Sample Rate of a Digital Audio Delivery Medium (i.e., CD)

<benjamin.couillard@gmail.com> wrote:

>There's an interesting paper written by Cirrus logic on that subject. 
>They call it "pre-echo"

>https://www.cirrus.com/en/pubs/whitePaper/DS668WP1.pdf

Thanks.  From what I can tell, "ringing" (Tim) "delay dispersion" (me)
and "pre-echo" (the Cirrus paper) are different names for the 
same artifact.

Steve

On Fri, 28 Oct 2016 18:52:13 +0000 (UTC), spope33@speedymail.org (Steve
Pope) wrote:

>Thanks.  From what I can tell, "ringing" (Tim) "delay dispersion" (me)
>and "pre-echo" (the Cirrus paper) are different names for the 
>same artifact.

I have always called it "pre-ring". And I first started raising red
flags about it way back in 1994, when I worked for Fostex. 

"Antialias Filters and System Transient Response at High Sample Rates",
by Peter Craven; JAES vol 52 no 3 March 2004.

Greg

Greg Berchin  <gjberchin@chatter.net.invalid> wrote:

>On Fri, 28 Oct 2016 18:52:13 +0000 (UTC), spope33@speedymail.org (Steve

>>Thanks.  From what I can tell, "ringing" (Tim) "delay dispersion" (me)
>>and "pre-echo" (the Cirrus paper) are different names for the 
>>same artifact.

>I have always called it "pre-ring". And I first started raising red
>flags about it way back in 1994, when I worked for Fostex. 

Right.  

Since many audio/musical sounds have a fast attack and a slow decay,
the perceived artifact is more likely to be audible before the main 
part of the sound, than afterwards when it is more likely to be
masked. (Even if the impulse response is symmetrical.)

>"Antialias Filters and System Transient Response at High Sample Rates",
>by Peter Craven; JAES vol 52 no 3 March 2004.

Thanks.

Steve

Greg Berchin  <gjberchin@chatter.net.invalid> wrote:

>On Fri, 28 Oct 2016 18:52:13 +0000 (UTC), spope33@speedymail.org (Steve

>>Thanks.  From what I can tell, "ringing" (Tim) "delay dispersion" (me)
>>and "pre-echo" (the Cirrus paper) are different names for the 
>>same artifact.

(I actually meant to say "delay spread" in the above.)

>I have always called it "pre-ring". And I first started raising red
>flags about it way back in 1994, when I worked for Fostex. 
>
>"Antialias Filters and System Transient Response at High Sample Rates",
>by Peter Craven; JAES vol 52 no 3 March 2004.
>
>Greg

Randy Yates wrote:
> Perhaps I'm forgetting something, but why would the sample rate of the
> digital audio delivery medium such as CD have _anything_ to do with the
> ease of sampling the original signal?
>

What is "ease of sampling?" The SR for CD was because 1) heritage of
A/D-D/A interfaces with NTSC based video tape implied certain
parameters and 2) some executive wanted a certain piece to fit on one
CD - the classic 74 minute CD, not the later 80 minute sized ones.

80 minute mighta been an artifact of CDR; don't remember.

> It is certainly true that it is better to oversample the original analog
> input so that the antialiasing filter requirements can be greatly
> relaxed. However, once we're in the digital domain we can run some very
> tight digital filter to resample to 44.1 kHz (or whatever). Right?
>

Yep. There might have been a good argument for something like that 
before good delta-sigma converters became common.

> Am I missing something? I bring this up because I see people arguing for
> a high sample rate in a new digital audio delivery medium (HRA) for this
> reason and I just don't see it.
>

There's nothing to get.

-- 
Les Cargill

Randy Yates wrote:
> Tim Wescott <tim@seemywebsite.really> writes:
>
>> On Wed, 26 Oct 2016 22:20:38 -0400, Randy Yates wrote:
>>
>>> Perhaps I'm forgetting something, but why would the sample rate of the
>>> digital audio delivery medium such as CD have _anything_ to do with the
>>> ease of sampling the original signal?
>>>
>>> It is certainly true that it is better to oversample the original analog
>>> input so that the antialiasing filter requirements can be greatly
>>> relaxed. However, once we're in the digital domain we can run some very
>>> tight digital filter to resample to 44.1 kHz (or whatever). Right?
>>>
>>> Am I missing something? I bring this up because I see people arguing for
>>> a high sample rate in a new digital audio delivery medium (HRA) for this
>>> reason and I just don't see it.
>>
>> Two reasons.  First, because you can't dip flash chips in LN2,
>
> You totally lost me with that statement! But if you're basically saying
> marketing, well then...
>
>> so you have to do _something_ to convince yourself that you're cooler
>> than the other guy. And second, because really steep filters ring. My
>> sisters and my boys can hear sounds significantly above 20kHz, which
>> is probably around where the ringing would happen if you're aiming to
>> sample at 44kHz.
>
> OK, so if ringing is an issue, sample a little higher. 64 kHz? Why do we
> need to go to 96 or 192?!?!
>


Because 48KHz is commonly used for film work. Having the available
rates be 48/96/192 would unify the music recording business with the 
film audio business.

-- 
Les Cargill

Greg Berchin wrote:
> On Wednesday, October 26, 2016 at 11:25:35 PM UTC-5, Randy Yates
> wrote:
>
>> OK, so if ringing is an issue, sample a little higher. 64 kHz? Why
>> do we need to go to 96 or 192?!?!
>
> This argument persisted in audiophile circles for decades. Now that
> 96 kHz, 192 kHz, and even 384 kHz are relatively common, one doesn't
> see much argument about it any more. So evidently 44.1/48 kHz really
> wasn't enough, but somewhere around double that is.
>
That's very far from even remotely settled. What's true is that some
classes of plugins work better at higher SR, and that realtime latency
can be lower at higher SR. But nobody who knows what they are talking 
about is proposing a higher end distribution SR.

> Those high sampling rates are not as wasteful of storage space or
> transmission bandwidth as one might expect. There is so much
> redundancy in the signals that lossless compression can be quite
> effective.
>
> Long ago I read an article that argued 64 kHz / 24 bit would be ideal
> for audio. I tend to agree with that idea myself, though I cannot
> hear 21 kHz like I used to. There will always be others for whom too
> much is never enough.
>

Sorry; it will always be audiphoolery to me. There is nothing of actual 
value above 14k in almost all program material. Besides; what will you
play it back on? Show me a speaker that can represent 25k accurately.

They exist, but they're not common.

> Greg
>

-- 
Les Cargill

rickman wrote:
> On 10/26/2016 10:20 PM, Randy Yates wrote:
>> Perhaps I'm forgetting something, but why would the sample rate of the
>> digital audio delivery medium such as CD have _anything_ to do with the
>> ease of sampling the original signal?
>>
>> It is certainly true that it is better to oversample the original analog
>> input so that the antialiasing filter requirements can be greatly
>> relaxed. However, once we're in the digital domain we can run some very
>> tight digital filter to resample to 44.1 kHz (or whatever). Right?
>>
>> Am I missing something? I bring this up because I see people arguing for
>> a high sample rate in a new digital audio delivery medium (HRA) for this
>> reason and I just don't see it.
>
> You don't mention anything about the anti-alias filter on the playback
> end.  Are you assuming there is another excellent resampling filter and
> a high sample rate DAC?  I believe the idea is to use the fancy stuff in
> the already expensive recording process to get the best possible audio
> on the disk (what you seem to be calling the "digital audio delivery
> medium") and minimizing the requirements on the much lower cost playback
> mechanisms.  A higher sample rate on the disk would do that.
>

But this is making the razor cheaper and the blades more expensive. 
Markets don't usually work that way.

-- 
Les Cargill

Steve Pope wrote:
> Randy Yates  <yates@digitalsignallabs.com> wrote:
>
>> Perhaps I'm forgetting something, but why would the sample rate of the
>> digital audio delivery medium such as CD have _anything_ to do with the
>> ease of sampling the original signal?
>>
>> It is certainly true that it is better to oversample the original analog
>> input so that the antialiasing filter requirements can be greatly
>> relaxed. However, once we're in the digital domain we can run some very
>> tight digital filter to resample to 44.1 kHz (or whatever). Right?
>
>> Am I missing something? I bring this up because I see people arguing for
>> a high sample rate in a new digital audio delivery medium (HRA) for this
>> reason and I just don't see it.
>
> A major issue is that those "very tight" digital filters have a
> very non-zero normalized delay spread.  The popular fiction is that
> if they are linear phase, they have no coloration.  To realize
> this is fiction, take it to the extreme -- could you have a delay
> spread of one minute and it is still inaudible?  One second?
>
> (I've pointed this out in the past, but it has fallen on deaf ears.)
>
> Steve
>

What is the expected effect of this? I can see how delay spread might 
matter for multipath in wireless comms because the frequencies are so 
much higher, but I fail to see how this would make any difference in audio.

The sheer distortion of transducers overwhelms the rest of the playback
chain anyhow.

-- 
Les Cargill

Les Cargill  <lcargill99@comcast.com> wrote:

>Greg Berchin wrote:

>> This argument persisted in audiophile circles for decades. Now that
>> 96 kHz, 192 kHz, and even 384 kHz are relatively common, one doesn't
>> see much argument about it any more. So evidently 44.1/48 kHz really
>> wasn't enough, but somewhere around double that is.

>That's very far from even remotely settled. What's true is that some
>classes of plugins work better at higher SR, and that realtime latency
>can be lower at higher SR. But nobody who knows what they are talking 
>about is proposing a higher end distribution SR.

That's quite an assessment.  It may be true for market reasons
(nobody ever lost money selling compressors, whereas they surely
lost money selling SACD), but this conversation is about the technical 
necessities.

Are you saying the observations presented here about pre-ringing, pre-echo
or delay spread are not technically informed?  (As opposed to, difficult
to sell, a point to which I will cotton to.)

That too-steep filters introduce artifacts is a basic fact for anyone
working in audio.  You only use them when you need to achieve some
purpose, you don't -- except under extreme pressure -- bake them into
your entire signal path.

That handful of those veterans who worked on the CD audio format and 
who were kind enough to apprise me of what transpired paint a picture of 
everyone believing it was a horrible underdesign, pushed on the 
standards engineers by fearful managers who didn't really understand 
what was technically possible at market cost.

That it took a decade or so for the industry to tame the artifacts
with workarounds lends credence to this history.

My own experience, way back when, in professional audio was that we
used more resolution (18 bits or more), and did not come nearly as 
close to Nyquist (e.g. typical might be 16 KHz bandwidth at 50 Ks/sec.)  
Of course that did not involve mass market with cost pressure yada yada.  
But it did involve engineers who would personally lose their jobs
if the thing didn't actually sound good enough.

Steve