Time-delay estimation using the Wigner-Ville distribution

This paper proposes a time-delay estimation method based on the Wigner-Ville Distribution (WVD) that outperforms traditional linear time-frequency approaches like the continuous wavelet transform by achieving higher accuracy and lower uncertainty, particularly for nonstationary signals in energetic frequency bands.

The Gist

Imagine you are trying to figure out exactly how long it takes for a sound to travel from a speaker to a listener in a very noisy, echoey room. This is the core problem of time-delay estimation. Scientists need to do this all the time: to find earthquakes, to listen to gravitational waves from black holes, or to see what's happening deep underground for oil exploration.

The problem is that signals (like sound or seismic waves) aren't perfect, steady beeps. They are messy, changing, and "non-stationary." They change their pitch and volume as they travel through different materials.

This paper introduces a new, sharper way to measure that travel time. Here is the breakdown in simple terms:

The Old Way: The "Blurry Flashlight" (CWT)

For a long time, scientists used a method called the Continuous Wavelet Transform (CWT).

• The Analogy: Imagine you are trying to read a sign in the dark using a flashlight with a very wide, soft beam. You can see the general shape of the sign, but the edges are blurry.
• How it works: This method looks at the signal by sliding a "wave" (a wavelet) over it. It's great at seeing when things happen, but because the flashlight beam is wide, it smears the details. It's hard to tell exactly what frequency (pitch) is happening at that exact moment.
• The Flaw: To make the picture clearer, scientists have to "smooth out" the blurry edges. But this smoothing is like putting a filter on a camera; it hides the tiny, important details and can sometimes make the signal look more connected than it really is.

The New Way: The "High-Definition Laser" (WVD)

The authors propose using the Wigner-Ville Distribution (WVD).

• The Analogy: Instead of a wide flashlight, imagine a laser pointer that is incredibly sharp. It can pinpoint the exact location and the exact color (frequency) of a dot on a wall simultaneously.
• How it works: This method doesn't use a sliding wave. Instead, it looks at the signal's energy in a way that respects the fundamental laws of physics (the Heisenberg uncertainty principle) to get the sharpest possible picture of time and frequency together.
• The Benefit: It doesn't need to "smooth" the data. It naturally sees the signal's true structure. It's like switching from a fuzzy sketch to a 4K photograph.

The "Ghost" Problem

There is one catch with the laser pointer (WVD). Because it is so sharp and mathematical, if you have two different notes playing at once, the math can create "ghosts" or "echoes" in the picture that aren't actually there. These are called cross-terms.

• The Fix: The authors developed a clever trick. They look at the signal in a special "shadow realm" (called the ambiguity domain) where the real signal is right in the center, and the "ghosts" are far away in the corners. They simply cut off the corners (filtering), keep the center, and then turn it back into a clear picture.

The Test Drive

The team tested both methods on two scenarios:

1. The "Static" Test: A signal traveling through a medium with random bumps (like a bumpy road).
   • Result: The old method (CWT) got the timing wrong at the end of the signal because it got confused by the noise. The new method (WVD) nailed it, especially in the loudest parts of the signal.
2. The "Twisting" Test: A signal where the delay changes in a complex, non-linear way (like a rubber band stretching and snapping).
   • Result: The old method smoothed out the twists, making it look like a gentle curve. The new method saw the sharp, sudden changes perfectly.

The Bottom Line

Think of the old method as trying to guess the speed of a race car by looking at a long-exposure photo (it's blurry, but you can see the general motion). The new method is like taking a high-speed burst photo that freezes the car in perfect detail, showing exactly where it is and how fast it's going at every split second.

Why does this matter?
If you are a doctor trying to locate a tumor, or a geologist trying to find oil, or a physicist listening to the universe, you need to know exactly when a signal arrived. This new method gives you a sharper, more accurate map of time and frequency, helping you make better decisions without needing to guess or smooth over the messy details.

Technical Summary

1. Problem Statement

Accurate time-delay estimation (TDE) between signals recorded by different sensors is critical in various physics applications, including gravitational wave detection, seismic inversion, and dynamical systems analysis.

• Limitations of Linear Methods: Traditional linear time-frequency representations, such as the Continuous Wavelet Transform (CWT), are widely used to compute cross-spectra for TDE. However, they suffer from inherent limitations:
  • Resolution Trade-off: The frequency resolution is constrained by the localized nature of the convolving wavelet (Gabor-Heisenberg uncertainty).
  • Improper Correlation Measure: The functional form of the CWT cross-spectrum is not a true correlation measure, often resulting in spectral leakage and requiring post-processing smoothing to reduce wavelet influence.
  • Window Sensitivity: Methods relying on windowed cross-correlation are sensitive to window length choices, potentially missing cycle-skipping effects or nonlinear distortions.
• Limitations of Non-Linear Methods: Techniques like Dynamic Time Warping (DTW) handle non-homogeneous media well but fail to provide a complete time-frequency delay map, lacking spectral details necessary for physical interpretation.
• The Gap: There is a need for a method that achieves joint time-frequency resolution approaching the theoretical limit while providing a robust, physically consistent measure of signal similarity without heavy reliance on user-defined smoothing.

2. Methodology

The authors propose a TDE method based on the Wigner-Ville Distribution (WVD), a quadratic time-frequency representation, and compare it against the standard CWT approach.

A. Theoretical Framework

• CWT Approach:
  • Uses a complex Morlet mother wavelet.
  • Computes a normalized cross-spectrum in the time-frequency domain.
  • Calculates time delay ($\delta t$) from the phase shift of the cross-spectrum.
  • Drawback: Requires smoothing operators to mitigate wavelet artifacts and spectral leakage.
• WVD Approach:
  • Utilizes the WVD, defined via the analytical signal $z(t)$ (Hilbert transform of the real signal).
  • Cross-Spectrum: Defines a cross-WVD ($WVD_{xy}$) and a normalized phase shift.
  • Advantage: The WVD provides a natural correlative interpretation and approaches the Gabor-Heisenberg limit for resolution.
  • Cross-Term Mitigation: A known issue with WVD is the presence of non-physical "cross-terms" (interference artifacts) in multi-component signals. The authors address this by:
    1. Computing the Ambiguity Function (AF) for the signals.
    2. Applying a 2D low-pass filter in the ambiguity domain to suppress cross-terms (which are located far from the origin) while retaining auto-terms (concentrated near the origin).
    3. Transforming the filtered AF back to the WVD domain to obtain a clean time-frequency representation.

B. Numerical Experiments

The study validates the method using two distinct wave-physics scenarios:

1. Example 1 (Stochastic Media): A 2D acoustic model with a stochastic velocity field (von Kármán correlation) inducing linear time delays. The signals exhibit coda waves and scattering.
2. Example 2 (Heterogeneous Media): A cropped section of the Marmousi model (a standard seismic benchmark) with a deterministic velocity gradient. The current signal is generated by applying a nonlinear time delay (Gaussian perturbation) to the reference signal, simulating complex medium heterogeneity.

3. Key Contributions

• Novel Application: This is the first study to employ the Wigner-Ville Distribution specifically for estimating time delays in nonstationary wave phenomena.
• Quadratic Representation: Demonstrates that quadratic representations (WVD) offer superior joint time-frequency resolution compared to linear representations (CWT) for TDE.
• Artifact Suppression Strategy: Proposes a specific workflow using the Ambiguity Function to effectively separate auto-terms from cross-terms, making WVD viable for complex, multi-component signals.
• Physical Consistency: Shows that WVD-derived time delays are physically consistent with wave dynamics without requiring the ad-hoc smoothing operators often necessary for CWT.

4. Results

The comparison between CWT and WVD yielded the following findings:

• Accuracy in Energy Bands:
  • In Example 1, the CWT overestimated cross-spectrum energy due to spectral leakage, leading to an underestimation of time delays in low-energy, high-frequency regions. The WVD accurately captured lower-energy features and provided more stable delay estimates.
  • In Example 2 (nonlinear delay), the CWT produced artificially smoothed time-delay maps, failing to resolve sharp transitions (e.g., near 0.9 s). The WVD successfully captured the timing of these transitions and the oscillating nature of the nonlinear delay.
• Uncertainty and Stability:
  • Marginalization of time-delay maps across frequency bands showed that WVD yields lower uncertainty (smaller standard deviation) compared to CWT.
  • The WVD confidence intervals consistently included the true time delay, whereas CWT intervals often deviated, particularly in regions of low signal energy.
• Resolution:
  • The WVD provided a clearer interpretation of the time-delay map, especially in the most energetic frequency bands.
  • While CWT resolution could be tuned by changing the wavelet's central frequency ($\omega_0$), this created a trade-off: higher $\omega_0$ gave better map resolution but poorer marginalized delay accuracy, and vice versa. WVD did not suffer from this specific trade-off to the same degree.

5. Significance

• Superior Performance: The WVD-based method outperforms the CWT in estimating time delays for nonstationary signals, particularly in scenarios involving scattering, attenuation, and nonlinear distortions.
• No Smoothing Required: The method eliminates the need for user-defined smoothing operators, which often obscure fine details in the time-frequency domain.
• Broad Applicability: The results suggest that WVD is a robust alternative framework for time-delay analysis. This has significant implications for:
  • Seismology: Improving the detection of subtle changes in Earth's heterogeneity and mitigating cycle-skipping in full-waveform inversion.
  • Gravitational Wave Physics: Enhancing the extraction of signals from space-based detectors.
  • General Wave Physics: Providing a tool for analyzing complex wave propagation where linear methods fail to resolve fine temporal-spectral features.

In conclusion, the authors establish that the Wigner-Ville Distribution, when properly filtered to remove cross-terms, offers a theoretically optimal and practically superior approach for time-delay estimation in complex, nonstationary wave environments.