Article: Digital Vinyl: Temporal Domain

[mansr]

Of course on any subject outside of people's own knowledge, the determination of whether someone with expertise is trustworthy must depend on other criteria. Should I automatically trust the 3% of climate scientists who have "taken the time to study" anthropogenic global warming and concluded it does not exist?

 

Are you suggesting there is anything but complete consensus among scientists and mathematicians regarding the validity of the sampling theorem?

 

In this case from what little I know as a layperson, you are mostly correct. It is absolutely true that the Sampling Theorem proves any sample rate adequate for a band-limited signal (above twice the highest "frequency of interest") defines the signal at all time points. However, the Sampling Theorem contains idealizing assumptions that don't exist in the real world - perfectly band-limited signals, infinite time to do the filtering to reconvert digital to analog, etc. So there is in fact in practicality a limit on how finely a real-world signal, digitally sampled and reconverted to analog, can be specified in time. It is much less than the time between samples. I believe Dennis (esldude) has this information and has mentioned it previously on the forum, but I don't remember it.

 

Yes, it's possible to calculate the effect of limited precision and whatever other imperfections there are in a practical system. For 16/44.1 the time accuracy is on the order of picoseconds, I don't remember the exact figure.

 

Perhaps a demonstration with a DAC and scope will convince. This is the left/right zero-crossings of an iFi Nano DAC playing the same sine wave at 44.1 kHz on both channels:

 

 

There's an inherent skew of about 27 ns, so this is just for reference.

 

Now we delay the right channel slightly:

 

 

The inter-channel difference has increased by about 8 ns which is quite substantially less than the 22 μs sample interval.

[mansr]

Wow.

 

Are you expecting me to recite all the maths necessary to fully understand the sampling theorem here? That would be a few hundred pages, and I doubt I'd do as a good a job of it as the numerous people who have written books on the subject. If you're interested, get one and read it.

[Fokus]

Half-truths and misrepresentations. This article is written by someone who does not understand signal theory, how digital works, and how analogue works, or someone with an agenda. A shame. It is perfectly fine to like the sound of vinyl. LP recording and playback is a triumph of electro-mechanical engineering (even though not without flaws). And that is all justification the format needs. No-one needs disinformation and lies.

[Jud]

Are you suggesting there is anything but complete consensus among scientists and mathematicians regarding the validity of the sampling theorem?

 

 

You couldn't have missed the mark by more if you were trying (or perhaps you were trying).

 

 

 

Nope, you were asked, after saying quite cryptically that the writer was mistaken due to the Sampling Theorem, to explain how the Sampling Theorem made what he said about timing accuracy of analog signals reconstructed from digital samples incorrect.

 

 

 

You chose to misinterpret this as Chris questioning the Sampling Theorem itself. I tried to give you an example of how unhelpful your "You just have to trust the people who know" response was, and you chose to misinterpret that.

 

 

 

This strikes me as a lot of work to avoid an explanation you've now mostly given in real-world terms with the graphs above, so I'm not sure exactly what all of it got you.

[mansr]

Nope, you were asked, after saying quite cryptically that the writer was mistaken due to the Sampling Theorem, to explain how the Sampling Theorem made what he said about timing accuracy of analog signals reconstructed from digital samples incorrect.

 

I referred to what esldude said: http://www.computeraudiophile.com/f3-article-comments/article-digital-vinyl-temporal-domain-31816/#post641392

[Jud]

I referred to what esldude said: http://www.computeraudiophile.com/f3-article-comments/article-digital-vinyl-temporal-domain-31816/#post641392

 

And you and he were quite correct. Chris then asked if you could put the reason for your conclusion on a level that would help others understand:

 

 

 

99.999% would really like help understanding how and why. It's over our heads.

 

 

 

All you had to do then was say, "I can't or won't boil down the way the Sampling Theorem works here, but look at these two graphs: One signal can be moved by a matter of nanoseconds in relation to the other, when both are reconstructed from a 44.1KHz-sampled signal (22 microsecond time interval). So nanosecond, or in fact picosecond, time accuracy is available from real world 44.1KHz digital sampling." After all, you wound up doing that anyway. But you made a great show of *not* explaining before you eventually did. Puzzling.

[mansr]

And you and he were quite correct. Chris then asked if you could put the reason for your conclusion on a level that would help others understand:

 

 

 

All you had to do then was say, "I can't or won't boil down the way the Sampling Theorem works here, but look at these two graphs: One signal can be moved by a matter of nanoseconds in relation to the other, when both are reconstructed from a 44.1KHz-sampled signal (22 microsecond time interval). So nanosecond, or in fact picosecond, time accuracy is available from real world 44.1KHz digital sampling." After all, you wound up doing that anyway. But you made a great show of *not* explaining before you eventually did. Puzzling.

 

I wasn't at my desk when I first replied. Also, theoretical explanations and experimental results are not mutually exclusive.

[KDinsmore]

Wow! This is denigrating into a "Cables-and-Interconnects-make-a-difference" conversation. I believe analog does sound better. I have a degree in Mathematics, studied the physics of sound (In 1973 I started by using the Helmholtz book "On the sensations of tone as a physiological basis for the theory of music" !?!), am a musician, recorded music in the two inch reel-to-reel days and......BIG DEAL.....I've heard all of my favorite music on vinyl and digital and vinyl wins. That being said I no longer have the physical space to store all the vinyl, all the equipment involved I would personally want, fussing over it.....I will stay with full on digital, hope for the best in future DAC's, enjoy many vinyl rips I own. But I still know that playing back The Allman Brothers "Hot Lanta" at full tilt never sounds as magnificent as it does on vinyl playback.

[cbee]

Are you suggesting there is anything but complete consensus among scientists and mathematicians regarding the validity of the sampling theorem?

 

 

 

Yes, it's possible to calculate the effect of limited precision and whatever other imperfections there are in a practical system. For 16/44.1 the time accuracy is on the order of picoseconds, I don't remember the exact figure.

 

Perhaps a demonstration with a DAC and scope will convince. This is the left/right zero-crossings of an iFi Nano DAC playing the same sine wave at 44.1 kHz on both channels:

 

[ATTACH=CONFIG]33665[/ATTACH]

 

There's an inherent skew of about 27 ns, so this is just for reference.

 

Now we delay the right channel slightly:

 

[ATTACH=CONFIG]33666[/ATTACH]

 

The inter-channel difference has increased by about 8 ns which is quite substantially less than the 22 μs sample interval.

Your example is for a constant running sine wave. It does not address the issue of resolving where in a sample a transient starts.

Are you suggesting there is anything but complete consensus among scientists and mathematicians regarding the validity of the sampling theorem?

 

 

 

Yes, it's possible to calculate the effect of limited precision and whatever other imperfections there are in a practical system. For 16/44.1 the time accuracy is on the order of picoseconds, I don't remember the exact figure.

 

Perhaps a demonstration with a DAC and scope will convince. This is the left/right zero-crossings of an iFi Nano DAC playing the same sine wave at 44.1 kHz on both channels:

 

[ATTACH=CONFIG]33665[/ATTACH]

 

There's an inherent skew of about 27 ns, so this is just for reference.

 

Now we delay the right channel slightly:

 

[ATTACH=CONFIG]33666[/ATTACH]

 

The inter-channel difference has increased by about 8 ns which is quite substantially less than the 22 μs sample interval.

 

 

Sent from my Nexus 7 using Computer Audiophile mobile app

[mansr]

Your example is for a constant running sine wave. It does not address the issue of resolving where in a sample a transient starts.

 

 

 

Sent from my Nexus 7 using Computer Audiophile mobile app

Everything is a sum of sines.

[cbee]

Everything is a sum of sines.

Yes, but can you resolve where the transient started within one 44Khz sample to the accuracy you were able to shift the constant sine wave in your example?

 

Sent from my Nexus 7 using Computer Audiophile mobile app

[Jud]

Yes, but can you resolve where the transient started within one 44Khz sample to the accuracy you were able to shift the constant sine wave in your example?

 

Sent from my Nexus 7 using Computer Audiophile mobile app

 

 

As long as the transient doesn't involve frequencies equal to or higher than 22.05KHz (for a 44.1KHz sample rate), yes.

 

 

Edit: A little more detail - As I mentioned in my previous attempt at an explanation, once you have adequate sampling (the "third sample point" in my prior explanation), you have mathematically established exactly where the signal had to be at every point along its length.

[Jud]

Wow! This is denigrating into a "Cables-and-Interconnects-make-a-difference" conversation. I believe analog does sound better. I have a degree in Mathematics, studied the physics of sound (In 1973 I started by using the Helmholtz book "On the sensations of tone as a physiological basis for the theory of music" !?!), am a musician, recorded music in the two inch reel-to-reel days and......BIG DEAL.....I've heard all of my favorite music on vinyl and digital and vinyl wins. That being said I no longer have the physical space to store all the vinyl, all the equipment involved I would personally want, fussing over it.....I will stay with full on digital, hope for the best in future DAC's, enjoy many vinyl rips I own. But I still know that playing back The Allman Brothers "Hot Lanta" at full tilt never sounds as magnificent as it does on vinyl playback.

 

Mastering is certainly involved, and perhaps differences in analog and digital systems in terms of design and parts quality - I don't know what your analog and digital systems look like.

 

 

 

As an example of mastering differences in my experience, the Who's Tommy and Steely Dan's Gaucho DVD-A versions weren't a patch on my LPs, but the SHM-SACD versions of each sound great.

[Jud]

I wasn't at my desk when I first replied. Also, theoretical explanations and experimental results are not mutually exclusive.

 

Thanks, my apologies for assuming it was simply stubbornness.

[cbee]

Jud, Are you sure? I am not asking if it can resolve the frequencies but where within the width of the sampling pulse the transient starts. MQA dont seem to think it can be done, that´'s why they have come up with different sampling and filtering methods

[Jud]

Jud, Are you sure? I am not asking if it can resolve the frequencies but where within the width of the sampling pulse the transient starts. MQA dont seem to think it can be done, that´'s why they have come up with different sampling and filtering methods

 

 

MQA has not found an exception to the Sampling Theorem.

 

 

 

What they are talking about is something different.

 

 

Mathematically, the more precise your filtering is in terms of time, the less precise it is in terms of frequency, which leads to the real world job of creating a filter that will find the appropriate balance between "time domain" distortions (ringing) and "frequency domain" distortions (aliasing and intermodulation distortion).

 

 

MQA has made a great show of moving its balance away from the place where the vast majority of people working in digital audio have chosen to put their emphasis. They have to some extent disdained trying to have correct response in the frequency domain. Whether this is enough to provide a better time domain response as MQA claims, we don't know, since we haven't seen any data/measurements thus far. What measurements we have seen in the frequency domain would indicate it is a good idea to be skeptical of MQA's claims.

 

 

None of this prevents you from listening to MQA and enjoying it if you like. But it doesn't appear to be any sort of game changer in technological terms.

[cbee]

As long as the transient doesn't involve frequencies equal to or higher than 22.05KHz (for a 44.1KHz sample rate), yes.

 

 

Edit: A little more detail - As I mentioned in my previous attempt at an explanation, once you have adequate sampling (the "third sample point" in my prior explanation), you have mathematically established exactly where the signal had to be at every point along its length.

I see what you are saying.

 

Sent from my Nexus 7 using Computer Audiophile mobile app

[cbee]

MQA has not found an exception to the Sampling Theorem.

 

 

 

What they are talking about is something different.

 

 

Mathematically, the more precise your filtering is in terms of time, the less precise it is in terms of frequency, which leads to the real world job of creating a filter that will find the appropriate balance between "time domain" distortions (ringing) and "frequency domain" distortions (aliasing and intermodulation distortion).

 

 

MQA has made a great show of moving its balance away from the place where the vast majority of people working in digital audio have chosen to put their emphasis. They have to some extent disdained trying to have correct response in the frequency domain. Whether this is enough to provide a better time domain response as MQA claims, we don't know, since we haven't seen any data/measurements thus far. What measurements we have seen in the frequency domain would indicate it is a good idea to be skeptical of MQA's claims.

 

 

None of this prevents you from listening to MQA and enjoying it if you like. But it doesn't appear to be any sort of game changer in technological terms.

Have you read their paper thar includes the triangular sampling? That is where they claim to be get better temporal accuracy in the ADC. The filtering is another part of their process.

I haven't heard any MQA'd music yet. I have a feeling I am going to be underwhelmed!

 

Sent from my Nexus 7 using Computer Audiophile mobile app

[Jud]

Have you read their paper thar includes the triangular sampling? That is where they claim to be get better temporal accuracy in the ADC. The filtering is another part of their process.

I haven't heard any MQA'd music yet. I have a feeling I am going to be underwhelmed!

 

Sent from my Nexus 7 using Computer Audiophile mobile app

 

 

Here's Miska talking about MQA at the ADC stage (on Roon's forum):

 

 

 

No, they just run decimation (sample rate reduction) using their filter (kernel) at mastering stage to reduce the source to 88.2/96 kHz rate for encoding. Then they use shaped high level dither to hide distortion from the leaky filter and trying to maintain dynamic range at lower frequencies despite only few bits. They use the entire top octave for filter roll-off (and aliasing), because they think that frequencies above 20 kHz are not useful so those can be sacrificed to keep ringing of the filter to minimum.

 

 

 

See where he talks about the filter kernel? MQA uses a "triangular" filter kernel. This is part of "decimating" (reducing) the sample rate of the source. It is not magic and it does not provide starting points of between-sample signals that others cannot. And when Miska talks about "leaky" filtering he means the poor frequency response performance of the MQA filters on the DAC side.

[esldude]

Hi esldude - Igor has used and researched many products for ripping vinyl. He has incredible passion for his project. He settled on the Pure Vinyl product as being the best, but he has absolutely zero relation to the company.

 

I'm not sure if that's what you were getting at, but I wanted to clarify for everyone.

 

Thanks for answering that question. I too have used Channel D software for needledrops. It is excellent software.

 

Frankly Chris I was dismayed to see this article on your site. If this had been a principle at Channel D it would have put a big dent in my respect for their company. Most of the article is untrue misinformation. Old myths that are factually not so. One need not perpetuate bad info to have a passion for vinyl.

[all300b]

From a non-engineering standpoint, there just has to be something different about sound that is born from mechanical vibration (analog) versus sound that is born from electricity (digital)- with "born" meaning how it starts in your living room. Yes, the turntable is powered- but only to spin the record. Somehow, translating vibration into electricity via a magnet just sounds different- and often better- than recreating an electrical waveform from a digital signal. I (remember) hearing a more realistic increase in loudness and image size when I had vinyl- does that reflect the way a magnet responds to larger vibrations? Have always hoped someone would delve more into these questions than samples rates and bits.

[The Computer Audiophile]

Thanks for answering that question. I too have used Channel D software for needledrops. It is excellent software.

 

Frankly Chris I was dismayed to see this article on your site. If this had been a principle at Channel D it would have put a big dent in my respect for their company. Most of the article is untrue misinformation. Old myths that are factually not so. One need not perpetuate bad info to have a passion for vinyl.

 

This is beyond my technical skill level. I honestly don't know what's right, wrong, or partially correct etc...

 

I saw the AES article and don't know if it discusses the same thing as discussed here.

 

Fortunately, we allow everyone to comment here and hopefully we'll get to the right place through "peer" review.

[esldude]

Hi esldude. The article does not describe the spacing of samples, but the temporal resolution. How close it can be placed in time one sound from another sound that the human ear could be the difference.

And if you want to be precise, 192 kHz is 5.2 microseconds,

96 kHz is 10.4 microseconds

48 kHz is 20.8 microseconds.

(Note, time resolution doesn't depend on the word length, e.g., 16/44.1 and 24/44.1 have exactly the same temporal resolution - 22.7 microseconds.)

 

Best

 

Pure Vinyl Club

 

Listen to short demos of the LP Records

and share your experience and observations.

 

 

As already said, you are incorrect about this. Trying to be respectful and courteous to you, but your information is wrong.

 

The correct theoretical number of time resolution with 16/44 is about 56 picoseconds. With dither it is even less.

 

Real world, jitter in clocks can be above 56 picoseconds, but very nearly all modern gear has jitter at or below 300 picoseconds which is still orders of magnitude lower than the 10 microseconds of human hearing ability. It is not difficult for instance to show time delay between one meter and two meters of cable with 16/44khz digital. A meter of time for the signal is around 3 nanoseconds or so.

[cbee]

Here's Miska talking about MQA at the ADC stage (on Roon's forum):

 

 

 

 

 

 

 

See where he talks about the filter kernel? MQA uses a "triangular" filter kernel. This is part of "decimating" (reducing) the sample rate of the source. It is not magic and it does not provide starting points of between-sample signals that others cannot. And when Miska talks about "leaky" filtering he means the poor frequency response performance of the MQA filters on the DAC side.

 

If you read Bob Stuarts Audio Engineering Society Convention Paper 9178 ( sorry no link ) the "triangular" filter ( actually a b spline kernel, whatever that is), they do actually claim it can resolve starting points between samples because the samples overlap.

 

There is also some mention of this in one of their patents, unfortunately the diagrams are missing:-

 

https://www.google.com/patents/WO2014108677A1

[Jud]

If you read Bob Stuarts Audio Engineering Society Convention Paper 9178 ( sorry no link ) the "triangular" filter ( actually a b spline kernel, whatever that is), they do actually claim it can resolve starting points between samples because the samples overlap.

 

There is also some mention of this in one of their patents, unfortunately the diagrams are missing:-

 

https://www.google.com/patents/WO2014108677A1

 

 

From the patent:

 

 

 

Preferably, the downsampler comprises a decimation filter specified at the first sample rate, wherein the asymmetric component of response of the decimation filter is characterised by an attenuation of at least 32dB at frequencies that would alias to the range 0-7 kHz on decimation. The range 0-7kHz is the range where the ear is most sensitive.

 

 

Holy s**t. Perhaps Miska and mansr have been too polite.

 

 

 

That being said: I have heard MQA sources where I also have the hi res files. I thought it sounded pretty good, not as good as the hi res. (mansr has mentioned that the bad frequency domain performance might not be audible.) And as I think is nearly always true, almost regardless of format, where Tidal has provided MQA masters that are superior to the masters for the RedBook-resolution files they had, I think the MQA version sounds better simply because of the mastering.