Time resolution of Redbook (16/44) PCM

[AdamG]

??? Not to @j_j I hope.

No of course not. It should be clear to j_j_. Or I may be looking for a new gig

 

[fieldcar]

Hey @mansr
I was just reading your article(below) and thought of something. Would you say that most of the issues with DSD (especially DSD64) could be fixed with a linear phase EQ capable of infinite attenuation at a LPF? Are most of the prominent issues with DSD just leaky filters?

https://troll-audio.com/articles/linear-and-minimum-phase/

 

[Wes]

"several orders of magnitude" does not compare well with a range of two dozen

 

[Wes]

https://asa.scitation.org/doi/10.1121/1.5087566
A recent paper available for everyone. It has lots of good references to see how the numbers were obtained in the past. Investigates various ways to measure human ITD capabilities.

Is this the sort of thing you are looking for or are you just saying more work should be based upon what we can hear with music vs other stimuli?

I am saying that more work needs to be done, and it needs to be done using complex signals.

Fundamentally, this is a "fight" among those who are trained in engineering and those trained in the neurosciences. We need to be very careful about extrapolating from the ambit of a study outward. This site is not just attracting trolls, it is under observation by unfriendly people, who would love to be able to broadcast mistakes to their commercial gain.

If you look at what I recommend to people for purchase or as ways to evaluate components, you will not find much space between me and the engineers here. If you look at claims, you will. I uses terms like unlikely, suggests, shows, etc - rather than absolutisms.

And BTW, a thank you to all the science people who sent me anonymous kudos, but I do not necessarily share your FUD about engineers, partly perhaps because I attended one of the xIT's as an undergrad. tho I majored in a couple of sciences, not engineering.

 

[j_j]

Of course no timing resolution can be gained but what is the mathematical loss with this particular upsampling?

Or is the loss completely negligible?

Unless it's been screwed up, the worst you can get is 2x the original. That's not necessary, either, because the bandwidth hasn't changed, but the details are complex, and involve how the original PCM was created, as well.

 

[j_j]

"several orders of magnitude" does not compare well with a range of two dozen

Eh? 1/(2*pi*20000*65536) = 12 ns
10 micros on the other hand, is looking like 3 orders of magnitude. What's with this 'two dozen' nonsense?

 

[j_j]

I am saying that more work needs to be done, and it needs to be done using complex signals.

No, it's not. If it's an audible signal, it's an audible signal. Your refusal to accept drumbeats with a standard 8" stereo pair is exactly something that shows you want to argue, not accept something that's actually in a recording.

As to 'exactly' nobody says 'exactly' and this varies by individual. I'm quite sure some people do better than others.

Interestingly the 10microsecond number corresponds startlingly well to the result suggested (but only theoretically) by the interaction of one of the very fast cochlear filters (i.e. wide bandwidth) with the 30dB SNR of an inner hair cell. The proof, of course, is in the actual results, which continue to be around 10 microseconds, regardless of what people would like to argue.

Oh, and some of us research scientists may have started as engineers, but that was long, long ago, before working with both engineers and psychologists, subjective testing experts, etc, for a career in Bell Labs research.

I smell the old "disparage the engineers" here again.

Science, I must remind you, Wes, is very simple at its heart. It's what you can demonstrate somehow or other with a test (yes, the tests for, say, the big bang, can be rather abstract, but that's how life is). Testable is the key. And when a subject demonstrates, say, 10 microsecond ITD sensitivity in ANY signal, and no repeated result shows better, we can take that as a reasonable likely bound. And we do, whatever you think I am.

You sound just like the guy on facebook who was going on about "engineers just follow rules", no, physics isn't a rule.

 

[danadam]

Seems related:
Here are some sample files with impulses with different delays: https://hydrogenaud.io/index.php?topic=107570.msg899713#msg899713
And here is an ABX result from someone passing the 10 µs sample: https://hydrogenaud.io/index.php?topic=107570.msg899732#msg899732

 

[Wes]

calling the range I posted nonsense indicates a lack of knowledge, much less disparagement

not to mention the facebook crack

for the record I am not on facebook, but I am on ResearchGate

With the tenor on this thread declining, I am out.

 

[DonH56]

"several orders of magnitude" does not compare well with a range of two dozen

Back to this... As a lowly, ignorant, uncreative, FUD-spreading engineer I am far too stupid to follow this sort of scientific claim. What exactly is "a range of two dozen" orders of magnitude? And why is time resolution of "several" orders of magnitude better than ITD with respect to timing resolution inconsequential in comparison?

 

[j_j]

calling the range I posted nonsense indicates a lack of knowledge, much less disparagement

not to mention the facebook crack

for the record I am not on facebook, but I am on ResearchGate

What range are you advocating for what? You are repeatedly creating whipsaws, and I'm starting to conclude YOU are the malice you meant.

There are 3 orders of magnitude between 10 microseconds and Redbook. What was your 'several dozens' in reference to, then?

Please be exact, specific, and do not equivocate.

You do, however, seem to be unwilling to specify what you ACTUALLY THINK, and are content to throw disparagement at the other guy. Please explain exactly what your "several dozens" refers to, so we don't have any more attempts at a magician's force.

 

[j_j]

Note: Quotes taken from a number of articles, not just one.

"several orders of magnitude" does not compare well with a range of two dozen

Exactly two dozen what, then? Are you claiming 24 orders of magnitude? The atmosphere's intrinsic noise at STP alone puts that one to death. What ARE you claiming?

I am saying that more work needs to be done, and it needs to be done using complex signals.

And you expect what outcome? What do you expect to happen when more masking energy is added?

for the record I am not on facebook, but I am on ResearchGate

I'm listed on both. I belong to enough professional organizations already, if you call Research Gate that. I suppose it might be. It says "membership is free" but sorry, I've enough. So? Ok, you're not any of the facebook trolls. Good for you.

Then, "Wes" please speak up and tell us just exactly WHO you are. You've made professional accusations here, so now it's time for you to show your credentials. Pony up. You've effectively claimed that you, personally, have the chops to engage in professional criticism. So let's see your research record here, along with just who you are.

As to10 microseconds? Is it too big, is it too small? Do enlighten us. It can't be "too small" because it's demonstrated. If it's too big, how much too big is it?

You're very good at casting extremely serious professional aspersions on others, say, like me and your "engineer" crack, but you seem remarkably short in the actual answer category.

You also seem somehow unable to say who you are, never mind you claim expertise in relevant fields. So, let's see that professional cred. Speak up, now. Stop hiding behind a vague nickname.

. This site is not just attracting trolls, it is under observation by unfriendly people, who would love to be able to broadcast mistakes to their commercial gain.

I think you need to be more specific there, "Wes". If someone is engaging in false disparagement of someone on this site, you are professionally obligated to speak up and say who and what? So evidence, man, the evidence.

(Note, quotes are collected from various posts by "Wes" in this thread.)

 

[earlevel]

The equation for this is given here. It can be simplified as:
tmin = 1/(π * fs * (2^b -1))
Where fs = sampling rate and b = no of bits

Needless to say, this is a lot less than the 1/πfs figure touted by those who claim resolution is simply limited by sampling rate, but it's not 0.

Boy, this thread woke up busy...hope I didn't miss something similar, but...

Mans' article is a good analysis, but it leaves out something, IMO. He shows how the time resolution is related to bit depth, and specifically calculates it for 16 bits. However, we normally dither from higher precision (24-bit or 32-bit float minimum). True that the average error doesn't get smaller, but it does get distributed as a noise floor. You could say the timing error is obscured by the noise floor, and that becomes the limit of being able to detect the exact timing of an event.

While I don't think putting up and math will make that more obvious, I think this makes it clear:

• By Mans' equation, 24-bit audio will have better time resolution than the same thing truncated to 16-bit. And that seems intuitive due to loss of precision.

• But we can (and do, typically) add dither and truncate for our conversions to 16-bit.

• Now, what is the difference between the original 24-bit and the dithered 16-bit version? Well, we can literally take the difference, and the answer is clear: it's the steady noise floor.

• So...listening to the 16-bit dithered version is precisely the same as listening to the original 24-bit version summed with that exact noise floor.

We already know that 16-bit has plenty of time resolution, from Mans' equation. And from it we know 24-bit is even better (~256 times smaller time resolution). But the 16-bit dithered version is precisely the 24-bit version, with added noise floor. So...we're really noise-floor limited—effectively, we didn't lose any time resolution, we just added noise making it a little less easy to hear (assuming we can anyway).

Is this any less true of magnetic tape recording? Any other form of analog recording? Why should anyone think that digital has a flaw regarding time resolution, compared with analog?

 

[j_j]

Is this any less true of magnetic tape recording? Any other form of analog recording? Why should anyone think that digital has a flaw regarding time resolution, compared with analog?

Tape suffers from a couple of mechanical issues, like skew, scrape flutter, and the like, as well as some intrinsic nonlinearities. This makes it hard to say exactly what the time resolution of tape is, as it varies by tape formulation, quality, head smoothness, recording level, and a bunch of other things, as well as having a substantially lower bandwidth.

Note that rather than FS, it's actually better to say 1/(2 * pi * bandwidth * noise-factors). In any case, it's not likely that tape does any better, to say the least. Part of the problem in defining tape time resolution is that two passes over the same track will not be the same, so it becomes necessary to figure out a mean, a variation, etc, as well as observe any degradation due to repeated playbacks.

This also reads on things like DSD, which have a very high "sampling rate" but the bandwidth is conventionally much more limited, and the SNR is not uniform across bandwidth. Knowing the noise shape and relevant bandwidth is necessary to complete the calculation.

 

[Music1969]

This also reads on things like DSD, which have a very high "sampling rate" but the bandwidth is conventionally much more limited, and the SNR is not uniform across bandwidth. Knowing the noise shape and relevant bandwidth is necessary to complete the calculation.

Does this apply to DSD64 rate mostly?

RME make affordable DSD256 recorders now so I would hope new DSD recordings these days aren't using DSD64.

Of course there are later issues with mixing with DSD (need to become DXD).

I think there are some recordings by Cookie M that are recorded and mixed live at high rate DSD (DSD128 and DSD256) with no processing later. But it's a small number of such productions out there.

 

[j_j]

Does this apply to DSD64 rate mostly?

RME make affordable DSD256 recorders now so I would hope new DSD recordings these days aren't using DSD64.

Of course there are later issues with mixing with DSD (need to become DXD).

I think there are some recordings by Cookie M that are recorded and mixed live at high rate DSD (DSD128 and DSD256) with no processing later. But it's a small number of such productions out there.

It reads at any level, in regard to bandwidth as a function of frequency and noise shape. It's very general.

 

[Music1969]

It reads at any level, in regard to bandwidth as a function of frequency and noise shape. It's very general.

Thanks but isn't DSD256 flat with high SNR at least to 80kHz ?

The quantisation noise starts above 80kHz?

I'm not an expert on this. All I know is from reading this:

https://positive-feedback.com/audio-discourse/raising-the-sample-rate-of-dsd-is-there-a-sweet-spot/

 

[j_j]

Thanks but isn't DSD256 flat with high SNR at least to 80kHz ?

The quantisation noise starts above 80kHz?

I'm not an expert on this. All I know is from reading this:

https://positive-feedback.com/audio-discourse/raising-the-sample-rate-of-dsd-is-there-a-sweet-spot/

Then one would use the actual snr, in linear (not dB) terms, say along with the 80kHz bandwidth, for a rough estimate. Probably best to use the peak noise inside that for a safe bound, otherwise you have to calculate all the contributions from the phase noise with the SNR at each frequency.

If it goes to 80kHz with 20 bit resolution, well, that pretty much exceeds the electronics in most anything in existence. So that would be pretty good, but still not 24 orders of magnitude better time resolution like "Wes" *MAY* have meant. (sorry, but vague statements intended to allow equivocation annoy me, no, you didn't do that, he did)

you'd get a factor of 1/4 in terms of time resolution for the band width, and a factor of 1/16 for the noise, so about 1/64th the time resolution of redbook (note, smaller is BETTER here) if we assume 80kHz and 20 bit effective noise level. The problem now is that we've hit clock accuracy issues, atmospheric transmission issues, etc, but that really does take us back to more than good enough. In terms of real physics and the auditory system, 20 bits at 64kHz is more or less a limit for human beings.

 

[amirm]

Time out. Folks, please take it easy with each other. It is just a technical topic, shouldn't get personal.