What is the meaning of wavelet coefficients and scaling coefficients? E.g. for a sequence I obtained the following wavelet coeffients.

enter image description here

How am I supposed to interpret them? I used wavelets package in R. The sequence concers price time series of 21 days. I used the Haar wavelet.

  • $\begingroup$ The Haar wavelet yields a kind of crude derivative at different scales. Here, there is only 18 coefficients, not enough to be a genuine invertible wavelet transformation. Why are you studying such short data, and why with Haar wavelet? $\endgroup$ – Laurent Duval Nov 6 '19 at 20:02

Hope this helps:...

The fast wavelet transform (FWT) is a mathematical transform which results in a series of orthogonal coefficients of varying resolution.

Assume an input signal or time series $S$ of length $n=2^{j}$. The length of $S$ must be equal to some power of 2, that is, $n$ equal to 8, 16, 32, 64, 128, 256, 512, 1024, etc. To transform a signal into wavelets, at each resolution the signal is convolved using a series of scaling coefficients, $a_0$, $a_1$, $a_2$, $\ldots$ representing a low pass filter and a sequence of $d_0$, $d_1$, $d_2$, $\ldots$ of a high-pass filter.

Scaling function coefficients, $\phi$. At the $j-1$th resolution, the convolution is taken over the signal's original length, $2^j$, in the form \begin{equation} \phi_k = a_{-1}s_{2k+1} + a_0 s_{2k}, \end{equation} where $k=0,1,2,\ldots,2^{j-1}-1$ are the indices of each scaling function, $s_0, s_1, \ldots, s_{2^{j}-1}$ are the original input signal elements, and $a_{-1}$ and $a_0$ are constants from the Haar scaling function. An example of expanding the notation for the $j-1$ resolution is \begin{equation} \begin{split} \phi_0 &= a_{-1}s_1 + a_0 s_0\\ \phi_1 &= a_{-1}s_3 + a_0 s_2\\ \vdots\\ \phi_{2^{j-1}-1} &= a_{-1}s_{(2^j-1)} + a_0 s_{(2^j-2)}.\\ \end{split} \end{equation} Consider a signal $S$ of length $n=1024=2^{10}$ $(j=10)$, and assume Haar scaling function values of $a_{-1}=a_0=0.5$. Although the length of $S$ is 1024, the notation used to represent the series of values is $s_0$, $s_1$, $s_2$, $\ldots$, $s_{1023}$. The first convolution of $S$ will result in 512 ($2^{10-1}=2^9$) scaling function coefficients, which are obtained using the relationship \begin{equation} \phi_k = 0.5 s_{2k+1} + 0.5 s_{2k} \end{equation} where $k=1,2,\ldots,511$. This translates to \begin{equation} \begin{split} \phi_0 &= 0.5 s_{1} + 0.5 s_{0}\\ \phi_1 &= 0.5 s_{3} + 0.5 s_{2}\\ \vdots\\ \phi_{511} &= 0.5 s_{1023} + 0.5 s_{1022}\\ \end{split} \end{equation} It is easily noticed that the scaling function coefficients are the average of each neighboring pair of signal elements.

Wavelet coefficients, $\psi$. The wavelet coefficient is essentially based on the difference between each neighboring pair of signal elements. At the $j-1$ resolution, these are \begin{equation} \begin{split} \psi_0 &= -d_{-1}s_1 + d_0 s_0\\ \psi_1 &= -d_{-1}s_3 + d_0 s_2\\ \vdots\\ \psi_{2^{j-1}-1} &= -d_{-1}s_{(2^{j-1}-1)} + d_0 s_{(2^{j-1}-2)}\\ \end{split} \end{equation} where $d_{-1}$ and $d_0$ are both 0.5 based on the Haar wavelet.

For our first convolution of signal $S$ of length 1024 at resolution $j-1$, the 512 wavelet coefficients are \begin{equation} \begin{split} \psi_0 &= -0.5 s_{1} + 0.5 s_{0}\\ \psi_1 &= -0.5 s_{3} + 0.5 s_{2}\\ \vdots\\ \psi_{511} &= -0.5 s_{1023} + 0.5 s_{1022}\\ \end{split} \end{equation}


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.