Abstract:The restoration of a non-stationary composite signal to its original components is a challenging and intriguing task in the field of signal processing. Various methods have been developed to tackle this problem. A traditional method for this problem is Singular Spectrum Analysis(SSA), while it needs to know the component number and costs too much time. Empirical Mode Decomposition(EMD) is also frequently applied, which decomposes the composite signal into a set of components known as Intrinsic Mode Functions(IMFs). However, the EMD method has been criticized for lacking a strong theoretical foundation and being more of a mathematical trick. Another popular method, Empirical Wavelet Transform(EWT), divides the frequency spectrum of the target signal into predefined segments and applies wavelet transforms using specific wavelet bases. Recently, a new method called Variational Mode Decomposition(VMD) has been proposed. This method is based on the variational principle and transforms the decomposition problem into an optimization problem. VMD has a robust mathematical foundation and exhibits excellent performance. However, it does have some limitations, such as the requirement for prior information on the modal number and the existence of a serious deviation in the end regions(named "end effect"). In this paper, we propose a novel method for sequentially separating non-stationary composite signals. Our method is inspired by the variational principle and can accurately and adaptively recover the original modes one by one from the raw mixture, without any prior knowledge or assumptions regarding the modal number. Moreover, our method can determine the modal number during the separation process, providing significant convenience in real-world applications. Additionally, we address the issue of end effects by conducting a new end elongation for the composite signal before the decomposition operation. This approach effectively reduces the end effect to a much lower level compared to the VMD method. To further enhance the accuracy of our method, we introduce a refinement approach after the coarse extraction. Combined these techniques, the final decomposition results demonstrate that our novel method outperforms VMD, EMD, EWT and SSA methods in most cases.

Modified Independent Component Analysis for Extracting Eigen-Modes of a Quantum System.

Multivariate Nonlinear Chirp Mode Decomposition

Estimating Eigenenergies from Quantum Dynamics: A Unified Noise-Resilient Measurement-Driven Approach

Eigen component analysis: A quantum theory incorporated machine learning technique to find linearly maximum separable components

Extraction and Identification of Noise Patterns for Ultracold Atoms in an Optical Lattice

The Quantum Cocktail Party Problem

A Quantum Algorithm for Dynamic Mode Decomposition Integrated with a Quantum Differential Equation Solver

The Devil is in the Details: Complexity Powered Machine Intelligent Classification of Quantum Many-Body Dynamics

High-dimensional Mode Analyzers for Spatial Quantum Entanglement

Detecting quantum entanglement with unsupervised learning

Unsupervised learning approach to quantum wavepacket dynamics from coupled temporal-spatial correlations

Pseudo-Bayesian Approach for Robust Mode Detection and Extraction Based on the STFT

Exploiting Maximally Mixed States for Spectral Estimation by Time Evolution

Uncovering Quantum Many-body Scars with Quantum Machine Learning

A queued Variational Mode Decomposition method

Data-driven discovery of statistically relevant information in quantum simulators

Diagnosing quantum transport from wave function snapshots

Quantum force sensing by digital twinning of atomic Bose-Einstein condensates

Characterization of decohering quantum systems: Machine learning approach

Quantum algorithm for MUSIC-based DOA estimation in hybrid MIMO systems

Quantum state estimation based on deep learning