Main Content


Multisignal 1-D denoising using wavelets

mswden is no longer recommended. Use wdenoise instead.


[XD,DECDEN,THRESH] = mswden('den',...)
THRESH = mswden('thr',...)
[...] = mswden(...,S_OR_H)
[...] = mswden(...,S_OR_H,KEEPAPP)
[...] = mswden(...,S_OR_H,KEEPAPP,IDXSIG)


mswden computes thresholds and, depending on the selected option, performs denoising of 1-D signals using wavelets.

[XD,DECDEN,THRESH] = mswden('den',...) returns a denoised version XD of the original multisignal matrix X, whose wavelet decomposition structure is DEC. The output XD is obtained by thresholding the wavelet coefficients, DECDEN is the wavelet decomposition associated to XD (see mdwtdec), and THRESH is the matrix of threshold values. The input METH is the name of the denoising method and PARAM is the associated parameter, if required.

Valid denoising methods METH and associated parameters PARAM are:


Principle of Stein's Unbiased Risk


Heuristic variant of the first option


Universal threshold sqrt(2*log(.))


Minimax thresholding (see thselect)

For these methods PARAM defines the multiplicative threshold rescaling:


No rescaling


Rescaling using a single estimation of level noise based on first level coefficients


Rescaling using a level dependent estimation of level noise

Penalization methods




Penal high, 2.5 ℜ≤ PARAM ℜ≤ 10


Penal medium, 1.5 ℜ≤ PARAM ℜ≤ 2.5


Penal low, 1 ℜ≤ PARAM ℜ≤ 2

PARAM is a sparsity parameter, and it should be such that: 1PARAM10. For penal method, no control is done.

Manual method


Manual method

PARAM is an NbSIG-by-NbLEV matrix or NbSIG-by-(NbLEV+1) matrix such that:

  • PARAM(i,j) is the threshold for the detail coefficients of level j for the ith signal (1jNbLEV).

  • PARAM(i,NbLEV+1) is the threshold for the approximation coefficients for the ith signal (if KEEPAPP is 0).

where NbSIG is the number of signals and NbLEV the number of levels of decomposition.

Instead of the 'den' input OPTION, you can use 'densig', 'dendec' or 'thr' OPTION to select output arguments:

[XD,THRESH] = mswden('densig',...) or [DECDEN,THRESH] = mswden('dendec',...)

THRESH = mswden('thr',...) returns the computed thresholds, but denoising is not performed.

The decomposition structure input argument DEC can be replaced by four arguments: DIRDEC, X, WNAME and LEV.

[...] = mswden(OPTION,DIRDEC,X,WNAME,LEV,METH,PARAM) before performing a denoising or computing thresholds, the multisignal matrix X is decomposed at level LEV using the wavelet WNAME, in the direction DIRDEC.

You can use three more optional inputs:

[...] = mswden(...,S_OR_H) or
[...] = mswden(...,S_OR_H,KEEPAPP) or
[...] = mswden(...,S_OR_H,KEEPAPP,IDXSIG)

  • S_OR_H ('s' or 'h') stands for soft or hard thresholding (see mswthresh for more details).

  • KEEPAPP (true or false) indicates whether to keep approximation coefficients (true) or not (false).

  • IDXSIG is a vector that contains the indices of the initial signals, or 'all'.

The defaults are, respectively, 'h', false and 'all'.


collapse all

Load the 23 channel EEG data Espiga3 [8]. The channels are arranged column-wise. The data is sampled at 200 Hz.

load Espiga3

Perform a decomposition at level 2 using the db2 wavelet.

dec = mdwtdec('c',Espiga3,2,'db2')
dec = struct with fields:
        dirDec: 'c'
         level: 2
         wname: 'db2'
    dwtFilters: [1x1 struct]
       dwtEXTM: 'sym'
      dwtShift: 0
      dataSize: [995 23]
            ca: [251x23 double]
            cd: {[499x23 double]  [251x23 double]}

Denoise the signals using the universal method of thresholding (sqtwolog) and the 'sln' threshold rescaling (with a single estimation of level noise, based on the first level coefficients).

[xd,decden,thresh] = mswden('den',dec,'sqtwolog','sln');

Plot an original signal, and the corresponding denoised signal.

idxA = 3;
hold on
grid on

Figure contains an axes object. The axes object contains 2 objects of type line. These objects represent Original, Denoised.


[1] Birgé, L., and P. Massart. “From Model Selection to Adaptive Estimation.” Festschrift for Lucien Le Cam: Research Papers in Probability and Statistics (E. Torgersen, D. Pollard, and G. Yang, eds.). New York: Springer-Verlag, 1997, pp. 55–88.

[2] DeVore, R. A., B. Jawerth, and B. J. Lucier. “Image Compression Through Wavelet Transform Coding.” IEEE Transactions on Information Theory. Vol. 38, Number 2, 1992, pp. 719–746.

[3] Donoho, D. L. “Progress in Wavelet Analysis and WVD: A Ten Minute Tour.” Progress in Wavelet Analysis and Applications (Y. Meyer, and S. Roques, eds.). Gif-sur-Yvette: Editions Frontières, 1993.

[4] Donoho, D. L., and I. M. Johnstone. “Ideal Spatial Adaptation by Wavelet Shrinkage.” Biometrika. Vol. 81, pp. 425–455, 1994.

[5] Donoho, D. L., I. M. Johnstone, G. Kerkyacharian, and D. Picard. “Wavelet Shrinkage: Asymptopia?” Journal of the Royal Statistical Society, series B, Vol. 57, No. 2, pp. 301–369, 1995.

[6] Donoho, D. L., and I. M. Johnstone. “Ideal denoising in an orthonormal basis chosen from a library of bases.” C. R. Acad. Sci. Paris, Ser. I, Vol. 319, pp. 1317–1322, 1994.

[7] Donoho, D. L. “De-noising by Soft-Thresholding.” IEEE Transactions on Information Theory. Vol. 42, Number 3, pp. 613–627, 1995.

[8] Mesa, Hector. “Adapted Wavelets for Pattern Detection.” In Progress in Pattern Recognition, Image Analysis and Applications, edited by Alberto Sanfeliu and Manuel Lazo Cortés, 3773:933–44. Berlin, Heidelberg: Springer Berlin Heidelberg, 2005.

Version History

Introduced in R2007a