The sound quality of DSD?

[bennetng]

The takeaway from the second one is that a pile of ultrasonic garbage might cause in-band distortion with some electronics.

It may also be caused by a DAC filter that doesn't roll-off below 20k (give or take a few percent) when playing 88.2kHz files. But still, the test fulfilled your original requirement that...

Again, has anyone demonstrated any audible difference using the same master? If so, can I get a reference?

as that member did not provide a loopback recording of the output of his DAC. Even if a loopback recording is provided, an ABX test can only demonstrate audible difference, but not explaining where the difference come from. It is really up to what the listener explains to others as there could be several differences.

 

[SIY]

It may also be caused by a DAC filter that doesn't roll-off below 20k (give or take a few percent) when playing 88.2kHz files. But still, the test fulfilled your original requirement that...

...someone show an audible difference between the same master encoded in DSD or standard PCM. And reading what they did, it does not appear to be such. There's a comparison of 16/44 to 24/96 where there's a timing error. And a guy claiming with no evidence, procedures, or controls that he hears a difference between DSD and PCM (and got banned for TOS violations). Then a reference to a Reddit post comparing 44.1 to 88.2, using a problematic resampler.

 

[bennetng]

Then a reference to a Reddit post comparing 44.1 to 88.2, using a problematic resampler.

No.
https://hydrogenaud.io/index.php?topic=110058.msg906791#msg906791

 

[SIY]

No.
https://hydrogenaud.io/index.php?topic=110058.msg906791#msg906791

One more time: there is NO evidence in these links of anyone showing an audible difference between the same master in DSD or PCM format.

 

[bennetng]

One more time: there is NO evidence in these links of anyone showing an audible difference between the same master in DSD or PCM format.

For such a test and within what the foobar ABX tool can do, both the 44k file and DSD file must be converted to something like 88.2k PCM for level matching, avoiding switching glitches and minimizing hardware dependent differences as somthing like the SoX resampler, even in default settings, is more precise than typical DACs. So ultimately this will become a PCM-only test.

 

[bennetng]

For those who did not read the thread carefully:

Of course SACD still has an advantage over CDDA in terms of ENOB and frequency bandwidth due to the sheer 4x data rate, but for data efficiency you only need about 3x the CDDA data rate to exceed what DSD64 can do, for example, 24-bit/88.2kHz, 22-bit/96kHz and so on.

So, for those who disagree that 24-bit/88.2kHz PCM (with additional adjustments like digital level matching) is good enough to represent SACD in listening tests (against 44.1kHz), they should provide their listening test results.

 

[mansr]

For what I've read, DSD is just a direct way to save a delta-sigma modulated signal in a format that doesn't introduce discrete quantization values as the values of the levels to be represented. DS instead of using discrete values as amplitude values used to recreate the original signal, uses the differential (the derivative) of the signal, and this is used to recreate an approximation of the signal. The speed in which the jumps between +1 and -1 are represented is the slope of the original signal at the interception of two points. I may be ignorant, but the methods are seemingly different from each other.

That is pretty much entirely wrong, if that.

Perhaps this is helpful: https://troll-audio.com/articles/pcm-and-dsd/

 

[KeithPhantom]

That is pretty much entirely wrong, if that.

Perhaps this is helpful: https://troll-audio.com/articles/pcm-and-dsd/

This is what Wikipedia says about the matter, I thought I understood correctly how it works:

"In a conventional ADC, an analog signal is sampled with a sampling frequency and subsequently quantized in a multi-level quantizer into a digital signal. This process introduces quantization error noise. The first step in a delta-sigma modulation is delta modulation. In delta modulation the change in the signal (its delta) is encoded, rather than the absolute value. The result is a stream of pulses, as opposed to a stream of numbers as is the case with pulse code modulation (PCM). In delta-sigma modulation, accuracy of the modulation is improved by passing the digital output through a 1-bit DAC and adding (sigma) the resulting analog signal to the input signal (the signal before delta modulation), thereby reducing the error introduced by the delta modulation ... Primarily because of its cost efficiency and reduced circuit complexity, this technique has found increasing use in modern electronic components such as DACs, ADCs, frequency synthesizers, switched-mode power supplies and motor controllers.[1] The coarsely-quantized output of a delta-sigma modulator is occasionally used directly in signal processing or as a representation for signal storage. For example, the Super Audio CD (SACD) stores the output of a delta-sigma modulator directly on a disk."

Maybe you misunderstood me, but, you do not store the discrete values used to calculate the delta between two samples, you store the delta between two samples. This is what I tried to say.

From your own link:

"This is all nonsense. A DSD signal is discrete in time and amplitude, ergo digital. That is all there is to it. A signal cannot be a little digital. The part about pulse density modulation is particularly devious. If rendered graphically as a square wave, the 1-bit signal may indeed resemble a series of pulses at varying intervals. This is, however, a flawed interpretation. Correctly, a 1-bit signal should be viewed just like any other sampled signal: as a sequence of pulses at fixed intervals, the height of each matching the sample value, in this case either +1 or -1. Moreover, the sigma-delta modulation process by which the signal is created operates on discrete-time samples. It is a method for quantising a sampled signal with noise shaping, nothing more. There is certainly not anything analogue about it. Actual PDM is in fact analogue; it just has nothing to do with DSD".

I never said the pulses are in variable intervals, they are fixed, but the change of the voltages between +1 and -1 represents the deltas encoded by the delta-sigma modulation (for a lack of a delta between samples, samples are held until the delta between two samples can be represented by a change in voltage). If DSD was PWM, it would be true that it is a special case of PCM (it would be considered DPCM). Albeit, Wikipedia is explicit by saying that DSD is encoded by delta-sigma modulation and PDM, being this different than DPCM:

"DSD uses pulse-density modulation encoding - a technology to store audio signals on digital storage media which are used for the SACD. The signal is stored as delta-sigma modulated digital audio, a sequence of single-bit values at a sampling rate of 2.8224 MHz (64 times the CD audio sampling rate of 44.1 kHz, but only at 1⁄32768 of its 16-bit resolution)".

Finally "actual" PDM is just "a form of modulation used to represent an analog signal with a digital signal" (Wikipedia). There is nothing analog from PDM, it is just another way to represent an analog signal using a digital signal. And it seems DSD has a lot to do with PDM.

This picture can help to visualize what is happening:

PS: I would like to add this is what I've understood, and it doesn't represent a totalitarian statement of the topic. If there is something I get wrong, corrections are welcome.

 

[mansr]

Wikipedia

Is occasionally wrong.

Maybe you misunderstood me, but, you do not store the discrete values used to calculate the delta between two samples, you store the delta between two samples. This is what I tried to say.

In that case, you tried to say the wrong thing.

I never said the pulses are in variable intervals, they are fixed, but the change of the voltages between +1 and -1 represents the deltas encoded by the delta-sigma modulation (for a lack of a delta between samples, samples are held until the delta between two samples can be represented by a change in voltage).

Again, completely wrong.

If DSD was PWM, it would be true that it is a special case of PCM (it would be considered DPCM).

That isn't even wrong.

Finally "actual" PDM is just "a form of modulation used to represent an analog signal with a digital signal" (Wikipedia). There is nothing analog from PDM, it is just another way to represent an analog signal using a digital signal. And it seems DSD has a lot to do with PDM.

Proper PDM has continuously variable pulse intervals. That makes it an analogue signal.

This picture can help to visualize what is happening:

The only thing that picture can "help" with is promoting misunderstandings.

DSD is 1-bit PCM with the quantisation noise pushed into ultrasonic frequencies. Why do people insist on concocting all these convoluted stories?

 

[KeithPhantom]

Is occasionally wrong.


In that case, you tried to say the wrong thing.


Again, completely wrong.


That isn't even wrong.


Proper PDM has continuously variable pulse intervals. That makes it an analogue signal.


The only thing that picture can "help" with is promoting misunderstandings.

DSD is 1-bit PCM with the quantisation noise pushed into ultrasonic frequencies. Why do people insist on concocting all these convoluted stories?

As I presented evidence to back up my claims, I would like to implore to peer-review and more expert knowledge to scrutinize my argument. I could not find anything related to your argument of DSD being simply 1-bit PCM. Actually, I found a paper that has the specs of the Scarlet Book, and they both show a graph pretty similar to the one I just posted: https://www.sonicstudio.com/pdf/dsd/SACD_FormatOverview.pdf page 13 (also, patent US006975257B2 page 2 shows something really similar). TI's own forum has this thread where they refer the generation of DSD signals as a process done by PDM: https://e2e.ti.com/support/audio/f/6/t/590672?PCM4204-How-to-measure-full-scale-PDM-DSD-signal-

All evidence that I can find says nothing about your claims, and more sources, not just Wikipedia, point to the same place. I do not think all engineers and experts are wrong about how PDM and DSD work. I even found a patent which explicitly states that "A good example of a PDM stream is the direct stream digital (DSD) used in Super audio applications (SACD). This stream is a 64 times over-sampled one-bit PDM stream" (Lutsen Ludgerus Albertus Hendrikus Dooper, US008681853B2). My definitions, sources, and interpretations are here and supported by evidence and can be criticized if necessary. Now, I do not see anything about your claims that deems them correct and valid; I would like to politely ask you to provide the evidence that supports your claims.

Another patent says that "Conventional SDMs are well-known for their insensitivity to analog imperfections, and therefore they are appropriate for a large number of applications. Their usefulness has led to the adoption of the Direct Stream Digital, or DSD, format (the single bit output of an SDM) as the data format for Super Audio CD (SACD)" (Reefman et al., US006975257B2). If these researches and engineers are wrong, I would like you to show how they are wrong and what can we improve about this matter.

Finally, I would like to thank you for your opinions, they have led me to learn more about this topic.

 

[Ron Party]

I seem to remember JJ posting either here or, more likely, at AVS and explaining the math underlying all of this. While my memory is fuzzy - seemingly a more common problem these days - I think mansr has it right. Can anyone reach JJ?

 

[Tks]

I seem to remember JJ posting either here or, more likely, at AVS and explaining the math underlying all of this. While my memory is fuzzy - seemingly a more common problem these days - I think mansr has it right. Can anyone reach JJ?

Could have sworn I've seen him post recently on this topic (like within a week or so), kept saying how wrong it was. My memory's hazy atm if the topic was actually DSD or not.

 

[KeithPhantom]

I seem to remember JJ posting either here or, more likely, at AVS and explaining the math underlying all of this. While my memory is fuzzy - seemingly a more common problem these days - I think mansr has it right. Can anyone reach JJ?

I would like to see that math, but even though that evidence is presented, all inventors, patent holders, and engineers refer to DSD as being the combination of both PDM and SDM. It could be a piece of evidence, but it has to go against the multiple papers I have found saying the same thing and supporting my argument. Papers where an AES-cited member such as Mr. Reefman supports my claims, at least to my interpretation. I am using the best of my reading and interpretation skills, but the evidence that I cite says otherwise.

And from his own link:

Even his own link says that the "more correct non marketing term" is PDM. I am saying the same thing.

 

[pkane]

I would like to see that math, but even though that evidence is presented, all inventors, patent holders, and engineers refer to DSD as being the combination of both PDM and SDM. It could be a piece of evidence, but it has to go against the multiple papers I have found saying the same thing and supporting my argument. Papers where an AES-cited member such as Mr. Reefman supports my claims, at least to my interpretation. I am using the best of my reading and interpretation skills, but the evidence that I cite says otherwise.

PDM, even according to Wikipedia, is the same as a 1 bit PCM quantizer with a carry-forward error propagation:

Analog-to-digital conversion
A PDM bitstream is encoded from an analog signal through the process of delta-sigma modulation. This process uses a one bit quantizer that produces either a 1 or 0 depending on the amplitude of the analog signal. A 1 or 0 corresponds to a signal that is all the way up or all the way down, respectively. Because in the real world, analog signals are rarely all the way in one direction, there is a quantization error, the difference between the 1 or 0 and the actual amplitude it represents. This error is fed back negatively in the ΔΣ process loop. In this way, every error successively influences every other quantization measurement and its error. This has the effect of averaging out the quantization error.

The D2A conversion is just a simple low-pass filter applied directly to the 1-bit PDM (PCM?) stream.

 

[KeithPhantom]

PDM, even according to Wikipedia, is the same as a 1 bit PCM quantizer with a carry-forward error propagation:



The D2A conversion is just a simple low-pass filter applied directly to the 1-bit PDM (PCM?) stream.

Thanks, I am trying to understand his argument, he may have a point with PDM being 1-bit PCM, but I haven't seen any reference of it anywhere. May you refer me somewhere to read more about this?

Edit: nevermind, Audio Precision explained the process (and here, page 3), @mansr, you were right, I didn't notice this little fact, thanks.

 

[pkane]

Thanks, I am trying to understand his argument, he may have a point with PDM being 1-bit PCM, but I haven't seen any reference of it anywhere. May you refer me somewhere to read more about this?

Follow the links to the Wikipedia article or sigma-delta modulation one from my post for more explanation and examples. The error feedforward mechanism is constructed so as to move the quantization noise out of the audible band through noise-shaping.

 

[KeithPhantom]

The error feedforward mechanism is created so as to move the quantization noise out of the audible band through noise-shaping.

I already knew that, but it was a good refresh of the topic, also thanks to you.

Edit: nevermind, Audio Precision explained the process (and here, page 3), @mansr, you were right, I didn't notice this little fact, thanks.

But it was pretty hard to find a more explicit explanation, but thanks to all. Just interpolate and reduce the wordlength of PCM and you get PDM, then apply noise shaping (using the error feedforward method you described) to move most of the quantization error energy out of the passband, and you get DSD. Interesting...

@pkane, I have a question: how do you increase word length in the conversion from PDM to multi-bit PCM, I know you have to decimate, but you do you recover the remaining "bits" out of an 1-bit signal? I don't find this anywhere, but I would like to know.

 

[JustAnandaDourEyedDude]

I already knew that, but it was a good refresh of the topic, also thanks to you.

But it was pretty hard to find a more explicit explanation, but thanks to all. Just interpolate and reduce the wordlength of PCM and you get PDM, then apply noise shaping (using the error feedforward method you described) to move most of the quantization error energy out of the passband, and you get DSD. Interesting...

@pkane, I have a question: how do you increase word length in the conversion from PDM to multi-bit PCM, I know you have to decimate, but you do you recover the remaining "bits" out of an 1-bit signal? I don't find this anywhere, but I would like to know.

I too learned quite a bit about DSD just by reading this thread and a few posts in one of the other DSD threads (one of which linked to Waldrep's 2013 interview of Siau in which Siau gives an excellent and balanced account of DSD-64 and multi-bit DSM for the audiophile layperson), mansr's Troll Audio DSD vs PCM page that he linked to, and also the same few Wikipedia pages that you read. It seems that in the case of 1-bit DSM, the delta of the multi-bit DSM comparators just happens to be the largest amplitude A of the analog signal, and is represented by a 1 for +A or a 0 for -A in the output of the 1-bit DSM. A raw 1-bit PCM bitstream (scaled from A down to 1) would be just 1 whenever the signal is positive and 0 whenever it is negative, with a signal-correlated quantization error. The output of a 1-bit DSM due to its error correction step, converts the samples of the analog signal to a (by definition 1-bit) PDM bitstream in which the quantization error is statistically uniformly distributed across the entire frequency bandwidth of the sampling. Then a noise-reshaping algorithm is used to shift the quantization error in the audio passband into the remaining high frequencies, giving the DSD (1-bit) bitstream as its output. So I think mansr is right when he says "DSD is 1-bit PCM with the quantisation noise pushed into ultrasonic frequencies." I do agree with you that PDM output as used in this process is in a digital sample form.

Trying to read between the lines of firedog's quotation of Roon's DSD manipulation process, I imagined that they start by converting the DSD bits to 64-bit reals or integers (thus padding with zeros), and then the decimation algorithms (and other audio processing / DSP) must fill in that padding through their operation, so the result looks like PCM. At the end they may do a 1-bit DSM and noise-shaping to recover the new DSD output to send to the DAC. But this last para is just my imaginations. pkane or mansr would be likely be able to clarify.

 

[watchnerd]

To me, DSD sounds like MQA.

 

[KeithPhantom]

To me, DSD sounds like MQA.

For me, that Tidal charges the same for regular lossless and MQA is a scam, they should charge less for just 44.1/16.