Article: Digital Vinyl: Temporal Domain

[Jud]

My point is that you are wrong.

 

Not sure whether I should "Like" this, but I did LOL.

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

I wanted to say that I like thoughtful reflections on listening preference. Particularly not draped in effusive comments and Frasier Crane-like pretentiousness. The use of examples is intriguing, too. Thank you.

 

What I might offer for your consideration(s) is that you start with a definition or two. Such as Temporal Domain.

Temporality is, to my knowledge, described as the ratios of, or relative intervals between events. The temporal domain carries no information about frequency or sequence.

 

That said, each of the domains- frequency, sequential, and temporal do overlap with the information about the time domain contained within their two counterpart component domains.

 

Next- imparting to the reproduction characteristics that cannot exist if they were not (authentically) available during production is a question here.

A perspective change might be useful from the micro-tech to the recording industry. The producers of the music product have very different agendas than the musicians. What you (we) get is a compromised, not optimized result.

Multiple mic-ing, 28 (or more) tracks, mixing boards, "pressing" or quantizing create a scale of complexity that is beyond the human span of control.

 

The recording industry is trying to 'train' our ears to producers interests. As Apple battles for its larger distribution access agenda, for example.

 

Live group musicians and their audiences have a very different 'training' than studio and over-dub musicians. Blending 28 events are limited to a very narrow subset of the domain attributions to give producers some control over their branding. Regardless if you like Phil Spector's, to name just one, signature sound or not.

 

Blending different performances of isolated individuals simply won't have authenticity to time or phase coherency. Let alone the "self-organization" and complexity between instruments and performers in a collective event. Even to the coupling of the environment of the "event".

 

Then there is the final mixing. Which is a marketing decision for the psycho-acoustic characteristic of the markets' producers expect to reach. Once it was phonograph, AM radio, or FM radio. Ear buds are very different than large rooms with no square corners. Many events that would have unfolded are sheared off. If you found something...its not my intent to marginalize it. It just isn't authentic- from the performance(s).

Last- a byproduct of my career, with some legacy collected media, one can use a LiDAR that creates a 3d map of vinyl. You don't have to destructively drag a diamond through the lacquered vinyl media. Even if everything was equal, which it is not, determinism is an assumption better left to Newtonian antiquity. Uncertainty (scientific) yields only more artifacts- not fewer.

 

There was a turntable that used a laser, but IIRC the surface noise was a problem.

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

the cepstrum is defined as the inverse DFT of the log magnitude of the DFT of a signal

 

 

OK, found this portion of the copy-and-paste from a Texas A&M University Intro to Speech Processing course. Wonder where the rest came from?

[Jud]

I really do dislike these sorts of graphics (the impulse ones as presented here, and the square wave ones I've seen as well), because they are fiction for an additional reason.

 

These impulses contain a very large frequency range that is not limited to the range of audibility (~20kHz or less), 24kHz, 48kHz, or even 96kHz. *That* is why you see the "ringing" (the additional waves) and the inability to reconstruct the full amplitude until you get to DSD frequency levels (2.8MHz).

 

Remember the Sampling Theorem: You must sample at above double the highest "frequency of interest."

 

All graphs like the above demonstrate is that sampling doesn't work well when the Sampling Theorem's conditions are violated. Duh. So please, folks reading marketing material that is couched in the form of technical papers: be careful.

[Jud]

In fact, this is a popular picture with a chart showing ANALOG (signal from the best microphone), as a reference standard.

 

Igor, I would be very surprised if this "click" was indeed a naturally produced acoustic signal picked up by a mic. Can you think of any naturally produced "click" sound in the real world with frequency components above 96kHz (as is obvious from the problems in reconstruction using a 192kHz sample rate) that has no reverberation at all, like this one?

[Jud]

Jud, I actually wrote with irony about this "popular" picture. Look again at the picture - there the peak of the DSD is even higher than the peak ANALOG ;-))

 

Thanks, Igor. Irony (which I love) is among the most difficult things to try to translate across even very minor language barriers.

[Jud]

Like many things, it comes down to assumptions and math. The big assumption is that the human auditory system does not respond to frequencies >20kHz. If this is true, then the Shannon-Nyquist sampling theorem provides a transformation between the time and frequency domains ... the Fourier transform ... which equates a signal in the time domain to the Fourier transform in the frequency domain.

 

So heres the thing... when you say that the human auditory system can respond to *transients* at x,y, or z microseconds, nanoseconds etc ... you are *exactly* saying that it is responding to certain frequencies, and if the 44kHz sampling rate cannot sample these transients sufficient for the auditory system to hear, then you are saying that the auditory system is responding to frequencies higher than 20 kHz. Again it comes down to your assumption. These two things are mathematically the same according to Shannon-Nyquist.

 

My own position is that just because people cannot generally hear an isolated tone > 20 kHz or even 16 kHz, that, because the system is non-linear, that does not imply that the system cannot respond in some fashion to a tone or combination of tones > 20 kHz i.e. a short transient. So the other big assumption in that argument is that the auditory system is fundamentally linear (but it is highly non-linear). Consider, as an example, IM distortion which, for an amplifier, might cause problems when fed a 100kHz or 1MHz signal. That's because of non-linearities in the electronics, and the human auditory system has its own non-linearities.

 

The old saying about: ASSUME => ASS-U-ME

 

So, the math is entirely correct. Its the assumptions behind the math that should be questioned.

 

Is this understandable?

 

Jonathan

 

I remember reading that different parts of the brain are used for processing transients vs. tones, but that doesn't answer whether the sensory end would be capable of detecting such fast-rise-time transients in the first place. I would guess there's got to be research, but I haven't been able to find anything right on point.

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

Those non-linear electronics that can display IM do have response to those frequencies even though non-linear. Even if non-linear, how can the ear create IM like distortions unless it responds to those higher frequencies? How could it even in theory respond to high frequency transients, but not high frequency steady tones?

 

Yes the brain processes transients differently that steadier sounds. But that doesn't free it from being limited by the frequencies the ear is able to put upon the auditory nerves. The stereocilia that convert sound to nerve impulses are tuned to respond no higher than a center frequency of 15 khz or so. That they are something like weakly tuned filters is why we have some response to slightly higher frequencies (near 20 khz when young). What about that would appear able to respond to faster transient events?

 

You can state that maybe it does, but what is the hypothetical mechanism behind how that would happen?

 

There may indeed be no response to transients with faster rise time than the highest sustained tone one can hear. But I sure would like to read a research publication that provides a definitive answer to the question.

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

That is one heck of a microphone, then. Tell us, in what sort of universe do you live?

Tell us more: what would the impulse response of something like a U47 look like?

 

'Sokay. He said in response to my question along the same lines that he was speaking ironically, but it was apparently lost in translation.

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

I wouldn't be so quick to question the microphone. Assuming the microphone is accurate and not knowing the specs assuming it is bandwidth limited to 100 kHz -- I would suspect that the PCM 96/192 recordings might have more jitter (specifically close in phase error) than optimal -- that would produce the pattern we are shown

 

It's the complete lack of any reverb that makes me question whether it could possibly be a mic pickup of a natural acoustic signal.

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

It all depends on what kind of synchronization tools you use when registering an analog signal in digital domain and then reconstructing it (playing it back in an analog way).

 

If you use a standard clock (quartz oscillator, possibly with the exception of the Grimm Audio CC1), then you can not completely solve the jitter problems even at a frequency of 44.1kHz. And a bigger problem will be when you use 96 or 192kHz.

 

We use for recording in the studio Antelope Audio Audiophile 10M (10 MHz Rubidium Atomic Reference Generator, Frequency Accurancy better than 0.03 PPB (parts per BILLION)) and therefore we have no problems, even when we record on the frequency 192kHz.

 

Though I am guessing you are likely not "recording at 192kHz," but rather the A/D converter runs the analog input through sigma-delta modulation internally, then internally decimates that to 192kHz.

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

Antelope Audio Pure2:

Burr-Brown (TI) A/D converter PCM4222 (High-Performance, Two-Channel, 24-Bit, from 8kHz to 216kHz Sampling Multi-Bit Delta-Sigma Analog-to-Digital Converter)

 

Yup, Delta-Sigma. A/D converters that go direct to PCM are very rare.

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

I'm not saying it's the case here, but if the recording device has, say, a non-flat frequency response, that can be corrected digitally. It might also be possible to remove some noise.

 

Likewise, a touch of EQ can compensate for acoustic issues in the room such as an unwanted resonance. Of course this will deviate from what was actually heard during the recording, but it may well sound better.

 

Just wondering if anyone knows - could something like the Plangent process (correction for tape speed variation) be used for DAT, or is that irretrievably baked into the original digital file?

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

Besides this, there are the possibilities of intermodulation distortion when ultrasonic frequencies are included.

 

But people always, always, ALWAYS want more. So, in five years it will be 48 bit and 7 gazillion kHz sampling rates, because "bigger numbers are better" if you don't understand the technology. And that's the main problem: People don't understand the Nyquist sampling theorem, so they really do think that a higher sample rate more closely approximates an analogue sound wave. But how can a signal that already includes everything contain more than everything?

The pictures we usually see are misleading and confusing. The only things that can exist between "the steps" on a digital signal are information at or above the Nyquist frequency (in the case of CDs, information at or above 22.05 kHz).

 

More than everything, I want it!

 

One other possibility with higher resolutions is intermodulation from ultrasonics that are actually signal rather than noise (e.g., cymbals, some of which have ~40% of their acoustic energy at ultrasonic frequencies). But that assumes a lot of good recording practices, and that playback systems could/would properly reproduce the intermodulation.

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

I'm so happy I can pick from a genius's brain !!

What happens, what does the sampling theory say when the signal is no more than 20K and sampled at 44.1 but varies every 5 microseconds ?

You can imagine a continuous decay but what would truly be worth of your genius would be bursts emitted every microsecond in a 20_20K range sampled at 44.1

 

A signal that is "no more than 20k" cannot by definition change faster than the maximum slope of a 20KHz sine wave. Any burst, pulse, square wave, with a vertical or near-vertical slope contains frequencies higher than 20KHz, so naturally a sample rate of approximately twice that would be insufficient.

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

Assuming the ultrasonics we are discussing are created by musical instruments and are present when listening to live music, there is no reason to believe that these specific ultrasonics are harmful in any way. The very fact that you can perceive ultrasonics at all suggests that their absence from a recoding may form part of the difference between hearing the performance "live" vs "recorded".

 

I don't know anyone who gets headaches when listening up front and close to a quartet who likes classical music, same for jazz for people who like jazz. Perhaps its the ultrasonics which sends the "chill down your spine" when listening up front and close to live music?

 

Ultrasonics contribute very little energy to the sound of most instruments, so my guess (it's only that) is *if* our experience is in any way diminished due to lack of ultrasonics, it would be because it is lacking just that last tiny bit to match the pattern that for us is the experience of a live violin, for example. (See my discussion of pattern matching in the other thread - was it the one about cables?). It would be something like CGI modeling of hair flowing and bouncing: so very, very close to perfect, but just not that last infinitesimal step to "real."

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

thank you. I'll sleep on your answers and maybe figure out. I did not have in mind that the bursts would stop and start in 1 microsecond. Just in mind the idea that the global envelope of an orchestra's sound is continuously changing and that in the context of this thread this should be represented by bursts of whatever audible frequencies emitted at a faster pace than the sampling rate. They don't have to stop and cycle in 1 microsecond : simply there's an event changing the global sound every microsecond. At the end of the day this might be as simple as the belief that more gives a better description but on the other hand why non quantic physic should be counterintuitive?

 

When you say that something can change the sound every microsecond, it depends by how much. When you see an orchestra's sound captured by mics and shown on an oscilloscope, the sound waves have a slope. In other words, the waves move a certain distance horizontally as they move vertically, and the ratio of these - how steep the steepest wave is - can't be steeper than a 20KHz sine wave, or a sample rate of around 44.1kHz won't be adequate to reproduce it.

 

A 20KHz sine wave won't move vertically very much at all in one millionth of a second horizontal distance on the oscilloscope, since the wave will take 1/20,000 of a second to complete. That is 50 microseconds, so only 1/50th of the total vertical path of the wave will be complete in a microsecond.

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

Bear in mind that suddenly starting or stopping a sine wave, even a low-frequency one, and even if done at a zero crossing, introduces a discontinuity in the first differential which amounts to a wideband burst. The same is true for any sharp corner in the waveform.

 

https://en.wikipedia.org/wiki/Gibbs_phenomenon

[Jud]

Yes that among other things shown in the video is why I have it in my signature. Why I have said at least once or more that so many of these ridiculous arguments about what digital can and cannot do would be avoided if everyone commenting would watch and take time to understand that 23 minute video.

 

Possibly number one is the idea repeated millions of times that digital audio loses whatever happens, starts or stops between the sample points. Yet the video demonstrated that as NOT so. And did so with analog inputs and analog monitored outputs.

 

So you get these articles like this about LP being better at timing that digital just can't touch, and it is demonstrably wrong and way off base. If your number one beginning premise is so wrong, everything that follows from it is highly suspect, and usually going to be wrong. Which is why I was dismayed to see you give home page space to this article.

 

If temporal resolution is as important as Pure Vinyl thinks, then digital is the way to go for sure.

 

Pure Vinyl had this image in the article to illustrate the steep transients that digital can't catch implying a vertical ramp up to impulsive signals:

[ATTACH=CONFIG]34032[/ATTACH]

 

It depends upon how close you look. Here is the plot of an actual cymbal recording at 176 khz, using a wide bandwidth Earthworks microphone. This cymbal is one of the quickest most impulsive available. It too looks nearly full vertical on the leading edge.

[ATTACH=CONFIG]34033[/ATTACH]

 

But looks closer:

[ATTACH=CONFIG]34034[/ATTACH]

 

And closer still you see it was not so vertical nor so steep that it can't be sampled. This is far steeper than you will normally see in any music except rarely.

[ATTACH=CONFIG]34035[/ATTACH]

 

Most musical instruments including cymbals are resonant devices. Even steep impulsive hits take a few cycles to ramp up to full level. Hit a cymbal and it takes time for the energy to travel over the metal surface and be reflected back to resonate. Just as in rooms at low frequencies room resonance doesn't reach full value in one cycle. Plucked strings are in front of a resonant box and also don't instantly reach max value though I think people have this picture in their mind.

 

Are you able to convert to 44.1kHz and show what it looks like there?

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

I sample rate converted to 44.1 and then back to 176 so the scale would match on both. Upper is the original 176 khz and lower is the 44 khz.

 

[ATTACH=CONFIG]34044[/ATTACH]

 

 

[ATTACH=CONFIG]34045[/ATTACH]

 

 

You can see in the latter image that the very beginning of the transient wasn't slow to start on 44 khz. You do see some peaks are rounded and lower in level due to the energy in ultrasonic frequencies that were not captured.

 

It is hard to say where the real fundamental modes of the cymbal are. The decaying tail has strong output at 6700 hz and 8 khz with a big drop off after that. The harmonics do extend to maybe 50 khz prior to 100 milliseconds. The microphone used is said flat to 40 khz and I am sure has some output beyond that. After about 100 milliseconds the strong harmonics beyond 10 khz suddenly die out leaving pretty much only below 10 khz sound.

 

Here is the spectrogram. It goes to the gray background at -80 db and uses 1024 bin FFT as I have it set here.

 

[ATTACH=CONFIG]34046[/ATTACH]

 

Thanks. What's interesting is that (at least as my eyes are seeing it - would be nice to get an overlay view if that's possible) that as you say the rise to peak doesn't seem slow; the differences I think I'm seeing are in the decay. (And these don't look to be large differences.)

 

Edit: What conversion filter did you use?

[Jud]

There's nothing wrong with having a preference for something – we all do. Many albums I would never, ever buy on CD. But I would have a lot more respect for the vinyl lovers if they just said "yes, I know vinyl is less transparent, but I simply like the coloration it gives to music", instead of spreading bullshit and lies.

 

 

* * *

 

If there's one category of CDs that are the easiest to generalize about it's CD from the 80s. I'm not a fan of 80s music, but I did compare a reasonable amount of CDs from that period. So far I have only found two CDs from the 80s that were better than their vinyl counterpart (as well as one album where the two medias were practically identical). The rest sounded thin, shrill, cold, clinical and hard.

I still have several hundred albums that either simply sound better on vinyl, or that I simply prefer. Unfortunately, there are many CDs, both old and new, that sound bad, and many albums from the analogue era sound lovely, as they are well-produced, unlike a lot of music nowadays.

 

So many vinyl lovers' preferences may come to some extent from the fact that at least part of the time they're listening to better mastering. I completely agree with the proposition that mastering trumps resolution or even vinyl vs. digital.

 

 

And then folks get sucked in by plausible-sounding hypotheses that are incorrect, but they don't examine too closely since these hypotheses are in line with their preconceptions. I'm still in agreement, in fact perhaps more so than you: that Lipshitz and Vanderkooy article is steadfastly cited by lots of people whose DACs are doing sigma-delta modulation internally (yourself as well, perhaps?), and they don't seem to mind.

[Jud]

Let me try to explain. The Fourier Transform itself is defined for continuous signals (without sampling). It states that for *any* continuous time dependent signal, there exists a transform into the frequency/phase dimension such that the signal can be described *only* by its constituent frequency/phase coordinates -- that means that *any* time dependent signal can be described entirely by a set of sinusoidal waves having a defined frequency, amplitude and phase.

 

This is mathematically true for every signal. *Why* its true involves some math: https://en.wikipedia.org/wiki/Fourier_transform but trust that this is true.

 

Now there is the discrete fourier transform (DCT) which is defined for discrete samples. This is equivalent to the general fourier transform when the sampling rate is infinite. So what to do? In practice suppose we take samples in the Ghz range. We can then examine the discrete fourier transform and suppose we see that *for that specific signal, e.g. a jazz band* that the amplitudes of the frequency components >20 kHz are *zero*. That tells us that *that particular signal* is bandwidth limited to 20 kHz. Now suppose we hit the cymbal ? is there a component > 20 kHz ... let's say there is, now for the sake of discussion up to 40 kHz and then zero above 40 kHz. That means that we can capture the transients in that signal by sampling at say 88 kHz. Or if there are components at 70 kHz then 176 kHz will contain those transients.

 

Do you follow?

 

So you can just look at the discrete fourier transform of the recording and see what the minimal sampling rate would be to capture any transients or *anything else* in that signal. You can determine the actual bandwidth of that signal.

 

Shannon-Nyquist https://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem just tells us what the sampling rate *needs to be* in order to capture the full fidelity of any particular signal (given its particular bandwidth).

 

So again *assuming the signal is bandwidth limited to 20 kHz* 44 kHz is adequate to capture any transients that exist *in that signal*, to answer your question. Not silly.

 

One thing further to what jabbr said, for our masked vegetable friend :

 

"Periodic" in the sense of sine wave-like signals is also not necessary. The Sampling Theorem also works fine for stuff that doesn't produce nice sine waves, like percussion or inharmonic attacks, such as the pluck of a string or (for brass players) "tonguing."

 

 

Sent from my iPhone using Computer Audiophile

[Jud]

That paper talks about inherent problems with 1-bit sigma-delta converters, and it is quite correct. It explicitly states that multi-bit sigma-delta is perfectly fine. Almost all modern DACs use multi-bit converters. Guess why.

 

Yes, as Miska is fond of saying, it shows you get bad results when you use crappy SDM.

[Jud]

Hi Guys - I just edited the article with the following statement at the top.

 

 

 

The following article contains information that has been deemed incorrect by leading digital audio engineers. I attempted to corroborate the findings of this article by asking several digital audio experts. I was unable to find anyone who could back up the statements made, with any scientific data or theory. Consider the following article retracted.

 

I am leaving the text of this article up on CA because it has enabled a good discussion to take place. By leaving it up, people can read what was claimed and read the followup arguments that the prove it incorrect. To remove the article completely only opens up a space for this to happen again, and again, and again.

 

I take full responsibility for the publishing of this article. I should have had a technical editor check it before publication. I apologize to the CA Community for the error in judgement.

 

- CC.

 

 

I think that's great, frankly. On how many other sites would the editorial content (1) take account of the comments, (2) ever bother to say "whoops, we goofed," and (3) bother to leave (2) in a position of some prominence?

[Jud]

The Lipshitz/Vanderkooy paper is about sigma-delta modulators, not the D/A stage. While excellent performance is possible, there is no denying that 1-bit modulators of 3rd order or higher are inherently unstable and exhibit various forms of distortion. A multi-bit modulator does not (necessarily) have these issues. Given the choice, a multi-bit architecture is therefore preferable.

 

Would you say "various forms of distortion" are characteristic of the DSD recordings on offer? What about audibility of such distortions if/where they do exist?

 

 

Sent from my iPhone using Computer Audiophile