Daubechies wavelet filter computation


W = dbaux(N,SUMW)
W = dbaux(N)
W = dbaux(N,0)


W = dbaux(N,SUMW) is the order N Daubechies scaling filter such that sum(W) = SUMW. Possible values for N are 1, 2, 3, ...

    Note   Instability may occur when N is too large.

W = dbaux(N) is equivalent to W = dbaux(N,1)

W = dbaux(N,0) is equivalent to W = dbaux(N,1)

Daubechies' Extremal Phase Scaling Filter with Specified Sum

This example shows to determine the Daubechies' extremal phase scaling filter with a specified sum. The two most common values for the sum are $\sqrt{2}$ and 1.

Construct two versions of the 'db4' scaling filter. One scaling filter sums to $\sqrt{2}$ and the other version sums to 1.

NumVanishingMoments = 4;
h = dbaux(NumVanishingMoments,sqrt(2));
m0 = dbaux(NumVanishingMoments,1);

The filter with sum equal to $\sqrt{2}$ is the synthesis (reconstruction) filter returned by wfilters and used in the discrete wavelet transform.

[LoD,HiD,LoR,HiR] = wfilters('db4');
ans =


For orthogonal wavelets, the analysis (decomposition) filter is the time-reverse of the synthesis filter.

ans =



The computation of the dbN Daubechies scaling filter requires the extraction of the roots of a polynomial of order 4N. Instability may occur when N is too large.

More About

expand all


The algorithm used is based on a result obtained by Shensa (see "References"), showing a correspondence between the "Lagrange à trous" filters and the convolutional squares of the Daubechies wavelet filters.

The computation of the order N Daubechies scaling filter w proceeds in two steps: compute a "Lagrange à trous" filter P, and extract a square root. More precisely:

  • P the associated "Lagrange à trous" filter is a symmetric filter of length 4N-1. P is defined by

    P = [a(N) 0 a(N-1) 0 ... 0 a(1) 1 a(1) 0 a(2) 0 ... 0 a(N)]

  • where

  • Then, if w denotes dbN Daubechies scaling filter of sum , w is a square root of P:

        P = conv(wrev(w),w) where w is a filter of length 2N.

    The corresponding polynomial has N zeros located at −1 and N−1 zeros less than 1 in modulus.

Note that other methods can be used; see various solutions of the spectral factorization problem in Strang-Nguyen (p. 157).


Daubechies, I. (1992), Ten lectures on wavelets, CBMS-NSF conference series in applied mathematics, SIAM Ed.

Shensa, M.J. (1992), "The discrete wavelet transform: wedding the a trous and Mallat Algorithms," IEEE Trans. on Signal Processing, vol. 40, 10, pp. 2464-2482.

Strang, G.; T. Nguyen (1996), Wavelets and Filter Banks, Wellesley-Cambridge Press.

See Also


Was this topic helpful?