Digital that sounds like analog

[LL21]

"Analog, it’s continuous. If you look at a sine wave, it’s a continuous wave. Whereas digital is an approximation, or a subset, of the analog waveform, it’s sampled, it’s steps, so you don’t get the full information. The resolution is not there. So with vinyl, even to neophytes, it’s immediately apparent the sound quality is so much better with vinyl" Shawn Britton – Senior Mastering Engineer at Mobile Fidelity Sound Labs

I totally disagree. He has no concept of how digital works..!

Hi Bruce, let me try to understand as a non-techie. Dredging out my old highschool math, there were 2 ways to figure out the area under a curve. In Trigonometry, we were always trying to take thinner and thinner slices of rectangular area under the curve. With calculus, it was everything under the curve.

Is digital a Trigonometry-like approximation of the curve, or is it calculus? Or is the better question:

- if film is [supposedly] analog and digital film is digital...then film is NOT in fact analog (ie, perfect sine wave)...because in the real world, film is about 12 megapixels due to the limitations of the pixelation of film, and digital has finally approached and possibly surpassed that mark at about 12-15 megapixels.

sorry for using such blunt analogies, but that's the limit of my tech.

 

[opus111]

Care to explain why?

Yes some information is lost but we choose enough bits and a high enough sample rate so that what's lost is negligible. Analog tape recording also loses information - you can see an example of how earlier in this thread where I'm showing a test tone recorded on analog tape.

Looks to me as if Shawn Britton is confusing the theory with the implementation - yes a lot of current digital implementations suck worse than vinyl, but as far as I can see, that's fixable. Ditch S-D converters, ditch opamps operating into HF overload and correctly band-limit the output.

 

[asiufy]

opus,

So, in theory, there IS a loss of information. And it's irrelevant that analog loses information as well, when the original source material, that's being reproduced in digital, is analog to begin with.
And I guess we come down to what kind of information is lost... You claim what's lost is negligible, but perhaps those "negligible" bits are just what made the original analog signal so real.
Since it's implementation we're talking about, what kind of ADC are normally used to transfer tapes to digital? My worst fear is that a lot of the music was already transfered to digital, way back then, with crappy ADCs, and labels are just working on those compromised digital files...


alexandre

 

[microstrip]

"Analog, it’s continuous. If you look at a sine wave, it’s a continuous wave. Whereas digital is an approximation, or a subset, of the analog waveform, it’s sampled, it’s steps, so you don’t get the full information. The resolution is not there. So with vinyl, even to neophytes, it’s immediately apparent the sound quality is so much better with vinyl" Shawn Britton – Senior Mastering Engineer at Mobile Fidelity Sound Labs

Bruce,

IMHO the sentence must be interpreted with care. If the it's steps refer to to the quantization error due the finite number of bits, than I understand you don't get the full information as meaning there is an error in the process, something was theoretical lost, and there is nothing wrong per se with the comment. And the comment about vinyl is purely subjective and his opinion.

I do not think that Shawn Britton believes that the output of a DAC looks like a stair! Although I think I read somewhere he prefers DSD to PCM.

Do you have the original text or a link to it? We need to know what digital he was addressing.

 

[TBone]

Trying to figure out why analogue sounds better than digital is akin to trying to figure out whether SNL character Pat is a man or a woman. What are you going to do with the answer?:b

so true ... although not that long ago I figured analog was without question superior based on my own system capability ... so much so ... that I wouldn't have entertained archiving/recording vinyl to 16/44 based on the "inevitable losses" ... that is ... until I started doing just that ... and to my surprise I soon realized I was very wrong ... they indeed can be made to sound quite similar.

tb1

 

[Orb]

Bruce,

IMHO the sentence must be interpreted with care. If the it's steps refer to to the quantization error due the finite number of bits, than I understand you don't get the full information as meaning there is an error in the process, something was theoretical lost, and there is nothing wrong per se with the comment. And the comment about vinyl is purely subjective and his opinion.

I do not think that Shawn Britton believes that the output of a DAC looks like a stair! Although I think I read somewhere he prefers DSD to PCM.

Do you have the original text or a link to it? We need to know what digital he was addressing.

Micro, technically it does look like a stair/step at the reconstruction unless you are using an integrated DAC with all the functionality embedded.
It is not something I have bothered debating at HA or even AVSF because it is more a technicality about chip architecture and where functionality resides, and end up in the analogue output as a sinewave so why bother arguing with anyone
Sort of like how Intel and AMD add more and more functions embedded to their processors, which could be done by a separate chip-function.

Cheers
Orb

 

[opus111]

So, in theory, there IS a loss of information.

Not only in theory, in practice too.

And it's irrelevant that analog loses information as well, when the original source material, that's being reproduced in digital, is analog to begin with.

Analog tape already lost information when it recorded the original signal. But yes, its irrelevant in the case where we're digitizing an analog tape. Preferably we'd record digitally in the first place.

And I guess we come down to what kind of information is lost... You claim what's lost is negligible, but perhaps those "negligible" bits are just what made the original analog signal so real.

Then we wouldn't be losing negligible information would we? We'd be losing significant information. Seems you don't want the normal meaning of 'negligible' here for some reason.

Since it's implementation we're talking about, what kind of ADC are normally used to transfer tapes to digital?

I don't think there is a normal ADC which is used, but mostly nowadays they are S-D type, which IME are less transparent. Implementations of S-D vary though, in consequence so does the transparency.

My worst fear is that a lot of the music was already transfered to digital, way back then, with crappy ADCs, and labels are just working on those compromised digital files...

In many cases your fears are realized, they did use crappy (in the sense of far from transparent) ADCs. Some examples I've given on this thread.

 

[DonH50]

Look up sampling theorem for a detailed discussion about the loss of information (or not).

 

[opus111]

Sampling theorem is the more commonly cited side of the digital coin, for which we have Shannon and Nyquist to thank.

http://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem

The 'darker side' is quantization theory and Bernard Widrow seems to me to be the father of that - http://oldweb.mit.bme.hu/books/quantization/index.html

 

[asiufy]

Not only in theory, in practice too.



Analog tape already lost information when it recorded the original signal. But yes, its irrelevant in the case where we're digitizing an analog tape. Preferably we'd record digitally in the first place.



Then we wouldn't be losing negligible information would we? We'd be losing significant information. Seems you don't want the normal meaning of 'negligible' here for some reason.



I don't think there is a normal ADC which is used, but mostly nowadays they are S-D type, which IME are less transparent. Implementations of S-D vary though, in consequence so does the transparency.



In many cases your fears are realized, they did use crappy (in the sense of far from transparent) ADCs. Some examples I've given on this thread.

opus,

So there is NO instance where that information that's lost could NOT be negligible?
You said:
"Yes some information is lost but we choose enough bits and a high enough sample rate so that what's lost is negligible."
OK, so how many bits are enough and what's a high enough sample rate, in order for the bits that are lost to be considered negligible? And it's safe to say that anything below that threshold will lose vital information?

And this loss of transparency by S-D ADCs, what causes it? Is it because of these same bits that are being thrown away?


alexandre

 

[Orb]

Of more interest is not what is lost (mathematical theory-model is that the reconstructed sinewave is perfect) but what said filters add.
The debate on whether anything is lost can only be decided if using a much more complex waveform comprising many instruments in different phase/where placed combined with the mastering to 2 channels.
Again mathematics would dictate this is going to be perfect as we are talking sum of sinewaves, but it would still be interesting.
Challenge would be creating a mathematical very complex tone with sustain and decay that could then be measured at the output by converting back to its synthetic digital form.
In a rush so cannot say more, but more of an academic interest rather than trying to prove anything apart from potential limitations of embedded functionality on a single chip compared to separated with individual-dedicated processors.
Cheers
Orb

 

[opus111]

You said:
"Yes some information is lost but we choose enough bits and a high enough sample rate so that what's lost is negligible."
OK, so how many bits are enough and what's a high enough sample rate, in order for the bits that are lost to be considered negligible?

Depends on the system - no universal answer for this. Bob Stuart has a paper where he reckons for replay that around 19bits and 60kHz or so is enough for anyone's possible needs. So for a few people, 16bits and 44k1 isn't quite enough, but I reckon there are only a few people with systems (and rooms) which exceed this.

And this loss of transparency by S-D ADCs, what causes it? Is it because of these same bits that are being thrown away?

Loss of transparency is always caused by addition of noise and/or distortion. In this case its additive noise through improperly dithered truncation - the throwing of bits away, yes. S-D converters work by truncating the original 16 (or more) bits to something at the 6bit or fewer level. The plan is that this quantization error is corrected on average, but that's not enough to fool our ears, we do hear the instantaneous changes in the noise floor that result. When I listen to DSD-mastered CDs I notice that the noise seems to be added at HF - bass dynamics remains fairly decent, but the top end gets congested, fuzzy.

 

[asiufy]

So, if I got it correctly, Bob Stuart thinks that, with 19 bits and a sampling frequency of 60Khz, he's able to capture enough of the analog signal without leaving anything behind, or at least capturing enough to make the digital identical to analog? I guess this is theoretical, because that means I could, for instance, convert analog tape to digital at that frequency/bitrate, and supposedly, not hear a difference between the original and the digital (also supposing the DAC won't mangle the playback of the digital recording).

Regarding the S-Ds, what's the reason that S-D ADCs would truncate and use only 6 out of the 16 bits? I mean, I thought our technology was perfectly able to capture 16 bits at 44Khz without needing to drop any of those bits... I can't understand why this kind of shortcut is being used...



alexandre

 

[opus111]

So, if I got it correctly, Bob Stuart thinks that, with 19 bits and a sampling frequency of 60Khz, he's able to capture enough of the analog signal without leaving anything behind, or at least capturing enough to make the digital identical to analog?

I think there's more backing it than just his opinion, he's done the math based on psychoacoustics. Here its important to understand the comparison isn't with analog tape, this is the analog mic feed from the recording studio. Analog tape doesn't come anywhere close to 19bits in terms of noise though it exceeds the equivalent of 60kHz in bandwidth terms.

Regarding the S-Ds, what's the reason that S-D ADCs would truncate and use only 6 out of the 16 bits? I mean, I thought our technology was perfectly able to capture 16 bits at 44Khz without needing to drop any of those bits... I can't understand why this kind of shortcut is being used...

To make DAC and ADC chips cheaper to manufacture - with this technique a digital process can be used, no expensive (time consuming) laser trimming is called for. Also to achieve higher SNRs and better THDs than the best multibit DACs.

 

[asiufy]

I think there's more backing it than just his opinion, he's done the math based on psychoacoustics. Here its important to understand the comparison isn't with analog tape, this is the analog mic feed from the recording studio. Analog tape doesn't come anywhere close to 19bits in terms of noise though it exceeds the equivalent of 60kHz in bandwidth terms.

Ah, I see, so his reference is the live mic feed. So 19 bit/60Khz would be completely transparent!
But back to analog tape, a 24/96 transfer then should be identical to the tape, no? Any discrepancies there should be attributed to the ADC used for the recording or the DAC used in the playback, correct?

To make DAC and ADC chips cheaper to manufacture - with this technique a digital process can be used, no expensive (time consuming) laser trimming is called for. Also to achieve higher SNRs and better THDs than the best multibit DACs.

That flew over my head I mean, it's cheaper to implement a whole process than to manufacture a chip that's able to handle the full 16 bits?
And multibit DACs have worse SNRs/THDs than S-D?

 

[opus111]

Ah, I see, so his reference is the live mic feed. So 19 bit/60Khz would be completely transparent!

Yes, that's what he's saying.

But back to analog tape, a 24/96 transfer then should be identical to the tape, no?

With the proviso that the bandwidth might be curtailed somewhat, whether that's audible is debatable. But no ADC has anything close to 24bit resolution, so 20bit/96k would be more than sufficient to be transparent to tape.

Any discrepancies there should be attributed to the ADC used for the recording or the DAC used in the playback, correct?

Yep, I reckon so.

That flew over my head I mean, it's cheaper to implement a whole process than to manufacture a chip that's able to handle the full 16 bits?

The word 'process' there is referring to the whole production line for a chip - because digital chips drive the improvements to semiconductor production, there's a desire to have as many 'analog' functions made using digital chip production lines to achieve economies of scale and improvements in cost and performance as the geometry shrinks.

And multibit DACs have worse SNRs/THDs than S-D?

Yes - take the ES9018 and the PCM1794 and compare them to the PCM1704. This latter part is the pinnacle of multibit in terms of measurements (not including the TDA1541A which is no longer in manufacture but only has 16bit capability).

 

[asiufy]

Thanks for all the info, much appreciated


alexandre

 

[NorthStar]

PCM-1794? ...Not the PCM-1792?

 

[Bruce B]

Loss of transparency is always caused by addition of noise and/or distortion. In this case its additive noise through improperly dithered truncation - the throwing of bits away, yes. S-D converters work by truncating the original 16 (or more) bits to something at the 6bit or fewer level. The plan is that this quantization error is corrected on average, but that's not enough to fool our ears, we do hear the instantaneous changes in the noise floor that result. When I listen to DSD-mastered CDs I notice that the noise seems to be added at HF - bass dynamics remains fairly decent, but the top end gets congested, fuzzy.

I just did a search on "opus111" and "S-D"...

I should have my NOS DAC in the system by next week.

 

[DonH50]

I think tape only provides about 70 dB or so dynamic range (seems like best I have ever seen was maybe 75 dB?) so even fewer bits should be enough for tape transfers.