Topological phase estimation method for reparameterized periodic functions

Thomas Bonis, Frédéric Chazal, Bertrand Michel, Wojciech ReisearXiv 2022 · DOI · arXiv

Introduction

Estimating the phase of a periodic signal is a fundamental task in signal processing. However, traditional methods like Fourier transforms or zero-crossing counts often fail when the underlying periodic function is unknown and subjected to non-linear "reparameterization"—where the signal is non-uniformly stretched or squeezed over time.

In "Topological phase estimation method for reparameterized periodic functions," Thomas Bonis and colleagues propose a robust framework using Topological Data Analysis (TDA). Their approach leverages 0-dimensional persistent homology to count cycles and segment signals, even when the shape of a single period is complex and the temporal warping is significant.

Problem Formulation

The authors address the challenge of analyzing a signal $S(t) = f(\gamma(t)) + W_t$, where $f$ is an unknown periodic function, $\gamma(t)$ is a non-linear reparameterization function, and $W_t$ represents noise. The primary goals are to determine the number of periods $N$ observed and to identify the specific time points marking the beginning of each new period (odometry).

Reparameterized signal (left) and its corresponding non-linear reparameterization function (right).

Standard spectral methods assume $f$ is a simple sinusoid, and geometric methods like zero-crossings fail if $f$ has multiple local extrema within a single period. Topology offers a solution because it is invariant to the "warping" caused by $\gamma(t)$.

Methodology: Persistent Homology

The core of the proposed method is the use of persistent homology on sublevel sets. By analyzing how local minima (valleys) are born and merge as a horizontal threshold rises across the signal's graph, the authors generate a persistence diagram.

A key theoretical insight is the homogeneity property: in a noiseless signal with $N$ periods, every point in the persistence diagram of a single period appears exactly $N$ times in the diagram of the full signal. To estimate $N$, the authors calculate the Greatest Common Divisor (GCD) of the multiplicities of points in the persistence diagram. To handle noise, they employ single-linkage hierarchical clustering to group nearby points before calculating the GCD, and they ignore low-persistence features that likely correspond to high-frequency noise.

Experimental Evaluation

The method was tested on both synthetic and real-world datasets. Synthetic signals included various template functions with multiple extrema per period, subjected to random non-linear warping and Gaussian noise.

Examples of noisy synthetic and real-world signals used for testing the topological estimator.

The topological estimator significantly outperformed traditional zero-crossing methods. In a real-world application of "magnetic odometry"—tracking a vehicle's movement via magnetic field perturbations—the topological method achieved a correct segment ratio of 0.96 compared to 0.69 for zero-crossings. Furthermore, the distance estimation was over four times more precise, and the reconstructed speed profiles closely matched high-end GPS reference data.

Conclusion

Topological phase estimation provides a powerful alternative to spectral analysis for complex, warped periodic signals. By focusing on the persistent structure of local extrema, the method remains robust to non-linear reparameterization and noise. Future work may address the current limitation of requiring integer period counts and explore multivariate extensions for even greater robustness in sensor fusion applications.

Summary updated

Abstract

We consider a signal composed of several periods of a periodic function, of which we observe a noisy reparametrisation. The phase estimation problem consists of finding that reparametrisation, and, in particular, the number of observed periods. Existing methods are well-suited to the setting where the periodic function is known, or at least, simple. We consider the case when it is unknown and we propose an estimation method based on the shape of the signal. We use the persistent homology of sublevel sets of the signal to capture the temporal structure of its local extrema. We infer the number of periods in the signal by counting points in the persistence diagram and their multiplicities. Using the estimated number of periods, we construct an estimator of the reparametrisation. It is based on counting the number of sufficiently prominent local minima in the signal. This work is motivated by a vehicle positioning problem, on which we evaluated the proposed method.
BibTeX
@misc{bonis2022topological,
  author = {Thomas Bonis and Frédéric Chazal and Bertrand Michel and Wojciech Reise},
  title = {{Topological phase estimation method for reparameterized periodic functions}},
  year = {2022},
  eprint = {2205.14390},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2205.14390},
}
Figure