Playing PCM to DSD Native with foobar2000

[ManniX]

Maxim Anisiutkin was kind enough to quickly release an interim version of the SACD plugin/DSD Processor to enable a flexible PCM to DSD conversion.

Super Audio CD Decoder - Browse /interim at SourceForge.net

Super Audio CD Decoder input plugin for foobar2000. Decoder is capable of playing back Super Audio CD ISO images, DSDIFF, DSF and DSD WavPack files.…

sourceforge.net

Now the DSD Processor can be selected in a DSP chain; previously enabling the PCM to DSD conversion was a global configuration.
Meaning that once enabled the conversion was active also with all other outputs which doesn't support DSD decoding (like ASIO, DS, Wasapi) resulting in just noise and hiss with any PCM content.

This is very useful if you have a DAC with PCM decoding issues (like the SMSL PO100 AK) or you prefer the DSD Native decoding which is basically unfiltered.
It's the nearest thing to a PCM NOS/Bypass mode which is mostly not available unless you have a high-end DAC.

You need to set to select the DSD Processor in the SACD Plugin:

Then you can create a a DSP profile with the DSD Processor in.
You can optionally set as default for the DSD Transcoder output or select it as needed.

Just set to convert to max DSD supported rate of your DAC.
And also to convert the unsupported DSD rate to the highest you can achieve.

You can also create another DSD profile only to play DSD and skip PCM conversion.

You just need to deselect the SDM Type.

There's a small bug, for me, as sometimes the first PCM file played will not be converted well and will produce a hiss.

 

[ManniX]

In case you were wondering what would be the impulse response in DSD mode, at least on an ESS ES9038Q2M DAC, it's corresponding to the PCM filter Slow Roll-off linear phase.
Of course there's quite a lot of HF noise in DSD mode, I had to enable the filter on the scope.

DSD512 Native:

PCM 768kHz Slow Roll-off linear phase:

 

[Snoopy]

In case you were wondering what would be the impulse response in DSD mode, at least on an ESS ES9038Q2M DAC, it's corresponding to the PCM filter Slow Roll-off linear phase.
Of course there's quite a lot of HF noise in DSD mode, I had to enable the filter on the scope.

DSD512 Native:

View attachment 289892

PCM 768kHz Slow Roll-off linear phase:

View attachment 289891

ESS DACs don't allow bypassing the internal modulators.

AKM and ROHM DACs are better for DSD upsampling.

 

[ManniX]

AKM and ROHM DACs are better for DSD upsampling.

I prefer a bit more the ESS 9038Q2M vs the AK4493S; but it's really by a thin margin.

ESS DACs don't allow bypassing the internal modulators.

I'm not sure I understand what you mean with internal modulators, sorry for the ignorance

I see they all have OSF Bypass mode which is bypassing the FIR and requires external oversampling, I guess is not what you are talking about:

 

[Snoopy]

I prefer a bit more the ESS 9038Q2M vs the AK4493S; but it's really by a thin margin.



I'm not sure I understand what you mean with internal modulators, sorry for the ignorance

I see they all have OSF Bypass mode which is bypassing the FIR and requires external oversampling, I guess is not what you are talking about:

View attachment 289981

I found this by someone that maybe explains things better than me.

"ESS chips send incoming DSD stream through IIR filter (volume processing is part of it), then through delta sigma modulator and finally through D/A conversion stage. It's not back converting to PCM rates but it also isn't direct DSD."

AKM , Rohm , burr brown , TI chips let you bypass this. by bypassing the DSM and using another one in HQplayer you can really "improve" or let's say adjust it to your liking. (because people here will say there is no difference etc)

 

[ManniX]

I found this by someone that maybe explains things better than me.

"ESS chips send incoming DSD stream through IIR filter (volume processing is part of it), then through delta sigma modulator and finally through D/A conversion stage. It's not back converting to PCM rates but it also isn't direct DSD."

AKM , Rohm , burr brown , TI chips let you bypass this. by bypassing the DSM and using another one in HQplayer you can really "improve" or let's say adjust it to your liking. (because people here will say there is no difference etc)

Ah ok, I get it now.
It's the volume control in my understanding.

I really have no idea how much it does affect the native DSD decoding but I suspect not that much.
I like a bit more the ESS DSD decoding but maybe it's just the different IIR, they both are not doing volume bypass.

The AKM4493S in theory supports it but I've never seen a low-end DAC giving you the option to enable/disable it.
Maybe some high-end DACs have it?

 

[voodooless]

The AKM4493S in theory supports it but I've never seen a low-end DAC giving you the option to enable/disable it.

The old E30 with pre-fire AK4493 had a DSD direct mode.

 

[ZolaIII]

ESS 32 DAC's are by design (separate better autended input line) better in DSD reproduction that PCM. Foobar2000 DSD convertor is unofficial and old plugin which doesn't work great. Every DSD conversion is CPU intensive and eats a lot of CPU cycles (actually only one whose ever written and it's old and rather bad optimised with SMP 4 suport) so keep that in mind (it's really more demanding then anything else so loosen up latency dependants as much as possible), other than that as long it's DSD 128 or more (you really, really don't need more) it doesn't matter (regarding added ultrasonic noise as in valid audible spectrum it will remain transport).
Edit: keep on mind that you will lose good portion of output gain (depending on design 3~6 dB) and possibility to adjust it (again dependant on software implementation but natively most definitely) and ask your self if it's worth it.

 

[ManniX]

The old E30 with pre-fire AK4493 had a DSD direct mode.

Nice to know, I'll keep an eye on this.
But I definitely don't like that much AKM with PCM, I'll have to look for something else.

ESS 32 DAC's are by design (separate better autended input line) better in DSD reproduction that PCM.

This means that they are not so good in PCM?
I wonder what else is worthwhile to consider, excluding AKM.

Foobar2000 DSD convertor is unofficial and old plugin which doesn't work great. Every DSD conversation is CPU intensive and eats a lot of CPU cycles (actually only one whose ever written and it's old and rather bad optimised with SMP 4 suport) so keep that in mind (it's really more demanding then anything else so loosen up latency dependants as much as possible),

It has been updated recently and quite often.
I can't complain that much honestly.
Yes it's quite CPU intensive and it doesn't work properly converting to DSD512 (I had forgot this and had to switch to DSD256 for the conversion).
After an initial struggle, I can playback for hours without a single hitch.
Massive problems with the new ASIO driver which comes with it (ASIO+DSD) but perfect with the standard ASIO v2.2.

When I'll have time I'll take a look, now I'm trying to do some practice updating the SoX resampler.

other than that as long it's DSD 128 or more (you really, really don't need more) it doesn't matter (regarding added ultrasonic noise as in valid audible spectrum it will remain transport).

What I wonder is if DSD128 it's really enough to convert from DxD PCM.
But I guess I'd hit the limits of my gear and my ear first

Edit: keep on mind that you will lose good portion of output gain (depending on design 3~6 dB) and possibility to adjust it (again dependant on software implementation but natively most definitely) and ask your self if it's worth it.

This is something I haven't yet fully understood...

I've read that DSD content is -6 dB and this is the reason why the SACSD plugin while generating the PCM stream adds +6 dB as default.
Is that what you mean?
Or that converting PCM to DSD there's a 6 dB loss?

Not sure if it's compensated by the plugin but I don't see any difference in output gain between DSD content and the rest.

Thanks for all the useful information!

 

[voodooless]

ESS 32 DAC's are by design (separate better autended input line) better in DSD reproduction that PCM.

Why would that be? All DSD gets convert to PCM. Why not use PCM in the first place?

 

[ManniX]

Why would that be? All DSD gets convert to PCM. Why not use PCM in the first place?

I think that is something that happened in the past.

For sure the latest ESS and AKM DACs are not converting to PCM, it's pretty clear from the datasheets.
But indeed for volume control they go over a 1-bit delta sigma converter.

 

[voodooless]

I think that is something that happened in the past.

For sure the latest ESS and AKM DACs are not converting to PCM, it's pretty clear from the datasheets.
But indeed for volume control they go over a 1-bit delta sigma converter.

Yes they are:

Here is the latest and greatest ES9039PRO.

Guess what the DSD FIR filter does…?

 

[ZolaIII]

Why would that be? All DSD gets convert to PCM. Why not use PCM in the first place?

Why would it be conveniented back? It's simply better autended input line (so it picks up less garbage) and it's line and not a PCM bus. Difference is not big (2~3 dB SINAD to my own measurements from an old ES9018K2M reference board based DAC trogh JRiver and only one I could confirm it for).
@ManniX part you didn't figure out is; normal output DAC provides (2V unbalanced, 4V balanced) - what DSD takes in dB in output voltage and that you can't control volume (output gain) of DSD as simple as I can tell you.

 

[voodooless]

Why would it be conveniented back?

Does it matter? It’s done

But if you must ask: the volume control on ESS cannot be bypassed, this already means you need data in PCM format. Secondly, as you can see from the diagram, the actual rate of the modulators is much higher than the DSD rate, and there is an (A)SRC in between to convert the clock domains and reduce jitter. Finally, the modulators are not 1-bit. They are multi-but delta sigma. 1-bit input doesn’t get best performance.

 

[ZolaIII]

Does it matter? It’s done

But if you must ask: the volume control on ESS cannot be bypassed, this already means you need data in PCM format. Secondly, as you can see from the diagram, the actual rate of the modulators is much higher than the DSD rate, and there is an (A)SRC in between to convert the clock domains and reduce jitter.

Don't use over simplified block diagrams and this whose already discussed hire. To my better knowledge it's multib DSD processing and of course you don't have volume control.

Choosing a DAC for DSD playback?

I’m giving you an actual, measured example where no audible high-frequency IMD exists in PCM or DSD case in the audible range I also shared Jussi's figures above, of lower IMD measured... in audible range.... at DSD512... I’m not sure why you feel the need to argue. Not arguing. Just a...

www.audiosciencereview.com

Unfortunately link to native DSD page is not available anymore. Feel free to correct me if I am wrong but I don't see a purpose of doing a such thing (for decoding path) in a first place.

 

[ManniX]

Guess what the DSD FIR filter does…?

I didn't check the datasheets of the new one... why would it need a FIR filter for DSD?

I'm using the DSD conversion exactly to avoid the PCM FIR filters and now they sneak back

@ManniX part you didn't figure out is; normal output DAC provides (2V unbalanced, 4V balanced) - what DSD takes in dB in output voltage and that you can't control volume (output gain) of DSD as simple as I can tell you.

Thanks for the effort in explaining
What it's not clear to me is why DSD takes away these dB in first place.
Volume control I can understand that it'd be better without.
But what is the cost of having it in practice? Is it really a big quality degradation?

 

[voodooless]

I didn't check the datasheets of the new one... why would it need a FIR filter for DSD?

I'm using the DSD conversion exactly to avoid the PCM FIR filters and now they sneak back

This is exactly to convert back to PCM. The idea is really simple: you expand your bit depth to whatever you like: 32 bits for instance. With this you make the DSD 0 the lowest value, and DSD 1 the highest (possibly some headroom for inter sample overs). Next have a low pass FIR filter. The result is the PCM version of your DSD data. Obviously a real implementation will have a bit more complexity, but the idea remains the same.

 

[ManniX]

This is exactly to convert back to PCM. The idea is really simple: you expand your bit depth to whatever you like: 32 bits for instance. With this you make the DSD 0 the lowest value, and DSD 1 the highest (possibly some headroom for inter sample overs). Next have a low pass FIR filter. The result is the PCM version of your DSD data. Obviously a real implementation will have a bit more complexity, but the idea remains the same.

Which looks very much alike what it's done normally for PCM content; but why converting back to PCM?
Just for Volume control?
Maybe I underestimated how important would be to have a bypass...

 

[voodooless]

Which looks very much alike what it's done normally for PCM content; but why converting back to PCM?
Just for Volume control?
Maybe I underestimated how important would be to have a bypass...

I’ve given plenty of reasons:

Playing PCM to DSD Native with foobar2000

Maxim Anisiutkin was kind enough to quickly release an interim version of the SACD plugin/DSD Processor to enable a flexible PCM to DSD conversion. https://sourceforge.net/projects/sacddecoder/files/interim/ Now the DSD Processor can be selected in a DSP chain; previously enabling the PCM to...

www.audiosciencereview.com

 

[ManniX]

I’ve given plenty of reasons:

I missed the edit.

Well, the reason is basically just volume control...
Thanks a lot for all the information, very interesting.

I think I'll test for a while PCM to DSD playback but the PCM decoding with the ESS sounds just as good to me.
If I still had the PO100 with the AKM, would have definitely switched 100% to DSD.
Ultimately, always a matter of taste.