Search All of the Math Forum:
Views expressed in these public forums are not endorsed by
Drexel University or The Math Forum.
|
|
|
|
Re: DWT decomposition level of EEG
Posted:
Feb 28, 2013 1:55 PM
|
|
On Thursday, February 28, 2013 10:32:08 PM UTC+13, Riheen wrote:
> TideMan <mulgor@gmail.com> wrote in message <fefb80f6-c00b-4fdc-bba3-f0a63ae8a64d@googlegroups.com>...
>
> > On Thursday, February 28, 2013 9:30:12 AM UTC+13, Riheen wrote:
>
> > > Hi ,
>
> > >
>
> > > I am doing EEG signal processing. Suppose, my data is sampled with 500Hz. So, according to Nyquist theorem there is 250 hz spectrum. I filtered the signal 0-64 Hz. I need 0-8Hz band. Now how much decomposition level is required?? 3 or 5??
>
> >
>
> > I find it easier to work in timescales, not frequencies:
>
> > dt=1/500; seconds
>
> > Level 1: 2*dt
>
> > Level 2: 2^2*dt
>
> > Level 3: 2^3*dt
>
> > etc
>
> >
>
> > Then,
>
> > 8 Hz => 1/8 s = 2^n*dt
>
> > 2^n=500/8
>
> > n=log2(500/8)=5.9
>
> > so you need to decompose to level 6, and the approximation will be 0 to 8Hz (approximately).
>
> >
>
> > Note, however, that these timescales are not exact.
>
> > In fact for ocean waves and using mother wavelet 'db5' (which fits ocean waves very nicely), I find (from zero crossing analysis) that the times scales go like:
>
> > Level 1: 1.5*2*dt
>
> > Level 2: 1.5*2^2*dt
>
> > Level 3: 1.5*2^3*dt
>
> > etc
>
> > whence:
>
> > n=log2(500/8/1.5)=5.4
>
>
>
> Hi TideMan,
>
> thanks for tour reply. you said i need level 6. But i told that, i filtered the signal 0-64 Hz . I think, now i need only 3 levels for 0-8 Hz coefficients. Here it is-
>
> 1st level: detail(33-63), approx.(0-32)
>
> 2nd level: detail(17-32), approx.(1-16)
>
> 3rd level: detail(9-16), approx(0-8)
>
>
>
> Here, you can see after 3 levels i can get 0 to 8 hz approximate coefficients. My data sampling frequency is 500Hz.
>
> Is there any problem in my concept?? If so please explain.
When you "filtered", did you also decimate?
If not, then what I wrote applies.
If you did, then the sampling frequency is not 500 Hz, but something else: 64 Hz perhaps?
BTW, there's no need to filter before wavelet decomposition.
Wavelet decomposition does a fine job of that itself.
|
|
|
|
