DSD Performance Criteria

[Sokel]

It certainly contains more samples that a 192kHz signal... The Fc of the 5.6MHz signal will just define the noise-shaping and filter for playback. A 1-bit 5.6MHz recording is just that... whether or not you believe it captures contextually relevant contextual audio energy is left to subjective opinion.

Whatever samples DSD or PCM carries vanish after processing and filtering and the output just presents a electrical signal so the pre or amp can attenuate, amplify, etc.
The digital characteristics of the incoming signal are just not there at the output, that's the purpose of the DAC.

 

[voodooless]

It certainly contains more samples that a 192kHz signal..

So what? More samples is not more better…

For that matter, that ESS DAC actually oversamples to an even higher rate than the puny 5 MHz DSD.

I don't see this DAC as DSD capable:

@Sokel , called it..

 

[Sokel]

I don't see this DAC as DSD capable:

Please, PLEASE, learn to read the chart.

Look at the top for the output device. It says (ASIO) Khadas TONE. That's the DAC, this one:

Review and Measurements of WesionTEK Khadas Tone Board DAC

This is a review and detailed measurements of Khadas Tone Board DAC from WesionTEK. It is on kind loan from the company. The retail price on Amazon is USD $99 including free shipping. As the name indicates, this is a bare board DAC: Don't confuse this with Raspberry Pi boards though. This...

www.audiosciencereview.com

Edit: O quote my previous post about the chain:

Once again, this is a real chain, PC>MTA>DAC>ADC>MTA>PC.

Where:
MTA= Multitone Analyzer, the software
DAC=Khadas tone board (ESS9038q2m)
ADC=E-MU 0204

Hope this will clears it out.

 

[Audicron]

Please, PLEASE, learn to read the chart.

Look at the top for the output device. It says (ASIO) Khadas TONE. That's the DAC, this one:

Review and Measurements of WesionTEK Khadas Tone Board DAC

This is a review and detailed measurements of Khadas Tone Board DAC from WesionTEK. It is on kind loan from the company. The retail price on Amazon is USD $99 including free shipping. As the name indicates, this is a bare board DAC: Don't confuse this with Raspberry Pi boards though. This...

www.audiosciencereview.com

Edit: O quote my previous post about the chain:



Where:
MTA= Multitone Analyzer, the software
DAC=Khadas tone board (ESS9038q2m)
ADC=E-MU 0204

Hope this will clears it out.

Okay the Khadas board is using the ESS chipset that decimates all DSD files to PCM for Hyperstream processing...

For that matter, that ESS DAC actually oversamples to an even higher rate than the puny 5 MHz DSD.

After it decimates the 1-bit DSD signal to PCM and subsequent of the Hyperstream processing the signal is truncated to multi-bit DSD-Wide.

 

[voodooless]

Please, PLEASE, learn to read the chart.

It’s not about the chart, it’s about the DACs architecture: it always does conversion to wide bit, oversample even more, and then apply it’s own multibit delta/sigma modulators. In contrast many AKM chips can bypass all that and just output the DSD as-is to the output.

 

[Sokel]

It’s not about the chart, it’s about the DACs architecture: it always dus conversion to wide but, oversample even more, and then apply it’s own multibit delta/sigma modulators. In contrast many AKM chips can bypass all that and just output the DSD as-is to the output.

There's also the easiest of all solution, like the diyaudio No-DAC
No processing at all whatsoever, just a filter.

 

[voodooless]

After it decimates the 1-bit DSD signal to PCM and subsequent of the Hyperstream processing the signal is truncated to multi-bit DSD-Wide.

That is incorrect. The processing path of the ESS DACs is at least 32-bits until it hits the modulators. There is no truncation happening, only decimation.

And actually this preserves very well the original spectrum of the original DSD modulation, as can be seen by the high rate plot that @Sokel showed were the start of the noise shaping is visibly in a area where the ESS chip would normally not show any.

 

[Sokel]

That is incorrect. The processing path of the ESS DACs is at least 32-bits until it hits the modulators. There is no truncation happening, only decimation.

And actually this preserves very well the original spectrum of the original DSD modulation, as can be seen by the high rate plot that @Sokel showed were the start of the noise shaping is visibly in a area where the ESS chip would normally not show any.

Yep.
And for further clarity, that's the original DSD incoming signal on its own:

 

[MaxwellsEq]

If the resolution sample-density of the ADC is 192kHz, it is decimating the 5.6MHz sample-density represented in the analog output signal of the DAC, I don't see this as being an accurate measurement of the 5.6MHz signal output of the DAC... what am I missing

You are missing an understanding of single-bit architectures and noise shaping.

When you wrote this question, did it not occur to you that it can not be relevant at human bounded audio frequencies?

 

[MaxwellsEq]

Of course the sarcasm.... I don't have the means like those available to ASR reviewers, otherwise I would not have posed the contextual question....

I was not being sarcastic.

Instead of criticizing the people on ASR who invest time and money on publishing measurements, why don't you make some measurements yourself? That way you will learn but also be able to add weight to your arguments.

 

[RandomEar]

I think this discussion is getting away from the topic of the thread which is "why no testing" and not "why DSD not better than PCM". I see three reasons for the lack of DSD in reviews on ASR:

• Time/effort: Amir has a long queue of equipment to be tested and very limited time. Therefore he focuses on the most important measurements and the most mainstream features. He also doesn't always test the performance of an EQ or tone controls if present, simply to save time.
• Interest: DSD is a rather niche format in the grand scheme of things. It's limited to SACD and some streaming sources, whereas PCM can be found pretty much anywhere (CDs, DVDs, steaming, YouTube, and so on). So it makes sense to test primarily with PCM, because pretty much all users will use that format for the vast majority of their listening time. PCM simply is the standard.
• Compatibility of the Audio Precision software: I checked the manual for the APx555 and didn't see a single mention of the term "DSD". I think the AP software might not even be compatible with that format, which would limit the testing options and make it much more cumbersome and time consuming to work with.

Maybe I'm wrong about the lack of software support - users of AP analyzers are welcome to correct me on this. The rest of the points still stands, though.

 

[Sokel]

I think this discussion is getting away from the topic of the thread which is "why no testing" and not "why DSD not better than PCM". I see three reasons for the lack of DSD in reviews on ASR:

• Time/effort: Amir has a long queue of equipment to be tested and very limited time. Therefore he focuses on the most important measurements and the most mainstream features. He also doesn't always test the performance of an EQ or tone controls if present, simply to save time.
• Interest: DSD is a rather niche format in the grand scheme of things. It's limited to SACD and some streaming sources, whereas PCM can be found pretty much anywhere (CDs, DVDs, steaming, YouTube, and so on). So it makes sense to test primarily with PCM, because pretty much all users will use that format for the vast majority of their listening time. PCM simply is the standard.
• Compatibility of the Audio Precision software: I checked the manual for the APx555 and didn't see a single mention of the term "DSD". I think the AP software might not even be compatible with that format, which would limit the testing options and make it much more cumbersome and time consuming to work with.

Maybe I'm wrong about the lack of software support - users of AP analyzers are welcome to correct me on this. The rest of the points still stands, though.

It's not only the lack of software by the AP.
Its main strength is the amazing signal generator, and that should be different for DSD.

Edit: Scrap the above, AP offers the APx PDM Module which i perfectly capable of generating DSD up to DSD256.

PDM Module for APx B Series Audio Analyzers

PDM module APx series audio analyzers provides direct connectivity for devices with a PDM output. Direct I/O, modulation for precise PDM device analysis.

www.ap.com

 

[MatrixS2000]

A 5.6 MHz 1-bit stream is not 5.6 MHz of useful audio detail. It’s mostly a high-rate noise-shaped encoding.

Exactly. Here is a screen capture of what that boat load of noise looks like:

 

[Audicron]

The processing path of the ESS DACs is at least 32-bits until it hits the modulators

The DSP is done at 32-bit... so, the 1-bit 5.6MHz PDM signal is decimated and quantized to 32-bit LPCM and running at whatever the working sample-rate is, where it is handed-off to the Hyperstream processing, and converted to multi-bit DSD-Wide... (See the attached functional block-diagram of the ESS9038q2m) ... It cannot possibly represent the sample-density of the 1-bit 5.6MHz PDM signal, it creates a decimated iteration of the PDM signal so to apply DSP....

 

[Scytales]

That is incorrect. The processing path of the ESS DACs is at least 32-bits until it hits the modulators. There is no truncation happening, only decimation.

Audicron: The DSP is done at 32-bit... so, the 1-bit 5.6MHz PDM signal is decimated ....

To my mind, it is very doubtful that ESS DACs perform decimation (=lowering of the sample rate) of DSD, because, according to the relevant DAC datasheets [1], the specifications of the built IIR low-pass filter that remove part of DSD HF noise show a passband + transition band up to the original Nyquist frequency of DSD (at 64x rate or 2.8224 MHz), i.e. 1.4112 MHz:

Recent ESS Sabre DACs accept DSD at much higher sample rates that 64x (2.8224 MHz), but, per the relevant datasheet [2] the specifications of the filters are just scaled depending on the DSD data input clock. That means that for each filter, the -3 dB corner frequency and the top end of the frequency response is pushed two times higher in frequency every time the DSD input sample rate is doubled. In any case, to my mind, that imply that no decimation filter is implemented in the DAC chips.

[1] ESS ES9008 DAC chip datasheet (Rev. 1.3 of February 15, 2015) and ES9012 and ES9018 DAC datasheets (February 11, 2010).
[2] ES9010K2M datasheet, April 27, 2021, and ES9038PRO datasheet, February 8, 2021.

 

[Audicron]

You are missing an understanding of single-bit architectures and noise shaping.

When you wrote this question, did it not occur to you that it can not be relevant at human bounded audio frequencies?

The sample-density of the 1-bit 5.6MHz PCM signal in the audible spectrum (subjective) is greater than any relevant LPCM sample-density... As far as FIR filter performance relevance? This is something to measure in the context of the DSD signal handled by the DAC architecture, as this is a typical functional operation available on a DSD-centric DAC.. For example, attached here is the functional cut-off frequency FIR filter of the ROHM BD34352EKV chipset employed in the S.M.S.L D200 DAC platform.... The contextual analysis of the DSD performance will be dependent on whether or not the file is native 1-bit PDM as recorded, versus a LPCM file being modulated to a 1-bit PDM target sample-rate.

 

[Audicron]

To my mind, it is very doubtful that ESS DACs perform decimation (=lowering of the sample rate) of DSD, because, according to the relevant DAC datasheets [1], the specifications of the built IIR low-pass filter that remove part of DSD HF noise show a passband + transition band up the original Nyquist frequency of DSD (at 64x rate or 2.8224 MHz), i.e. 1.4112 MHz:

Recent ESS Sabre DACs accept DSD at much higher sample rates that 64x (2.8224 MHz), but, per the relevant datasheet [2] the specifications of the filters are just scaled depending on the DSD data input clock. That means that for each filter, the -3 dB corner frequency and the top end of the frequency response is pushed two times higher in frequency every time the DSD input sample rate is doubled. In any case, to my mind, that imply that no decimation filter is implemented in the DAC chips.

[1] ESS ES9008 DAC chip datasheet (Rev. 1.3 of February 15, 2015) and ES9012 and ES9018 DAC datasheets (February 11, 2010).
[2] ES9010K2M datasheet, April 27, 2021, and ES9038PRO datasheet, February 8, 2021.

The multi-bit DSD-Wide signal is output at the Hyperstream working frequency...

 

[voodooless]

To my mind, it is very doubtful that ESS DACs perform decimation (=lowering of the sample rate) of DSD

Yes, correct! This is my bad! Sample rate is never lowered. Only bit depth reduction is done by modulation.

 

[voodooless]

The sample-density of the 1-bit 5.6MHz PCM signal in the audible spectrum (subjective) is greater than any relevant LPCM sample-density

If you understand the sampling theorem, you understand that this doesn’t matter.

 

[Audicron]

If you understand the sampling theorem, you understand that this doesn’t matter.

Then why have a sample-rate greater than 2Hz?