What is the advantage on high-sampling rate ?

Erik de Castro Lopo wrote:

> Arthur wrote:

>> Hello all,

>> Recently, there is a tendency to use high sampling rate 96 KHz, 192
>> KHz in the audio application.

Numbers for the sake of numbers.

>  To my knowledge, about ~ 20 KHz,
>> everything will be cut off from the loudspeaker, amplifier, etc..

Something like that. But the killer is that old bugabear the human ear.
Typically, a well-done brick wall filter @ 16 KHz or so is sonically
innocuous. People can typically hear sine waves at higher frequencies, but
with real world sound sources, you have to consider masking that doesn't
show up in tests based on isolated pure tones.

It's not the least bit unusual for perceptual coders to dump everything
above about 16 KHz. Why waste the bandwidth coding it when nobody notices
when it is gone?

>> So, why 96 KHz, 192 KHz instead of 48 KHz..

> Analogue audio amplifiers typically have a flat frequency response to
> 100kHz or higher.

Nope. In fact it's pretty stylish to start a fractional-dB roll-off starting
at 20 KHz, with rated load. One reason being that it is exceedingly common
to enhance amplifier reliability by putting a series choke on the output.
This choke is typically sized for response that is about 0.5 dB down at 20
KHz.

> Tweeters on mid to high price range loudspeakers
> are also capable of reproducing frequencies well above 20kHz.

OTOH a very high proportion of high quality speakers start rolling off
pretty fast above 15-16 KHz. While tweeters with roll-offs in the 25 KHz
range are not rare, move the corner frequency up another 10 KHz and you can
count 'em on one hand, maybe two.

Move into the recording studio, and while there are a few mics that are
reasonably flat at or above 30 KHz, the vast majority of microphones in
day-to-day use, start rolling off someplace between 13 and 22 KHz. In
particular, while cardioid mics are very widely used, the ones that are flat
above 25 KHz can again be counted on one or two hands.

> There have also been double blind tests which prove that even
> untrained listeners can tell the difference between a 10kHz sine wave
> and a 10kHz square wave (components at 10kHz, 30kHz, 5okHz etc) even
> after the two signals have been normalised to have the same perceived
> volume.

I'd like to see a formal cite for that. This is not a very tough listening
test to do, and IME it always comes out negative.

There's a web site that provides no-cost software and data files for testing
this sort of thing on a PC with a suitable audio interface.

Site introduction:

http://www.pcabx.com/

Related pages:

http://www.pcabx.com/technical/low_pass/index.htm

http://www.pcabx.com/technical/sample_rates/index.htm

french_student wrote:
> Hi Arthur,
> 
> For me the advantage to have a high sampling rate is to obtain more
> samples in one second and as a consequence to reproduce with more
> efficiency the analog signal you sample.
> For me it's the only reason why sample audio frequency like 96KHz or
> 192 KHz have appeared. Maybe I'm wrong but i give you my point of
> view.
> The advantage is certainly not to be able to sample signals containing
> higher frequencies than 20 KHz since the human beings can't hear
> sounds with frequency higher than about 16 KHz.
> 

As it turns out, the higher frequencies are the only 
advantage.  No precision in reproduction other than that is 
achieved.  Reconstruction done from a properly lowpassed 
sampling perfectly reconstructs the original in theory.  In 
practice there are lots more considerations.  Read Dan 
Lavry's excellent paper.  He just posted a link.


Bob
-- 

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

                                              A. Einstein

In comp.dsp, "Arthur" <arthurc99@yahoo.com> wrote:

>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..

   This is a recurring discussion on rec.audio.pro. Rather than
comment, I'll just point to one of the latest threads there:

http://groups.google.com/groups?hl=en&lr=&ie=UTF-8&oe=UTF-8&c2coff=1&threadm=673b149b.0404191340.4f108efb%40posting.google.com&rnum=2&prev=/groups%3Fq%3Dgroup:rec.audio.pro%2Bauthor:lavry%26hl%3Den%26lr%3D%26ie%3DUTF-8%26oe%3DUTF-8%26c2coff%3D1%26scoring%3Dd%26selm%3D673b149b.0404191340.4f108efb%2540posting.google.com%26rnum%3D2

or 

http://makeashorterlink.com/?D26B11918

>Regards
>Arthur
>

danlavry@mindspring.com (dan lavry) writes:
> www.lavryengineering.com

Hi Dan,

I just tried this link and, while the home page comes up OK, if
I click on either HTML or Flash version links I get a 404 error
from your web server.
-- 
Randy Yates
Sony Ericsson Mobile Communications
Research Triangle Park, NC, USA
randy.yates@sonyericsson.com, 919-472-1124

Works for me, but here is the direct link to the article in question:
http://www.lavryengineering.com/documents/Sampling_Theory.pdf

"Randy Yates" <randy.yates@sonyericsson.com> wrote in message
news:xxpekqf92wj.fsf@usrts005.corpusers.net...
> danlavry@mindspring.com (dan lavry) writes:
> > www.lavryengineering.com
>
> Hi Dan,
>
> I just tried this link and, while the home page comes up OK, if
> I click on either HTML or Flash version links I get a 404 error
> from your web server.
> -- 
> Randy Yates
> Sony Ericsson Mobile Communications
> Research Triangle Park, NC, USA
> randy.yates@sonyericsson.com, 919-472-1124

Arny Krueger wrote:
> 
> > There have also been double blind tests which prove that even
> > untrained listeners can tell the difference between a 10kHz sine wave
> > and a 10kHz square wave (components at 10kHz, 30kHz, 5okHz etc) even
> > after the two signals have been normalised to have the same perceived
> > volume.
> 
> I'd like to see a formal cite for that. This is not a very tough listening
> test to do, and IME it always comes out negative.

I have only seen second hand references to this study, but I believe
the original is this one:

   Tsutomi Oohashi, Emi Nishina, Norie Kawai, Yoshitaka Fawamoto, 
       Hiroshi Imai 
   "High Frequency Sound Above the Audible Range Affect Brain Activity 
   and Sound Perception."
   Audio Engineering Society preprint No 3207 (91st convention, NYC)

Erik
-- 
+-----------------------------------------------------------+
  Erik de Castro Lopo  nospam@mega-nerd.com (Yes it's valid)
+-----------------------------------------------------------+
Linux: the only OS that makes you feel guilty when you reboot
    -- Kenneth Crudup in comp.os.linux.misc

Erik de Castro Lopo <nospam@mega-nerd.com> writes:

> Arny Krueger wrote:
> > 
> > > There have also been double blind tests which prove that even
> > > untrained listeners can tell the difference between a 10kHz sine wave
> > > and a 10kHz square wave (components at 10kHz, 30kHz, 5okHz etc) even
> > > after the two signals have been normalised to have the same perceived
> > > volume.
> > 
> > I'd like to see a formal cite for that. This is not a very tough listening
> > test to do, and IME it always comes out negative.
> 
> I have only seen second hand references to this study, but I believe
> the original is this one:
> 
>    Tsutomi Oohashi, Emi Nishina, Norie Kawai, Yoshitaka Fawamoto, 
>        Hiroshi Imai 
>    "High Frequency Sound Above the Audible Range Affect Brain Activity 
>    and Sound Perception."
>    Audio Engineering Society preprint No 3207 (91st convention, NYC)

Oh, that. Yep, they convinced me. The other 4 billion people on the
planet must be wrong. They must have discovered something that no
other researcher has been able to find. Yessirr, that's a taker. 
-- 
Randy Yates
Sony Ericsson Mobile Communications
Research Triangle Park, NC, USA
randy.yates@sonyericsson.com, 919-472-1124

Jerry Avins wrote:

(snip of discussion regarding high sample rate audio)

> That's true, of course. I think that the important question is, "How
> much is enough?" and the answer is very dependent on context. Good
> engineering involves compromise; many are implicit here. Given a fixed
> number of bits in a medium, apportion them among word size, error
> recovery, sample rate, and playing time to give the user the best
> perceived value.

Some of it is that there are a lot of bits on a DVD, and figuring
out how to use them.   Four hour long CD's might not be too popular.

To me, though, there are separate questions:  What is a good sample
rate for recording and mastering?   How should the results be
distributed?  and finally, how should they go through the D/A converter
in the end.

Higher rates make filtering easier, and computing power is cheap enough
to apply various compression algorithms.

Say, for example, a performance was mastered at 192kHz, 24bit,
10 channels (just an example), and mixed down to two channels.

Assume also that player will run the D/A and 192kHZ, though maybe
only 16 bits.  What is the best way to get those bits to the player?

Can it be compressed with reasonably small loss down to 1.4Mb/s
(about the CD bit rate)?

-- glen

Jerry Avins wrote:

(snip)

> Many people who decide on their own end up buying centimeter-diameter
> gold-plated loudspeaker wire. Having spent an outrageous amount for the
> cables, they will inevitably hear the improvement these cables make.
> (Many of those same people might swear that their car's gas mileage and
> acceleration is improved by after-market gear-shift knobs.)

There was discussion about gas saving devices on Car Talk not so
long ago.  Most promise about 20% gas savings, likely within the
distribution of mileage normally seen.  Also, they figured that after
someone bought one they might drive more carefully.  It would be hard
to prove that the savings weren't 20%.  Audio improvements are
much more difficult to measure than gas savings.

I remember from the 1980's only one device ever passed the EPA
gas mileage tests.  It was a device to turn off the air-conditioner
compressor at high acceleration based on engine vacuum.  Such logic
is now included in the computer system of most cars.  (Possibly
using DSP techniques!)

-- glen

I agree with you, and 88.2KHz and 96KHz are certainly fast enough, an
overkill in fact.

The argument that filtering near 20KHz while at 44.1KHz is true, but
going to 88-96KHz solve it. In fact, while douing that (going faster),
the ability to work nera zero (say 20Hz or other low frequencies)
takes gets tougher. So 96KHz is ok with me. 192KHz is not.

BR
Dan Lavry


Jerry Avins <jya@ieee.org> wrote in message news:<4087f15f$0$16463$61fed72c@news.rcn.com>...
> french_student wrote:

> Assuming suitable filters, a sample rate so high that more than two 
> samples occur during one period of the highest frequency, the 
> reproduction can be perfect. Sampling at a rate that allows 2.5 samples 
> makes the filter design easier.

> When I wrote "the reproduction can be perfect" above, I meant it. 
> Perfection can't be improved. As for the upper limit of hearing, by 
> daughter could hear up to 22,000 Hz. It was a curse, not a blessing.
> 
> > Bye.
> > Jerome
> 
> Jerry