The discrete wavelet transform is a bank filter actually, I guess that you may first look at the spectrum of your signal and focus on the frequency region you want to save or suppress. Then, you may convert this frequency band to the wavelet octave scale (2^j). The design of your filter bank may help.
Choosing coefficients for wavelet filter is a broad area.
One of the way I followed is to devise a basis function and there by deriving the coefficients.
For instance, consider a biorthogonal wavelet. It has a pair of filters at decomposition and reconstruction.
1. Here, design procedure is first to choose the length of the filter.
2. Then, the properties which our wavelet should posses.
3. After choosing the properties, express these properties in terms of the filter coefficient of our chosen length.
4. Now chose a basis function or modify an existing basis function to get the coefficients for one set of filters. The standard biorthogonal wavelet use spline function as the basis function.
5. After substituting the values of the coefficients from the chosen basis function, we will get few equations with the same number of unknown variables which are the filter coefficients of other set of filters.
6. Solve these equation to get the second set.
This completes the design of a wavelet.
A project is going on similar work. So, I could not reveal more that this. Anyway, if you have any more queries, on these steps, you are welcome.