1: \begin{abstract}

2: In this paper, we introduce a new adaptive data analysis method

3: to study trend and instantaneous frequency of nonlinear and 

4: non-stationary data. This method is inspired by 

5: the Empirical Mode Decomposition method (EMD) and the recently 

6: developed compressed (compressive) sensing theory. The main

7: idea is to look for the sparsest representation of multiscale

8: data within the largest possible dictionary consisting of

9: intrinsic mode functions of the form

10: $\{ a(t) \cos(\theta(t))\}$, where $a \in V(\theta)$, $V(\theta)$

11: consists of the functions smoother than $\cos(\theta(t))$ and 

12: $\theta'\ge 0$. This problem can be formulated as a nonlinear 

13: $L^0$ optimization problem. In order to solve this optimization 

14: problem, we propose a nonlinear matching pursuit method by 

15: generalizing the classical matching pursuit for the $L^0$ 

16: optimization problem. One important advantage of this nonlinear 

17: matching pursuit method is it can be implemented very efficiently 

18: and is very stable to noise. Further, we provide a convergence 

19: analysis of our nonlinear matching pursuit method under certain 

20: scale separation assumptions. Extensive numerical examples will 

21: be given to demonstrate the robustness of our method and comparison 

22: will be made with the EMD/EEMD method. We also apply our method to 

23: study data without scale separation, data with intra-wave frequency 

24: modulation, and data with incomplete or under-sampled data. 

25: 

26: \end{abstract}

27: