MODWT
Also known as: MODWT, Stationary wavelet transform, Undecimated DWT
The maximal overlap discrete wavelet transform (MODWT) is a translation-invariant wavelet decomposition method that addresses a key limitation of the standard DWT: lack of shift invariance. Introduced by Percival and Walden (1995), MODWT applies the same wavelet filters at each scale without downsampling, producing an undecimated decomposition. Each detail and approximation coefficient array maintains the full length of the input signal, enabling both robust multi-scale analysis and translation-invariant feature extraction.
Key highlights
- Translation-invariant decomposition—shifting the signal shifts the coefficients proportionally, preserving feature alignment
- Full-length coefficient arrays at each scale, preserving temporal localization
- Enables unbiased variance decomposition across scales
- Suitable for multi-scale correlation and coherence analysis between signals
- No arbitrary boundary handling or padding required
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
Use MODWT when you need translation-invariant wavelet analysis or when you require the full-length coefficient array at each scale. It is ideal for wavelet coherence and cross-covariance analysis of multiple signals, wavelet variance decomposition, and whenever alignment across scales matters. MODWT is superior to DWT when shift invariance is critical.
Strengths & limitations
- Translation-invariant decomposition—shifting the signal shifts the coefficients proportionally, preserving feature alignment
- Full-length coefficient arrays at each scale, preserving temporal localization
- Enables unbiased variance decomposition across scales
- Suitable for multi-scale correlation and coherence analysis between signals
- No arbitrary boundary handling or padding required
- Computationally more expensive than DWT—requires O(N J) operations for J decomposition levels instead of O(N)
- Produces J times more coefficients than DWT, increasing memory usage and storage requirements
- Full-length arrays can be redundant for data compression or when downsampling to coarser scales is acceptable
- Interpretation of scale-dependent features requires careful analysis of coefficient correlations
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
What is the relationship between MODWT and DWT?
DWT downsamples (decimates) coefficients at each level. MODWT keeps all samples by upsampling the filters. This makes MODWT translation-invariant and redundant, but more computationally expensive.
Why is MODWT translation-invariant?
Because filters are upsampled, not signals downsampled. If you shift the input by k samples, all coefficient arrays shift by the same k samples—there is a one-to-one correspondence. DWT, by contrast, downsamples, so shift-dependent aliasing occurs.
Can I use MODWT coefficients directly in machine learning?
You can, but beware: coefficients at the same scale are highly correlated across time and across scales. Use dimensionality reduction (PCA, feature selection) or exploit the multiresolution structure (e.g., energy per scale) to avoid multicollinearity.
How do I compute wavelet variance from MODWT?
Wavelet variance at scale j is the mean squared of the detail coefficients: Var_j = (1/N) sum(d_j^2). The sum of all scale variances equals the total variance of the signal.
Is MODWT invertible?
Yes, the sum of all detail coefficients plus the approximation coefficients reconstructs the original signal exactly. The redundancy allows perfect inversion despite the increased data.
Sources
- 1.Percival, D. B., & Walden, A. T. (1995). Wavelet Methods for Time Series Analysis. Cambridge University Press.
- 2.Percival, D. B. (2000). Wavelet methods for time series analysis. Cambridge University Press.
- 3.Whitcher, B., Guttorp, P., & Percival, D. B. (2000). Wavelet analysis of covariance with application to atmospheric time series. Journal of Geophysical Research, 105(D11), 14941–14962.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). MODWT. ScholarGate. https://scholargate.app/time-series/modwt