Audio D/A converter

[mtrycrafts]

Hi mtry,

As I said in my earlier post, I find this all very confusing. I've heard the S/N ratio of 16 bit given as 91.3 dB before.

At the end of the day, I just listen depending on how good the performance is. A lot of the time, I'm more than satisfied with LP, despite the occasional crackle and distortion etc. Yesterday I listened to a very good LP, a Decca digital recording - a Prokofiev violin concerto with the London Philharmonic - superb sound quality. One thing that CD has going against it are the botched up, over-processed releases of some classical CD's. If you'd only ever heard rubbish like this, then it would be no surprise if you were put off CD.

I don't have a SACD or DVD Audio player at the moment, but I might get one in the future out of sheer curiosity. I bet the SACD Telarc 1812 Overture would be worth a listen.

Back to the original question, I do think it's worth getting a 24 bit DAC, because as the Rane Note said, this ensures at least 18 bit (and hence 16 bit) accuracy. You should review the specs for the CD player or digital receiver's digital output component as well, looking for the usual things (signal-to-noise, harmonic distortion, frequency response). On wikipedia, there are other measures given for DAC performance, but I've never seen these quoted in specifications.

Bit rate and S/N are not the same. Bit rate is the ability to encode between the loudest signal and the softest signal. So far, there has yet to be a recording much more than 70dB dynamic range on the master.

 

[mtrycrafts]

Dynamic range is one of those terms that is used in different contexts to mean different things but the real definition is the difference between the softest and loudest parts. The human ear is said to have a dynamic range of 130 dB which would mean we can hear from 0 dB (the softest perceivable sound in this context) all the way up to 130 dB (the threshold of pain).

The Dolby definition stated is exactly what I understand to be crest factor.

Oh well, it's the same in other disciplines. The term 'proxy' means many different things in computer science but ask anyone with at least some familiarity with computers and they will most often tell you that a proxy is a firewall.

Yes, while the ear has that dynamic range, special conditions need to be in place for real world Since the surrounding space is not anechoic, the noise floor establishes the low limit at any given time.
I think he is confusing the max spl levels with the equivalent bit rates. DD doesn't need that 18 bits at all as you can still get plenty volume with less bits recorded. Two different issues as an amp can amplify a very small signal to any level you want, practically speaking.

 

[tbewick]

mtrycrafts/WmAx/MDS,

I know that this is off-topic, but do any of you know of any papers or books that have measured (understandable) data on digital audio performance? I did pick up a book today written by tech people at Sony (Digital Audio Technology, 4th edition, J. Maes & M. Vercammen, Focal Press 2001) which maintains that SACD and DVD Audio sound better than 16 bit CD audio.

The Sony book did give definitions of dynamic range (DR) and signal-to-noise:

DR = max. signal level (RMS)/RMS level of quantization noise without signal

S/N = signal level (RMS)/RMS level of quantization noise with signal

It states the dynamic range of CD as 'more than 90 dB', with a S/N of 90 dB.

 

[WmAx]

I know that this is off-topic, but do any of you know of any papers or books that have measured (understandable) data on digital audio performance?

You can find SNR and DR measurements for many cd players in Stereophile's product archive and you can find many 3rd party measurements of soundcards in 16 bit mode online with a search.

I did pick up a book today written by tech people at Sony (Digital Audio Technology, 4th edition, J. Maes & M. Vercammen, Focal Press 2001) which maintains that SACD and DVD Audio sound better than 16 bit CD audio.

What do they specify as perceptual references, to back themselves up? If they do not back themselves up, this reduces their credibility substantially, especially in a published technical paper or book. Makes you wonder, if they will go out on a limb and make this up, what else is being falsified?

It states the dynamic range of CD as 'more than 90 dB', with a S/N of 90 dB.

You can routinely find DACs that can achieve -96dB noise level and DR of 94 - 95 dB in modern equipment. But even 90dB is more than sufficient for any realistic playback purposes. Not even the best quality classical recording is going to require 90dB[except for the possible special exceptions I made clear earlier] for playback purposes.

-Chris

 

[mtrycrafts]

I did pick up a book today written by tech people at Sony (Digital Audio Technology, 4th edition, J. Maes & M. Vercammen, Focal Press 2001) which maintains that SACD and DVD Audio sound better than 16 bit CD audio.
.

Foster, Edward J. "Evolutionary or Super Audio Revolutionary? Audio, Nov 1999, page 40-47.

This is a Sony demo of SACD. Unfortunately they were caught cheating...
"When I pushed the point, the engineer admitted to adding a bit of reverb and otherwise toying around to make the sound 'more natural.'

 

[tbewick]

'You can find SNR and DR measurements for many cd players in Stereophile's product archive and you can find many 3rd party measurements of soundcards in 16 bit mode online with a search.' - Chris

I was thinking more of papers or books that detail the sound quality of 16 bit or of DAC performance/testing. How do Stereophile's tests do this? What are the characteristics of the noise produced when dithering and how does it change with programme? Most importantly, how audible is this noise? I would rather have text references as information on the internet is less likely to be accurate.

Most books I've looked at on the subject of hi-fi do not provide any experimental data at all. Some, like the Audio Amplifier Design Handbook, D. Self, present rants/philosophical ramblings and dismissive remarks that you would not find in any other serious scientific textbook.

'Foster, Edward J. "Evolutionary or Super Audio Revolutionary? Audio, Nov 1999, page 40-47.

This is a Sony demo of SACD. Unfortunately they were caught cheating...
"When I pushed the point, the engineer admitted to adding a bit of reverb and otherwise toying around to make the sound 'more natural.'' - mtrycrafts

I would hardly call this a scientific way of dismissing SACD. Does this book have anything more substantive?

 

[WmAx]

I was thinking more of papers or books that detail the sound quality of 16 bit or of DAC performance/testing.

I don't know what you mean by 'detail the sound quality of 16 bit'. The bit depth has to do with signal to noise; nothing more.

How do Stereophile's tests do this?

You can coorelate measurements with established perceptual papers on each parameter(frequency response, thd, noise floor, etc.). However, in most cases, the measurements are better than is required for human hearing. But this still, is a slew of other parameters. Since we are on the specific subject of bit depth; I do not understand how one can 'detail the sound quality of 16 bit', except for the noisefloor. Depending on the specific dither used, the lowest level sounds could be slightly effected; but for this to be audible would require extaordinary circumstances(program material encoded in the first few bits only, turning up volume to levels well beyond what would be used for listening to actual music)

What are the characteristics of the noise produced when dithering and how does it change with programme?

It does not change with program. It is a steady state noise applied to the bottom bit(exact bit depth dependant on the particular system used to capture the audio, or if a reduced bit; 24-->16; this is specified and controlled by the audio engineer. The characteristic of the noise shape spectral distribution also varies, dependant on the same. A high quality dither will apply most of the noise to very high frequencies, so that it is less audible than if a constant white noise was applied to all bands, for example. )

Most importantly, how audible is this noise?

That depends on the specific dithering algorythm used. However, in a good process, it sounds exactly like a constant white noise, if you amplify it enough to make it audible in absence of other signals. If you desire, I can prepare a file with a standard high quality dither(or multiple dithers) amplified to where it is clearly audible.

I would hardly call this a scientific way of dismissing SACD. Does this book have anything more substantive?

It's really a dead end to look for refutations. What is there to refute? DSD was never proven to be an advantage in the first place. Where/when did Sony prove DSD was better sounding compared to PCM?

-Chris

 

[tbewick]

'You can coorelate measurements with established perceptual papers on each parameter(frequency response, thd, noise floor, etc.). However, in most cases, the measurements are better than is required for human hearing. But this still, is a slew of other parameters. Since we are on the specific subject of bit depth; I do not understand how one can 'detail the sound quality of 16 bit', except for the noisefloor.'

Does this mean that two 16 bit systems with identical S/N and dynamic range will be audibly identical? I would like references to papers or books which prove this. One of the claims of the Sony textbook is something to do with SACD versus CD and the need for D/A converters:

'the need for high-precision D/A converter circuitry versus the possiblity to even replace D/A circuitry by simple low-pass filtering'

and 'the need for noise shaping, digital filters, interpolation versus much reduced need for such circuitry'.

I don't know enough about CD to comment on these claims, that was the point of my earlier question - do you know of any papers or books that have measured (understandable) data on digital audio performance?

'It's really a dead end to look for refutations.'

Rather than dismissive remarks about SACD, I'm looking for proper tests comparing it to CD.

 

[mtrycrafts]

'Foster, Edward J. "Evolutionary or Super Audio Revolutionary? Audio, Nov 1999, page 40-47.

This is a Sony demo of SACD. Unfortunately they were caught cheating...
"When I pushed the point, the engineer admitted to adding a bit of reverb and otherwise toying around to make the sound 'more natural.'' - mtrycrafts

I would hardly call this a scientific way of dismissing SACD. Does this book have anything more substantive?

This is not a book. An article written by Foster about a Sony attempt to demonstrate the SACD vs CD. That sentence is a sample from that demonstration. It is telling that Sony had to cheat to demonstrate SACD.

http://www.stereophile.com//features/374/index.html

Another article on SACD and its technical problems.

If you are looking for science papers you better dismiss that Sony as well; it isn't scientific anymore than the articles posted. You won't find one. No one wants to do a credible DBT of the two. Might give the wrong answer.

 

[Buckle-meister]

In all likelihood your loudspeakers are almost always the weakest component.

No, 99.9% of the time the room itself will almost certainly be the weakest component. I understand why they don't, but people need to get into their head that the room is part of the hi-fi, not just an entity 'the hi-fi' is placed in.

 

[tbewick]

hi mtrycrafts,

Thanks for the link. It's very technical (and way above my head) but does not present audibility tests. Are these available in the referenced papers? I find this utterly perplexing. If you were to read a medical journal on drugs testing, the author would not state that drug X is better than drug Y and then move on smoothly. They would either provide data supporting this claim in the paper itself, or they would provide a reference where such data is available. They would view experimental testing with the same importance as the theoretical description of how the drug works. Otherwise the theoretical model would be useless. In fact, the experimental data is far more important than the theory itself.

The books I read on CD digital audio state that it is 'transparent' or quote the low harmonic distortion and high dynamic range/signal-to-noise figures (the Sony textbook which talks-up SACD does state the case for CD as well). As I asked Chris, are these the only measures of performance you need (THD, s/n, dynamic range, frequency response)?They also say that filter and ADC/DAC design are technically difficult. If so, I would appreciate references explaining how these tests are useful and reliable measures of real-world performance. As is implied in the Stereophile link you gave, one book states in no uncertain terms that the THD spec (sinusoid) test does not reveal all the possible distortion which may be added during ADC/DAC. As most CD players have very good specs, why do authors concentrate attention on how DAC and filter designs produce different results (e.g. some external DAC's producing worse performance)? The only data I've seen is the one Chris posted on jitter audibility.

'if you are looking for science papers you better dismiss that Sony as well; it isn't scientific anymore than the articles posted. You won't find one. No one wants to do a credible DBT of the two. Might give the wrong answer.'

Why, are Sony going to put them under house-arrest? If this is true, then hi-fi must be stuck in the Dark Ages.

 

[tbewick]

Hi Robbie,

I absolutely agree. Unfortunately though most of us have to share the usage of the living room, which makes installing ugly acoustical treatment problematic. It's much easier to buy a nice new set of speakers.

As I'm sure you know, you can, to some extent, avoid acoustical problems by sitting nearer to the speakers. This was suggested in one of Audioholics THX articles on acoustics. It was reported that for the average living room, you have to sit around 6 feet away from the speakers. This is in order to meet the THX guideline for proportionally hearing the room acoustics and the original, (hopefully uncoloured) speaker sound.

 

[WmAx]

Does this mean that two 16 bit systems with identical S/N and dynamic range will be audibly identical? I would like references to papers or books which prove this.

The bit depth refers to the SNR. The noise is identical to steady state white noise(just like analog sourced noise).What is there is to prove regarding the digital noise? A properly dithered digital system will have noise identical to an analog circuit. Therefor the audibility of such is a matter of general SNR thresholds; not some specific digital noise.

One of the claims of the Sony textbook is something to do with SACD versus CD and the need for D/A converters:

'the need for high-precision D/A converter circuitry versus the possiblity to even replace D/A circuitry by simple low-pass filtering'

Sony is perhaps one of the least credible sources for information. Realize that any excuse they give is going to be centered around what can make their product appear to be superior. PCM is a highly refined technology. All considerations for proper design have been well known for some time. It might sound complex when you read about it, but it's such routine today, that one can buy an essentially technically perfect performing ADC/DAC PCM system for a computer for $200(E-MU 1212M PCI Card). Transparent audio performance does not require near the performance offered by this example device; it's given as example simply for it's low price in relation to the superb technical performance that is basicly at the limits of possibility, including both ADC and DAC.

I don't know enough about CD to comment on these claims, that was the point of my earlier question - do you know of any papers or books that have measured (understandable) data on digital audio performance?

You have to be very specific. Each parameter is treated as an individual subject. If you want measured final performance, Stereophile is fine for that. The papers discussing ADC/DAC design will refer to theoretical technical limits and considerations, and will not usually refer to perceptual studies. You have to coorelate technical performance with perceptual studies in a seperate step.

Rather than dismissive remarks about SACD, I'm looking for proper tests comparing it to CD.

If you find such a test, let me know. I don't know of a single valid comparison. I am aware of rumors of half-baked tests that do not use a scientific protocol; but these are not reliable tests.

When CD came about; it was developed from scientificly valid perceptual tests that examined the relevant parameters to human audibility. When SACD came about, it was not based upon such testing.

-Chris

 

[Buckle-meister]

Unfortunately...most of us have to share the usage of the living room...

Simple; chuck them out!

 

[WmAx]

As I asked Chris, are these the only measures of performance you need (THD, s/n, dynamic range, frequency response)?

Harmonic distortion should be analyzed seperately, not as a total value, unless that total value is known to be a typical spectral distribution. It's important to point this out becuase the THD number can be very low, and distortion can be audible, if it's distributed into unusually high levels into very high harmonics. This will not normally be an issue; but if equipment is defective, or very badly designed(refer to some exotic high-end gear). Also, IMD should be tested at narrow widths, at many individual points at high frequency limitations of the digital system in order to examine technical performance related to the anti-alias filter and the resample system used. Jitter is also of relevance, but will almost always be well under known audibility thresholds. If one really want to nitpick, the clock frequency of a particulare device should be measured if one suspects slight pitch change; as the system could be defective(out of tolerance parts). This should almost never happen, but it's a possibility.

-Chris

 

[tbewick]

Simple; chuck them out!

I have

The papers discussing ADC/DAC design will refer to theoretical technical limits and considerations, and will not usually refer to perceptual studies. You have to coorelate technical performance with perceptual studies in a seperate step.

Thanks for your reply Chris.

To me, this seems an odd way of going about things. I would have thought that once Sony/Panasonic/Philips etc. had developed digital equipment that is audibly transparent, they would stop further research. When Intel design a new CPU, it runs faster. When a new CD player is released, it does what exactly? It's like re-inventing the wheel.

Harmonic distortion should be analyzed seperately, not as a total value, unless that total value is known to be a typical spectral distribution. It's important to point this out becuase the THD number can be very low, and distortion can be audible, if it's distributed into unusually high levels into very high harmonics. This will not normally be an issue; but if equipment is defective, or very badly designed(refer to some exotic high-end gear). Also, IMD should be tested at narrow widths, at many individual points at high frequency limitations of the digital system in order to examine technical performance related to the anti-alias filter and the resample system used. Jitter is also of relevance, but will almost always be well under known audibility thresholds. If one really want to nitpick, the clock frequency of a particulare device should be measured if one suspects slight pitch change; as the system could be defective(out of tolerance parts). This should almost never happen, but it's a possibility.

I'll look out for these specs in the future reviews. I've only ever seen harmonic distortion specs presented like this for high-end loudspeakers.

I might do some further reading on this subject. Principles of Digital Audio by Pohlmann is meant to be good, but it'll probably be incomprehensible.

 

[mtrycrafts]

To me, this seems an odd way of going about things. I would have thought that once Sony/Panasonic/Philips etc. had developed digital equipment that is audibly transparent, they would stop further research.

Well, this is a marketplace where something new sells, no matter if there is an added benefit or not. Would you want last years body in next years cars?

When Intel design a new CPU, it runs faster.

Yes, in the lab. Will you notice it typing faster? Your server allow you faster up/down loads to take advantage? Not to say when you have huge number crunching to do, it will be faster and maybe noticeable. I haven't noticed much in my comp upgrade.

When a new CD player is released, it does what exactly? It's like re-inventing the wheel.

Yes, that is what it is. Will audiophiles stand for last years CD player? With no improvement on it, even if it is only psychological? No, the public is well conditioned to buying the latest and greatest, no matter what. How else would the marketplace expand? If you don't replace but once every 10 years, how many can they sell? Market saturation and stagnation sets in pretty quick.


I might do some further reading on this subject. Principles of Digital Audio by Pohlmann is meant to be good, but it'll probably be incomprehensible.

Some parts will be, lots of it is understandable.

 

[mtrycrafts]

hi mtrycrafts,

Thanks for the link. It's very technical (and way above my head) but does not present audibility tests. Are these available in the referenced papers? I find this utterly perplexing. If you were to read a medical journal on drugs testing, the author would not state that drug X is better than drug Y and then move on smoothly. They would either provide data supporting this claim in the paper itself, or they would provide a reference where such data is available. They would view experimental testing with the same importance as the theoretical description of how the drug works. Otherwise the theoretical model would be useless. In fact, the experimental data is far more important than the theory itself.

.

Its like this in audio. It is about selling a product. Since it is not deadly as medications can be, the FTC, FCC, whoever regulates has set only basic limits, amp power ratings. In medicine, the Food and Drug agency regulates pretty well, at times as we see some drugs are recalled by them, from time to time. Not all Journal research are created equal. One only has to look at some of the Journal stuff that came out of Harvard on the effects of prayer. It was even fraudulent. But, that is off topic

Back to audio. If the industry used DBT listening to compare components, I'd be in fat city and out of here There would not be an audio market you would recognize That would be the only meaningful testing and is not conducted for an obvious reason, to me at least

So, I don't know, other than to pull out other documents on audibility, threshold of detection and see if what that article found would be audible.

He could have just gone straight to the DBT comparison with no technical investigation other than the set up and testing procedures.
I don't have a good answer for you. It isn't quite like medicine.