Re: non gaussian noise

[Robert Malek]

[...]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

[santosh nath]

[email protected] (Robert Malek) wrote in message news:<[email protected]>...
> [...]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.

I am again lost! your comment "correlation between noise and signal"
made me to think what is called colored noise(signal). I don't know if
they can be called in a different way. I guess that follows to evalute
the covariance matrix between noise and the signal. The offdiagonal
elemenets of the matrix becomes zero if the signal becomes white
otherwise the signal remains colored. At least "Prewhitening" is done
to remove the colorness from the signal. As I mentioned it is very
important for the signal to be white for efficient receiver
processing.

Any pointer is appriciated if I am missing somewhere.
Santosh.

>
> > > 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

[Jerry Avins]

santosh nath wrote:
>
...
>
> I am again lost! your comment "correlation between noise and signal"
> made me to think what is called colored noise(signal).
...

I don't think so. For a hint, look at
http://ccms.ntu.edu.tw/~karchung/Phonetics%20II%20page%20sixteen.htm

There are more comprehensive lists, but not at my fingertips.

Jerry
--
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

[SJ (James) Kim]

santosh nath wrote:
> ... As I mentioned it is very
> important for the signal to be white for efficient receiver
> processing.

That's the really important point, which I had also wanted to emphasis
on my below posting.

However, I'm so wondering wheather multiuser interfere signal is also
the AWGN like signal in communication environments. For instance, the
received signal r(t) which has the interfere signal i(t) as well as the
desired signal d(t) and the plain noise signal n(t) is generally denoted
as below:

r(t) = d(t) + i(t) + n(t).

In summary, I agreed with the importance of the noise whitening, and I
asked one question about the property of the interfere signal, which is
treated as one of the essential impairments of the wireless
communications.
--
Direct access to this group with http://web2news.com
http://web2news.com/?comp.dsp