Is native DSD playback a myth?

[ecwl]

I've been wondering about this for a long time but I wouldn't mind if more experts chime in.

 

I've seen many people request that their DACs do native DSD playback. But to me, in many ways, the whole thing is a myth because it is rare that we truly see a DAC nowadays that does 1-bit DSD playback of the original DSD signal using 1 element. Some DACs do have a single element DSD playback but they would upsample the DSD signal first.

 

But more commonly, the so-called native DSD playback DACs often involve many elements, e.g. 32 elements at a much higher sampling rate, e.g. 8x or 16x DSD.

 

But to me, the bigger issue is that ultimately, the "original" DSD signal is not really played back. Let's think of a hypothetical scenario of a 1xDSD signal: 1010

And let's imagine a DAC with 4 elements and plays back at 4xDSD.

 

In theory, at 4xDSD, the equivalent signal would be 1111000011110000 and if you truly have 4 elements playing that, you're getting 4444000044440000

 

However, my understanding of most "native" DSD playback actually involve a shift register FIR so that means the 4 elements would actually play back

1111000011110000

01111000011110000

001111000001110000

0001111000011110000

So their sum would actually be 1234321012343210 which is not "native" to the original signal of 1111000011110000 or 4444000044440000

 

Now let's take another scenario where you took your original PCM or 1xDSD signal and upsampled it to 4xDSD via your computer using sophisticated software. And the 4xDSD signal is actually 1010101010101010. Once again, if you're playing the signal back with 4 elements, a "native" playback would be 404040404040404040

However, in reality, most of the time native DSD DACs actually do a shift register FIR so what you actually get are the 4 elements playing:

1010101010101010

01010101010101010

001010101010101010

0001010101010101010

So their sum would actually be 112222222222222211 which is not the same as 4040404040404040

 

In some ways, none of this to me is important because I personally don't believe that DACs should play back the original sampled (or upsampled) signal fed from the PC/CD because I think the role of a DAC is to actually try to upsample and reconstruct the original analog signal as best as possible. This is why I personally don't believe in or like R2R DACs or so-called native DSD playback. Of course, the shift register FIR actually improves the playback of an upsampled DSD signal coming from a PC from a reconstruction of the original analog signal standpoint. 

 

But since I see so many people insist that they want their DACs to have native DSD playback, I thought it might be worthwhile to reflect on what people really mean. Sure, if people are using their PC to upsample for 16xDSD play back, they probably don't want their DACs to resample that signal and would rather just have the DAC do a shift register FIR to produce great sound. But by no means are they truly playing back the 16xDSD signal.

 

Is my understanding of how native DSD playback truly works inaccurate? Since I have no expertise in DACs or even engineering, I thought more experts can maybe explain all of this better to me.

 

[Miska]

30 minutes ago, ecwl said:

But to me, the bigger issue is that ultimately, the "original" DSD signal is not really played back. Let's think of a hypothetical scenario of a 1xDSD signal: 1010

And let's imagine a DAC with 4 elements and plays back at 4xDSD.

 

In theory, at 4xDSD, the equivalent signal would be 1111000011110000 and if you truly have 4 elements playing that, you're getting 4444000044440000

 

No you have understood it wrong. I recommend that you take a look at my DSC1 DAC schematics. If you have 4 elements playing it back the element states would be

1010

 

This presents value 2 out of 5 possible output levels.

 

30 minutes ago, ecwl said:

However, my understanding of most "native" DSD playback actually involve a shift register FIR so that means the 4 elements would actually play back

 

Again, you have understood it wrong.

 

If you have 4 elements and your sequence is they are set to:

1010

 

On purpose I don't prefix it with 0's or 1's, because you need to actually fill it with data.

 

30 minutes ago, ecwl said:

1111000011110000

01111000011110000

001111000001110000

0001111000011110000

So their sum would actually be 1234321012343210 which is not "native" to the original signal of 1111000011110000 or 4444000044440000

 

It is native, each element converts the original bits of the DSD stream. It is totally bit-perfect conversion process. For four elements this results in output value ranging from 0 - 4 so total 5 possible levels, equivalent to 2.3219 PCM bits.

 

Essentially you get moving average over N bits. This is sometimes called DWA.

 

30 minutes ago, ecwl said:

But since I see so many people insist that they want their DACs to have native DSD playback, I thought it might be worthwhile to reflect on what people really mean. Sure, if people are using their PC to upsample for 16xDSD play back, they probably don't want their DACs to resample that signal and would rather just have the DAC do a shift register FIR to produce great sound. But by no means are they truly playing back the 16xDSD signal.

 

There is no rate conversion, the converter works exactly at the sample rate, and every bit coming in is converted exactly as it is.

 

Any D/A conversion requires low-pass filter for reconstruction. This FIR will act as totally non-ringing linear-phase low-pass filter that will do first stage of reconstruction that must be further followed by second analog stage.

 

For example ESS Sabre also operates in a similar way.

 

[ecwl]

Great @Miska. Thanks for explaining. I get that each element is still playing the native signal. You can see that in the example I gave. So yes, if the original signal is 1010, each element is still playing 1010. But because there is the shift register, as I pointed out, the moving average/DWA means that the combination of all the elements playing would end up playing say 2222 instead of 4040. This is why you said that in this example, there are 5 possible levels of output (0, 1, 2, 3, 4). Unless I misunderstood what you are saying.

 

To me, what you're saying is that each element is playing the native signal so it's native DSD. And perhaps that is the definition of DSD, that each element plays natively. Not that the combined output of all the elements is native to the DSD input signal?

 

It's just that when I see other people talk about native DSD playback, they always seem to get the impression that combined output of all the elements are playing back the native DSD input signal which my understanding is not true? Because I think in most people's minds, if their DSD is playing native signal of 1010, they expect the combined output to be 4040 if there are 4 elements, they don't expect it to play 2222.

 

[bogi]

1 hour ago, ecwl said:

Not that the combined output of all the elements is native to the DSD input signal?

Is that considered yet to be in digital domain? In my understanding those elements are already part of D/A conversion stage (DAC in narrower sense). In my view 'native' is about the form how digital data is reaching D/A conversion stage. The way how current or voltage outputs of those elements are combined or summed together looks to me more as analog than digital processing. Maybe you are looking too deeply into D/A stage implementation. Consider it to be a black box, so you don't know about that there are equally weighted elements inside. I would look at it as to an implementation detail, one of more possible approaches how to do the D/A thing.

[ecwl]

8 minutes ago, bogi said:

Is that considered yet to be in digital domain? In my understanding those elements are already part of D/A conversion stage (DAC in narrower sense). In my view 'native' is about the form how digital data is reaching D/A conversion stage. The way how current or voltage outputs of those elements is combined or summed together looks to me more as analog than digital thing. Maybe you are looking too deeply into D/A stage implementation. Consider it to be a black box, so you don't know about that there are equally weighted elements inside. I would look at it as to an implementation detail, one of more possible approaches how to do the D/A thing.

Well, I would say it's more analog than digital because each element is digitally getting the same native DSD signal and then each element is converting the signal into the analog domain and the sound we hear is the sum of the analog signal of all the elements.

 

But now I see @Miska's point. Clearly, I'm wrong in assuming that most people asking for native DSD playback is expecting an analog reconstruction of the native digital signal. Like you, @bogi, you're happy as long as the digital signal enters the DAC stage unaltered. What the DAC actually outputs doesn't matter as long as the DA conversion didn't change the digital signal. In other words, @bogi is perfectly happy that 1010 is 2222 when it comes out of the DAC as an analog signal if every element is getting the digital 1010 signal.

[Miska]

1 hour ago, ecwl said:

To me, what you're saying is that each element is playing the native signal so it's native DSD. And perhaps that is the definition of DSD, that each element plays natively. Not that the combined output of all the elements is native to the DSD input signal?

 

Also the the output is native since each element is native. Function of a DAC is to convert digital data into analog waveform the data represents, as accurately as possible. For delta-sigma DAC that conversion is a low-pass filter.

 

1 hour ago, ecwl said:

It's just that when I see other people talk about native DSD playback, they always seem to get the impression that combined output of all the elements are playing back the native DSD input signal which my understanding is not true?

 

There would be absolutely no point in having multiple elements in such case. No benefit...

 

They are still playing back the native signal, they are just not all switching at the same time for the same bit.

 

[Miska]

16 minutes ago, bogi said:

Is that considered yet to be in digital domain? In my understanding those elements are already part of D/A conversion stage (DAC in narrower sense). In my view 'native' is about the form how digital data is reaching D/A conversion stage. The way how current or voltage outputs of those elements is combined or summed together looks to me more as analog than digital thing. 

 

Yes, like in DSC1, you have 32 current sources (elements) summed up, producing 33 possible different current levels. This then goes to I/V stage that converts this varying current into voltage. Which then goes to 4th order transient optimized analog post-filter. And finally to a cable capacitance compensated output buffer.

 

[Miska]

6 minutes ago, ecwl said:

But now I see @Miska's point. Clearly, I'm wrong in assuming that most people asking for native DSD playback is expecting an analog reconstruction of the native digital signal. Like you, @bogi, you're happy as long as the digital signal enters the DAC stage unaltered. What the DAC actually outputs doesn't matter as long as the DA conversion didn't change the digital signal. In other words, @bogi is perfectly happy that 1010 is 2222 when it comes out of the DAC as an analog signal if every element is getting the digital 1010 signal.

 

Yes, point is to have a DAC that is directly driven by the digital input we give to it. Without altering the digital data. Instead the data as such directly drives the conversion elements. It doesn't matter whether this is R2R for PCM or element array for SDM. There are many possible ways to implement element arrays for SDM. Sometimes you even have couple of different choices to choose from how the array is arranged.

 

In addition, either type of DAC also must have a proper analog post-filter and output buffer.

 

ESS and TI/BB use switched resistors (like DSC1), Cirrus Logic and AKM used switched capacitors, but now in latest top of the line chips AKM also moved to switched resistors. dCS and Chord use switched resistors too.

 

[bogi]

5 minutes ago, ecwl said:

In other words, @bogi is perfectly happy that 1010 is 2222 when it comes out of the DAC as an analog signal if every element is getting the digital 1010 signal.

Analog output accuracy is the only important thing. If you consider how dynamic element matching works, it also violates your 'native' view, because in fact it is changing ordering of bits. But it is intentional to get more accurate analog output.

[Nkam]

Can anyone explain in measurements how DSD compares to PCM in the time domain ? 
 

i feel and have heard that PCM is better in time domain function

[kelvinwsy]

9 hours ago, Miska said:

 

Yes, point is to have a DAC that is directly driven by the digital input we give to it. Without altering the digital data. Instead the data as such directly drives the conversion elements. It doesn't matter whether this is R2R for PCM or element array for SDM. There are many possible ways to implement element arrays for SDM. Sometimes you even have couple of different choices to choose from how the array is arranged.

 

In addition, either type of DAC also must have a proper analog post-filter and output buffer.

 

ESS and TI/BB use switched resistors (like DSC1), Cirrus Logic and AKM used switched capacitors, but now in latest top of the line chips AKM also moved to switched resistors. dCS and Chord use switched resistors too.

 

I am still burning in my Gustard A26 with twin AK4499EX dac chips.. It sounds more elegant than its predessor A22 with twin AK4499EQ chips - I had the SMSL VMS D1 b4 and A26 really has the best of my previous 2 Dac's SQ! The added AK4191EX demodulator chip allows the separation of digital and analogue signals in the digital signal conversion process

i dont even pretend to understand the AKM Data sheet explaining the benefits over the previous generation of integrated AK4499 dac chips

 

All I can say is it sure sounds so much better! 
And the A22 had I believe a pure linear powered Class A output stage( it burns yr hand after an hour's operation even in an airconditioned music room

VS 

The A26 which has a mix of op amps and discrete especially in the IV conversion stage 

It runs A LOT Cooler and sounds as good uf not better (quicker and more nimble sounding)

So its the total product that matters for SQ I feel

 

 

 

[Miska]

7 hours ago, Nkam said:

Can anyone explain in measurements how DSD compares to PCM in the time domain ? 
 

i feel and have heard that PCM is better in time domain function

 

It is the opposite. Since PCM is strictly band-limited in frequency domain, it is also rise time limited in time domain. DSD doesn't have such bandwidth limitation.

 

Also remember that unless you have R2R ladder DAC, your data is converted to DSD-type format for D/A conversion in any case (all delta-sigma DACs). There are very few DACs out there for which PCM would be native format.

 

[Nkam]

31 minutes ago, Miska said:

 

It is the opposite. Since PCM is strictly band-limited in frequency domain, it is also rise time limited in time domain. DSD doesn't have such bandwidth limitation.

 

Also remember that unless you have R2R ladder DAC, your data is converted to DSD-type format for D/A conversion in any case (all delta-sigma DACs). There are very few DACs out there for which PCM would be native format.

 

maybe DSD needs very good gear to reproduce properly.  Because I don’t hear that in my DSD 128 with your filters and modulators compared to your filters and modulators in PCM. 
 

 

[Miska]

1 minute ago, Nkam said:

maybe DSD needs very good gear to reproduce properly.  Because I don’t hear that in my DSD 128 with your filters and modulators compared to your filters and modulators in PCM. 

 

Or you mean you prefer your DAC's modulators instead of mine? Since there are no modulators in PCM.

 

[Nkam]

9 minutes ago, Miska said:

 

Or you mean you prefer your DAC's modulators instead of mine? Since there are no modulators in PCM.

 

Replied in mistake 

 

[Nkam]

No I use your modulators in my DAC running DSD.  
 

my DAC is the Marantz HD DAC1.   Which we have talked about through email, which you have as well.   I think you told me it use the cirrus chip direct DSD Path 

[Miska]

4 minutes ago, Nkam said:

oh, you mean in PCM I’m not bypassing the cirrus chip modulators? 

 

Of course not, and since the maximum PCM input is 192k, it will still go through the on-chip oversampling filter to 8x rate (384k) before going through ugly linear interpolation before heading to the on-chip modulator.

 

4 minutes ago, Nkam said:

in AKM DACs are the modulators bypassed?   Or just the filters there as well? 

 

If you input PCM, only the digital filter. Then follows the S/H oversampling (copying same sample N times) and their on-chip modulator.

 

 

In both cases, the digital filter oversampling ratio remains just 8x. While if you do DSD output from HQPlayer the digital filters are running to 128x ratio in case of the Cirrus chip and 256x or 512x in case of the AKM chip.

 

[Miska]

1 minute ago, Nkam said:

my DAC is the Marantz HD DAC1.   Which we have talked about through email, which you have as well.   I think you told me it use the cirrus chip direct DSD Path 

 

Yes, to use that direct path, you must input DSD data.

 

[Nkam]

Just now, Miska said:

 

Of course not, and since the maximum PCM input is 192k, it will still go through the on-chip oversampling filter to 8x rate (384k) before going through ugly linear interpolation before heading to the on-chip modulator.

 

 

If you input PCM, only the digital filter. Then follows the S/H oversampling (copying same sample N times) and their on-chip modulator.

 

 

In both cases, the digital filter oversampling ratio remains just 8x. While if you do DSD output from HQPlayer the digital filters are running to 128x ratio in case of the Cirrus chip and 256x or 512x in case of the AKM chip.

 

 

yeah I’m on the fence about getting the RME pro BE or a used Bryson BDA-3 which uses an AKM chip as well.  

[Nkam]

Just now, Miska said:

 

Yes, to use that direct path, you must input DSD data.

 

 

Just now, Miska said:

 

Yes, to use that direct path, you must input DSD data.

 

 

and that’s where I say that I like PCM on the Marantz a bit more. 
It has more attack , is denser and more instrument separation.    

[Miska]

Just now, Nkam said:

yeah I’m on the fence about getting the RME pro BE or a used Bryson BDA-3 which uses an AKM chip as well.  

 

I'd recommend to give different kinds of DACs a listen. RME, Holo, T+A (if it's not too expensive).

 

Unfortunately we don't know if Bryston uses the AKM in DSD Direct mode or not.

 

[Nkam]

1 minute ago, Miska said:

 

I'd recommend to give different kinds of DACs a listen. RME, Holo, T+A (if it's not too expensive).

 

Unfortunately we don't know if Bryston uses the AKM in DSD Direct mode or not.

 

Yeah in the end I have to take the chance and listen to some of those. 
unfortunately now the T+A is out of my budget.  
RME PRO BE uses direct DSD for playback correct?

 

i have to write to Bryson to see if their BDA - 3 uses the AKM in DSD direct. 

[Miska]

1 minute ago, Nkam said:

and that’s where I say that I like PCM on the Marantz a bit more. 
It has more attack , is denser and more instrument separation.    

 

For me, I prefer DSD128 through ASDM7ECv2 modulator instead. Same filters in both cases.

 

In addition for 44.1/48k inputs the digital filter in Marantz is pretty leaky, just about 76 dB of attenuation with 1 kHz input, which is pretty weak.

 

With DSD128 input, the THD is lower due to less extended distortion harmonics.  Mostly just second and third harmonic. With PCM inputs there is also fairly strong 5th for example. Also multitone is cleaner with DSD128 input. IMD is also lower, for DSD128 1 kHz difference tone is at minuscule -135 dB while PCM inputs give about -116 dB.

[Miska]

4 minutes ago, Nkam said:

RME PRO BE uses direct DSD for playback correct?

 

There's a configuration switch for it in the menus yes. And there is also switch to choose between the two conversion stage configurations (low-pass filter frequency).

 

[Nkam]

1 minute ago, Miska said:

 

For me, I prefer DSD128 through ASDM7ECv2 modulator instead. Same filters in both cases.

 

In addition for 44.1/48k inputs the digital filter in Marantz is pretty leaky, just about 76 dB of attenuation with 1 kHz input, which is pretty weak.

 

With DSD128 input, the THD is lower due to less extended distortion harmonics.  Mostly just second and third harmonic. With PCM inputs there is also fairly strong 5th for example. Also multitone is cleaner with DSD128 input. IMD is also lower, for DSD128 1 kHz difference tone is at minuscule -135 dB while PCM inputs give about -116 dB.

is that why your filters on the Marantz in PCM make such a large difference in sound quality from their own filter?