24bit vs 16bit, the myth exploded!
Dec 14, 2019 at 10:40 PM Post #5,416 of 6,480

old tech

500+ Head-Fier
Joined
Jul 8, 2015
Posts
830
Likes
414
Location
Sydney, Australia
There is infinite inwards... Within the parameters of frequency response and noise floor dictated by Nyquist it is infinite resolution. You can't "blow it up and see the halftone dots". In this sense, it is like vector graphics. It doesn't matter what scale you look at it, within the boundaries of Nyquist, it is perfect.
I can't be infinite backwards if SNR/DR doesn't measure as infinity.

No one is suggesting that you can "blow it up and see half tone dots", but rather that there is always quantisation errors or implemented dithered noise which put a limit on resolution. If I broadly understood @gregorio 's post, then the "infinite resolution" part cannot in practice exist separately to the noise and/or distortion part. Intuitively it makes sense and goes a long way to explain why increasing the bit depth or dithering and shaping the noise to outside the range where human hearing is most sensitive will increase resolution (and hence, increase SNR/DR), but infinite resolution? - I would be keen to see some real world measurements of digital audio no matter how much the bit depth is or how well it is noise shaped, that show SNR/DR approaching infinity.
 
Last edited:
Dec 14, 2019 at 10:50 PM Post #5,417 of 6,480

old tech

500+ Head-Fier
Joined
Jul 8, 2015
Posts
830
Likes
414
Location
Sydney, Australia
That "gap" is tricky to fill because it requires more than a "laymanish" understanding of digital audio. We experience acoustic sound and can relate that experience to analogue audio precisely because it's analogous. However, digital audio is not analogous to acoustic sound (or analogue audio) and therefore requires a conceptualisation (thinking about it) that is different. This requires going beyond "layman", and is somewhat difficult because it's not especially intuitive. It is possible to explain in terms that are "beyond layman" but still "laymanish", but bare in mind that such an explanation is rather like an analogy that is only somewhat accurate:
Not sure if this helps?
G
Thanks for the explanation, it does somewhat expand my "laymanish" understanding of audio, though I did already understand the concept behind digital audio. I appreciate that a key difference between analog and digital signals is that noise and error is intrinsic to an analog signal while it exists somewhat separately with digital. Dealing with noise in analog production, eg pre/de emphasis (like Dolby) will increase its resolution, just like noise shaping can increase resolution of a digital production. In that sense they are similar, though analog will always be at a disadvantage because dealing with the noise also messes with the signal.
 
Last edited:
Dec 15, 2019 at 12:21 AM Post #5,418 of 6,480

castleofargh

Sound Science Forum Moderator
Joined
Jul 2, 2011
Posts
9,541
Likes
4,941
I can't be infinite backwards if SNR/DR doesn't measure as infinity.

No one is suggesting that you can "blow it up and see half tone dots", but rather that there is always quantisation errors or implemented dithered noise which put a limit on resolution. If I broadly understood @gregorio 's post, then the "infinite resolution" part cannot in practice exist separately to the noise and/or distortion part. Intuitively it makes sense and goes a long way to explain why increasing the bit depth or dithering and shaping the noise to outside the range where human hearing is most sensitive will increase resolution (and hence, increase SNR/DR), but infinite resolution? - I would be keen to see some real world measurements of digital audio no matter how much the bit depth is or how well it is noise shaped, that show SNR/DR approaching infinity.
It's just a different way to look at the same phenomenon, of course you never end up with anything perfect IRL. But thinking of a sampled signal as the perfect original(when within Nyquist theorem's conditions!!!!!!!) plus whatever amount of noise added by quantization error, is accurate. It's also a model that makes a few digital behaviors more intuitive IMO.

About your question, if you count on real life examples, the S9038PRO sabre chip brags about a dynamic range of 140dB in mono on a full moon with the wind in their back while buried 2kilometers underground. And they give a more realistic THD+N at -122dB. That's nominal conditions and only the chip itself, most complete DACs will do worst.
So your request rapidly falls back to pure math simulation. And then, so long as we're within Nyquist theorem's conditions, we already know that bringing quantization errors near zero will also bring near infinite resolution. I mean, the quantization noise is literally the difference between the perfect sample and the quantized approximation.
 
Dec 15, 2019 at 12:45 AM Post #5,419 of 6,480

old tech

500+ Head-Fier
Joined
Jul 8, 2015
Posts
830
Likes
414
Location
Sydney, Australia
It's just a different way to look at the same phenomenon, of course you never end up with anything perfect IRL. But thinking of a sampled signal as the perfect original(when within Nyquist theorem's conditions!!!!!!!) plus whatever amount of noise added by quantization error, is accurate. It's also a model that makes a few digital behaviors more intuitive IMO.

About your question, if you count on real life examples, the S9038PRO sabre chip brags about a dynamic range of 140dB in mono on a full moon with the wind in their back while buried 2kilometers underground. And they give a more realistic THD+N at -122dB. That's nominal conditions and only the chip itself, most complete DACs will do worst.
So your request rapidly falls back to pure math simulation. And then, so long as we're within Nyquist theorem's conditions, we already know that bringing quantization errors near zero will also bring near infinite resolution. I mean, the quantization noise is literally the difference between the perfect sample and the quantized approximation.
Cheers for that.

That is what I sort of had understood, ie assuming Nyquist's conditions hold, we would have infinite resolution (or close to it) if there were no quantisation errors. I certainly agree that resolution can be practically perfect within sample rate and bit depth parameters, particularly in regard with human end use, but not infinite. Your example of the Sabre chip "only" being able to achieve a DR of 140db under the most contrived conditions illustrates the point. Now 140db is spectacular DR and way beyond any relevance to reproducing music but even so, it is nowhere near infinity.
 
Dec 15, 2019 at 1:00 AM Post #5,420 of 6,480

castleofargh

Sound Science Forum Moderator
Joined
Jul 2, 2011
Posts
9,541
Likes
4,941
A file can be sold claiming a resolution of 24/96 when the recording conditions never reach that, and the playback chain is never able to come close to output that. So maybe we can just as well decide that a sampled signal is perfect, and ignore quantization noise as merely another extra stuff we ignore along the way so we can continue to discuss fantasy HiFi XL. :imp:
 
Dec 15, 2019 at 1:42 AM Post #5,421 of 6,480

old tech

500+ Head-Fier
Joined
Jul 8, 2015
Posts
830
Likes
414
Location
Sydney, Australia
A file can be sold claiming a resolution of 24/96 when the recording conditions never reach that, and the playback chain is never able to come close to output that. So maybe we can just as well decide that a sampled signal is perfect, and ignore quantization noise as merely another extra stuff we ignore along the way so we can continue to discuss fantasy HiFi XL. :imp:
I agree. My interest is in the science behind it, not some audiophile belief that resolution beyond some amount matters.

It is also being pendantic. Us claiming that digital audio production has infinite resolution is almost as bad as those analog guys making that claim. I say almost as bad because the resolution of analog production doesn't exceed the equivalent of 14 bits digital at best, let alone a fantasy that it is infinite.
 
Dec 15, 2019 at 7:14 AM Post #5,422 of 6,480

gregorio

Headphoneus Supremus
Joined
Feb 14, 2008
Posts
4,235
Likes
2,360
[1] Dealing with noise in analog production, eg pre/de emphasis (like Dolby) will increase its resolution, just like noise shaping can increase resolution of a digital production. In that sense they are similar,
[1a] though analog will always be at a disadvantage because dealing with the noise also messes with the signal.
[2] That is what I sort of had understood, ie assuming Nyquist's conditions hold, we would have infinite resolution (or close to it) if there were no quantisation errors.
[3] Your example of the Sabre chip "only" being able to achieve a DR of 140db under the most contrived conditions illustrates the point. ...
[3a] Now 140db is spectacular DR and way beyond any relevance to reproducing music but even so, it is nowhere near infinity.

1. No, they're not really similar at all. Pre/de emphasis does NOT increase the resolution of the signal and doesn't attempt to, what it does is decrease the amount of noise that will be added to the signal downstream (EG. That occurs from transmitting the analogue signal). If we take a theoretically perfect implementation of pre-de emphasis, what you would end up with is slightly lower resolution, IE. Exactly the same signal/resolution with some additional thermal noise. In contrast, dither linearises all the quantisation errors, effectively giving us perfect accuracy (infinite resolution) and then it's a case of moving the resultant dither noise to reveal that resolution (with noise-shaping).
1a. Other forms of noise reduction for acoustic/analogue do, as you say, "mess with the signal", IE. Introduce distortion, and the more noise you try to remove, the more distortion you introduce. Introducing distortion is not increasing resolution, it's reducing resolution. The trick of applying noise reduction is to reduce the amount of noise somewhat (typically by no more than a few dB), without the added distortion becoming audible/objectionable.

2. And that's why dither is an intrinsic part of digital audio, because it results in "no quantisation errors" and therefore effectively "infinite resolution" (but with dither noise).

3. Unfortunately, that figure doesn't really tell us much. Where it's measured/calculated makes a big difference because a DAC (or ADC) chip is by definition partly an analogue device and therefore subject to the physics of analogue signals (EG. At least some thermal noise) but of course now we're effectively talking about analogue, not digital.
3a. Yes, in practice we cannot achieve infinity resolution or even close to infinity because we always have to enter the analogue realm (and then the acoustic realm).
Us claiming that digital audio production has infinite resolution is almost as bad as those analog guys making that claim.

Not really, although we do have to be careful about what we mean, for example that we are in fact talking about digital audio and not about analogue audio, EG. Not about ADCs or DACs, which are partly analogue devices. And obviously, we're not talking about a digital audio system, which of course isn't actually a digital audio system, it's an analogue/acoustic system with some digital components/processes, which is therefore constrained by the limits of analogue circuitry and the laws of motion in creating an acoustic signal. And even when talking specifically about digital audio, there are still conditions to such a claim, for example, infinite resolution within a limited/specified audio frequency bandwidth. Furthermore, at the dynamic ranges and bit depths employed in digital audio, we run into the problem of what audio frequencies actually are, and therefore exactly what is meant by an infinite resolution of them. For example, a sound wave is a pressure wave travelling/propagating through a medium (air in our case), IE. A sound wave is the compression and rarefaction (movement) of air molecules, but what happens when we have a massive dynamic range? In the case of say 24bit (or 16bit with aggressive noise-shaping) the amount of energy represented at the bottom of the dynamic range (EG. -144dB) cannot compress and rarefy enough air molecules by a sufficient amount to propagate a sound wave and therefore a sound wave doesn't exist. In other words, a sound wave is itself a finite entity and therefore can't have infinite resolution. To rephrase your statement, we could say that digital audio production has a resolution/DR which exceeds the DR possible for a sound wave to actually exist and therefore effectively has infinite resolution. This is in stark contrast to analogue guys making that claim because in the case of an analogue audio production, we have numerous analogue units/signal paths each of which introduce cumulative distortion/noise, which brings the effective dynamic range of an analogue music production down to about 80dB at best (but more likely <60dB) and -80dB represents way more energy than is needed to propagate a sound wave. Digital audio also has numerous processes, each of which also introduces cumulative distortion/noise but professional audio production environments are 64bit float, so each process is adding noise/distortion at around the -350dB level and even a thousand or more still would not add up to enough for a sound wave to exist (and would therefore still effectively be infinite resolution).

G
 
Last edited:
Dec 15, 2019 at 7:59 AM Post #5,423 of 6,480

old tech

500+ Head-Fier
Joined
Jul 8, 2015
Posts
830
Likes
414
Location
Sydney, Australia
@gregorio thanks for the explanation. Thinking of resolution as being infinite in terms of reproducing a sine wave is a good way of looking at it. What did my head in was that infinite resolution doesn't exist in the natural world, ie even the propagation of air molecules in the atmosphere will result in some noise and distortion along the way. so how could it in digital production when there is quantisation errors or alternatively, dithered noise that obscures the true signal (however below human thresholds that may be)?

I'm a bit surprised though that noise reduction cannot increase resolution (whether it be analog or digital processes). I thought that it was a trade-off, ie if the denominator (N) could be reduced by a greater amount than the reduction effect in the numerator (S) then, assuming distortion is kept at bay, by definition SNR must increase. I recall SNR specs on end tape recorders usually quoting a higher SNR if NR is used (like Dolby A or DBX etc ). I forget the brand, but a final reiteration of a well known studio tape recorder claimed to achieve nearly 90db with NR.
 
Dec 15, 2019 at 3:11 PM Post #5,424 of 6,480

bigshot

Headphoneus Supremus
Joined
Nov 16, 2004
Posts
23,420
Likes
4,941
Location
A Secret Lab
Noise reduction just reduces noise. It doesn't extend signal down below the noise floor.

I'm not going to consider any sound reproduction format to be infinitely perfect unless it covers the range of 100GHz to 10THz, the point where sound becomes light.

Now that we've answered this one, can we get an answer to how many angels can dance on the head of a pin?
 
Last edited:
Dec 15, 2019 at 3:37 PM Post #5,425 of 6,480

TheSonicTruth

1000+ Head-Fier
Joined
Dec 19, 2014
Posts
1,018
Likes
126
Noise reduction just reduces noise. It doesn't extend signal down below the
noise floor.

I'm not going to consider any sound reproduction format to be infinitely perfect unless it covers the
range of 100GHz to 10THz, the point where sound becomes light.

Now that we've answered this one, can we get an answer to how many angels can dance on the
head of a pin?


I hereby declare you well enough to be relocated from Acutes back to Chronics in Nurse Ratched's ward. Congratulations! :D
 
Dec 15, 2019 at 6:09 PM Post #5,426 of 6,480

Arjey

100+ Head-Fier
Joined
Nov 2, 2019
Posts
156
Likes
106
Location
Belarus
So I have a question. 24 bit is useless for listening to music. Does that make DSD useless too? Because I could swear I can hear a difference..
Or is there a difference, because DSD isn't just more bits? What's the difference between DSD and other formats?
 
Dec 15, 2019 at 6:40 PM Post #5,428 of 6,480

bigshot

Headphoneus Supremus
Joined
Nov 16, 2004
Posts
23,420
Likes
4,941
Location
A Secret Lab
On SACDs the DSD track is generally at a higher volume level than the redbook layer. That skews comparison tests if you don't do careful level matching. Even worse, there are quite a few SACDs that have different masterings on the two layers, and I even found one that had a completely different mix. It is very difficult to do a controlled test between SACD and CD. I struggled for a month to find an SACD that was a fair comparison. I finally found one with a DSD recording that had the exact same master on the redbook layer. A sound engineer friend of mine and I racked them up side by side and couldn't detect any difference between them at all.

If someone has a report of a well conducted blind listening test that found otherwise, I would like to hear about it. But I doubt any such thing exists.

If you'd like to set up a controlled test to find out for yourself, I would be happy to help you with advice and tips on how to do that.
 
Last edited:
Dec 15, 2019 at 6:51 PM Post #5,429 of 6,480

Arjey

100+ Head-Fier
Joined
Nov 2, 2019
Posts
156
Likes
106
Location
Belarus
Just an unrelated to the topic question. Is there a way to remove noise from old records? I know about the feature in audacity and other apps, but it doesn't really work.. I like some old jazz, but it has so much noise, it's basically impossible to listen to :frowning2:
 
Dec 15, 2019 at 6:57 PM Post #5,430 of 6,480

old tech

500+ Head-Fier
Joined
Jul 8, 2015
Posts
830
Likes
414
Location
Sydney, Australia
So I have a question. 24 bit is useless for listening to music. Does that make DSD useless too? Because I could swear I can hear a difference..
Or is there a difference, because DSD isn't just more bits? What's the difference between DSD and other formats?
As I understand it, DSD is only 1 bit but sampled in the megahertz range to create a wide enough band to shape all that 1 bit noise into (this is an oversimplification...)

I don't doubt that you hear a difference between DSD and other formats, I often do as well.

The question is what is causing the difference? Typically the main reasons are that different or higher quality masters were used for the DSD version or just plain expectation biases - ie you expect the DSD version to sound better so your brain is fooled into perceiving the sound as better. Then there are a myriad of other reasons such as not level matching the listening samples, not testing blind and so on.

All I know is that after more than 30 years that DSD and other hi res formats have been available to the public, why are we still debating this while, for example, no-one really debates whether LP records sound better than 8 Track cartridges? Surely the jury would have been back long ago, yet there are no controlled tests that convincingly demonstrate that DSD or other hi res formats sound different to 16/44 and many studies supporting the claim that like 16/44, the hi res formats are equally transparent to the human ear (such as in the link below).

http://drewdaniels.com/audible.pdf
 

Users who are viewing this thread

Top