page 1 of 48 one, two or many frequencies:synchrosqueezing sierra - lyon, 13 th march 2013 steve...
TRANSCRIPT
![Page 1: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/1.jpg)
Page 1 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
One, Two or Many Frequencies:
Synchrosqueezing and Multicomponent Signal Analysis
Steve McLaughlin
Heriot-Watt, School of Engineering & Physical Sciences
with thanks to
Patrick Flandrin, Mike Davies, Yannis Kopsinis, Sylvain Meignen, Thomas Oberlin
![Page 2: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/2.jpg)
Page 2 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
![Page 3: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/3.jpg)
Page 3 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
![Page 4: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/4.jpg)
Page 4 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Structure
• Introduction• Signals, Time and Frequency• Empirical Mode Decomposition• SynchroSqueezing
![Page 5: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/5.jpg)
Page 5 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
What are signals?
A signal is a time (or space) varying quantity that can carry information. The concept is broad, and hard to define precisely.
(Wikipedia)
![Page 6: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/6.jpg)
Page 6 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
What are signals made of?
The Frequency viewpoint (Fourier):
Signals can be built from the sum of harmonic functions (sine waves)
Joseph Fourier
k
kk tctx )()(
signal
Fourier coefficient
s
Harmonic functions
![Page 7: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/7.jpg)
Page 7 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
What is a Signal?
W. Tecumseh Fitch, “Calls out of chaos: the adaptive significance of nonlinear phenomena in mammalian vocal production”, Animal Behaviour, 2002
![Page 8: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/8.jpg)
Page 8 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
What is a Signal Frequency modulation (FM) implies the instantaneous frequency varies. This contrasts
with amplitude modulation, in which it is the amplitude which is varied while its frequency remains constant. Many signals can be modeled as a sum of AM and FM components.
0 1000 2000 3000 4000 5000 6000-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
![Page 9: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/9.jpg)
Page 9 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Empirical Mode DecompositionBasic idea is very simple:
signal = fast oscillation + slow oscillation
(& iteration)
- Separation “fast vs slow’ data driven
- Local analysis based on neighbouring extrema
- Oscillation rather than frequency
![Page 10: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/10.jpg)
Page 10 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Empirical Mode Decomposition
• Identify local maxima and local minima• Deduce an upper envelope and a lower
envelope by interpolation (cubic splines)– subtract the mean envelope from the signal– iterate until “mean envelope = 0" (sifting)– subtract the obtained mode from the signal– iterate on the residual
![Page 11: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/11.jpg)
Page 11 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Empirical Mode Decomposition: Sifting
![Page 12: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/12.jpg)
Page 12 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
EMD - Summary
• Locality — The method operates at the scale of one oscillation• Adaptive — The decomposition is fully data-driven • Multi-resolution — The iterative process explores sequentially the
“natural” scales of a signal• Oscillations of any type — No assumption on the nature of
oscillations (e.g., harmonic) 1 nonlinear oscillation = 1 mode⇒• No analytic definition — The decomposition is only defined as the
output of an algorithm analysis and evaluation ?⇒
![Page 13: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/13.jpg)
Page 13 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
What is frequency?Frequency is the number of occurrences of a repeating event per unit time. It is also referred to as temporal frequency. The period is the duration of one cycle in a repeating event, so the period is the reciprocal of the frequency.
(Wikipedia)
![Page 14: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/14.jpg)
Page 14 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Fatima Miranda
Macaque
Amolops tormotus
What is frequency?
![Page 15: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/15.jpg)
Page 15 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Hilbert Spectrum
Can we produce TF-representation plots based on the IF, IA estimates?
Hilbert Spectrum: IA and IF are combined in order to form a spectrogram-like plot, H(t,f ).
Specifically, for any time instant t, H(t,f )=α(t) if IF(t) equals to f and zero at the rest of the frequencies.
In practice, a Time – Frequency grid is adopted in order to construct H(t,f ).
![Page 16: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/16.jpg)
Page 16 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
On Instantaneous Frequency(Monocomponent signals)
In the case of monocomponent signals, IF coincides with the actual frequency of the signal.
![Page 17: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/17.jpg)
Page 17 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Instantaneous Frequency Multicomponent signals
IF is a non-symmetric oscillation exhibiting spikes that point toward the component with the larger amplitude.
The strength of the IF oscillations (that is the height of the spikes) depends on the relative amplitudes of the components.
Smoothing, , forces IF to “lock” at the
component with the higher amplitude
![Page 18: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/18.jpg)
Page 18 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Instantaneous Frequency Multicomponent signals
The further apart in frequency the components are, the larger the
amplitude of oscillations become. The frequency of oscillations is relative
to the frequency difference.
![Page 19: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/19.jpg)
Page 19 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Beating• When a signal is composed of two components with close
instantaneous frequencies, EMD exhibits a beating phenomenon [Rilling and Flandrin (2008)].
• More precisely, if a signal is composed of two harmonics, i.e., f(t) = cos(2πt)+a cos(2πξ0t), Rillingand Flandrin [2008] show an interesting zone in the a vs. ξ plane (amplitude vs. frequency plane) where EMD misidentifies the sum of two components as only a single component; the precise shape of this zone depends on the value ξ0.
• They also quantified this phenomenon and termed it beating, because EMD “identified” the two harmonics as a single oscillating (i.e., beating) signal.
![Page 20: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/20.jpg)
Page 20 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Time Frequency Reassignment and SSQ• Time-Frequency Reassignment originates from a study of the STFT,
which smears the energy of the superimposed Instantaneous Frequencies around their center frequencies in the spectrogram. TFR analysis “reassigns” these energies to sharpen the spectrogram
• Synchrosqueezing is a special case of reassignment methods [Auger and Flandrin (1995); Chassande–Mottin et al. (2003,1997)].– SSQ reallocates the coefficients resulting from a continuous wavelet transform
based on the frequency information, to get a concentrated picture over the time-frequency plane, from which the instantaneous frequencies are then extracted
– It is adaptive to the signal;– Reconstruction of the signal from the reallocated coefficients is feasible– Solid Theoretical foundations
![Page 21: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/21.jpg)
Page 21 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
SynchroSqueezing
• Idea from Daubechies & Maes 1996– Concentrate (squeeze) wavelet coefficients, at fixed
times, on the basis of local frequency estimation (synchronised)• Guarantees a sharply localised representation• Allows for reconstruction of identified modes• Offers a mathematical tractable alternative to EMD
![Page 22: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/22.jpg)
Page 22 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Synchrosqueezing – Reconstruction Approach
We are interested in the retrieval of the components fk of a multicomponent signal f defined by:
The problem can be viewed from two perspectives.
The first consists of computing an approximation of the modes, either by using EMD or via wavelet projections.
The second uses the SST to reallocate the WT before proceeding with multicomponent retrieval
![Page 23: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/23.jpg)
Page 23 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Underlying theory First recall some results first proposed by Daubechies et al 2011. For
a signal defined as a superposition of intrinsic-mode-type functions (IMT):
![Page 24: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/24.jpg)
Page 24 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Underlying theory
Where d is a separation condition.
![Page 25: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/25.jpg)
Page 25 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Theorem
![Page 26: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/26.jpg)
Page 26 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Theorem - Conditions
![Page 27: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/27.jpg)
Page 27 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Theorem – A Remark
• Component retrieval using the SST is thus first based on the computation of the synchrosqueezing operator and then on its integration in the vicinity of curves defined by ..
![Page 28: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/28.jpg)
Page 28 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Retrieval of Multiple Components (RCM):Practical Implemetation (Brevdo et al 2011)
The strategy developed by Brevdo et al for the RCM is to proceed on a component by component basis. The idea is to find a curve ,in the time-frequency plane such that it maximizes the energy which forces the modes to be smooth through a total variation term penalization and is obtained by computing the following quantity:
![Page 29: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/29.jpg)
Page 29 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
SST Parameters - WT of Pure Harmonic Signals
When dealing with a signal made of pure harmonics
The separation of the components using the WT is equivalent to the support of the WT of each component being disjoint that is for all k>1 :
Finally, using the definition of the separation condition introduced in Definition 2, when there is no interference between the components and
![Page 30: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/30.jpg)
Page 30 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
SST Parameters - WT of an ε-IMT
So, when is high, a low approximation error requires that
must be small. One way to ensure this is to impose that has a
fast decay or equivalently, that be large.
Let f(t) be a mode as defined in Definition 2, then the WT of f is now approximated by:
the approximation error being controlled by (C1(f)B1+C2(f)B2+(C3(f)B3), (see Brevdo et al 2011), where Ci, 1≤i≤3, does not depend on the wavelet choice, while
![Page 31: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/31.jpg)
Page 31 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
SST Parameters - WT of a Superposition of ε-IMT
For a multi-component signal as defined in Definition 2 we wish to have the following approximation:
So, following the study on a single component, a low approximation error requires
that be large enough. Then, we also want each component to be well separated from the other components (which is compulsory if we are to use the WT for mode retrieval), that is for each (a,t) the sum above should be reduced to a single term which is true provided that:
Taking into account the separation condition (see Definition 2), we can deduce that the above inequality is true as soon as
![Page 32: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/32.jpg)
Page 32 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
![Page 33: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/33.jpg)
Page 33 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Related Work
It is worth noting that the behavior of the SST was studied in detail by Wu et al 2011 for two tone signals. However, no conclusions can be drawn from the latter regarding modulated signals because the modulation considerably alters the wavelet representation of such signals and consequently the results given by the SST.
A very similar study was carried out by Mallat, ([14, p. 102]), but using a mother wavelet which was symmetric and compactly supported in the time domain. In such a case, the second order terms in the approximation would be negligible only if for all k:
provided
![Page 34: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/34.jpg)
Page 34 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Observation
The size of the frequency support of the mother wavelet plays a fundamental role when dealing with multi-component signals; adjusting this parameter enables us to tune between good separation and good localization of the modes. In the sequel we will use the previous remark regarding the size of the frequency support of the mother wavelet to build a new RCM algorithm.
This second constraint when applied to an -IMT implies that Φk is such that ,provided that , which is true when
,or equivalently .
If we assume that both Ak and are bounded below, the quality of the approximation is better if is small when is large. Since one must also have (
where is the frequency bandwidth of the wavelet) to ensure separation of a different component, cannot be taken arbitrarily small.
In contrast to the previous approach, the trade-off parameter between separation and localization of the components is now .
![Page 35: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/35.jpg)
Page 35 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
A New RCM
Code at: http://www-ljk.imag.fr/membres/Sylvain.Meignen/recherche/index.html
The algorithm aims at identifying the ridge associated with each component from the magnitude of the WT and then reconstructing the components by using the information in the vicinity of the different ridges. It is based on three distinct steps:
- first determine adaptively the number of modes
- compute an optimal wavelet threshold
- use threshold to retrieve the modes.
![Page 36: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/36.jpg)
Page 36 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Automatic detection of the number of modes
Since, for a given time t, the scales of interest in the SST are those where is large enough, we aim at detecting the modes using the following sets:
The natural expectation is that given t and when γ is appropriately chosen, each element will be associated with a single mode as defined above).
Note: The normalization of the wavelet transform we use is necessary to give sense to the above set. We compute an estimate of the number of modes at time t is:
The number of modes is defined as:
![Page 37: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/37.jpg)
Page 37 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Wavelet Threshold
Once the number of modes is determined, we first consider the sets:
and define T0 as the times t associated with non empty S0(t). A threshold for the WT and t in T0 is then computed as follows:
![Page 38: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/38.jpg)
Page 38 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
The Algorithm for Retrieving the Modes
We consider for each t in T0 the set Inf-k+1(t), which enables us to retrieve the mode k by applying the following approximation formula:
Note that the index of I in the sum comes from the fact that Ik for small k corresponds to a high frequency component and that in Definition (2) the components are arranged in increasing frequency order.
This reconstruction procedure is fully automatic and avoids the need to narrow down the curve searching in the time-scale space before proceeding with the mode retrieval.
![Page 39: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/39.jpg)
Page 39 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Example of a signal made of non-trivial intrinsic mode type signals
![Page 40: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/40.jpg)
Page 40 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Modulus of the corresponding Wavelet Transform
![Page 41: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/41.jpg)
Page 41 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Reconstruction using a bump wavelet
![Page 42: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/42.jpg)
Page 42 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Reconstruction as before but using alternative reconstruction formulae
![Page 43: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/43.jpg)
Page 43 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
![Page 44: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/44.jpg)
Page 44 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Non Uniform Sampling
To adaptively choose non-uniform samples to minimize the interpolation error is a thorny issue. A method inspired by the EMD consists of selecting the extrema of the signal, such that they depend locally on the strength of the oscillations. Another possibility would be to choose equi-spaced sample points over the whole signal duration.
Methods
1. Consider non-uniform sample points for the location of the extrema of the high frequency mode given by the RCM proposed and then piecewise cubic Hermite interpolation of f at these points.
2. Follow the same framework as the above, but the sample points being equi-spaced with the same average sampling rate.
![Page 45: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/45.jpg)
Page 45 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
![Page 46: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/46.jpg)
Page 46 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Further Work
• Strong modulation• Non-uniform sampling• Signals with a piecewise regular TF representation
![Page 47: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/47.jpg)
Page 47 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
Thank You
![Page 48: Page 1 of 48 One, Two or Many Frequencies:Synchrosqueezing SIERRA - Lyon, 13 th March 2013 Steve McLaughlin One, Two or Many Frequencies: Synchrosqueezing](https://reader038.vdocument.in/reader038/viewer/2022110303/5516cb3d550346f0208b5cbc/html5/thumbnails/48.jpg)
Page 48 of 48
One, Two or Many Frequencies:Synchrosqueezing
SIERRA - Lyon , 13th March 2013
Steve McLaughlin
References
E. Brevdo, N. S. Fuckar, G. Thakur, and H.-T. Wu, “The synchrosqueezing algorithm for time-varying spectral analysis: Robust properties and new plaeoclimate applications,” 2011, arXiv preprint ID: 1105.0010.
E. Chassande-Mottin, I. Daubechies, F. Auger, and P. Flandrin, “Differential reassignment,” IEEE Signal Process. Lett., vol. 4, no. 10, pp. 293–294, 1997.
I. Daubechies, J. Lu, and H.-L. Wu, “Synchrosqueezed wavelet transforms: An empirical mode decomposition-like tool,” Appl. Computat. Harmon. Anal., vol. 20, no. 2, pp. 243–261, 2011.
H.-T. Wu, P. Flandrin, and I. Daubechies, “One or two frequencies? The synchrosqueezing answer,” Adv. Adapt. Data Anal., vol. 3, no. 1–2, pp. 29–39, 2011.
S. Mallat, A Wavelet Tour on Signal Processing. NewYork:Academic,1998.
G. Rilling, P. Flandrin, and P. Goncalves, “On empirical mode decomposition and its algorithms,” in Proc. IEEE-EURASIP Workshop on Nonlin. Signal and Image Process. (NSIP), Grado (I), Jun, 2003.