DSD Battle Royale!

[tailspn]

And when it is explained that clearly, it is obviously wrong.

DSD is just as digital as PCM, in that the signal is stored in discrete, quantized values at a constant sample rate. Just because one system uses a higher sample rate but a smaller number of bits (1, in the case of DSD) doesn't make one "analog" and the other "digital".

DSD and PCM, from a digital storage standpoint are both quantized, in that they are two level digital in form at a fixed clock rate, but only PCM has an actual quantized weight that expresses a value. Like a voltmeter, it can be decoded to an actual voltage at that sample time. DSD (1-bit two level Pulse Density Modulation) expresses only relative change indicators each sample time, and no weight information. It is analog in that the DSD (1-bit two level) pulse stream is a continuous modulated pulse density stream of relative amplitude values when decoded, ie, low pass filtered. Follow the sequence in the table in the middle of the page to understand the workings of a first order Pulse Density Modulation Delta-Sigma Modulator:

http://www.embedded.com/design/debu...-of-sigma-delta-analog-to-digital-converters-


DSD (1-bit two level Pulse Density Modulation) is not Pulse Width Modulation, although they are in the same family.

"DSD Wide" is not PCM. It is a multi-bit (8) two level Pulse Density Modulation pulse stream, like 1-bit two level PDM, where there is still no quantized value ala PCM, but a 8-bit word based relative value from sample to sample like 1-bit two level PDM (DSD), where there is a binary relationship between the 8 bits. Therefore there is a binary weighted relationship between samples that is processable in a computer. Of course PDM has to operate at a much higher sampling rate than PCM, relative to the signal bandwidth, to allow enough samples to recognize the fastest transients. PCM is a frame/sample based system of 2's compliment binary format of actual amplitude values. It is MUCH less efficient than PDM due to all the redundant data carried from sample to sample, and therefore less resolution than PDM for the same data rate. It's the primary no one outside audio uses it.

 

[audioarcher]

Does the Lampi do multiples of DSD or is it designed for just 1x DSD? And if it can do 2xDSD wouldn't it need different filtering than is required for 1xDSD?

 

[wisnon]

Lampi does DSD-128 and its a different circuit.

In my early prototype, its manually switched. Since about a month ago, its autosensing.

 

[Julf]

It is analog in that the DSD (1-bit two level) pulse stream is a continuous modulated pulse density stream of relative amplitude values

No. It is not continuous. It has steps because of the limited resolution, and it has a limited number of values that the output waveform can take. You can't make a pulse shorter than one sample interval. An analog PWM/PDM signal can be modulated smoothly, continuously and without steps between 0 and 100% pulse width. DSD PDM can not, it can only take set pulse width values. Yes, the resulting wave is smooth and continuous after filtering, but so is PCM.

Follow the sequence in the table in the middle of the page to understand the workings of a first order Pulse Density Modulation Delta-Sigma Modulator

I do know how delta-sigma modulation works. You don't seem to understand the quantized nature of the process.

DSD (1-bit two level Pulse Density Modulation) is not Pulse Width Modulation, although they are in the same family.

Correct. DSD is strictly speaking PDM. PWM is a special case of PDM.

"DSD Wide" is not PCM. It is a multi-bit (8) two level Pulse Density Modulation pulse stream, like 1-bit two level PDM, where there is still no quantized value ala PCM, but a 8-bit word based relative value from sample to sample like 1-bit two level PDM (DSD), where there is a binary relationship between the 8 bits. Therefore there is a binary weighted relationship between samples that is processable in a computer. Of course PDM has to operate at a much higher sampling rate than PCM, relative to the signal bandwidth, to allow enough samples to recognize the fastest transients. PCM is a frame/sample based system of 2's compliment binary format of actual amplitude values.

PCM is a way to represent an audio waveform in the time domain as a series of amplitudes. 2's complement Linear PCM is just one of the many encoding systems for PCM.

As far as I understand it, the 8-bit encoding of DSD-wide is not two-level, it is 256-level (8 bit).

It is MUCH less efficient than PDM due to all the redundant data carried from sample to sample, and therefore less resolution than PDM for the same data rate. It's the primary no one outside audio uses it.

"Much" is a rather vague term. How much less efficient is PCM compared to PDM in your view? I have already given you one number based on some widely accepted parameters.

 

[PhilipAC]

"DSD Wide" is not PCM. It is a multi-bit (8) two level Pulse Density Modulation pulse stream, like 1-bit two level PDM, where there is still no quantized value ala PCM, but a 8-bit word based relative value from sample to sample like 1-bit two level PDM (DSD), where there is a binary relationship between the 8 bits. Therefore there is a binary weighted relationship between samples that is processable in a computer. Of course PDM has to operate at a much higher sampling rate than PCM, relative to the signal bandwidth, to allow enough samples to recognize the fastest transients. PCM is a frame/sample based system of 2's compliment binary format of actual amplitude values. It is MUCH less efficient than PDM due to all the redundant data carried from sample to sample, and therefore less resolution than PDM for the same data rate. It's the primary no one outside audio uses it."

Hi tailspn

This is very interesting. I think I understand (at a basic level) what PCM is (as you describe) and what one bit DSD is (1 if the signal goes up and 0 if it goes down from the previous sample).

But this is the first description I've seen of DSD-wide. This obviously holds the key to whether the processing done by most DSD DACs really does mean staying in the "DSD domain" or not. You are clearly saying it does. Could you please explain "DSD wide" in a bit more detail (no pun intended), keeping things very simple for an old codger like me?

Thanks in hope.

Philip

 

[wisnon]

Logically it would be the mathematical calculation of the entire DSD pulse (total relative change) compared to the quantity of the PCM pulse that represents the change and not the previous state (larger portion of the pulse), times the sample rate per period. If this is true, MUCH would be even an understantement.

 

[tailspn]

Just curious, but don't they do any EQ or processing at all? Or do you count "DSD-wide" and/or DXD as DSD?

Channel Classics approach to DSD recording is somewhat unique in that a stereo analog mix/balance is made at the session, and that recorded in DSD. For the majority of their recordings, that session master is only edited in post processing. No EQ, reverb, or re-balancing. Most other labels doing DSD recordings separately track all the microphones at full level, then mix and balance in post. The difference is that Channels recordings, which are recorded and edited on a Pyramix workstation, are only transferred to DXD for the 100 or so millisecond interval of the edit crossfade, leaving the remainder of the DSD content untouched.

Labels doing mixing and balancing in post must convert the entire content to DXD for those processes. As long as they're there, then it's open season on whatever other DXD processing tools they choose to sweeten the mix.

Occasionally, Channel does have to process a recording in DXD to repair a un-releasable album, usually with a noise problem not detected at the session. But that's the exception.

Labels using a Sonoma, or the rare Genex Mixer, can change levels in post using DSD Wide, staying at the same sample rate, and not suffer sonically from the X8 decimation converting to 352.8KHz DXD.

 

[tailspn]

But this is the first description I've seen of DSD-wide. This obviously holds the key to whether the processing done by most DSD DACs really does mean staying in the "DSD domain" or not. You are clearly saying it does. Could you please explain "DSD wide" in a bit more detail (no pun intended), keeping things very simple for an old codger like me?

Thanks in hope.

Philip

Hi Philip,

Multi-bit PDM is used broadly in virtually all DAC's and A/D converters for signal processing today. I'm not knowledgeable enough to say how any one DAC or ADC implements those operations, but in general, it was the breakthrough that allowed going higher than 16 or so bits economically.

There's nothing wrong with any of the formats. They're just digital expressions of analog values at some sample rate. While PCM is archaic and inefficient, PDM has its own set of problems, not the least of which are shifted noise issues. The problem arises when converting from one format to another, when the two formats operate at different sample rates. The greater the difference, the greater the problem. Since all ADC's are front ended with a multi-bit, or to a much smaller degree, 1-bit Delta Sigma Modulators, the purest way to achieve the the highest signal fidelity is to stay in the format and sample rate originally used for the A/D process.

The majority of DAC's available today use multi-bit PDM as their actual end D/A conversion. In DAC's like the ESS9016/18, that operation is running as fast as 40MHz, and up to 6 bits wide. I think your question is really about what happens in the DAC chip prior to the output stage, and that's dependent on the specific chip used, and the input format serving it.

DSD Wide, on the other hand is another Sony/Philips marketing term used to describe the data format used in a proprietary Sony designed chip named E Chip. It's was used in the Sonoma and SADiE DSD workstations, in every Sony VAIO laptop, some PS3 game boxes, and in many DSD/DVD players, like the Denon 1900 series. It's single purpose, using Delta-Sigma Modulators was to convert 1-bit two level PDM (DSD) to 8-bit two level PDM, do math operations on the result, and convert back to 1-bit two level PDM. The important aspect is the sample rate stays the same through all the processes.

 

[tailspn]

No. It is not continuous. It has steps because of the limited resolution, and it has a limited number of values that the output waveform can take. You can't make a pulse shorter than one sample interval. An analog PWM/PDM signal can be modulated smoothly, continuously and without steps between 0 and 100% pulse width. DSD PDM can not, it can only take set pulse width values. Yes, the resulting wave is smooth and continuous after filtering, but so is PCM.

I agree. It's the weakness of my overly simplified Delta-Sigma Modulator explanation. Neither PDM, nor PCM are continuous. Both are made of sequential discrete samples. But where PCM is a whole 2's compliment binary value, like separate stand alone picture frames in a movie film strip, with the majority of the data the same from frame to frame, PDM is not. PDM is simply the result of a n order Delta-Sigma Modulator seeking null with the analog input through successive clock cycles (16 in the linked example I posted earlier), and in that process, outputting successive 1 or 0 levels as a result of the hunting/nulling process. Those 1's and 0's have no weight, as opposed to a PCM sample. Their density is a measure of the relative CHANGE of value impressed on its input. Any one sample is related to its neighbor(s), unlike PCM samples. Please excuse my extension, but that's allot more analog like to me than PCM samples.

 

[Julf]

Neither PDM, nor PCM are continuous. Both are made of sequential discrete samples.

Glad we are in agreement

Any one sample is related to its neighbor(s), unlike PCM samples. Please excuse my extension, but that's allot more analog like to me than PCM samples.

I agree with your explanation, but I just don't see why a relative coding scheme is any more "analog" than an absolute value coding scheme. The result is just the same - a staircase function that is smoothed into a continuous function using a reconstruction filter.

 

[Julf]

Labels using a Sonoma, or the rare Genex Mixer, can change levels in post using DSD Wide, staying at the same sample rate, and not suffer sonically from the X8 decimation converting to 352.8KHz DXD.

I still don't think "decimation" is appropriate to describe a resampling (as opposed to a downsampling), and I have not really seen any evidence of "suffering" either

 

[Julf]

There's nothing wrong with any of the formats. They're just digital expressions of analog values at some sample rate.

Exactly.

While PCM is archaic and inefficient

I think I have to ask you to expand on that one

the purest way to achieve the the highest signal fidelity is to stay in the format and sample rate originally used for the A/D process.

Right - unless the record label / studio has already converted the data into another format. Then you would prefer to stay in that format, instead of reconverting back to the "original" format.

 

[Bruce B]

and I have not really seen any evidence of "suffering" either

it may not "suffer", but it is audible

 

[Julf]

it may not "suffer", but it is audible

I would love to see the lab notebook log of that test

 

[tailspn]

I agree with your explanation, but I just don't see why a relative coding scheme is any more "analog" than an absolute value coding scheme. The result is just the same - a staircase function that is smoothed into a continuous function using a reconstruction filter.

Oh, I agree! The coding systems aren't the issue to me. As I said above to Philip, there's nothing sonically wrong with any of them. They each have their strengths and issues. It's the getting to them that's the issue I think. The original Pacific Microsonics PM1 and PM2 are/were terrific sounding ADC's. They'rs just not made anymore, and those ADC's that are available are all high sample rate Delta-Sigma modulator front ended. So to get to today's PCM sample rates, decimation filtering must take place. That's where the sound quality degradation, if noticeable, happens. It certainly doesn't get better sounding

 

[Bruce B]

I would love to see the lab notebook log of that test

Michael Bishop has said this many times...

Is HE credible enough for you?

 

[Julf]

Michael Bishop has said this many times...

But... That doesn't compute, Captain!

 

[wisnon]

To me DSD is like a photograph of the signal waveform that must be filtered to get rid of the dross and music is all that is left, even if you consider the analog filter also reconstructs to TRUE analog. (Note that mr Lampi has also done this with ONLY tube filtration and except for a few sound artifacts, its pure music. These artifacts are likely resonances from the components he uses which would need a TON of work to optimize). PCM is more like a morse code depiction of the waveform that first needs to be decoded back to human music language and THEN reconstructed to the analog waveform.

 

[wisnon]

Fun and informative discussion. Thanks to all.

 

[Julf]

To me DSD is like a photograph of the signal waveform that must be filtered to get rid of the dross and music is all that is left, even if you consider the analog filter also reconstructs to TRUE analog. (Note that mr Lampi has also done this with ONLY tube filtration and except for a few sound artifacts, its pure music. These artifacts are likely resonances from the components he uses which would need a TON of work to optimize). PCM is more like a morse code depiction of the waveform that first needs to be decoded back to human music language and THEN reconstructed to the analog waveform.

Both PDM and PCM are ways to depict an analog waveform in a binary coded form. It's like writing down a recipe for lasagne in both Spanish and German, and arguing that italian is more "natural" because Spain is closer to Italy