From: Randy Yates on
Randy Yates <yates(a)ieee.org> writes:
> [...]
> My point wasn't that such tasks don't exist, but that, in general (e.g.,
> without assuming such a task), the SNR of a signal that has been
> dithered is actually MORE than the SNR of a signal that has not been
> dithered.

Bah! Correction:

...the SNR of a signal that has been dithered is actually LESS than
the SNR of a signal that has not been dithered.

and for

> There are many signal processing tasks where only the longish term
> result really matters. In those cases the average effect across many
> samples is usually all you need to see a huge win from dithering.

the correction:

...is usually well-worth the slight DECREASE in overall SNR.

--Randy

(Obviously I was thinking of the noise power, not the SNR...)

--
Randy Yates % "The dreamer, the unwoken fool -
Digital Signal Labs % in dreams, no pain will kiss the brow..."
mailto://yates(a)ieee.org %
http://www.digitalsignallabs.com % 'Eldorado Overture', *Eldorado*, ELO
From: Randy Yates on
<begin red-faced demeanor>

Randy Yates <yates(a)ieee.org> writes:
> [...]

This is misquoted:

> and for
>
>> There are many signal processing tasks where only the longish term
>> result really matters. In those cases the average effect across many
>> samples is usually all you need to see a huge win from dithering.
>
> the correction:
>
> ...is usually well-worth the slight DECREASE in overall SNR.


The correction is this:

and for

That's not to say that the dither was bad - the decorrelation, or
more intuitively, the removal of a bunch of "spikes" in the
frequency spectrum (and the corresponding increase in SFDR), is
usually well-worth the slight increase in overall SNR.

the correction:

...is usually well-worth the slight DECREASE in overall SNR.

<end red-faced demeanor>
<begin search for more coffee...>
--
Randy Yates % "The dreamer, the unwoken fool -
Digital Signal Labs % in dreams, no pain will kiss the brow..."
mailto://yates(a)ieee.org %
http://www.digitalsignallabs.com % 'Eldorado Overture', *Eldorado*, ELO
From: Greg Heath on
On Nov 8, 3:23 am, Greg Heath <he...(a)alumni.brown.edu> wrote:
> On Nov 7, 5:16 pm, Jerry Avins <j...(a)ieee.org> wrote:
>
>
>
>
>
> > Greg Heath wrote:
> > > On Nov 5, 1:04 am, "BCLIM" <boonchun_...(a)yahoo.com> wrote:
> > >> Hi,
> > >>    I have a doubt on jitter generation. Wonder is the method to generate
> > >> jitter signal is that same as dithering generation method? Understand that
> > >> both are usign random signal generation with different distribution. For
> > >> example, triangular, rectangular etc.
>
> > > I always thought that for a signal defined over a finite length of
> > > time,
> > > jittering added a randomness to the amplitude whereas dithering
> > > added a randomness to the starting time.
>
> > > Hope this helps.
>
> > It probably hurts. Jitter randomizes the timing of pulse edges. When you
> > look at such a signal on an oscilloscope, the trace jitters.
>
>  Given your additional comment. it helps emphasize that
> terminology is application dependent.
>
> In the fields of classification and regression, the terminology
> is as I have indicated.

Unfortunate circumstance. From the comp.ai.neural-neta FAQ:

ftp://ftp.sas.com/pub/neural/FAQ3.html#A_jitter

Hope this helps.

Greg

From: wazerface on
Wow good dithering discussion!

Yes, using a pure energy metric | |^2, dithering will always increase
noise power and decrease overall SNR. But whitening often eases
detection / estimation / human audio perception.

The real discussion is, what is the best way to choose quantization
levels for representing a given signal? P. T. Boufounos and A. V.
Oppenheim, "Quantization noise shaping on arbitrary frame expansion,"
investigates this in general.

-Martin
From: dbd on
On Nov 9, 5:33 am, Greg Heath <he...(a)alumni.brown.edu> wrote:
> On Nov 8, 3:23 am, Greg Heath <he...(a)alumni.brown.edu> wrote:
>
>
>
> > On Nov 7, 5:16 pm, Jerry Avins <j...(a)ieee.org> wrote:
>
> > > Greg Heath wrote:
> > > > On Nov 5, 1:04 am, "BCLIM" <boonchun_...(a)yahoo.com> wrote:
> > > >> Hi,
> > > >>    I have a doubt on jitter generation. Wonder is the method to generate
> > > >> jitter signal is that same as dithering generation method? Understand that
> > > >> both are usign random signal generation with different distribution. For
> > > >> example, triangular, rectangular etc.
>
> > > > I always thought that for a signal defined over a finite length of
> > > > time,
> > > > jittering added a randomness to the amplitude whereas dithering
> > > > added a randomness to the starting time.
>
> > > > Hope this helps.
>
> > > It probably hurts. Jitter randomizes the timing of pulse edges. When you
> > > look at such a signal on an oscilloscope, the trace jitters.
>
> >  Given your additional comment. it helps emphasize that
> > terminology is application dependent.
>
> > In the fields of classification and regression, the terminology
> > is as I have indicated.
>
> Unfortunate circumstance. From the comp.ai.neural-neta FAQ:
>
>  ftp://ftp.sas.com/pub/neural/FAQ3.html#A_jitter
>
> Hope this helps.
>
> Greg

So there exist neural net folks who made sloppy use of signal
processing terminology too. Is that supposed to be a surprise to
anyone?

Dale B. Dalrymple
First  |  Prev  |  Next  |  Last
Pages: 1 2 3 4 5 6
Prev: STUPIDENT::Re: Floting Point Saturation
Next: MPEG-Xtreme