Empirical Wavelet Transform
Also known as: EWT, Empirical wavelets
The empirical wavelet transform (EWT) is a data-driven wavelet decomposition method that automatically defines wavelet bases adapted to the frequency content of the signal. Introduced by Jérémie Gilles (2013), it overcomes a key limitation of classical wavelets—which use fixed, predefined bases—by constructing custom wavelets from the signal's own spectrum. This adaptive approach is particularly effective for analyzing non-stationary signals with complex, multi-component structures.
Key highlights
- Data-driven basis construction adapts perfectly to the signal's frequency content
- Provides sharp frequency localization by design, avoiding cross-component interference
- Produces interpretable modes corresponding to physically meaningful oscillations or phenomena
- Works well for signals with well-separated frequency bands
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Apply EWT when analyzing non-stationary signals with multiple, well-separated frequency components. It is ideal for mechanical vibration analysis, power system monitoring, biomedical signals, and seismic data where fixed wavelet bases are inadequate. Use EWT when you need an interpretable, multi-scale decomposition that adapts to the signal characteristics.
Strengths & limitations
- Data-driven basis construction adapts perfectly to the signal's frequency content
- Provides sharp frequency localization by design, avoiding cross-component interference
- Produces interpretable modes corresponding to physically meaningful oscillations or phenomena
- Works well for signals with well-separated frequency bands
- Requires clear frequency separation in the spectrum—overlapping frequency bands lead to ambiguous mode definitions
- The initial spectrum segmentation is heuristic-based and can be sensitive to noise or spectral estimation errors
- Computational cost is higher than fixed-wavelet transforms due to custom basis construction
- Performance degrades significantly in low-signal-to-noise ratio scenarios
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How does EWT differ from empirical mode decomposition (EMD)?
EMD uses an iterative sifting process to extract modes based on extrema, which is computationally expensive and prone to mode-mixing. EWT is faster, data-driven in the frequency domain, and produces components that align with true frequency bands in the spectrum.
What happens if frequency bands overlap significantly?
EWT's segmentation algorithm will struggle to clearly separate overlapping bands, potentially merging them into single modes or creating spurious modes. For such signals, consider preprocessing to isolate frequency bands first, or use variational mode decomposition (VMD) instead.
Can EWT handle signals with time-varying frequency content?
Standard EWT assumes stationary frequency structure. For time-varying content, you may decompose the signal into windows and apply EWT locally, or use adaptive variants that re-estimate the frequency segmentation over time.
How sensitive is EWT to the initial spectrum segmentation?
EWT automatically detects spectrum peaks and segments between them, but the quality depends on spectral estimation accuracy. Use high-resolution spectral methods (e.g., multitaper, Welch's method) and consider denoising the spectrum if noise is present.
Is the reconstruction from EWT modes always perfect?
Yes, EWT is designed to be perfectly invertible—the sum of empirical wavelet coefficients reconstructs the original signal exactly (up to numerical precision). This invertibility is a key advantage over some heuristic decomposition methods.
Sources
- 1.Gilles, J. (2013). Empirical wavelet transform. IEEE Transactions on Signal Processing, 61(16), 3999–4010.
- 2.Gilles, J. (2015). Empirical wavelet transform for multiscale analysis of signals. IEEE Signal Processing Magazine, 32(6), 125–130.
- 3.Dragomiretskiy, K., & Zosso, D. (2014). Variational mode decomposition. IEEE Transactions on Signal Processing, 62(3), 531–544.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Empirical Wavelet Transform. ScholarGate. https://scholargate.app/time-series/empirical-wavelet-transform