dsd DSD Offshoot Discussion From MQA Topic

[tmtomh]

1 hour ago, Lee Scoggins said:

 

Chris, 

 

There were two replies that successfully rebutted your personal attack here.  There was no prior DSD discussion like we were having and John Atkinson and one other person pointed that out to you.  Therefore my comments regarding Mansr not adding value to the discussion with his reply were valid.  Did you feel bad enough about those comments to not include them on this thread?  

 

If you are going to selectively edit the discussion here, then there is no way to have a "level playing field" discussion.

 

Also, you left out the run-up discussion I had with mansr on this topic.  Why did you leave that out?  That sets important context.

 

Don't worry, @Lee Scoggins. There's more than sufficient context here and in the other thread for folks to accurately gauge the strength of your arguments and the accuracy of your claims and summaries.

[Lee Scoggins]

1 hour ago, Jud said:

@The Computer Audiophile moves a bunch of stuff out of a thread where it's off topic to this thread, in which we've had an excellent discussion about the technical side of DSD, and your deep concern, @Lee Scoggins, is that your complaint about another member who has been tremendously helpful in this very thread might no longer have the proper context?

 

Go listen to some music and chill, dude.

 

It's selective editing though that preserves his attack on me but edits out the good responses of others.  If it's off topic then Chris should have removed it all.

 

The fact that you don't see the issue here is more evidence that you are taking a partisan stand against me.

[mansr]

1 minute ago, Lee Scoggins said:

It's selective editing though that preserves his attack on me but edits out the good responses of others.  If it's off topic then Chris should have removed it all.

 

The fact that you don't see the issue here is more evidence that you are taking a partisan stand against me.

I made the list of posts to move (though Chris may have added to or removed from it), starting where it lost relevance to the MQA thread. Posts comparing DSD to MQA are relevant to that thread and should stay there. The selection had nothing to do with any supposed attacks on you. Trust me, you're not that interesting.

[Jud]

1 hour ago, Lee Scoggins said:

more evidence that you are taking a partisan stand against me.

 

No it’s more evidence that I’m far more interested in audio than in Lee Scoggins.  Which in turn makes me think that I don't wish to waste more time on your personal concerns, so welcome to my ignore list.  It's a quite exclusive club with very few members, but you made it.

[jabbr]

18 hours ago, Lee Scoggins said:

 

It's selective editing though that preserves his attack on me but edits out the good responses of others.  If it's off topic then Chris should have removed it all.

 

The fact that you don't see the issue here is more evidence that you are taking a partisan stand against me.

 

[mansr]

For those demanding references, here's Bruno Putzeys and Eelco Grimm: https://www.grimmaudio.com/site/assets/files/1088/dsd_faq-1.pdf

"DSD is another word for 1-bit PCM sampled at 2.8224MHz."

[Paul R]

2 hours ago, mansr said:

For those demanding references, here's Bruno Putzeys and Eelco Grimm: https://www.grimmaudio.com/site/assets/files/1088/dsd_faq-1.pdf

"DSD is another word for 1-bit PCM sampled at 2.8224MHz."

 

 

Quote

DSD is another word for 1-bit PCM sampled at 2.8224MHz. As a release format, SACD is the only one actually using this 1-bit data, but as an internal processing format it’s surprisingly common. The vast majority of audio converters operate at 1 (or a few bits) at megahertz sampling rates. To get to PCM, digital low-pass filtering and decimation is required. This is mostly invisible to the user because converter ICs contain the decimation filter on-board, and put out PCM. But even if you’re not aware of it, your PCM converter has quite a bit of DSD- like processing going on inside.

 

Oh yeah, DSD is 1 bit PCM but you have to decimate and low pass filter it to get to PCM.  Huh - sounds to me like Bruno is saying that DSD is a different data format, something like a clock signal modulated by an analog signal, and producing PDM like data.. 

 

In in other words your argument is like saying a Digital music file and a Word document are the same thing because they are both stored on a spinning disk as ones and zeros. In a sense that is true, but it in the sense one needs to interpret the data it is certainly not. 

 

[pkane2001]

36 minutes ago, Paul R said:

Oh yeah, DSD is 1 bit PCM but you have to decimate and low pass filter it to get to PCM.  Huh - sounds to me like Bruno is saying that DSD is a different data format, something like a clock signal modulated by an analog signal, and producing PDM like data.. 

 

In in other words your argument is like saying a Digital music file and a Word document are the same thing because they are both stored on a spinning disk as ones and zeros. In a sense that is true, but it in the sense one needs to interpret the data it is certainly not. 

 

 

To say these are different is like saying 24 inches of snow is a completely different thing than 1 inch because you have to shovel harder.

 

24 bit PCM has the same problems as 1-bit PCM, just at a different magnitude. Quantization noise is hugely reduced, so we don't need noise shaping with low-pass filtering as much. 16 bit PCM can benefit from dither (noise shaping) and low-pass filtering, and to a greater degree 8 bit PCM, and definitely 1 bit PCM. They are still the same technology, all made out of snowflakes

[Paul R]

31 minutes ago, pkane2001 said:

 

To say these are different is like saying 24 inches of snow is a completely different thing than 1 inch because you have to shovel harder.

 

24 bit PCM has the same problems as 1-bit PCM, just at a different magnitude. Quantization noise is hugely reduced, so we don't need noise shaping with low-pass filtering as much. 16 bit PCM can benefit from dither (noise shaping) and low-pass filtering, and to a greater degree 8 bit PCM, and definitely 1 bit PCM. They are still the same technology, all made out of snowflakes

 

What you are saying is the equivalent of saying that a book written in French and translated to English is the same book. I agree, but the format of the information in each book is quite different, requiring different “filters” to become useful to the reader. All the same alphabet, all the same music, but stored differently. 

 

1bit PCM stores information in a different way than multibit PCM, and that way just happens to be a PDM format. Just as the same book may be stored in French or English, Or Morse Code, or what would generically be called a constrained code. It is fundamentally a different representation of the data than multibit PCM. This would seem to be made obvious by the need to decimate and convert it to get PCM.

 

Are they similar in terms of one can be transcoded into the other with little effective data loss? Sure. Does that make them the same? No. 

 

We we can agree to disagree, I do understand quantization noise, dither, decimation, and the other subjects being brought up. I find it somewhat bemusing that the format of the information storage is such a hot topic. Seems rather basic to me. 

 

More importantly, does music stored in both formats sound different when played back? Does one sound better than the other? If it does sound different,  is it a general case, or specific to individual systems? 

 

[mansr]

1 hour ago, Paul R said:

Oh yeah, DSD is 1 bit PCM but you have to decimate and low pass filter it to get to PCM.

Going from high-rate PCM to low-rate PCM always requires a low-pass filter.

[pkane2001]

11 minutes ago, Paul R said:

1bit PCM stores information in a different way than multibit PCM, and that way just happens to be a PDM format. Just as the same book may be stored in French or English, Or Morse Code, or what would generically be called a constrained code. It is fundamentally a different representation of the data than multibit PCM. This would seem to be made obvious by the need to decimate and convert it to get PCM.

 

What's different except for the number of bits? 1 bit PCM, 16 bit PCM, or 24 bit PCM is still PCM, regardless of how we decide to process it after the fact.

 

[Paul R]

Just now, pkane2001 said:

 

What's different except for the number of bits? 1 bit PCM, 16 bit PCM, or 24 bit PCM is still PCM, regardless of how we decide to process it after the fact.

 

 

Okay, what does one sample represent in a DSD bitstream, and what does one sample represent in a 24 bit data stream? Does one hold intrinsic information about the state of the audio signal when taken in isolation? 

 

I am specifically using the word information here, not data.  How do you extract information from a bit vs. from a word? 

[mansr]

2 minutes ago, Paul R said:

Okay, what does one sample represent in a DSD bitstream, and what does one sample represent in a 24 bit data stream? Does one hold intrinsic information about the state of the audio signal when taken in isolation? 

 

I am specifically using the word information here, not data.  How do you extract information from a bit vs. from a word? 

A 1-bit sample obviously contains at most 1 bit of information while a 24-bit sample can hold more (up to 24 bits). Still, a single 24-bit sample doesn't really tell you much. It doesn't say whether the signal is rising or falling, let alone at what rate. To learn anything meaningful, you must look at multiple consecutive samples.

 

What's important to realise here is that the information you're talking about is actually information about signal components below some cut-off frequency. To obtain such information, we must examine the signal for a minimum duration, the exact amount depending on what we're looking for. Whatever this duration happens to be, the DSD signal has many more samples than a 48/24 PCM signal, so even though the contribution from each sample is smaller, the total is still comparable.

[Paul R]

21 minutes ago, mansr said:

A 1-bit sample obviously contains at most 1 bit of information while a 24-bit sample can hold more (up to 24 bits). Still, a single 24-bit sample doesn't really tell you much. It doesn't say whether the signal is rising or falling, let alone at what rate. To learn anything meaningful, you must look at multiple consecutive samples.

 

What's important to realise here is that the information you're talking about is actually information about signal components below some cut-off frequency. To obtain such information, we must examine the signal for a minimum duration, the exact amount depending on what we're looking for. Whatever this duration happens to be, the DSD signal has many more samples than a 48/24 PCM signal, so even though the contribution from each sample is smaller, the total is still comparable.

Yep. Basic sampling theory is one of the most amazing things in the universe. Just the symmetry alone is a beautiful and awe inspiring thing. And of course, duration is a fundamental concept of it.  I agree totally with what you are saying there. 

 

However, a single multi-bit PCM frame contains a numerical representation of the amplitude of a single channel at the sample time. For stereo, you of course need two samples to make up a single frame. If those samples are greater than 8bits each, then you have to be aware of endianess. There are more considerations of course, including as you point out, the timing. 

 

A single DSD sample does not contain a representation of the amplitude at any one time.  The information it encodes is the vector of change during that sample period. 

 

Best analogy I know of (in English, not technoSpeak) is that PCM and DSD are two different languages saying the same thing. “Different” appears to be what we disagree about, if indeed we do disagree. 

 

What are your thoughts on how DSD sounds compared to say, 24 bit PCM at various sample rates?

 

[mansr]

24 minutes ago, Paul R said:

However, a single multi-bit PCM frame contains a numerical representation of the amplitude of a single channel at the sample time. For stereo, you of course need two samples to make up a single frame. If those samples are greater than 8bits each, then you have to be aware of endianess. There are more considerations of course, including as you point out, the timing. 

 

A single DSD sample does not contain a representation of the amplitude at any one time.  The information it encodes is the vector of change during that sample period.

That's where you are wrong. Whether 24-bit or 1-bit, a sample encodes an approximation of the signal value at one time. The only difference is the accuracy of the approximation.

[JoeWhip]

Personally as I noted in a previous article here at AS, I find pcm and Dsd to sound similar. I do find dsd played back natively through a DSd DAC sounds a tad softer generally, which may be a better match for some systems and not others. I find that dsd converted to pcm sounds great. The same with pcm to dsd. The engineering is way more important than the format, imho.

[Paul R]

45 minutes ago, mansr said:

That's where you are wrong. Whether 24-bit or 1-bit, a sample encodes an approximation of the signal value at one time. The only difference is the accuracy of the approximation.

 

No, this is where you are quite wrong.  DSD, what the sample encodes is simply the vector of change, not the value of the amplitude.  

 

To convince me otherwise, you would need to be able to take a single DSD sample and show me the amplitude of the signal at the time of the sample, using only that single sample. Obviously you can do this with PCM.

 

It doesn't matter if you call DSD 1 Bit PCM, PDM, or even virtual PWM.  The information DSD conveys is *encoded* differently than multi-bit PCM. You can not use the same filters, the same algorithms to *decode* the signal from DSD as you use with multi-bit PCM.

 

[mansr]

2 minutes ago, Paul R said:

DSD, what the sample encodes is simply the vector of change, not the value of the amplitude.

That would be delta modulation. DSD is created using sigma-delta modulation.

 

3 minutes ago, Paul R said:

o convince me otherwise, you would need to be able to take a single DSD sample and show me the amplitude of the signal at the time of the sample, using only that single sample. Obviously you can do this with PCM.

One sample only gives a signal value as accurate as the bit depth permits. A 16-bit or 24-bit sample is obviously a lot more accurate than a 1-bit sample, but there is no difference in principle.

 

4 minutes ago, Paul R said:

You can not use the same filters, the same algorithms to *decode* the signal from DSD as you use with multi-bit PCM.

Oh yes, you can. I do it all the time.

[Paul R]

1 minute ago, mansr said:

That would be delta modulation. DSD is created using sigma-delta modulation.

 Correct me if I am wrong here, but the input to Sigma-Delta modulation, what the Sigma part integrates, is delta modulation, no? 

 

1 minute ago, mansr said:

One sample only gives a signal value as accurate as the bit depth permits. A 16-bit or 24-bit sample is obviously a lot more accurate than a 1-bit sample, but there is no difference in principle.

To me it is a fundamental difference. But then, my background is more in sensors than with playback. I see DSD as the output of essentially a flat line of sensors. 

I suppose we should just agree to disagree here, since neither of us will ever convince the other. 

 

1 minute ago, mansr said:

Oh yes, you can. I do it all the time.

 

Here 's a value - "1"   - what is the signal amplitude at that moment?   (grin) Okay, that's unfair, but that is the "fundamental" difference.

 

One bit samples can not carry enough information, so all they can give you is a vector for change. You can certainly integrate that into a point in time value from the datastream, assuming the 0db is 50%. (I think, I could easily be wrong about where 0db is.) 

 

-Paul 

 

 

[mansr]

3 minutes ago, Paul R said:

Correct me if I am wrong here, but the input to Sigma-Delta modulation, what the Sigma part integrates, is delta modulation, no? 

No.

 

4 minutes ago, Paul R said:

Here 's a value - "1"   - what is the signal amplitude at that moment?   (grin) Okay, that's unfair, but that is the "fundamental" difference.

The value is 1 with a large amount of uncertainty.

[Jud]

22 minutes ago, Paul R said:

 

 Here 's a value - "1"   - what is the signal amplitude at that moment?

 

If I've been understanding the discussion:

 

The maximum possible amplitude.  But the signal doesn't get all the way there because of the low-pass filter, same as it doesn't get all the way to the lowest amplitude when the bitstream value is 0.

 

And if I haven't properly understood, I will as always welcome a correct explanation.

[Miska]

With PCM, the values are always wrong by some amount, while with SDM they are correct most of the time...

 

Anyway, looking just digital domain representations is pretty much pointless. For either one you need ADC and DAC for it to be useful for our purposes.

 

So, for example, the question is more like PCM with R2R/SignMagnitude DAC, or SDM with element array DAC.

 

In the old days low rate PCM was easy to deal with given low computing resources available. But in modern world we wouldn't need to make trade-offs because of computational complexity or amount of data anymore.

 

[fas42]

What's in a name? From https://en.wikipedia.org/wiki/Pulse-density_modulation,

 

Quote

In a PDM signal, specific amplitude values are not encoded into codewords of pulses of different weight as they would be in pulse-code modulation (PCM); rather, the relative density of the pulses corresponds to the analog signal's amplitude. The output of a 1-bit DAC is the same as the PDM encoding of the signal.

 

[asdf1000]

26 minutes ago, Miska said:

With PCM, the values are always wrong by some amount, while with SDM they are correct most of the time...

 

Is this referring to recordings? 

 

What about PCM-to-SDM (DSD) playback conversion?

[Miska]

7 minutes ago, Em2016 said:

Is this referring to recordings? 

 

What about PCM-to-SDM (DSD) playback conversion?

 

Nothing in particular, just from purely mathematical point of view. It is kind of joke...