Forums

Wavelet filters

Started by DWT February 3, 2009
Hello,

Does anyone know the relation between the signal and mother wavelet
function? In other words, what wavelet (db2, db6, sym,...) would better
decompose the signal into subbands. 

Thanks
On 3 Feb, 14:44, "DWT" <zwfilte...@yahoo.com> wrote:
> Hello, > > Does anyone know the relation between the signal and mother wavelet > function? In other words, what wavelet (db2, db6, sym,...) would better > decompose the signal into subbands.
As far as I know, all the wavelet bases are complete. So you can choose any one of them. It's up to you to decide if any one of them is particularly beneficial (or unsiutable) for your particular purpose. Rune
>As far as I know, all the wavelet bases are complete. >So you can choose any one of them. It's up to you to >decide if any one of them is particularly beneficial >(or unsiutable) for your particular purpose. > >Rune
God morgen Rune, Thanks for the reply. I'm working on the simple speech denoising module (orthonormal DWT subband thresholding). I'm using Daubechies family (inc. Symmlet & Coiflet) wavelets only. Based on the experiments in Matlab, I've found that all of them are showing quite similar results (+/- 0.03dB SNR) after denoising. On the other hand, there are numerous papers discussing the best wavelet packet (WP) basis selection taking in account the signal's entropy. E.g. http://www.eurasip.di.uoa.gr/eurasip/Proceedings/Eusipco/Eusipco2000/sessions/FriAm/PO1/cr1567.pdf I'm using simple filter-bank (not splitting the detailed coefficients). Can I uses the entropy estimation for my DWT wavelet selection? Serge
On 4 Feb, 08:22, "DWT" <zwfilte...@yahoo.com> wrote:

> I'm using simple filter-bank (not splitting the detailed coefficients). > Can I uses the entropy estimation for my DWT wavelet selection?
Of course you can. You can select them by playing spin-the-bottle or reading Tarot cards as well. It's up to you, the analyst or system designer, to choose how to proceed. But it's also up to you to argue in favour of your choise (so spin-the-bottle or Tarots might not be the best method, if you want to stay in favour with your bosses...). If you have several alternatives that perform more or less similar with the data, then other constraints come into consideration: - Are there significant differences in computation time? - Are there significant differences in storage? - Are any approaches in more wide-spread use than others? - Are any methods faster or easier to implement? and so on. If you start out in a field, it's usually a good idea to first implement what everybody else do, both to quickly get an overview of the field and to check implementations. Once you have done that, you can start replacing components in the system, by testing different wavelets, including entropy- based selectors, order estimators (if applicable) etc. The main idea is to start out wit the simple straight-forward stuff, that is easy to understand and quick to implement. Once you have done that, you can start playing with alternatives. Rune
On Feb 4, 3:50&#2013266080;am, Rune Allnor <all...@tele.ntnu.no> wrote:
> On 4 Feb, 08:22, "DWT" <zwfilte...@yahoo.com> wrote: > > > I'm using simple filter-bank (not splitting the detailed coefficients). > > Can I uses the entropy estimation for my DWT wavelet selection? > > Of course you can. You can select them by playing spin-the-bottle > or reading Tarot cards as well. > > It's up to you, the analyst or system designer, to choose how to > proceed. But it's also up to you to argue in favour of your choise > (so spin-the-bottle or Tarots might not be the best method, > if you want to stay in favour with your bosses...). > > If you have several alternatives that perform more or less similar > with the data, then other constraints come into consideration: > > - Are there significant differences in computation time? > - Are there significant differences in storage? > - Are any approaches in more wide-spread use than others? > - Are any methods faster or easier to implement? > > and so on. If you start out in a field, it's usually a good > idea to first implement what everybody else do, both to quickly > get an overview of the field and to check implementations. > > Once you have done that, you can start replacing components > in the system, by testing different wavelets, including entropy- > based selectors, order estimators (if applicable) etc. > > The main idea is to start out wit the simple straight-forward > stuff, that is easy to understand and quick to implement. > Once you have done that, you can start playing with alternatives. > > Rune
Rune Are you going to the signal processing conference in Oslo this July? Maurice Givens
On 4 Feb, 19:19, maury <maury...@core.com> wrote:

> Rune > Are you going to the signal processing conference in Oslo this July?
Would this be the NORSIG? No. I have no plans about that. It's been a long time since I went to conferences. I used to get a few 8-10 days 'vacations' in return for giving 15-minute lectures back when I was a student, but I quit doing that after I had my dissertation. Rune