A grid-based nonlinear approach to noise reduction and deconvolution for coupled systems
Introduction
Since the goal of experimental measurements is to more fully understand the state of a physical system, dealing with the ubiquitous presence of noise is a constant challenge for experimental design. From a strictly theoretical perspective, noise is completely independent from the system being measured; however, nature provides virtually limitless coupling pathways which prevents absolute decoupling of the two. Therefore, this paper is limited to considering experimental noise as having a sufficiently weak coupling as to not significantly affect the dynamics of the physical system under study. This distinction causes challenges in the context of the infinitesimal sensitivity of chaotic systems and will therefore require the additional constraint of application to bounded systems. Nevertheless, independent of the distribution of noise (e.g. random, Gaussian, etc.), so long as it has an algebraic mean of zero, measurement of sufficient amount of data is the underpinning of virtually all strategies to effectively denoise experimental measurements.
The traditional approach to noise reduction in a single signal is to simply assume that it is faster than the true signal; this opens up a huge variety of linear “smoothing” techniques. Two simple examples of linear smoothing include performing an FFT decomposition of the signal and reconstructing it with only low frequency components of the decomposition (i.e. choosing a cutoff frequency) or convoluting the signal with some sort of filter (e.g. top-hat filter) in time. A variety of nonlinear noise reduction technique have also been successfully proposed. The majority of this work has focused on denoising through projection of time-series data onto a high-dimensional phase space (i.e. inspired by time-delay embedding) and subsequent averaging/filtering of the resulting manifold [1], [2], [3], [4], [5], [6], [7], [8], [9].
Beyond single signal filtering techniques, it is also possible to leverage additional information embedded in other signals gathered from the same system to perform noise reduction. A simple, yet highly effective example of this was demonstrated by Sternickel et al. [10], who used two signals measured simultaneously. One of the measurement contained a clean signal contaminated by noise, and the other only contained only the noise. This “local” approach to denoising has been used broadly and with great success, but only works if the measurements are taken at the same spatial volume and only if the system state can be accurately synchronized to the acquisition of the desired measurement.
The crux of this paper is to demonstrate a new strategy for leveraging the availability of high-fidelity reference probe from one part of the system to denoise low-fidelity measurement signal sampled from a different spatial location in the system. This technique has particular promise for quasi-periodic bounded systems commonly associated with chaotic limit-cycle behavior, and with sufficient data, a straightforward extension also allows for temporal deconvolution of the measurement signal. The structure of the paper is as follows. In Section 2, a short discussion of the relevance of quasi-periodic bounded systems and previous denoising strategies for a prototypical chaotic limit-cycle system, the Hall-effect thrusters (HETs), is discussed. Next, Section 3 discusses the motivation for high-dimensional phase-space representations from dynamical systems theory and the underlying basis for temporal deconvolution. This is followed by Section 4, which provides details on both algorithms and numerical implementations used in this work. In Section 5, both nonlinear noise reduction and temporal deconvolution are applied to signals from three different coupled dynamic systems (sine wave system, Lorenz system, and Hall-effect thruster) and discussion of their performance relative to linear filtering methods is provided. Finally, in Section 6, further research areas and other potential applications of these methods are identified and discussed.
Section snippets
Background
A particularly challenging application of denoising is for filtering time-dependent signals from quasi-periodic bounded systems. These signals readily emerge from many real-world systems, from plasma thrusters and rocket engines to atmospheric convection and biological dynamics. These systems often display limit-cycle type behaviors around low-dimensional attractors and display finite sensitivity to even small amounts of noise. Thus, the signals extracted from these systems, while they are
Theory
The foundation underlying the approach described in this paper is the idea that nonlinear maps between multiple general measurements (observables) from the same causally coupled system can be constructed using time-delay embedding. However, existing techniques to construct these cross maps that use a fixed number of near-neighbor samples such as the maps constructed in convergent cross-mapping (CCM) [24] and the SMI techniques were designed with uncorrupted smooth dynamics and observations in
Numerical methods
This section covers the core and support algorithms necessary to perform both denoising and temporal deconvolution. It begins with the procedure of lifting a temporal signal to generate a high dimensional phase-space representation. Next, the algorithm for ensemble-averaging in phase space, the remainder of the nonlinear denoising algorithm, is described. The extension to accomplish temporal deconvolution is then discussed and, finally, a support algorithm for data smoothing in phase space is
Results
The nonlinear noise reduction method is applied to three different dynamic systems: 5.1 sine wave, 5.2 Lorenz system, and 5.3 Hall-effect thruster. For each system, two different scenarios are considered. In the first scenario, the target signal corrupted by noise, , is available along with a reference signal, , and the goal is simply to remove noise in the target signal to recover the true signal, . In the second scenario, a temporally underresolved synthetic measurement, , is
Conclusions and future work
This paper described novel approaches to perform (1) a grid-based nonlinear noise reduction and (2) deconvolution of underresolved measurements to reveal the underlying time-resolved signal, when reference data representing the dynamics of system were available along with noisy time-series data from the same system. The notional relationships among the various signals studied are summarized in Fig. 18.
Lagged coordinates on the reference data are used to map the signal into phase space, and a
CRediT authorship contribution statement
Samuel J. Araki: Methodology, Software, Validation, Formal analysis, Investigation, Writing - original draft, Writing - review & editing, Visualization. Justin W. Koo: Conceptualization, Methodology, Writing - original draft. Robert S. Martin: Writing - original draft, Writing - review & editing. Ben Dankongkakul: Investigation.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The experimental data used was obtained from the EPTEMPEST experimental program supported by AFOSR grant FA-9550-17QCOR497 (Program Officer: Dr Brett Pokines) and additional support was provided by AFOSR grant FA-9550-18RQCOR107 (Program Officer: Dr Fariba Fahroo) and FA-9550-20RQCOR098 (Program Officer Dr. Frederick Leve).
References (33)
A noise reduction method for chaotic systems
Phys. Lett. A
(1990)A noise reduction method for signals from nonlinear systems
Physica D
(1992)- et al.
Generalized phase space projection for nonlinear noise reduction
Physica D
(2005) - et al.
Optimal shadowing and noise reduction
Physica D
(1991) - et al.
On noise reduction methods for chaotic data
Chaos
(1993) Extremely simple nonlinear noise-reduction method
Phys. Rev. E
(1993)- et al.
Nonlinear noise reduction
Proc. IEEE
(2002) Smooth local subspace projection for nonlinear noise reduction
Chaos
(2014)- et al.
Improvements to local projective noise reduction through higher order and multiscale refinements
Chaos
(2015) - et al.
Nonlinear noise reduction using reference data
Phys. Rev. E
(2001)
Low frequency oscillations in a stationary plasma thruster
J. Appl. Phys.
Fundamentals of Electric Propulsion: Ion and Hall Thrusters
High-speed dual langmuir probe
Rev. Sci. Instrum.
A Time-Resolved Investigation of the Hall Thruster Breathing Mode
Validation and evaluation of a novel time-resolved laser-induced fluorescence technique
Rev. Sci. Instrum.
Development of a Time-Resolved Laser-Induced Fluorescence Technique for Nonperiodic Oscillations
Cited by (7)
Local projective noise reduction algorithm based on fuzzy recurrence and optimal hard threshold
2023, Xi Tong Gong Cheng Yu Dian Zi Ji Shu/Systems Engineering and ElectronicsOptimal transport for parameter identification of chaotic dynamics via invariant measures∗
2023, SIAM Journal on Applied Dynamical SystemsSparse Representation Denoising of Roadsid Acoustic Signals Based on Wavelet-domain RLS Filters
2023, Proceedings - 2023 China Automation Congress, CAC 2023An Unstructured Mesh Approach to Nonlinear Noise Reduction for Coupled Systems
2023, SIAM Journal on Applied Dynamical SystemsA Cascaded Adaptive Local Projection Denoising Method
2022, Dongbei Daxue Xuebao/Journal of Northeastern University