Window:

Blackman Window: Hann Window: Gaussian Window:

Part I: fundamentals Part II: Fourier series Part III: Wavelets

classification definition why wavelet

STFT  f , t '  = ∫ [ x t t−t ' ]e −i 2  f t

 ... translation parameter , s ... scale parameter

x j , k = ∫ x t *j , k t  dt WL­transformation

Das könnte Ihnen auch gefallen

LsqFit: minimize Qb=∑ u n −u n 

x j , k = ∫ x t *j , k t  dt WLtransformation