You are on page 1of 5

The ASTRES Toolbox for Mode Extraction of

Non-Stationary Multicomponent Signals
Dominique Fourer? , Jinane Harmouche, Jérémy Schmitt, Thomas Oberlin,
Sylvain Meignen, François Auger and Patrick Flandrin
?
dominique@fourer.fr

Abstract—In this paper, we introduce the ASTRES∗ toolbox TFRs Data−driven methods
which offers a set of Matlab functions for non-stationary multi-
component signal processing. The main purposes of this proposal STFT CWT S−transform EMD SSA
is to offer efficient tools for analysis, synthesis and transformation
of any signal made of physically meaningful components (e.g.
sinusoid, trend or noise). The proposed techniques contain some reassignment synchrosqueezing
recent and new contributions, which are now unified and theoret-
ically strengthened. They can provide efficient time-frequency or
time-scale representations and they allow elementary components ridge detection Mode extraction
extraction. Usage and description of each method are then
detailed and numerically illustrated.
Fig. 1. ASTRES toolbox content description freely available online2 .

I. I NTRODUCTION
Real world signals are often non-stationary and made of of the considered transforms and their TFRs are presented. In
several components. They require advanced techniques to be Section III, reassigned and synchrosqueezed versions of each
efficiently processed. Unfortunately, the Short-Time Fourier transform are described with two ridge detection methods. Sec-
Transform (STFT) and the Continuous Wavelet Transform tion IV describes data-driven methods, EMD and SSA, with
(CWT) which belong among the most usual approaches, are their corresponding new developments. Finally, the proposed
limited due to a poor energy localization in the time-frequency methods are illustrated by numerical experiments on real world
plane [1]. signal in Section V, before concluding the paper in Section VI.
Two solutions, the reassignment and the synchrosqueez- II. T IME - FREQUENCY REPRESENTATIONS
ing methods [2], [3], [4], can improve the readability of a A. Short-time Fourier transform (STFT)
time-frequency representation (TFR). More particularly, the Let Fxh (t, ω) denote the STFT of signal x, using a differen-
synchrosqueezing can provide sharpened and invertible TFRs tiable analysis window h, defined as
allowing many applications like noise removal and signal Z
decomposition into modes [3], [5], [6]. In addition, the Fxh (t, ω) = x(u)h(t − u)∗ e−jωu du, (1)
R
Empirical Mode Decomposition (EMD) [7] and the Singular
Spectrum Analysis (SSA) [8], which belong to data-driven z ∗ being the complex conjugate of z. Thus, the spectrogram
methods, allow unsupervised signal decomposition into a sum can be computed by |Fxh (t, ω)|2 (cf. tfrstft, tfrgab). Eq.
of physically meaningful components (e.g. periodic functions, (1) admits the following synthesis formula when h(t0 ) 6= 0,
trend or noise). Several recent developments were proposed to allowing to recover signal x with a delay t0 ≥ 0 (cf.
enhance these techniques [9] and to enable a fully automatic rectfrgab)
Z +∞
version of SSA, in the same flavor as EMD. 1 dω
x(t − t0 ) = Fxh (t, ω) ejω(t−t0 ) . (2)
This paper proposes a collection of Matlab functions related h(t0 ) −∞ 2π
to the new methods developed in the “Analysis, Synthesis, An efficient recursive implementation (suitable for real-time
Transformation by Reassignment, EMD and Synchrosqueez- computation), is proposed in [10], [12]. It uses a specific
ing” (ASTRES) research project∗ . This new toolbox (cf. Fig. 1), causal (one-sided) analysis window which can be expressed
designed for automatic signal mode extraction, can be viewed k−1
as hk (t)= T kt(k−1)! e−t/T U (t), with k ≥ 1, where U (t) is the
as an extension of the previously proposed Time-Frequency
Heaviside step function and T a time spread parameter (cf.
ToolBox (TFTB)1 . It contains implementations of some recent
recursive_stft and stft_rec).
contributions related to reassignment, synchrosqueezing [10],
[11], [6], [12], EMD [13], [14], and SSA [9] methods. This B. Continuous wavelet transform (CWT)
paper is organized as follows. In Section II, proper definitions The CWT of a signal x is defined for an admissible mother
wavelet function Ψ as [1]
∗ This research was supported by the French ANR ASTRES project (ANR- Z +∞  ∗
13-BS03-0002-01). 1 τ −t
1 http://tftb.nongnu.org, 2 https://github.com/dfourer/ASTRES toolbox WxΨ (t, s) = p x(τ )Ψ dτ (3)
|s| −∞ s

(15) 2πT WxΨ (t. ω) + µI2 Rx (t.If we define the scale as s = ωω0 .s0 )2 (17) (cf.ω) s ∗ |ω| +∞ Z  ω and ω̂ (t. s) = =− . Ω)) dτ dΩ (18) R2 Reassignment and synchrosqueezing are sharpening tech- niques designed to improve TFRs. proofs in §8.ω) = Re t̃(t. tfrrst. Each synchrosqueezed transform can be deduced from the chrosqueezing and a quite poorer time-frequency localization simplified synthesis formula given by Eqs (2). Eq. R EASSIGNMENT AND SYNCHROSQUEEZING |STx (τ. The corresponding TFR. substitut- ing MWx in Eq. ω) = Ch (ω0 T ) = Fh (ξ − ω0 ) .ω)  h  |ω| − with Rxh (t.ω) ∂t x (t. s) transform by s̃(t.ω) = Im (ω̃ (t. The synthesis formula of Eq. ω) = s with T h(t) = th(t) and Dh(t) = dh dt (t). Ω)) dτ dΩ. compared to the reassignment. where t̂(t. (13) (cf. we finally obtain MWx (t. Ω)| δ t − t̂(τ. s) where h is a Gaussian window defined as h(t) = e . the CWT |ω| +∞ Z   ω h STx (t. respectively for the STFT. and recursive_lmrsp). Synchrosqueezing tween these methods is the reconstruction capability of syn. Reassignment frequency. (13)  given by (cf. ω) t −1 h = − ∇t Rxh (t. ∇tRxh (t. ω).ω) = ∂R ∂Rxh STx (t. the reassigned scalogram can be computed as (cf. called Stockwellogram.ω) = Im(s̃(t. ω) by Z SThx (t.ω) F Dh (t. (6) and (10). ω). ω) (14) signal x as (cf. [18] ω0 −∞ ω0 2 WxT Ψ (t. tfrlmrgab STFT using a frequency-varying window width. s) The S-transform can also be related to the Morlet wavelet ω0 sω0 WxΨ (t. ω̂ (t. ω) 7→ (t̂(t. ω) ω   Z +∞ ω 2 (t−τ )2 t − t̂(t. s) WxDΨ (t. FxT h by STTx h and FxDh by STDh x (cf. ∂t ∂ WxΨ (t. Thus. ω) = x(τ )h (t − τ ) e−jωτ dτ (8) reassignment operators can be expressed by [2].  ized phase-shifted Morlet wavelet transform. [1]. (4) reassignment operators. ω) = √ e−jωt MWx (t. (3) is |Fxh (τ. ω)|2 .ω). The main difference be. s) |ω| STx (t.ω) = . 2) CWT: Since. s)|s|−3/2 ds . ω) ejωt . ω) = ZZ (cf. tfrrsp and √ x(τ ) e 0 e−jω(τ −t) dτ (5) ω0 T π −∞ recursive_rsp) RFx (t. expressed as FxT h (t. (6) leads to the S-transform synthesis formula R2 ŝ(τ. s)|2 or |CWx (t. ŝ(t.s0 ) δ (s − ŝ(τ.ω) = t − h . ω̂µ ) in Eq. Thus. As a result. are the CWx (t. the TFR values are moved according transform as to the map (t. based on the x(t) = WxΨ (t. ω) = x(τ )Ψ (τ − t) dτ. parameter µ [17]. which is not invertible. tfrst) ω̂µ (t. ω̃ (t. recMW) R2 Z Z 1 dω A recent extension of the reassignment process. √1 t − 2T 2 t̃(t. the ω = ω0 . s) = − 1s WxDΨ (t. B. . Ω)|2 δ t − t̂(τ. tfrscalo and MW). ω̂) by (t̂µ . by replacing Fxh (t.ω) Fx (t. Ω) δ (ω − ω̂(τ. Thus. (11) R |ξ| ZZ 2  III. the reas- +∞ ω 2 (t−τ )2 |ω| Z − 2(ω T )2 signed spectrogram can be computed as (cf.ω)). Stockwell transform (S-transform) computed as     The S-transform [15] can be defined for a given zero-mean t̂µ (t. (10) Ch (ω0 T ) R |ω| and Levenberg-Marquardt reassigned Stockwellograms are ob- where Ch (ω0 T ) is a proportionality factor defined by tained through Eqs. (16) s ω̃(t. rectfrst2) [16] Z 3) S-transform: Reassignment operators can be computed 1 dω as for the STFT (cf. s) = which shows that the S-transform can be viewed as a normal- s02 ZZ |WxΨ (τ. (3) can now be expressed as a time-frequency For each transform. is Levenberg-Marquardt reassigned spectrogram can be obtained provided by |STx (t.ω) ∂ω (t. Thus we have RSTx (t. both classical x(t) = STx (t. allows to adjust the energy (6) localization in the time-frequency plane through a damping where FΨ (ω) denotes the Fourier transform of Ψ(t). (12) and (13). (9) Then.ω) t̃(t. with Cψ = FΨ (ω)∗ Cψ R R ω Levenberg-Marquardt algorithm. [16]).ω) 2πω0 T −∞ where T is the width of the Gaussian analysis window when where I2 is the 2 × 2 identity matrix.m dξ and tfrlmrst). s) = t + s . ω0 being an arbitrary A. ω) = √ x(τ ) e 2(ω0 T )2 e−jωτ dτ (7) ω − ω̂ (t. s0 )|2 δ t − t̂(τ. ω0 −∞ ω0 1) STFT: Reassignment operators can be computed as [2] The scalogram can be computed by |Wx (t. If we use the Morlet wavelet [1]. The new reassignment operators can be C. ω)|2 . (12) −1/4 π√ −t2 jω0 t Fx (t.s0 )) dτ ds0 .ω))).ω)) (resp. Ω) δ (ω − ω̂(τ. s).ω) Ψ(t) = T e 2T 2 e .The S-transform can be viewed as a by replacing (t̂.ω) = jω + xh . rMW) 2 πω0 T RWx (t. the CWT and the S-transform.

Re FxD h (t. tfrvsst). ω) can be computed for each transform Its squared modulus provides a sharpened TFR. (20) h(t0 ) R 2π 2. ω) .ω) SWx (t.ω) 2 t 2 q̃x (t. Second-order synchrosqueezing 3) S-transform: A new local modulation estimator can Second-order synchrosqueezing was first proposed by Ober.ω) = Dn−1 h n−1 h .ω) − STx (t. (31) for the sake of clarity.ω) = Z Z . (22) q̃x(ωn) (t.ω) − STx (t.ω) = n−1 hF T h − F T nhF h . (19) R An estimation of q̃x (t. ω 0 ))δ(ω − ω̂(τ. valid for any differentiable analysis window. (23) terms of accuracy for n = 2. we obtain (cf. vsMW) ∂t (t.ω)Fxh (t. by restricting tfrosgab) can be viewed as a time-frequency reassignment the integration area to the vicinity of the ridge. (32) Wx (t. Assuming a signal model expressed as 2 STD h x Dh (t. (14) allows to make a vertical synchrosqueezed STFT transform (cf. and can be improved using a slightly different recursive_sstft) estimator expressed as (when |q̃x (t. s) = DΨ . ω 0 ) ejω t δ (ω − ω̂(t.ω) 2 D. ridge_detect_brvmask) ment operator). we proposed several local modulation 1 dω estimators. (29) Im FxT Dn−1 h (t. extracted from its synchrosqueezed transform. each mode xi (t) = ai (t) e can be Oblique synchrosqueezing (only proposed for the STFT) (cf. ω)FxDn−1 h (t. s0 )|s0 |−3/2 δ(s − ŝ(t. ω 0 )) dτ dω 0. The ridge Ωi (t) can be directly estimated from a TFR α̂x using a ridge detector as: OSFx (t. (25) STx (t. ω) = |ω| STx (t. tfrlmsgab.ω) = Fxh (t.s) Ch (ω0 T ) R |ω| q̃x (t. ∀n ≥ x̂(t − t0 ) = ∗ SFhx (t. (33) x(t) = a(t) ejΦ(t) .ω)Fx (t.s) − Wx (t. s) = ∂ ω̃ ∂t (t.ω0 )) dω 0 . ω 0 )) 0 . in each reconstruction formula given by Eqs. with a phase correction term [11]. tfrlmsst) In [12]. ω) = Dh Th T Dh . also be derived to compute the vertical synchrosqueezed S- lin et al.ω) was omitted in Eq. defined as [16] (cf. which maximizes the energy with a smoothness constraint Vertical synchrosqueezing uses an improved estimation of through a total variation penalization term expressed as (cf. denoted Ωi (t). and it can be as follows: inverted by Z 1) STFT: In [12]. (24) s2 Wx (t. recsMW) (30) n−1 n−2 n−1 1 Z (FxT Dh + (n−1)FxT h )Fxh − FxT h FxDh x̂(t) = SWx (t. tfrsgab.s)WxT Ψ (t. tfrvsgab. (27) or in (28) to obtain Replacing ω̂ by ω̂µ computed by Eq. ω)FxD h (t.ω) − FxT D (t. R Fx (t. the instantaneous frequency (instead of the classical reassign. s) ds. Ridge detection for mode extraction where a(t) and Φ(t) stand for the time-varying amplitude and phase.s) C. it is shown in [12] that for any t0 ≥ 0 such that h(t0 ) 6= 0 (cf. sMW) n n−1 FxD h − FxD h (t.s)WxΨ (t. Then. 1) STFT: The synchrosqueezed STFT can be defined For amplitude modulated signals.ω) = . synchrosqueezing adjustable as for the Levenberg-Marquardt recursive_vstft)  n n−1  reassignment (cf. s) = Wx (t.ω)STx (t. ω 0 ) ej(ω− 2 (t−τ ))(t−τ ) 1) Brevdo et al. ω) . (22) and ZZ (24).ω)FxT h (t. i=1 xi (t).s) − WxT DΨ (t.ω) and can be inverted by (cf. method: This technique [19] aims at R2 δ(t − t̂(τ. these estimators were implemented and compared in dω 0 Z 0 SSTx (t. rectfrsst) ∂ t̃ proposed for the CWT [18]. CΨ R FxT x x x (31) 3) S-transform: The synchrosqueezed S-transform can be where (t.s0 )) ds0 . ω 0 ) ejω (t−t0 ) δ(ω − ω̂ (t.ω)STx (t. (21) q̃x(tn) (t. signal.ω)STx (t. recursive_lmsstft).s)Wx (t. in [11] to improve TFRs of strongly modulated chirps transform [16] (cf. which can be used in Eq. tfrsst.ω) + q̃x (t. ω) = Fxh (τ. ω)∗ 2) CWT: The synchrosqueezed CWT can be defined as [3] α̂xKn (t.ω)(t − t̃(t. R |ω | 2) CWT: Using the local modulation estimator q̃x (t.ω)FxDh (t. with Φ(t) = ϕx + ωx t + αx . thanks to the chirp rate estimation α̂x (t. (27) is biased.ω)) .ω)| < ∞) ω̂ (3) (t.s) and can be inverted by (cf. Second-order synchrosqueezing proposes to use PLet us consider a multicomponent signal denoted by x(t) = I jφi (t) local modulation estimation to improve the resulting TFR. ω) = Im ω̃ (t. (26) finding the best frequency curve Ω(t) in the TFR Sx . (20).ω) h Z (t. Z  0 (28) h SFx (t. ω)∗ (cf.s) Z 1 dω   1 D2 Ψ Ψ DΨ 2 x̂(t) = SSTx (t.

.

dΩ .

2 .

2 Im(q̃x (t.ω)) (see [12] for a further investigation). The improved Ω̂ = argmax |Sx (t. Ω(t))| dt − λ .

.

(t).

.

ω)| < ∞ For multi-component extraction (i. be iterated after setting Sx to zero in the vicinity of the (27) previously detected ridge. . dt. (2) ω̂ (t.ω)) if |α̂x (t.ω) + α̂x (t. ω) otherwise. (34) frequency estimator can be computed as Ω R dt R ( where λ controls the importance of the smoothness constraint. ω) = ω̂(t.ω)(t − t̂(t.e. this method can ω̂ (t. I > 1).

Flandrin. S. H. a sampled signal) s = {sn . Mag. “The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series them to build a new class. t2 x∈Ci . Long. This to paper size limitation.. with Xi = σi ui vi T . z 2 These TFRs are computed using respectively. May 1995. [14]. of size L × K. d(x. 2013. pp. R. C. This operation is iterated analysis. (37) these methods designed for efficient TFRs computation and i=1 mode extraction. and H. no.n−m+1 for K + 1 ≤ n ≤ N. . 1971. The main idea of this algorithm [7] N. Future works consist in theoretically σi being the singular values of X and ui (resp. ssa) is described by: The ASTRES toolbox was introduced as a collection of Matlab functions for processing non-stationary and multicom- 1) Construction from x of the trajectory Hankel matrix X ponent signals. 60. 5. 30. “Synchrosqueezed wavelet trans-    X L forms: An empirical mode decomposition-like tool. were also illustrated by application examples on real world signals. vi ) the strengthening these tools. H. spz_delaunay_dom4a). pp. Liu. no. resulting from a binary black hole collision. J. 2012. . V ). Shih. M. the Gabor trans- e zn + 2 ( zn ) . τ )x(τ )dτ. ponents. 43. 2. to obtain R developed into the ASTRES project. through Bargmann Thus. C. x̂) = outlier edges (cf.- [20] is to look for the two nearest classes and to merge C. N } and a user parameter L. with their respec- Then. unifies into the same framework several recent techniques 2) Singular Value Decomposition (SVD) of X. with Fx (z) = A(z. Nov. For a particular Gaussian analysis window D(Ci . 2137–2141. 11.” Proc. Fig. Daubechies. of the Royal Society of London A: Math. m=1    and H. Flandrin. with the Brevdo et al. 5787–5798.n−m+1 for L ≤ n ≤ K [4] F.    N −n+1  [5] S. 11. Auger. Lett. C.. the zeros zn = ωn + jtn can be determined from the STFT . Auger and P. pp.. N. 6. “Time–frequency filtering based on spectrogram zeros. S.-T. 2015. based on Hierarchical Clustering (HC) IEEE Signal Process. [2] F.. It considers a finite length time VI.   m=1  [3] I.” Applied and Com- (i) 1 putational Harmonic Analysis. ssa_decomp). and in proposing new applications.y∈Cj ||x|| ||y|| g(t) = π −1/4 e− 2 . τ ) = π e . 1068–1089. 454. Wu. no. 22. Syn- A.DRIVEN METHODS The ASTRES toolbox also includes several implementations B. vol. Time-Frequency/Time-Scale analysis. McLaughlin. Z. E. Lin. Process. pp. Physical and until the desired number of classes (i. Its algorithm (cf. pp. 1  (Xi )m. y >| trogram zeros. McLaughlin. 3). trends and noise. other methods which are more likely to create interference. we consider a real-world and extended for two dimensional signals [13]. Flandrin. method described in Section III-D1. Shen. [9] (cf. This toolbox which is freely available online. P. pp. 2) Delaunay triangulation method: Frandrin proposes in To compare two classes Ci and Cj containing several [6] a new way to disentangle the different components in time series. Time-frequency representations It admits a Weierstrass-Hadamard factorization such as [6]: Fig. the dissimilarity is defined as the minimal the time-frequency plane through a simplified representation distance between two distinct time series of each class provided by the Delaunay triangulation attached to the spec. H.n−m+1 for 1 ≤ n < L    Trans. The and the domains attached to signal components are related to Reconstruction Quality Factor (RQF) given by RQF(x. vol. |z|2 Fxg (t. N UMERICAL SIMULATIONS 2 2 −1/4 − τ2 −jτ z+ z4 and A(z. and S. xn = L (Xi )m. with K = N − L + 1.. Meignen. ω) = e− 4 Fx (z). Singular spectrum analysis chrosqueezing leads to better separation results compared to SSA [8] can expand a signal in a sum of periodic com. DATA . Zheng. Tung. each extracted mode is obtained by summing the ele- Z transform as: mentary components included into the resulting classes Ci . Wu. vol. 903–995. T. 30. Empirical mode decomposition of the well known EMD method [7]. R V. Signal Process. Oberlin.” ponents grouping. Engineering Sciences. Y. 1998. Flandrin. (35) A. 243–261. Oberlin. the EMD method and the proposed SSA method [9]. y) = 1− . the Delaunay triangulation is performed over zeros zn tive first. I) is reached. Some capabilities of X = U ΣV T = X Xi . Press. no. “TF reassignment and synchrosqueezing: An overview. IV.”  L  X IEEE Signal Process. 1998. R EFERENCES 3) Elementary components reconstruction from each Xi by anti-diagonal averaging through [1] P. Yen. 2 compares the TFRs of the gravitational wave signal ∞   GW150914 [21]. Lu. C ONCLUSION AND FUTURE WORKS series (e. Wu.g. . 32–41. Meignen. vol.e.” IEEE Trans. Cj ) = min d(x. vol. y). and H. Q.  ||x||2  10 ln10 ||x−x̂|| 2 is indicated above each invertible TFR. Acad. [6] P. no. T. which was reformulated For this experiment (cf. Huang. “A new algorithm for m=n−K+1 multicomponent signals analysis based on synchrosqueezing: with an application to signal sampling and denoising.and second-order synchrosqueezed versions. Signal 4) Unsupervised mode extraction through elementary com. Nov. |< x. we chose to focus on more recent signal is analyzed using the synchrosqueezed STFT combined developments allowing an unsupervised usage of SSA. the Morlet Wavelet transform and the S-transform. S. n = 1 . vol. “Improving the readability of time-frequency  Xn and time-scale representations by the reassignment method. Y z 1 z Fx (z) ∝ 1− (36) n=1 zn form (STFT using a Gaussian analysis window). column vectors of U (resp. no. Due multi-component audio signal recorded from a cello.” IEEE  n1 (Xi )m. 2011.

STFT.015 0. Mar. 5. 4880–4884. vol.03 normalized frequency normalized frequency normalized frequency 0. STFT T=0. Oberlin and S. IEEE ICASSP.1 0.12 0. K. Auger. p. two components ? the singular spectrum analysis answers.. [21] P. F.” in Proc. 2011.02 0. “Reassigning and synchrosqueezing the [10] D.025 0. S. Fourer. 1996. E.1 0. “Second-order synchrosqueezing adjustable: the Levenberg-Marquardt approach.025 0.02 0. mod.32 dB squ..06 0. Lett.12 0. N.08 0. S.015 0. F.02 0.02 0. “Making reassignment [11] T.04 0. RQF=26.-T.80 dB squ.04 0. synch.14 0 0. 2017.14 0 0.1 0. Borgnat. Perrier. Meignen. Dec. 1963. Auger.01 0.025 0. 2017. Tech. Meignen. vertical synch. and R. Brevdo.04 0.04 0. L. vertical synch. prony-huang transform: A new tool for 2D spectral analysis.14 time [s] time [s] time [s] (a) synchrosqueezed STFT (b) EMD (c) SSA Fig. 236–244. Flandrin.02 0..02 0.02 0. mod. F.08 0.12 0. and V.08 0. Plenum Press.01 0.08 0.1 0. Comparison of the extracted modes resulting from a real-world audio signal.03 0.06 0. S.03 normalized frequency normalized frequency normalized frequency 0. Flandrin. vol.0223 Morlet scalogram.12 0. Flandrin. Schmitt. Morlet CWT and the Stockell transform applied on the Livingston GW150914 signal [21]. 1996. 3. Fourer.12 0.01 0. Oberlin.80 dB squ. spectrum.12 0. [17] F. the Levenberg-Marquardt reassigned spectrogram and of the syn. f0T=1. 2015. no. pp.025 0. Elsner and A. “Recursive versions of Stockwell transform: Complementary proofs.005 500 1000 1500 2000 2500 3000 500 1000 1500 2000 2500 3000 500 1000 1500 2000 2500 3000 time samples time samples time samples squ. and P. Low-Rank Approximation (SLRA’2015). Sep. Rep. Singular Spectrum Analysis.02 0. Mar.01 0. Apr.1 0. Image Process. Signal Process.06 0. Borgnat. and V. pp.11. Meignen.fr/ensl-01370441 . “Chirp [19] E. Czarnecki.06 0.89 dB squ.1 0. “2D École normale supérieure de Lyon.11. mod.12 0. 2016. Thakur.03 0.02 0.06 0. 60.02 0.37 dB 0. 44. and P.25 dB 0.” IEEE Available: https://hal-ens-lyon.06 0. 2014. mod.06 0. N. pp.01 0.015 0.005 500 1000 1500 2000 2500 3000 500 1000 1500 2000 2500 3000 500 1000 1500 2000 2500 3000 time samples time samples time samples Fig. vol. [8] J.015 0. [13] T. pp. transform or invertible reassignment? Towards ideal time-frequency Mar.04 0. Comparison between the Gabor STFT. Auger.015 0. Time Series Analysis. RQF=44.04 0.02 0.archives-ouvertes. RQF=66.11 Stockwellogram. f0T=1. f0T=1.08 0. 4. Available: http://hal. Flandrin.02 0.1 0.14 frequency [Hz] frequency [Hz] frequency [Hz] 4000 4000 4000 2000 2000 2000 0 0 0 0 0. A New Tool in Trans.14 0 0.015 0. P. Mansinha.” [14] J. [12] D. 2012.14 0 0. no.” IEEE Trans. f0T=1.0223.08 0. and J. pp.02 0. P.. 2015.” IEEE Signal Process.005 0. “A note on the time-frequency analysis of GW150914. Condat.” in Proc.1 0.03 0.” IEEE Trans.02 0. 1335– [18] T.005 500 1000 1500 2000 2500 3000 500 1000 1500 2000 2500 3000 500 1000 1500 2000 2500 3000 time samples time samples time samples squ. vertical synch. and P.03 0.01 0. mod.03 normalized frequency normalized frequency normalized frequency 0. G. 2012. and L. 12.14 0 0.015 0.” Tech. Fourer.015 0.02 0.11.04 0. 23.005 0.04 0. pp.025 0. Auger.025 0.archives-ouvertes.02 0. 998–1001. Signal Process. synch. Gabor spectrogram. vol..025 0. Ward.14 0 0. [16] D.025 0. 2236–2246. Rep.08 0.08 0.14 0 0. Meignen. RQF=32.04 0.08 0. transform. EMD and the proposed SSA method.14 frequency [Hz] frequency [Hz] frequency [Hz] 4000 4000 4000 2000 2000 2000 0 0 0 0 0.1 0. ST.015 0.06 0. MW. France. Perrier. Hu. (to appear). Chassande-Mottin. vol. H. 2015. 58.02 0.01 0. F.” IEEE Trans.03 0. Flandrin. JASA.02 0.02 0. Lowe. Stockwell. 3889–3892. Flandrin. Mar. and H.1 0.fr/hal-01467244 chrosqueezed STFT.12 0. Grenoble.01 0. “Hierarchical grouping to optimize an objective function.” in Proc.08 0.02 0. mod.005 0. “Localization of the complex [9] J.005 0. 63. Oberlin. 2. Wu. “One or spectrum: the S-transform. Fuckar. Jun.12 0.11 0. RQF=35. “An alternative formulation [20] J. no. IEEE ICASSP.. and P. synch.02 0. representations.12 0. using synchrosqueezed STFT. Pustelnik.04 0. T=0.08 0.11. [Online].14 0 0. Tsonis.12 0. [Online].06 0. “The synchrosqueez- rate and instantaneous frequency estimation: application to recursive ing algorithm: a robust analysis tool for signals with time-varying vertical synchrosqueezing. MW. T=0. Signal Process.03 0. 2016. Auger.” in Structured pp.06 0. no. ST.1 0. “The second-order wavelet synchrosqueezing 1344.025 0.” for the empirical mode decomposition. D. synchrosqueezing extracted components EMD extracted components SSA extracted components frequency [Hz] frequency [Hz] frequency [Hz] 4000 4000 4000 2000 2000 2000 0 0 0 0 0.06 0.12 0..005 0. RQF=26. P. Fourer. f0T=1. IEEE ICASSP. [15] R.06 0.08 0.0223. f0T=1.” Arxiv preprint.04 0. Harmouche.04 0.14 frequency [Hz] frequency [Hz] frequency [Hz] 4000 4000 4000 2000 2000 2000 0 0 0 0 0. 5.1 0.005 0.01 0. 5233–5248. S.