Abstract
We observe a large number of signals, all of them with identical, although unknown, shape, but with a different random shift. The objective is to estimate the individual time shifts and their distribution. Such an objective appears in several biological applications like neuroscience or ECG signal processing, in which the estimation of the distribution of the elapsed time between repetitive pulses with a possibly low signal-noise ratio, and without a knowledge of the pulse shape is of interest. We suggest an M-estimator leading to a three-stage algorithm: we first split our data set in blocks, then the shift estimation in each block is done by minimizing a cost function based on the periodogram; the estimated shifts are eventually plugged into a standard density estimator. We show that under mild regularity assumptions the density estimate converges weakly to the true shift distribution. The theory is applied both to simulations and to alignment of real ECG signals. The proposed approach outperforms the standard methods for curve alignment and shift density estimation, even in the case of low signal-to-noise ratio, and is robust to numerous perturbations common in ECG signals.
Introduction
In many biological fields, researchers observe repetitive signals—such as heartbeats (ECG) or neural spikes—that share a similar underlying morphology but occur at slightly different times. To analyze these signals effectively, they must be "aligned" or registered. A significant challenge in this process is the "template problem": to align signals perfectly, one typically needs a template of the true shape, but to find that true shape, the signals must first be aligned.
Standard methods often fail when the signal-to-noise ratio (SNR) is low or when the underlying shape is completely unknown. This paper introduces a robust semiparametric method to align large sets of signals by leveraging shift-invariant spectral properties, allowing for the accurate estimation of both individual shifts and their overall probability distribution without prior knowledge of the signal's shape.
Methodology
The authors propose a semiparametric M-estimator that treats the unknown signal shape as a "nuisance parameter." Their approach follows a three-stage algorithm:
- Block Splitting and Reference Alignment: The dataset is divided into smaller blocks. To ensure the alignment stays anchored to a common baseline and to address the problem of identifiability (where shifts are only defined relative to one another), a "reference" signal $y_0(t)$ is included in every block.
- Spectral Cost Function: Instead of comparing signals in the time domain, the algorithm works in the frequency domain. It minimizes a cost function based on the Energy Spectral Density (ESD). Because the power spectrum of a signal is invariant to time shifts, the authors can compare the average of individual spectra against the spectrum of the averaged signal. If the signals are perfectly aligned, these two values should match.
- Density Estimation: Once the individual shifts are estimated, they are plugged into a standard kernel density estimator to reconstruct the probability density function (PDF) of the shifts.
Experimental Results
The researchers validated their model using two primary sources: simulated neural responses generated via the Hodgkin-Huxley model and real-world ECG signals obtained from the Hadassah Ein-Karem Hospital.
Key findings include:
- Superiority over Standard Methods: In simulations, the proposed method consistently achieved a lower Mean Integrated Squared Error (MISE) than standard Functional Data Analysis (FDA) techniques, especially as noise increased.
- ECG Clarity: When applied to real ECG data, the method produced a much cleaner "average heartbeat." Features like the P-wave, QRS complex, and T-wave became significantly more visible and distinct compared to standard alignment methods.
- Robustness: The algorithm proved remarkably resilient to common ECG "noise," including baseline wander (low-frequency drift from breathing) and power-line interference (50/60 Hz hum).
Conclusion
By moving to the spectral domain, the proposed method bypasses the need to estimate signal shape simultaneously with shifts, simplifying the optimization problem significantly. While the model assumes an identical underlying shape—an assumption that may be challenged by certain medical conditions—it provides a mathematically rigorous and robust framework for signal registration in high-noise biological environments. Future research may focus on data-driven bandwidth selection for the density estimator to further improve convergence rates.
Summary updated