Robert Malek
06-25-2003, 02:58 PM
[...]ifferent spectral segments. This can be constant (white
> > noise) or different (blue, red, pink, orange, whatever noise.)
>
> Do you mean color noise since I am not accustomed with red/blue/pink
> etc?
Yes, that's what I'm alluding ...
> > What you seem to aim at is the a "local" distribution of noise,
> > which can be very different, depending on your application. First:
> > is there a correlation between noise and signal?
>
> I guess again you try to hint colored signal which could be studied by
> covariance martix and prewhitened the signal to make it white.
No, there are systems (see immediately below!) where noise is
a function of the amplitude of the signal in the corresponding
spectral segment (a very annoying "feature", sometimes).
This is in no respect related to color or pdf.
> > In many systems (search for poisson noise) noise increases with signal
> > (a very nasty property, which is eg typical for counting devices).
[...]
> > rayleigh (a special but frequent
> > case of the above distributions)
> > And so on...
>
> I guess you try to induce different test here(Apology if I am
> wrong!).
> Validation of specific distribution can be studied by statistical
> analysis.
> For example popuplar K-S/W-S/chi square fit or Anderson Darling
> goodness of fit
> tests are excercised for validation of exploratory data to match a
> specific distribution. I do not know whether original author is
> intended to mean that or
> tries to see the Gaussianity Vs Non-Gaussianity in communication
> system. I guess
> I tried to explain little bit and wonder why you did not notice that.
> Sorry if
> I could not follow your message.
>
> Regards,
> Santosh
>
That's ok for me. If you could not follow I guess the original
poster didn't so at least as much. And this means I missed the
purpose :o(
I should add that the purpose was less to give a concise
explanation but more to provide an annotated collection of
search tags for google :o)
Robert
> > noise) or different (blue, red, pink, orange, whatever noise.)
>
> Do you mean color noise since I am not accustomed with red/blue/pink
> etc?
Yes, that's what I'm alluding ...
> > What you seem to aim at is the a "local" distribution of noise,
> > which can be very different, depending on your application. First:
> > is there a correlation between noise and signal?
>
> I guess again you try to hint colored signal which could be studied by
> covariance martix and prewhitened the signal to make it white.
No, there are systems (see immediately below!) where noise is
a function of the amplitude of the signal in the corresponding
spectral segment (a very annoying "feature", sometimes).
This is in no respect related to color or pdf.
> > In many systems (search for poisson noise) noise increases with signal
> > (a very nasty property, which is eg typical for counting devices).
[...]
> > rayleigh (a special but frequent
> > case of the above distributions)
> > And so on...
>
> I guess you try to induce different test here(Apology if I am
> wrong!).
> Validation of specific distribution can be studied by statistical
> analysis.
> For example popuplar K-S/W-S/chi square fit or Anderson Darling
> goodness of fit
> tests are excercised for validation of exploratory data to match a
> specific distribution. I do not know whether original author is
> intended to mean that or
> tries to see the Gaussianity Vs Non-Gaussianity in communication
> system. I guess
> I tried to explain little bit and wonder why you did not notice that.
> Sorry if
> I could not follow your message.
>
> Regards,
> Santosh
>
That's ok for me. If you could not follow I guess the original
poster didn't so at least as much. And this means I missed the
purpose :o(
I should add that the purpose was less to give a concise
explanation but more to provide an annotated collection of
search tags for google :o)
Robert