Is the dynamic range of CD sufficient?

[j_j]

JJ so why do all technical/engineering articles say SACD uses PDM, which is related to PWM?
Thanks
Orb

If they do, they're wrong. And, ALL the articles most certainly do not say that. Some lay articles that clearly don't understand how it works have equivocated about PWM or thereabouts, if you want to know why they said that, ask the journalist who wrote the article.

The original SACD presentation did, in fact, mention delta-sigma conversion methods, which are well known to be a particular kind of oversampled PCM. The DAC's in most computers, etc, are in fact using SACD-like technology, they use low-bit-count convertors and oversample and implement noise shaping digitally, and then push the data out of a low-res convertor, leading directly to PCM output.

6 of one, half a dozen of the other.

J. C. Candy and A. Huynh, “Double integration for digital-to-analog
conversion,” IEEE Trans. Commun., vol. 34, no. 12, pp. 1746–1756,
1986.

S. R. Norsworthy, R. Schreier, and G. C. Temes, Delta-Sigma Data
Converters. IEEE Press, 1997.

 

[Orb]

Just to add,
Julf you also have stated DSD/SACD is PDM, others also chime in:
http://www.whatsbestforum.com/showthread.php?11944-DSD-Battle-Royale!&p=235108&viewfull=1#post235108

Cheers
Orb

 

[Julf]

Just to add,
Julf you also have stated DSD/SACD is PDM, others also chime in:
http://www.whatsbestforum.com/showthread.php?11944-DSD-Battle-Royale!&p=235108&viewfull=1#post235108

In that thread I specifically point out the quantified nature of the "PDM" of DSD. I have to agree with JJ that it is better seen as 1-bit oversampled PCM. Both are of course just name labels, the important this is how they behave.

 

[Julf]

Can you provide a real world example how this happens with actual music released?

Sure. Take a 16-bit file. Convert it to 24 bits using zero-fill. Run "flac -ac" on the file. It will show "wasted_bits" as 8, and the HiFi News transition analysis program will show the same. Now either do some gentle low pass filtering (or do a sample rate conversion), so that the data values get recalculated and interpolated. Now flac -ac won't show any wasted bits, and a transition analyzer will not show anything untoward either.

 

[j_j]

Just to add,
Julf you also have stated DSD/SACD is PDM, others also chime in:
http://www.whatsbestforum.com/showthread.php?11944-DSD-Battle-Royale!&p=235108&viewfull=1#post235108

Cheers
Orb

It is again obvious that what you really want to do is to impugn the reputation of others by creating dissent.

SACD is a form of PCM. So is DSD, but DSD is not a single-bit form. That's what the math shows. You can argue whatever you like, but PWM is not a linear system, and SACD is a linear system with added noise. There is a fundamental difference.

I really don't care how much you like it or not, the mathematics is plain as day. Did you read the Jim Candy paper I cited at you? Or Steve Norseworthy's.

 

[j_j]

Sure. Take a 16-bit file. Convert it to 24 bits using zero-fill. Run "flac -ac" on the file. It will show "wasted_bits" as 8, and the HiFi News transition analysis program will show the same. Now either do some gentle low pass filtering (or do a sample rate conversion), so that the data values get recalculated and interpolated. Now flac -ac won't show any wasted bits, and a transition analyzer will not show anything untoward either.

This isn't surprising. On the other hand, you shouldn't see a huge difference in the resulting file size, either. You could get the same results by applying an allpass filter (expressed in matlab form) as bb=** .1 1} and aa={1 .1} Try it.

 

[Orb]

In that thread I specifically point out the quantified nature of the "PDM" of DSD. I have to agree with JJ that it is better seen as 1-bit oversampled PCM. Both are of course just name labels, the important this is how they behave.

So are you saying PCM and PDM in general are same in behaviour and the structure in time domain?
Context being digital.
I take it we are looking at this differently because the only similarity IMO comes down to modern DAC SDM that converts everything to 1-bit stream anyway, which does not encompasse all digital audio or related digital theory as we still have NOS,ladder,DAW architecture,etc.
But as I started I did say I was being picky with the way it was being described to clarify "stairs"/graining that another poster had the wrong meaning on.

Cheers
Orb

 

[Orb]

It is again obvious that what you really want to do is to impugn the reputation of others by creating dissent.

SACD is a form of PCM. So is DSD, but DSD is not a single-bit form. That's what the math shows. You can argue whatever you like, but PWM is not a linear system, and SACD is a linear system with added noise. There is a fundamental difference.

I really don't care how much you like it or not, the mathematics is plain as day. Did you read the Jim Candy paper I cited at you? Or Steve Norseworthy's.

Your arguing semantics here JJ and I am NOT impunging or creating dissent!
IS PCM and PWM the same?
Yes or No.
Is PDM a "kind of subset" of PWM, yes or no?
DSD and SACD is PDM, I am having difficulty in why you will not say it is but then using technical argument with a set scope saying how PCM and PDM are the same; agree yes but in a limited scope (in same way Julf is debating the HiFiNews tool has a limited scope).
Will drop it as we are not looking at this in the same context or even scope criteria, I guess you will correct all the true audio engineering companies working within this structure (such as dCS-MSB-etc), Sony, and also DAW manufacturers such as Merging-etc, and analysis companies such as AP-Agilent.
Orb

 

[Julf]

This isn't surprising. On the other hand, you shouldn't see a huge difference in the resulting file size, either.

Indeed. The "clean" zero-padded "24-bit" file is almost the same size as the original 16-bit one. The filtered/re-interpolated one is significantly larger.

You could get the same results by applying an allpass filter (expressed in matlab form) as bb=** .1 1} and aa={1 .1} Try it.

Will do! Thanks!

 

[Julf]

I take it we are looking at this differently because the only similarity IMO comes down to modern DAC SDM that converts everything to 1-bit stream anyway, which does not encompasse all digital audio or related digital theory as we still have NOS,ladder,DAW architecture,etc.

No. That really has nothing to do with it.

 

[Julf]

in same way Julf is debating the HiFiNews tool has a limited scope

Not sure your "limited scope" scenarios are in any way comparable. What I am saying is that just like the flac encoder, the HiFi News tool will detect simple zero-padding from 16 to 24 bits, but any transformation that forces a re-interpolation/resampling of the file will hide the 16->24 conversion from either tool.

 

[Orb]

No. That really has nothing to do with it.

Ok so you are going on record to say PDM is identical to PCM even if we do not limit comparisons?
So therefore PDM in reality is PCM is what your saying....therefore DSD is PCM.....
Do not think so, nor do a lot of engineering companies and DAW manufacturers.
Cheers
Orb

 

[Orb]

Sure. Take a 16-bit file. Convert it to 24 bits using zero-fill. Run "flac -ac" on the file. It will show "wasted_bits" as 8, and the HiFi News transition analysis program will show the same. Now either do some gentle low pass filtering (or do a sample rate conversion), so that the data values get recalculated and interpolated. Now flac -ac won't show any wasted bits, and a transition analyzer will not show anything untoward either.

So are you saying they can never know if a file was bit depth changed from 16-bit to 24?
Just curious how they have reported this in the past then, and still do.
Cheers
Orb

 

[Julf]

So are you saying they can never know if a file was bit depth changed from 16-bit to 24?
Just curious how they have reported this in the past then, and still do.

As I keep writing, they *can* know if the file is simply zero-padded from 16 to 24 bits. I have come across a number of supposedly 24-bit recordings where this was the case.
If there has been re-interpolation or resampling, they can not tell for sure. They can still make educated guesses.

 

[Julf]

Ok so you are going on record to say PDM is identical to PCM even if we do not limit comparisons?
So therefore PDM in reality is PCM is what your saying....therefore DSD is PCM.....

We are not talking about PDM in general. We are talking about DSD, a very specific quantized format.

 

[Orb]

As I keep writing, they *can* know if the file is simply zero-padded from 16 to 24 bits. I have come across a number of supposedly 24-bit recordings where this was the case.
If there has been re-interpolation or resampling, they can not tell for sure. They can still make educated guesses.

So if it is zero padded for lsb, you would expect that to be noise yes and the structure be different across the dynamic range?
Specifically regarding noise, for lsb last 8 bits what do you think the structure is with regards to TPDF analysis and does it match actual real content (such as musical notes-etc)?
I will try to find one of the hirez analysis that is a resampled with bit depth change hirez download they identified; they identified a CD quality that was resampled and bit depth change to 24bit and 96khz; maybe too much effort and time though to find.

Cheers
Orb

 

[Orb]

We are not talking about PDM in general. We are talking about DSD, a very specific quantized format.

No, this all started when SACD (so therefore include DSD) was suggested to be like PCM, I just pointed out it is actually closer to PWM due to being PDM and this is not the same as PCM.
It has snowballed to this arguing ever since, due to both you and JJ disagreeing with my comment about that correction.
I still cannot see how one can say in general from a technical standpoint DSD is same as PCM (you said yourself it has nothing to do with modern 1-bit SDM DAC chips such as ESS Sabre when using the comparison); that is the conclusion one reaches reading both you and JJ's responses
Cheers
Orb

 

[Julf]

I still cannot see how one can say in general from a technical standpoint DSD is same as PCM (you said yourself it has nothing to do with modern 1-bit SDM DAC chips such as ESS Sabre when using the comparison); that is the conclusion one reaches reading both you and JJ's responses

I'll try one more time, and then I leave it to JJ, who is much better at this than I am.

DSD has a fixed sample rate and pulse width (there is no pulse width modulation per se). You have 2822400 data values per second. In normal 1-bit DSD that is a 1-bit data value, but there are forms of DSD ("dsd-wide") that use multibit data values. Regular DSD is thus a 1-bit PCM data stream produced using a delta-sigma coding process that happens to have the property that the data stream can in theory be converted into analog by simply passing it through a low-pass filter (that is of course not true for multi-bit DSD). Yes, the resulting DSD data stream can be viewed as discrete-time-value PDM, but that is merely a side effect of the delta-sigma encoding process.

This has absolutely nothing to do with modern (or less modern) DAC chips.

 

[JackD201]

Ey Julf, So the coding methods differ but the data is the same?

Sorry this is definitely not my area of expertise. Just looking for clarification. This is over my head not to mention news to me.

 

[Julf]

So if it is zero padded for lsb, you would expect that to be noise yes and the structure be different across the dynamic range?

Not sure I understand your question.

Specifically regarding noise, for lsb last 8 bits what do you think the structure is with regards to TPDF analysis and does it match actual real content (such as musical notes-etc)?

Are you talking about the simple zero-padding case, or the one where there has been a step of re-interpolation or resampling?

I will try to find one of the hirez analysis that is a resampled with bit depth change hirez download they identified; they identified a CD quality that was resampled and bit depth change to 24bit and 96khz;

Please do. If not, you can easily produce one in the way I described earlier (I recommend use of SoX).

If you can't find any, I have several examples I am happy to share, but I still don't know what it is you want an example of - one where the zero-padding is detectable with the flac and HiFi News method, or one where it has been masked by re-interpolation?