Digital Vinyl: Temporal Domain

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?

With a decent modulator, the distortion is well below audible levels, at least with realistic inputs.

Sorry I was referring to the Grimm white paper.

 

My impression has been that Lipshitz was referring to the ADC and storage stages at least primarily. Hard to know how many chip based DACs use 3rd order modulators as they don't typically publish their internals.

Why do you bring up the Grimm paper when everybody else is talking about Lipshitz? Or did I misunderstand something?

The conclusions of Lipshitz are applicable to any SDM regardless of position in the production chain.

With a decent modulator, the distortion is well below audible levels, at least with realistic inputs.

Out of curiosity, how many bits are the Sox modulators? (I know they're open source, and I've looked at the source, but since I'm not a programmer I may as well not have.)

Out of curiosity, how many bits are the Sox modulators? (I know they're open source, and I've looked at the source, but since I'm not a programmer I may as well not have.)

1 bit, obviously. That's what DSD is.

1 bit, obviously. That's what DSD is.

You are remarkably sanguine for someone who has loosed on the world's Sox, Daphile, and A+ users "1-bit modulators of 3rd order or higher [that] are inherently unstable and exhibit various forms of distortion."

You are remarkably sanguine for someone who has loosed on the world's Sox, Daphile, and A+ users "1-bit modulators of 3rd order or higher [that] are inherently unstable and exhibit various forms of distortion."

Such is the nature of the beast. Blame Sony and Philips.

Why do you bring up the Grimm paper when everybody else is talking about Lipshitz? Or did I misunderstand something?

 

The conclusions of Lipshitz are applicable to any SDM regardless of position in the production chain.

There were several issues being discussed and different papers. I'm responding while waiting for some timing analyses to be done i.e. multitasking -- oh well..

You stated that DACs are mostly multibit "for a reason" -- a NOS DAC which accepts DSD input generally has no reason to do SDM, and the inputs I know of are 1-bit ... one could transmit multibit over the network or USB using a proprietary protocol but not one known to ALSA for example ... so I am not understanding why the Lipshitz issue generally affects the DAC? (unless the DAC is upsampling or converting from PCM)

There were several issues being discussed and different papers. I'm responding while waiting for some timing analyses to be done i.e. multitasking -- oh well..

Timing analyses, something related to this thread?

You stated that DACs are mostly multibit "for a reason" --

The vast majority of audio DACs use multi-bit SDM internally for the simple reason that it gives the best results.

a NOS DAC which accepts DSD input generally has no reason to do SDM, and the inputs I know of are 1-bit ... one could transmit multibit over the network or USB using a proprietary protocol but not one known to ALSA for example ... so I am not understanding why the Lipshitz issue generally affects the DAC? (unless the DAC is upsampling or converting from PCM)

I'm not sure what you're trying to say here. NOS PCM DACs are entirely outside this discussion. If the signal is at any point in 1-bit format, something must have created it, most likely a sigma-delta modulator subject to the issues raised by Lipshitz.

There were several issues being discussed and different papers. I'm responding while waiting for some timing analyses to be done i.e. multitasking -- oh well..

 

You stated that DACs are mostly multibit "for a reason"

The vast majority of audio DACs use multi-bit SDM internally for the simple reason that it gives the best results.

 

The central point of confusion. mansr is not referring to "multibit" in the Schiit PCM DAC sense. He's talking about the internal SDM stage.

Here's Wikipedia about a similar thing on the A/D side:

Because it has been extremely difficult to carry out DSP operations (for example performing EQ, balance, panning and other changes in the digital domain) in a one-bit environment, and because of the prevalence of solely PCM studio equipment such as Pro Tools, the vast majority of SACDs—especially rock and contemporary music, which rely on multitrack techniques—are in fact mixed in PCM (or mixed analog and recorded on PCM recorders) and then converted to DSD for SACD mastering.

 

To address some of these issues, a new studio format has been developed, usually referred to as DSD-wide, which retains the high sample rate of standard DSD, but uses an 8-bit, rather than single-bit digital word length, yet still relies heavily on the noise shaping principle. DSD-wide is PCM with noise shaping—and is sometimes disparagingly referred to as "PCM-narrow"—but has the added benefit of making DSP operations in the studio a great deal more practical. The main difference is that "DSD-wide" still retains 2.8224 MHz (64Fs) sampling frequency while the highest frequency in which PCM is being edited is 384 kHz (8Fs). The "DSD-wide" signal is down-converted to regular DSD for SACD mastering. As a result of this technique and other developments there are now a few digital audio workstations (DAWs) that operate, or can operate, in the DSD domain, notably Pyramix and some SADiE systems.

I'm pretty sure, though, that given the nature of sound they are a bunch of nice sine waves

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

"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." Indeed. And you can go up to 22K and still be kosher per sampling theorem. But if there are cymbals and a up to 44K bandwidth you need a 88K sampling, don't you ?SN theorem certainly is guaranteed theorem quality ; that actual music can perfectly fit a 22K bandwidth is much less certain. Thus the need for larger bandwidth hence sampling rate to capture reality.

What I'm not sure of is the equivalence of a limited to start with 20K bandwidth capture with a 2O/20K bandwidth loudspeakers output

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.

Timing analyses, something related to this thread?

FPGA, clocks, network, DAC so peripherally.

I'm not sure what you're trying to say here. NOS PCM DACs are entirely outside this discussion. If the signal is at any point in 1-bit format, something must have created it, most likely a sigma-delta modulator subject to the issues raised by Lipshitz.

I meant NOS DSD DAC and the DAC just converts what it's given both good and bad

I meant NOS DSD DAC and the DAC just converts what it's given both good and bad

The term NOS typically refers to PCM DACs. DSD is already oversampled.

I'm pretty sure, though, that given the nature of sound they are a bunch of nice sine waves

Nope. Please read about what "inharmonic" means. Inharmonicity is characteristic of various instrumental attacks, and of a lot of percussion instruments. If you do Fourier analysis, it shows that all these sounds are built from sine waves, but without the analysis that isn't at all apparent - on a scope they don't look like "nice sine waves."

Sent from my iPhone using Computer Audiophile

"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." Indeed. And you can go up to 22K and still be kosher per sampling theorem. But if there are cymbals and a up to 44K bandwidth you need a 88K sampling, don't you ?SN theorem certainly is guaranteed theorem quality ; that actual music can perfectly fit a 22K bandwidth is much less certain. Thus the need for larger bandwidth hence sampling rate to capture reality.

 

What I'm not sure of is the equivalence of a limited to start with 20K bandwidth capture with a 2O/20K bandwidth loudspeakers output

There are (at least) two possibilities:

- All the effects of ultrasonic frequencies are already "baked into" what we can hear. That is, intermodulation - interference - between ultrasonics and what we can hear, and among the ultrasonics themselves, if the result is audible, is in the recording already, and therefore 20KHz is enough.

- Or, in order to have the most faithful rendition, perhaps the interference between ultrasonics and the audible range should be recreated in your room. Note that this would require speakers to be relatively flat to ultrasonic range, the room and speaker drivers to be able to reproduce the interference as it occurred live.... None of this is trivial stuff, if it is even necessary (i.e., if the first alternative above isn't good enough).

Sent from my iPhone using Computer Audiophile

Glad to see Chris add the note to this article.

I wonder however, is this a good time for one of the engineering advisers to write the proper article. One that lays out the basics of how digital works, what the parameters are. One that compares it to LP, and reel tape. One that shows the ways in which digital PCM is superior. That one can prefer or enjoy these other formats. Just don't resort to myths or fallacies about why that might be the case. It isn't about superior fidelity. An article that illuminates rather than obfuscates.

Wouldn't that be a good article one likely to draw equal interest to this article? One which is doing a service to make clear how this digital audio really works. A light in the darkness of myth and superstition which is all too pervasive. Seems like such an article would be feather in the cap of Computer Audiophile if you ask me.

?? "If you do Fourier analysis, it shows that all these sounds are built from sine waves" Yes

Nope. Please read about what "inharmonic" means. Inharmonicity is characteristic of various instrumental attacks, and of a lot of percussion instruments. If you do Fourier analysis, it shows that all these sounds are built from sine waves, but without the analysis that isn't at all apparent - on a scope they don't look like "nice sine waves."

 

 

Sent from my iPhone using Computer Audiophile

There's a trivial hypothesis : does a DAC ouputing at say 192 is outputing the same stuff with the exact same timing when it's been fed 44 , that is when some events have not been sampled, rather than 192?

BTW Fremer impaired the US career of the wonderful Cabasse Sphere with considerations that what he was hearing was too good so had to be kind of rolled of sweetened by the (then limited) 16/44 filtering of the Sphere

I use old active studio monitors that deliver 20/20K and are top guns from vinyl/experimental PCM age

Note that my TEAC 501 I feed DSD 128 and can be 384K fed as well starts to roll off at 20K

Is it really the same to filter 20/20K at speakers level and to not sample beyond a 20K bandwidth ? Is it the same reconstruction of the actual musical event? at DAC level?

There are (at least) two possibilities:

 

- All the effects of ultrasonic frequencies are already "baked into" what we can hear. That is, intermodulation - interference - between ultrasonics and what we can hear, and among the ultrasonics themselves, if the result is audible, is in the recording already, and therefore 20KHz is enough.

 

- Or, in order to have the most faithful rendition, perhaps the interference between ultrasonics and the audible range should be recreated in your room. Note that this would require speakers to be relatively flat to ultrasonic range, the room and speaker drivers to be able to reproduce the interference as it occurred live.... None of this is trivial stuff, if it is even necessary (i.e., if the first alternative above isn't good enough).

 

 

Sent from my iPhone using Computer Audiophile

There's a trivial hypothesis : does a DAC ouputing at say 192 is outputing the same stuff with the exact same timing when it's been fed 44 , that is when some events have not been sampled, rather than 192?

 

BTW Fremer impaired the US career of the wonderful Cabasse Sphere with considerations that what he was hearing was too good so had to be kind of rolled of sweetened by the (then limited) 16/44 filtering of the Sphere

 

I use old active studio monitors that deliver 20/20K and are top guns from vinyl/experimental PCM age

Note that my TEAC 501 I feed DSD 128 and can be 384K fed as well starts to roll off at 20K

Is it really the same to filter 20/20K at speakers level and to not sample beyond a 20K bandwidth ? Is it the same reconstruction of the actual musical event? at DAC level?

Probably helpful to use numbers instead of abstract concepts, so just for our present purposes, let's say we have some hypothetical instrument that has strong harmonics at 38kHz and 40kHz, and due to interference between the two we get audible energy from the instrument at 2kHz.

In order to get that 38 and 40kHz energy through the DAC, we need a sample rate more than double those frequencies, so at least 88.2 or 96kHz, right from the A/D at the recording end through the DAC, which nicely provides the same 38 and 40kHz output that was recorded. Fine, so let's say we've done that, and your electronics pass all those frequencies. But your speakers only have useful response to 30kHz. Whoops, no more 38 and 40kHz energy, and so there's nothing at 2kHz, right? But wait - didn't the DAC also pick up the 2kHz energy along with everything else up to 44 or just under 48 kHz? And your speakers can reproduce that 2kHz just fine. So have we lost anything? I don't know the answer.

Now let's say instead of being recorded at an 88.2 or 96kHz sample rate, the input from our hypothetical instrument is band-limited by filtering it to capture just the components up to 20kHz, so we can do A/D with a 44.1kHz sample rate. No 38 or 40kHz energy. But we have captured the result of that interference in the audible band at 2kHz, so the question is exactly the same: Have we lost anything by capturing only the audible result of the interference (2kHz), rather than the energy that created the interference in the first place (38 and 40kHz)? Again, I don't know the answer.

The other thing I think you have asked is that assuming we have recorded a signal limited just to the audible band, just up to 20kHz, will sampling at 44.1kHz give us the same timing as sampling at 192kHz? The answer is yes - that's just math, the Sampling Theorem we've been talking about. The interesting issues here don't have anything to do with capturing events that happen between samples. That may not be intuitive, but it's mathematical fact. Rather, the interesting issues have to do with how easily and cheaply you can make an excellent filter for a signal if you sample it at 44.1kHz versus sampling at a higher rate. Sampling at a higher rate makes good filtering easier and cheaper, which is why the vast, vast majority of DACs oversample (in fact, oversampling became pretty well the standard thing shortly after the introduction of CD players, before separate DACs even existed).

Your last paragraph is sheer interpretation of my question

Probably helpful to use numbers instead of abstract concepts, so just for our present purposes, let's say we have some hypothetical instrument that has strong harmonics at 38kHz and 40kHz, and due to interference between the two we get audible energy from the instrument at 2kHz.

 

 

In order to get that 38 and 40kHz energy through the DAC, we need a sample rate more than double those frequencies, so at least 88.2 or 96kHz, right from the A/D at the recording end through the DAC, which nicely provides the same 38 and 40kHz output that was recorded. Fine, so let's say we've done that, and your electronics pass all those frequencies. But your speakers only have useful response to 30kHz. Whoops, no more 38 and 40kHz energy, and so there's nothing at 2kHz, right? But wait - didn't the DAC also pick up the 2kHz energy along with everything else up to 44 or just under 48 kHz? And your speakers can reproduce that 2kHz just fine. So have we lost anything? I don't know the answer.

 

 

Now let's say instead of being recorded at an 88.2 or 96kHz sample rate, the input from our hypothetical instrument is band-limited by filtering it to capture just the components up to 20kHz, so we can do A/D with a 44.1kHz sample rate. No 38 or 40kHz energy. But we have captured the result of that interference in the audible band at 2kHz, so the question is exactly the same: Have we lost anything by capturing only the audible result of the interference (2kHz), rather than the energy that created the interference in the first place (38 and 40kHz)? Again, I don't know the answer.

 

 

The other thing I think you have asked is that assuming we have recorded a signal limited just to the audible band, just up to 20kHz, will sampling at 44.1kHz give us the same timing as sampling at 192kHz? The answer is yes - that's just math, the Sampling Theorem we've been talking about. The interesting issues here don't have anything to do with capturing events that happen between samples. That may not be intuitive, but it's mathematical fact. Rather, the interesting issues have to do with how easily and cheaply you can make an excellent filter for a signal if you sample it at 44.1kHz versus sampling at a higher rate. Sampling at a higher rate makes good filtering easier and cheaper, which is why the vast, vast majority of DACs oversample (in fact, oversampling became pretty well the standard thing shortly after the introduction of CD players, before separate DACs even existed).

 

My basic thought is your premise of the question isn't true a majority of the time. It isn't generally true that CDs of analog recordings sound like crap. For one thing I have digitally recorded both reel tape and LP with the result sounding like reel tape or LP. So the digital CD isn't a bottleneck.

Me too. Very good digital recordings of needle drops are indistinguishable on playback from the original vinyl - played back on the same system. I've personally experienced this and demonstrated it with others.

A properly done digital recording of vinyl has all those "vinyl like" qualities that vinyl lovers love. So the digital medium itself isn't the culprit or the limitation. I refer all of you to John Atkinson's review of the Ayre ADC, where he said he compared digital conversions of his own analog recordings to the original- in fact compared them "till his ears bled" - and said he couldn't tell them apart.

Your last paragraph is sheer interpretation of my question

I guessed at what you meant incorrectly, then.

Probably helpful to use numbers instead of abstract concepts, so just for our present purposes, let's say we have some hypothetical instrument that has strong harmonics at 38kHz and 40kHz, and due to interference between the two we get audible energy from the instrument at 2kHz.

 

 

In order to get that 38 and 40kHz energy through the DAC, we need a sample rate more than double those frequencies, so at least 88.2 or 96kHz, right from the A/D at the recording end through the DAC, which nicely provides the same 38 and 40kHz output that was recorded. Fine, so let's say we've done that, and your electronics pass all those frequencies. But your speakers only have useful response to 30kHz. Whoops, no more 38 and 40kHz energy, and so there's nothing at 2kHz, right? But wait - didn't the DAC also pick up the 2kHz energy along with everything else up to 44 or just under 48 kHz? And your speakers can reproduce that 2kHz just fine. So have we lost anything? I don't know the answer.

 

 

Now let's say instead of being recorded at an 88.2 or 96kHz sample rate, the input from our hypothetical instrument is band-limited by filtering it to capture just the components up to 20kHz, so we can do A/D with a 44.1kHz sample rate. No 38 or 40kHz energy. But we have captured the result of that interference in the audible band at 2kHz, so the question is exactly the same: Have we lost anything by capturing only the audible result of the interference (2kHz), rather than the energy that created the interference in the first place (38 and 40kHz)? Again, I don't know the answer.

The answer depends on where the intermodulation takes place. If the 2 kHz product is registered by the microphone, and there are no further effects of the ultrasonic tones inside our ears, then removing them will have no audible effect.

The answer depends on where the intermodulation takes place. If the 2 kHz product is registered by the microphone, and there are no further effects of the ultrasonic tones inside our ears, then removing them will have no audible effect.

Right. To address @Jud, there should be no assumption that nonlinear effects are the same. Intermodulation is a type of nonlinear effect. Air, bone, various tissues each have unique nonlinear transmission qualities. The nervous system also has memory and computation capabilities that enable nonlinearities that go beyond what is being discussed hear eg "synesthesia" The reason I bring this up is to make the point that many linear assumptions are not necessarily valid.

Me too. Very good digital recordings of needle drops are indistinguishable on playback from the original vinyl - played back on the same system. I've personally experienced this and demonstrated it with others.

 

A properly done digital recording of vinyl has all those "vinyl like" qualities that vinyl lovers love. So the digital medium itself isn't the culprit or the limitation. I refer all of you to John Atkinson's review of the Ayre ADC, where he said he compared digital conversions of his own analog recordings to the original- in fact compared them "till his ears bled" - and said he couldn't tell them apart.

I'm not disputing that a digital copy can and should sound as good as the original analog recording but in reality if you listen to the 16/44 content of artist's like Frank Sinatra on Tidal, it does sound like utter crap compared to the analog original.