What is the advantage on high-sampling rate ?

"Max Hauser" <maxREMOVE@THIStdl.com> writes:
> [...]
> In a later tutorial review in 1992, I cited contemporary authors treating
> oversampling A/D conversion (for example) as solely a filtering problem,

Max, do you mean "straight" oversampling (as in the late 80s' CD players'
"8x Oversampling" outputs), or delta sigma conversion, where there is  
noise-shaping involved?

If the former, then, apart from brass-tacks things like how to scale
your filters, it IS just a filtering problem. If the latter, then I
would find it hard to believe that any designer would not know that
design of the modulator architecture is crucial in addition to the
(in the case of D/A) interpolation filtering. 
-- 
%  Randy Yates                  % "Rollin' and riding and slippin' and
%% Fuquay-Varina, NC            %  sliding, it's magic."
%%% 919-577-9882                %  
%%%% <yates@ieee.org>           % 'Living' Thing', *A New World Record*, ELO
http://home.earthlink.net/~yatescr

"Randy Yates" in message news:1xmd30eg.fsf@ieee.org...
> "Max Hauser" <maxREMOVE@THIStdl.com> writes:
> > [...]
> > In a later tutorial review in 1992, I cited contemporary authors
treating
> > oversampling A/D conversion (for example) as solely a filtering problem,
>
> Max, do you mean "straight" oversampling (as in the late 80s' CD players'
> "8x Oversampling" outputs), or delta sigma conversion, where there is
> noise-shaping involved?
>
> If the former, then, apart from brass-tacks things like how to scale
> your filters, it IS just a filtering problem. If the latter, then I
> would find it hard to believe that any designer would not know that
> design of the modulator architecture is crucial in addition to the
> (in the case of D/A) interpolation filtering.

Yes, I found it hard to believe also at the time, that is why I cited it
then, and now.  But I stress that I was not describing real systems and
equipment (so very unforgiving of overlooked reality) but engineering
papers, often theoretically flavored, on curent research problems in "A/D"
interfaces.  The broad problem that they all addressed was the complete
interface from continuous waveforms in continuous time to discrete-valued
discrete-time signals.  As you know, this always entails both a time
interface (antialias filtering and sampling, or waveform generation and
reconstruction filtering) together with a quantization issue.  In modern
oversampling data converters (with noise shaping), the hardware executing
the filtering and quantization-management tasks actually overlaps (which was
unprecedented in mainstream A/D/A work, and was not anticipated by the
engineering textbook block diagrams depicting time interface and value
interface as separable tasks, as they may still often do).  Rapid R&D on
this subject in the 1980s and early 1990s -- exploring architectures,
implementations, trade-offs -- entailed noise-shaping A/D or D/A interfaces
that were approached nevertheless by some authors with a selective interest
or focus, and therefore selective identification of issues.  I go into
specifics in the JAES paper especially, and also cite many references on the
development of this specialty.

The language, fraught with seductive pitfalls, is confounded by things like
at least two simultaneous filtering tasks in the digital filter inside an
oversampling noise-shaping A/D interface.  It suppresses frequency-shaped
quantization error, and it consummates the antialias-filtering function
begun prior to sampling.  In addressing a system like this, some researchers
will consider only one of these filtering tasks.  And they may develop very
useful results thereby.  (Problems arise only when selective partitioning of
the problem partitions away something important.)

Rick Carley and I have used the terms "signal acquisition system" (SAS) and
"signal regeneration system" (SRS) for complete chains entailing time and
value interfaces, to distinguish them from the older "data-acquisition
systems" (DAS) that used A/D conversion for sample-by-sample "data"
quantization and were, though you wouldn't know it from this newsgroup,
commercially much more common until a few years ago.   (For example, I
pointed out to Sheingold in 1990 that his published textbook-size Analog
Devices "data-conversion handbook" lacked the term "quantization noise," or
error, in its index; he got me back though by pointing out that the word
could actually be found once inside, though not important enough for the
index.)  In the 1980s and 1990s I routinely encountered engineering seminar
audiences where the engineers had a fixed idea of an A/D interface as a
circuit that grabs a single sample and thinks about it then spits out bits.
No filters, no sample-to-sample memory, no "quantization noise."  Some of
them demanded to know where the "convert command" input was in an
oversampling A/D interface (which makes no sense).  Another contact at
either Analog or Crystal quipped in exasperation circa 1991 that they'd
considered a dummy "convert command" input pin on oversampling converters,
to satisfy that persistent intuition.  Intuition is often the issue here,
indeed.

DSP filtering to simplify the analog requirements on the time interface
_alone_ is ancient and venerable and was, I believe, a motivation for early
"decimation" and "interpolation" filtering in DSP.  Peled and Liu's thin
1976 DSP textbook, and Crochiere and Rabiner's writing, brought out that
point I think.

Speaking of language, you probably know (but some recent authors appear not
to) that "oversampling" and "noise shaping" are wide terms in connection
with A/D conversion, while "delta-sigma" (coined and popularized by Inose et
al., 1962 -- the alternative form "sigma-delta" appeared later, about 1974)
always meant systems with specifically _one-bit_ quantization.  That was the
reason for existence of the term.  Multi-bit oversampling noise-shaping
quantization is older (Cutler in 1954, patented 1960).  One of the later
Inose papers was even titled "A unity-bit coding method ..."  (The
particular reason why the one-bit subclass became practically important for
data conversion, a different task than what Inose et al. were doing, is the
unique ability to interpolate valuewise between two points in a DC
input-output plane to get finer resolution, without introducing
nonlinearity, as I pointed out in 1986.)  Authors including me have
respected the inventors' definition ever since 1962, distinguishing
"delta-sigma" from more general classes of oversampling quantization (or
coding).  At the time of the 1991 JAES review, exceptions to this in the
literature were seldom.  Despite the inventors' publications that are still
cited today, despite periodic tutorial explanations, some newer authors use
"delta-sigma" to mean anything with noise shaping and oversampling (as if
previous ambiguity in this specialty were insufficient -- I inventoried it
in the conclusion of the JAES tutorial).  New authors doing things like that
always implicitly raise the question of what else about the subject they
have missed.

(In the event that any authors are surprised by that, and they stick to the
usual script, the customary next step will be to cast about first for
rationalizations of their existing understanding, rather than to examine it
searchingly, with the possible aid of the pointers above.  But if I may be
of any service here in pointing out facile reationalizations, please write,
I do have much literature.  ;-)

Max Hauser

Max Hauser wrote:


> 
> Yes, I found it hard to believe also at the time, that is why I cited it
> then, and now.  But I stress that I was not describing real systems and
> equipment (so very unforgiving of overlooked reality) but engineering
> papers, often theoretically flavored, on curent research problems in "A/D"
> interfaces.  The broad problem that they all addressed was the complete
> interface from continuous waveforms in continuous time to discrete-valued
> discrete-time signals.  As you know, this always entails both a time
> interface (antialias filtering and sampling, or waveform generation and
> reconstruction filtering) together with a quantization issue.  In modern
> oversampling data converters (with noise shaping), the hardware executing
> the filtering and quantization-management tasks actually overlaps (which was
> unprecedented in mainstream A/D/A work, and was not anticipated by the
> engineering textbook block diagrams depicting time interface and value
> interface as separable tasks, as they may still often do).  Rapid R&D on
> this subject in the 1980s and early 1990s -- exploring architectures,
> implementations, trade-offs -- entailed noise-shaping A/D or D/A interfaces
> that were approached nevertheless by some authors with a selective interest
> or focus, and therefore selective identification of issues.  I go into
> specifics in the JAES paper especially, and also cite many references on the
> development of this specialty.
> 
> The language, fraught with seductive pitfalls, is confounded by things like
> at least two simultaneous filtering tasks in the digital filter inside an
> oversampling noise-shaping A/D interface.  It suppresses frequency-shaped
> quantization error, and it consummates the antialias-filtering function
> begun prior to sampling.  In addressing a system like this, some researchers
> will consider only one of these filtering tasks.  And they may develop very
> useful results thereby.  (Problems arise only when selective partitioning of
> the problem partitions away something important.)
> 
> Rick Carley and I have used the terms "signal acquisition system" (SAS) and
> "signal regeneration system" (SRS) for complete chains entailing time and
> value interfaces, to distinguish them from the older "data-acquisition
> systems" (DAS) that used A/D conversion for sample-by-sample "data"
> quantization and were, though you wouldn't know it from this newsgroup,
> commercially much more common until a few years ago.   (For example, I
> pointed out to Sheingold in 1990 that his published textbook-size Analog
> Devices "data-conversion handbook" lacked the term "quantization noise," or
> error, in its index; he got me back though by pointing out that the word
> could actually be found once inside, though not important enough for the
> index.)  In the 1980s and 1990s I routinely encountered engineering seminar
> audiences where the engineers had a fixed idea of an A/D interface as a
> circuit that grabs a single sample and thinks about it then spits out bits.
> No filters, no sample-to-sample memory, no "quantization noise."  Some of
> them demanded to know where the "convert command" input was in an
> oversampling A/D interface (which makes no sense).  Another contact at
> either Analog or Crystal quipped in exasperation circa 1991 that they'd
> considered a dummy "convert command" input pin on oversampling converters,
> to satisfy that persistent intuition.  Intuition is often the issue here,
> indeed.
> 
> DSP filtering to simplify the analog requirements on the time interface
> _alone_ is ancient and venerable and was, I believe, a motivation for early
> "decimation" and "interpolation" filtering in DSP.  Peled and Liu's thin
> 1976 DSP textbook, and Crochiere and Rabiner's writing, brought out that
> point I think.
> 
> Speaking of language, you probably know (but some recent authors appear not
> to) that "oversampling" and "noise shaping" are wide terms in connection
> with A/D conversion, while "delta-sigma" (coined and popularized by Inose et
> al., 1962 -- the alternative form "sigma-delta" appeared later, about 1974)
> always meant systems with specifically _one-bit_ quantization.  That was the
> reason for existence of the term.  Multi-bit oversampling noise-shaping
> quantization is older (Cutler in 1954, patented 1960).  One of the later
> Inose papers was even titled "A unity-bit coding method ..."  (The
> particular reason why the one-bit subclass became practically important for
> data conversion, a different task than what Inose et al. were doing, is the
> unique ability to interpolate valuewise between two points in a DC
> input-output plane to get finer resolution, without introducing
> nonlinearity, as I pointed out in 1986.)  Authors including me have
> respected the inventors' definition ever since 1962, distinguishing
> "delta-sigma" from more general classes of oversampling quantization (or
> coding).  At the time of the 1991 JAES review, exceptions to this in the
> literature were seldom.  Despite the inventors' publications that are still
> cited today, despite periodic tutorial explanations, some newer authors use
> "delta-sigma" to mean anything with noise shaping and oversampling (as if
> previous ambiguity in this specialty were insufficient -- I inventoried it
> in the conclusion of the JAES tutorial).  New authors doing things like that
> always implicitly raise the question of what else about the subject they
> have missed.
> 
> (In the event that any authors are surprised by that, and they stick to the
> usual script, the customary next step will be to cast about first for
> rationalizations of their existing understanding, rather than to examine it
> searchingly, with the possible aid of the pointers above.  But if I may be
> of any service here in pointing out facile reationalizations, please write,
> I do have much literature.  ;-)
> 
> Max Hauser
> 

Max, try as I might, I can't figure out what your point is 
with all this.


Bob
-- 

"Things should be described as simply as possible, but no 
simpler."

                                              A. Einstein

Bob Cain wrote:
> Arny Krueger wrote:
>
>> If so, then do YOUR OWN  DBTs with the provided software and data
>> file, and a computer with a 24/96 audio interface and the speakers
>> or headphones of your choice.  24/96 and even 24/192 audio
>> interfaces are now plentiful and relatively inexpensive. The M-Audio
>> "Revolution 7.1" is a good card to use as is their "Audiophile
>> 24/96" or Echo's Mia. The SoundBlaster Audigy stuff tends to be so
>> flawed (high frequency IM) as to be questionable for this
>> application.
>>
>> If you don't trust my files, analyze them to double-check my work,
>> or even make some for yourself. This is not rocket science, it's
>> just a walk in the park for people who even know just a little bit
>> about audio production.
>>
>> It's safe to predict that you'll sing at least a slightly different
>> tune after you get your hands dirty with the real world!
>
> Arny, in my case this would be a total waste of time so
> would you mind providing the results that you have accumulated?

Not only are high sample rates a waste of processing and storage, but even
16/44 is overkill as a distribution medium. People start hearing artifacts
due to sample rate and sample size some place around 14 bits and 32 KHz
sampling.

"Bob Cain" in news:c6ff6t0oe3@enews2.newsguy.com:
> Max Hauser wrote:
>
> >  [91 quoted lines snipped]
>
> Max, try as I might, I can't figure out what your point is
> with all this.


RFC1855:

. Content of a follow-up post should exceed quoted content.

"Arny Krueger" <arnyk@hotpop.com> writes:

> Bob Cain wrote:
>> Arny Krueger wrote:
>>
>>> If so, then do YOUR OWN  DBTs with the provided software and data
>>> file, and a computer with a 24/96 audio interface and the speakers
>>> or headphones of your choice.  24/96 and even 24/192 audio
>>> interfaces are now plentiful and relatively inexpensive. The M-Audio
>>> "Revolution 7.1" is a good card to use as is their "Audiophile
>>> 24/96" or Echo's Mia. The SoundBlaster Audigy stuff tends to be so
>>> flawed (high frequency IM) as to be questionable for this
>>> application.
>>>
>>> If you don't trust my files, analyze them to double-check my work,
>>> or even make some for yourself. This is not rocket science, it's
>>> just a walk in the park for people who even know just a little bit
>>> about audio production.
>>>
>>> It's safe to predict that you'll sing at least a slightly different
>>> tune after you get your hands dirty with the real world!
>>
>> Arny, in my case this would be a total waste of time so
>> would you mind providing the results that you have accumulated?
>
> Not only are high sample rates a waste of processing and storage, but even
> 16/44 is overkill as a distribution medium. People start hearing artifacts
> due to sample rate and sample size some place around 14 bits and 32 KHz
> sampling.

Ahh, the voice of reason. Thank you, Arny. (In case anyone is wondering,
I'm stating this in all seriousness.)
-- 
%  Randy Yates                  % "The dreamer, the unwoken fool - 
%% Fuquay-Varina, NC            %  in dreams, no pain will kiss the brow..."
%%% 919-577-9882                %  
%%%% <yates@ieee.org>           % 'Eldorado Overture', *Eldorado*, ELO
http://home.earthlink.net/~yatescr

Max Hauser wrote:

>>> [91 quoted lines snipped]
>>
>>Max, try as I might, I can't figure out what your point is
>>with all this.
> 
> RFC1855:
> 
> . Content of a follow-up post should exceed quoted content.

Wait...  Are you pulling Bob's ear and telling him "bad Bob"
for quoting too much of your message for the little he wrote,
or are you answering his question saying that you wrote that
much because you had to exceed what you had quoted?

:-)

Carlos
--

"Carlos Moreno" in news:uSYic.24731$2V6.341047@wagner.videotron.net...
> Wait...  Are you pulling Bob's ear and telling him "bad Bob"
> for quoting too much of your message for the little he wrote,
> or are you answering his question saying that you wrote that
> much because you had to exceed what you had quoted?
> :-)
> Carlos


Exactly!

:-)


Max

danlavry@mindspring.com (dan lavry) wrote in message news:<673b149b.0404231149.142e40ed@posting.google.com>...
> mark_ovchain@yahoo.com (Mark Ovchain) wrote in message news:<d61ab25b.0404221910.11b30855@posting.google.com>...
> > "Arthur" <arthurc99@yahoo.com> wrote in message news:<c65haq$1dgf$1@news.hgc.com.hk>...
> > > Hello all,
> > Well, I suspect that it helps the manufacturers sell more equipment.
> > 
> > Think of it as a form of "trickle-down", why don't you?
> 
> Think of it as "trickle-down"? Like - they are standing on the roof
> with the zipper open, trickling down on their customers? :-)
> 
> Dan Lavry

Hush, now, I wasn't going to explain that part yet!

"Arthur" <arthurc99@yahoo.com> wrote in message news:<c65haq$1dgf$1@news.hgc.com.hk>...
> Hello all,
> 
> Recently, there is a tendency to use high sampling rate 96 KHz, 192 KHz in
> the audio application.  To my knowledge, about ~ 20 KHz, everything will be
> cut off from the loudspeaker, amplifier, etc..  So, why 96 KHz, 192 KHz
> instead of 48 KHz..
> 
> Regards
> Arthur

Thank you for all the peoples participated in the dicussion..
The conclusion seems to be : 96 KHz + 24 bits should be enough for all
sorts of reasons.  But there is a theory around stated that sampling
higher will lower down the noise floor..  Any thoughts about that..

Regards
Arthur