Which DACs bypass digital filtering?

[stefano_mbp]

15 minutes ago, davide256 said:

many DACs convert DSD to PCM such as the Chord Mojo I use

All Chord dacs do, it is their “peculiarity and this is well known, but many other don’t 

[Yurii]

Well, if everyone agrees to dig this deep, then here are the detailed characteristics of the AK4191 DSD filters. You can see a wide range of values in direct DSD mode. Which of them the manufacturer will choose and what he will call them, I do not know. And here is the biggest surprise for perfectionists. Direct path is not supported in DSD1024 mode. I'm here without comment.

A block diagram in DSD mode can also be useful. It's nice that special attention is paid to eliminating clicks in pauses.

[bogi]

2 hours ago, davide256 said:

Gustard R26 claims it’s native DSD, + R2R NOS for PCM. Is it truly native DSD?

From R26 manual:

 

It contains two discrete DAC implementations in one case. l7audiolab measurements show high noise floor of DSD part. Wolf used Roon to upsample 48k/24b test signal to DSD256. I don't know if (how much) it could influence measurement results.

[Nkam]

14 minutes ago, bogi said:

From R26 manual:

 

It contains two discrete DAC implementations in one case. l7audiolab measurements show high noise floor of DSD part. Wolf used Roon to upsample 48k/24b test signal to DSD256. I don't know if (how much) it could influence measurement results.

well Roon will upsample PCM 48khz to DSD even if the DAC does not support it. 
 

so go figure. 
 

if I try and do that with HQplayer, it will properly  tell me it’s a no go 

[bogi]

5 minutes ago, Nkam said:

well Roon will upsample PCM 48khz to DSD even if the DAC does not support it. 

I had test signal quality in my mind. It may impact measurement results. But I don't know how much.

One could generate higher resolution PCM test signal and then convert it to DSD in HQPlayer Pro.

[Miska]

17 hours ago, Nkam said:

what do you mean it doesn’t do anything?   
 

it doesn’t implement any filters it has and modulation?  

 

If you change the DSD filter, nothing changes in the output. IOW, the respective output doesn't match that of what is specified on the AKM datasheet. Overall, this is a good test for DSD Direct functionality with this chip combination, because the DSD filter selection is only available in DSD Direct mode. And now it isn't working. But switching between DSD Direct  enabled/disabled does change the output.

 

[Miska]

1 hour ago, Nkam said:

well Roon will upsample PCM 48khz to DSD even if the DAC does not support it. 
 

so go figure. 
 

if I try and do that with HQplayer, it will properly  tell me it’s a no go 

 

Roon always converts to 44.1k multiple DSD rates. AFAIK it never produces DSD at multiples of 48k.

 

HQPlayer can do either one, it is up to you to choose. If you choose a combination that is not possible, HQPlayer will refuse to play.

 

[Miska]

1 hour ago, bogi said:

It contains two discrete DAC implementations in one case. l7audiolab measurements show high noise floor of DSD part. Wolf used Roon to upsample 48k/24b test signal to DSD256. I don't know if (how much) it could influence measurement results.

 

Modulator choice has big impact on any SDM performance.

 

But R26 seems to have some bug where DSD at multiples of 44.1k produce noisy output, while DSD at multiples of 48k doesn't. (and roon produces just multiples of 44.1k, AFAIK)

 

[Nkam]

45 minutes ago, Miska said:

 

If you change the DSD filter, nothing changes in the output. IOW, the respective output doesn't match that of what is specified on the AKM datasheet. Overall, this is a good test for DSD Direct functionality with this chip combination, because the DSD filter selection is only available in DSD Direct mode. And now it isn't working. But switching between DSD Direct  enabled/disabled does change the output.

 

so if I understood correctly it doesn’t do DSD direct as it says it does? And thus won’t be optimized for use with HQplayer?

 

if so that changes my verdict to try that DAC.  
 

ugh, why all these bugs , which seem like ‘ childhood illnesses ‘ they should have seen from the start? 
that bug seems to me as a ‘rookie mistake ‘

[Miska]

29 minutes ago, Nkam said:

so if I understood correctly it doesn’t do DSD direct as it says it does? And thus won’t be optimized for use with HQplayer?

 

Maybe it does, because the DSD Direct switch has effect. But now it's a strange beast and I cannot make up my mind what to think about it.

 

But in my opinion, all the provided options should actually work.

 

Hopefully a hopefully upcoming firmware update doesn't brick the entire device like happened to my SMSL D300.

 

[Nkam]

5 hours ago, Miska said:

 

Maybe it does, because the DSD Direct switch has effect. But now it's a strange beast and I cannot make up my mind what to think about it.

 

But in my opinion, all the provided options should actually work.

 

Hopefully a hopefully upcoming firmware update doesn't brick the entire device like happened to my SMSL D300.

 

 

you can’t measure it to see if it’s actually doing direct DSD?

 

that is, there is no measurements to verify it?

 

thanks 

[davide256]

7 hours ago, Miska said:

 

Modulator choice has big impact on any SDM performance.

 

But R26 seems to have some bug where DSD at multiples of 44.1k produce noisy output, while DSD at multiples of 48k doesn't. (and roon produces just multiples of 44.1k, AFAIK)

 

Trying to follow this. Are you talking about up-sampling 44.1 PCM to DSD?

[bogi]

@Miska could you please answer yet this post? https://audiophilestyle.com/forums/topic/31941-which-dacs-bypass-digital-filtering/page/23/#comment-1240758

[bogi]

1 hour ago, davide256 said:

Trying to follow this. Are you talking about up-sampling 44.1 PCM to DSD?

> Modulator choice has big impact on any SDM performance.
This was about creating DSD test signal from PCM test signal. There are no DSD test signals publicly available for download so one has to find a software tool and generate one. Miska told that software delta sigma modulator used in PCM to DSD conversion has big impact on conversion result.

> (and roon produces just multiples of 44.1k, AFAIK)

Not every PCM to DSD capable tool/program is able to generate 48k based DSD content. That explains that Wolf could not measure 48k based DSD with Roon (if he would want to).

 

> But R26 seems to have some bug where DSD at multiples of 44.1k produce noisy output, while DSD at multiples of 48k doesn't.

That's about Miska's R26 measurements with DSD input (Miska of course can generate 48k DSD test signal).

[Miska]

4 hours ago, Nkam said:

you can’t measure it to see if it’s actually doing direct DSD?

 

that is, there is no measurements to verify it?

 

Yes and no, not a simple way for a previously unknown black box chip. Other than comparing the results with the information from datasheet and knowledge how these converters technically work.

 

[Miska]

3 hours ago, bogi said:

@Miska could you please answer yet this post? https://audiophilestyle.com/forums/topic/31941-which-dacs-bypass-digital-filtering/page/23/#comment-1240758

 

As @semente said, is the filter analog or digital? TI/BB chips have analog FIR (four options), so does my DSC1.

 

Essentially, the DWA section in 4499EX is also analog FIR.

 

[bogi]

15 minutes ago, Miska said:

Essentially, the DWA section in 4499EX is also analog FIR.

 

I was pointing to AK4191 'DSD Filter 2', not to AK4499EX. AK4191 is digital circuit. Gustard menu contains "DSD digital filter adjustment". When DSDD bit is set to "1", there is no way to skip 'DSD Filter 2', which is digital FIR filter described as "FIR filter that reduces high frequency noise of DSD input data". 'DSD Filter 2' has clearly digital input and digital output.

 

 

This thread is named "Which DACs bypass digital filtering". My question remains. Can be DSD input stream processing through such a FIR filter considered lossless? Does DSD signal path containing such a FIR filter fit to our understanding of the term 'direct DSD path'? ESS DSD path also contains FIR filter, but followed by additional modulation before D/A stage. So there is my question what can be yet considered to be direct DSD path and why.

 

I could understand your word "essentially" this way:

'DSD filter 1' is digital filter, but it is essentially used for the same thing like analog one. DSD signal needs nothing more than low pass filtering to get analog signal. Low pass filtering can be done in both digital and analog domains. If we do some preliminary low pass filtering (removing of high frequency content) in digital domain and then follow it with analog filtering, in principle it brings the same type of ananlog result as pure analog filtering. Did I understand your "essentially" well?
 

[Miska]

31 minutes ago, bogi said:

Gustard menu contains "DSD digital filter adjustment".

 

Which does nothing, as I told earlier...

 

31 minutes ago, bogi said:

When DSDD bit is set to "1", there is no way to skip 'DSD Filter 2', which is digital FIR filter described as "FIR filter that reduces high frequency noise of DSD input data". 'DSD Filter 2' has clearly digital input and digital output.

 

 

This thread is named "Which DACs bypass digital filtering". My question remains. Can be DSD input stream processing through such a FIR filter considered lossless? Does DSD signal path containing such a FIR filter fit to our understanding of the term 'direct DSD path'?

 

There are two options for the Filter 2. For the Filter 1 there are no options.

 

It is clear that Filter 1 path cannot be considered direct DSD path, it will do quite a bit of DSP processing for the data.

 

 

Here, you can conclude that the Filter 2 essentially rearranges the data to match input requirements of the 4499EX. Note that 4499EX always runs at either 5.6/6.1 MHz or 11.2/12.3 MHz rate. This is controlled by OSR bit, where OSR=1 is 5.6/6.1 MHz. Then there is also control for how many bits are transferred to the 4499EX per clock. This is OBIT setting, where OBIT=00 is seven, OBIT=01 is six and OBIT=10/11 is five. This affects how the DWA filter in 4499EX cuts. You can see that at the lower conversion rate, the 4191 cannot pack enough DSD data to the transfer unit.

 

The Gustard DSD filter menu option is supposed to switch between the two DSDF settings. Now it doesn't work.

 

31 minutes ago, bogi said:

ESS DSD path also contains FIR filter, but followed by additional modulation before D/A stage.

 

ESS DSD path contains IIR filter. That IIR filter is also on it's PCM path as second stage digital filter.

 

31 minutes ago, bogi said:

I could understand your word "essentially" this way:

'DSD filter 1' is digital filter, but it is essentially used for the same thing like analog one. DSD signal needs nothing more than low pass filtering to get analog signal. Low pass filtering can be done in both digital and analog domains. If we do some preliminary low pass filtering (removing of high frequency content) in digital domain and then follow it with analog filtering, in principle it brings the same type of ananlog result as pure analog filtering. Did I understand your "essentially" well?

 

No, because you have the modulator in the path. So you are not converting the original data to analog.

 

[Miska]

6 hours ago, davide256 said:

Trying to follow this. Are you talking about up-sampling 44.1 PCM to DSD?

 

It doesn't matter for this context how the DSD data came to be.

 

[bogi]

50 minutes ago, Miska said:

No, because you have the modulator in the path. So you are not converting the original data to analog.

 

I'm sorry I intended to write 'DSD Filter 2' not 1 in my previous post. I did not mean the path with volume control and modulator in it.

 

55 minutes ago, Miska said:

Here, you can conclude that the Filter 2 essentially rearranges the data to match input requirements of the 4499EX.

 

That's helpful explanation. Is the "internal DSD filter cut off frequency" result of data re-arrangement (to binary weighted 5 to 7 bit with possible rate lowering to fit the rate between chips) or do you expect any additional data processing algorithm yet to be used to filter high frequency noise, as suggested by "reduces high frequency noise in DSD input data" from the datasheet?

[Miska]

54 minutes ago, bogi said:

That's helpful explanation. Is the "internal DSD filter cut off frequency" result of data re-arrangement (to binary weighted 5 to 7 bit with possible rate lowering to fit the rate between chips) or do you expect any additional data processing algorithm yet to be used to filter high frequency noise, as suggested by "reduces high frequency noise in DSD input data" from the datasheet?

 

There cannot be much, because there's so little value space. And the "high frequency" noise is fundamental need for the converter to operate in first place, same for the four modulators that are available in 4191. The primary noise filter is the conversion stage itself. Data arrangement in 4191 combined with the 4499EX operation forms the desired filter characteristics. Similar way as in TI/BB chips it gives the four analog FIR response options depending on how the data is arranged.

 

Binary weighted encoding is just used to reduce number of chip pins needed. Otherwise you would need 128 pins on both sides instead of just 7. That would make the chip much larger physically and would take more board space. This is why the DWA needs to be at the 4499EX side.

 

[bogi]

6 minutes ago, Miska said:

Data arrangement in 4191 combined with the 4499EX operation forms the desired filter characteristics. Similar way as in TI/BB chips it gives the four analog FIR response options depending on how the data is arranged.

 

Thanks. That's much more understandable than the datasheet formulation.

[davide256]

3 hours ago, Miska said:

 

It doesn't matter for this context how the DSD data came to be.

 

Still unclear. What does 1 bit multi-megabit encoded DSD recordings have to do with N x CD rate? I see the 48K option on HQPlayer output choices but have

never really understood why its there.

[bogi]

49 minutes ago, davide256 said:

What does 1 bit multi-megabit encoded DSD recordings have to do with N x CD rate? I see the 48K option on HQPlayer output choices but have never really understood why its there.

 

DSD is 1 bit signal at high rates. Something like very very quickly blinking LED lamp, where average density of '1' (light on) defines the light intensity in a time interval the density is related to. If LED lamp operates on very high frequencies so we cannot distinguish individual blinks, our eyes do such an averaging. In fact our eyes do low pass filtering with high frequency light content, similar like delta sigma DACs do with upsampled and delta sigma modulated audio content.

 

DSD signal bit rate can be of different values. Usual values are multiples of 44.1k base. For example DSD64 rate (used for SACD) is 64 x 44.1k = 2.8224M. That's what DAC may display as incoming sample rate. DSD512 is at 512 x 44.1k rate. High resoultion DSD downloads in DSF format are by a convention at 44.1k based rates (44.1k x N, N is usually 64, 128 or 256).

 

But then there are people like we who are doing PCM to DSD upsampling in HQPlayer or any other player capable of that. Many of us apply PCM to DSD conversion for every PCM content, not only for 44.1k based one. For N x 48k based PCM content we may configure HQPlayer if we want it to output for example DSD256 at 256 x 44.1k rate or at 48k x 256 rate. Why at all to use those 48k based rates for DSD? For me, from two reasons:

1) Integer upsampling ratio (like 48k -> 48 x 256) is less CPU demanding than when the ratio is non integer (like 48 -> 44.1 x 256 ). With non integer ratios you easier get dropouts because of insufficient CPU power.

2) 48 x 256 is more than 44.1 x 256. It may sound a bit better. From the same reason why DSD512 may sound better than DSD256. Of course, these things depend on concrete DAC implementation so it need not be a general rule.

 

Not all DACs support 48k based DSD rates. If that support is missing, it may be harder to upsample 48k based PCM content to DSD than 44.1k based content - because of that non integer ratio. People owning such a DAC may prefer 44.1k based PCM content over 48k based one because it may be easier playable in HQPlayer when PCM to DSD conversion is used.

[Miska]

1 hour ago, davide256 said:

Still unclear. What does 1 bit multi-megabit encoded DSD recordings have to do with N x CD rate? I see the 48K option on HQPlayer output choices but have

never really understood why its there.

 

That option is particularly useful with R26, because it seems to allow DSD playback to work without background noise issues... So instead of using DSD at 44.1k base rates, better use DSD at 48k base rates with that particular DAC.