PCM vs DSD

[Sokel]

Have a look at this too. DSD256 not only resulted in higher noise floor, but also higher distortion than PCM.
Khadas tone board is known to have hump, but there are also many other ES9038 DACs with no hump. So deliberately use a DAC with hump for comparison cannot draw very useful conclusion.

REVIEW: E1DA #9038D6K Dongle DAC / Headphone Amp (Part II): Headphone Amp Power, DSD, Subjective Impressions, AMPT & Perfectionistic Tweaking!

A blog for audiophiles about more objective topics. Measurements of audio gear. Reasonable, realistic, no snakeoil assessment of sound, and equipment.

archimago.blogspot.com

Specifically, Archimago converted a DSD512 signal to PCM then sent to the DAC, the figures immediately improved a lot. Pay attention to the Realtime DSD -> PCM annotation.

Also clear sign of higher distortion in DSD256 playback, similar to Archimago's previous tests on Oppo UDP205 and Topping D90SE.

I have seen this results.
I don't see higher distortion,only noise,H2 and H3 are on the -125 to -135db ballpark for all of them.

Problem is I can't see other that 1Khz measurement or an IMD one other than Multitone which strangely is somehow forgiving.
Hopefully not all ESS based DACs are like this but it's not only about ones who have the hump.

Take a look for example this hump-free one:



It's IMD distortion is still worst than the above E-MU at -30db for example.
Yes,DSD256 and DSD512 are hard to do right,I'm about to measure and older AKM based one with nice clocks,etc.
I'm curious too!

Edit:distortion figures

 

[Sokel]

MQA looks to be on the ropes, but DSD seems like it is not going to take a deserved trip to the dustbin of history.

Nope

 

[bennetng]

For example, Creative AE-5 (ES9016) is fine while the ironically more expensive AE-9 with ES9038 has the hump.

soundcard help

I got the soundblaster ae 5 soundcard to power my Sennheiser hd 660s headphones. I like it, however, I learned that the soundblaster ae 9 is for audiophiles? should i pay the extra 200$ to get that to drive my headphones and will the dac used in the ae9 be better than the ae 5 and make a major...

www.audiosciencereview.com

I have seen this results.
I don't see higher distortion,only noise,H2 and H3 are on the -125 to -135db ballpark for all of them.

The problem is Archimago's DSD plots have much more irregular spikes than PCM, so even when the height of individual spike is similar, more spikes result in higher distortion, as displayed by the total THD+N value.

View attachment 314161

It's IMD distortion is still worst than the above E-MU at -30db for example.

I am talking about relative performance of PCM vs DSD when using the same product, the E-MU does not support DSD (the chip itself may support DSD, but the whole product does not), so there is no relative comparison.

 

[Sokel]

I am talking about relative performance of PCM vs DSD when using the same product, the E-MU does not support DSD (the chip itself may support DSD, but the whole product does not), so there is no relative comparison.

Of course there isn't,only the fact that they are divided by 20-25 years of supposed evolution should make comparison a joke.
DSD of AK4396 is comparable to it's PCM performance,I've seen comparisons back in the day when they called it "miracle" DAC.

What it's really interesting is that what they where seeking back in the day was not the single 1Khz at 0db nice measurement but an overall decent performance.
I have measured E-MU for example in any possible way and the results are always decent,THN+N across all audible spectrum,IMD,all kind of sweeps,etc.
For example lowering the input level in it distortion goes down too,at -15db is vanishingly low and that specific level is critical as I think of it as music's main bulk is usually there.

Of course all the above are only for fun,probably inaudible with music,but evolution should have carried the whole industry with it and not only a few bright examples with good overall performance,specially after 25 years.

 

[bennetng]

Older chips like CS4398 in EMU1212m are also simpler too which makes DSD more relevant. I guess, even a CS43198/43131 dongle these days can have more consistent PCM vs DSD performance regardless of having "direct" path or not.

EMU interfaces are good of course, so good that Creative discontinued the brand.


[edit] There are tests using -60dBFS tone and IMD+N dual tone sweep too.

 

[Sokel]

[edit] There are tests using -60dBFS tone and IMD+N dual tone sweep too.

IMD sweep is across frequency,not level,that's where things are interesting as they approach some (said by some) audible levels of IMD distortion with real use case (say,reduced level by 10db by user and another 13-20db down by the recording) .
(still waiting for someone to saw me papers about the established thresholds about it)

 

[bennetng]

RMAA uses -3dB for THD, IMD and IMD sweep tests. IMD and IMD sweep tests have different test conditions, one is 60Hz/7kHz pair, another one is two sweeping tones separated by a distance of 1kHz.

Now, if a 0dB, -3dB sine, -60dB sine, -3dB IMD, a -3dB IMD sweep and a 32-tone test, all of them show that PCM works better than DSD on this product, but only a mid level IMD test has different result, it still does not mean DSD is more suitable for this product because all other tests show that PCM is better. Not to mention DSD64 and 128 have much more ultrasonic noise than PCM, and strange rise of noise and distortion within 20kHz when using DSD256.

 

[Sokel]

RMAA uses -3dB for THD, IMD and IMD sweep tests. IMD and IMD sweep tests have different test conditions, one is 60Hz/7kHz pair, another one is two sweeping tones separated by a distance of 1kHz.

Now, if a 0dB, -3dB sine, -60dB sine, -3dB IMD, a -3dB IMD sweep and a 32-tone test, all of them show that PCM works better than DSD on this product, but only a mid level IMD test has different result, it still does not mean DSD is more suitable for this product because all other tests show that PCM is better. Not to mention DSD64 and 128 have much more ultrasonic noise than PCM, and strange rise of noise and distortion within 20kHz when using DSD256.

Of course it does do better,no question about it.
At my first post I already say that this test is product specific.

Is more of an urge for people to test stuff,individually,in real life conditions as much as possible and if possible whole rigs end to end (I do that too,I have measurements on MTA thread) with every material available.

And another.It's close to impossible to find decent DSD test files,even Archimago's ones are not perfect and back in the day there was a discussion about it.
On top of that a scheme to apply level is also needed as DSD cannot be handled the same way with PCM,reducing level generates noise and artifacts.,not sure how many use it properly.
At least MTA takes care of that so it's a good chance for people with nice measurement stuff to measure properly.
That is this thread more about,despite it's click-bait title.

 

[bennetng]

And another.It's close to impossible to find decent DSD test files,even Archimago's ones are not perfect and back in the day there was a discussion about it.
On top of that a scheme to apply level is also needed as DSD cannot be handled the same way with PCM,reducing level generates noise and artifacts.,not sure how many use it properly.

Yes, I asked him about this. Why, because the Topping D90SE was all the hype at that time but I suspected there could be issues with DSD, so I interrupted him before he was about to give a final verdict.

MEASUREMENTS: A look at DSD and using SoX-DSD as a standard for test signal conversion. The 1/10-decade Multitone 32 test. And retro-pop remixes... ;-)

A blog for audiophiles about more objective topics. Measurements of audio gear. Reasonable, realistic, no snakeoil assessment of sound, and equipment.

archimago.blogspot.com

Turns out the distortion in Archimago's DSD test results were caused by some specific DACs instead of the DSD encoders because the distortion does not happen with other DACs he tested, including AKM and TI.

Revisiting the TEAC UD-501 DAC (2013): THD(+N), DSD output, Sweeps, Jitter, 1/10 Decade Multitone 32. (And on SSD transition, Corsair MX500 SSD, and Hi-Fi+ closes comments...)

A blog for audiophiles about more objective topics. Measurements of audio gear. Reasonable, realistic, no snakeoil assessment of sound, and equipment.

archimago.blogspot.com

In the case of test signals, they can be directly generated at a specific level before converting to DSD. Now you mentioned it, for typical use case of music playback, a DAC's direct DSD path does not support volume control, which is another issue inherent to the DSD format, which is not nice. If one uses an indirect DSD path with the DAC's built -in volume control, it is a DAC specific feature and should be tested too.

 

[Sokel]

In the case of test signals, they can be directly generated at a specific level before converting to DSD. Now you mentioned it, for typical use case of music playback, a DAC's direct DSD path does not support volume control, which is another issue inherent to the DSD format, which is not nice. If one uses an indirect DSD path with the DAC's built -in volume control, it is a DAC specific feature and should be tested too.

Here's the strange thing now:
Khadas Tone doesn't have volume control on it's own but it happens I have Ian Canada ESS controller so I can control it from there.
Strangely it seems to control nicely DSD too,normally we have to put the DAC level to 0db to play DSD properly,but see here

(levels adjusted from DACs side,analyzer's gain is the same,all test sinals are DSD128)




-2db



-10db



-30db (the...danger zone)



-60db

I don't know how it does it but it's not bad at all (notice that the units are dbr so plot doesn't seem as nice as it would in dbFS but seems more proper to me that way)

Edit,here's one at -60db level in dbFS units too for easy comparison with Archimago's one

 

[Snoopy]

Why even do all these tests with ESS chips instead of chips that actually support bypassing the SDM? And from there use HQPlayer to choose a different modulator and upsample to a DSD rate that works best for the DAC chip.

I'm not claiming any "Vails lifted" difference and night and day improvements but otherwise you are just sending the pcm or DSD signal through the same SDM modulator in the ESS chip.

 

[Sokel]

Why even do all these tests with ESS chips instead of chips that actually support bypassing the SDM? And from there use HQPlayer to choose a different modulator and upsample to a DSD rate that works best for the DAC chip.

I'm not claiming any "Vails lifted" difference and night and day improvements but otherwise you are just sending the pcm or DSD signal through the same SDM modulator in the ESS chip.

Don't you think I would know that?
The test here is about how each device can handle things differently.
There's a gazillion of ways to upsample to DSD and after that you may not even use a chip at all,diyaudio has some fun projects about it but that's not the point here.

 

[Music1969]

I don't know how it does it but it's not bad at all (notice that the units are dbr so plot doesn't seem as nice as it would in dbFS but seems more proper to me that way)

With ESS, with DSD input you get 1-bit DSD converted to 32-bits for digital volume control.

But the original sample rate is maintained, so you bypass all ESS oversampling DSP.

So it stays at the DSD MHz original sample rate.

 

[Music1969]

Why even do all these tests with ESS chips instead of chips that actually support bypassing the SDM?

I think its still interesting to see DACs that still measure better with DSD input , like the one here, regardless of whether you can bypass the modulator.

And some DACs do not measure better with DSD input

But we won't know unless people measure and share

 

[bennetng]

-30db (the...danger zone)

Interesting, this one indeed has highest distortion, same as the PCM hump position in Khadas measurements.

 

[Sokel]

Interesting, this one indeed has highest distortion, same as the PCM hump position in Khadas measurements.

That's another strange,yes,cause if I apply the same gain from the analyzer's side while having DACs volume control at normal here is what we get;

Totally different distortion structure (and some 20db better THD but at the expense of 3db worst noise)

 

[bennetng]

That's another strange,yes,cause if I apply the same gain from the analyzer's side while having DACs volume control at normal here is what we get;

View attachment 314353

Totally different distortion structure (and some 20db better THD but at the expense of 3db worst noise)

What if the signal is generated at -10dB with an additional -20dB volume applied using the ESS volume control?

 

[Sokel]

What if the signal is generated at -10dB with an additional -20dB volume applied using the ESS volume control?

Low side of the middle,big H3:

It seems that the best way is to keep it high on DACs side.

 

[Sokel]

It also seems to follow PCM's pattern at the same settings:

 

[bennetng]

Your DSD128 plot with -60dB ESS volume control looks close enough to Archimago's DSD128 DNR plot which did not use the ESS volume control, but Archimago's DSD256 plots in all tests (THD/IMD/DNR) are pretty pathetic.

Anyway, according to your tests, using the ESS volume control makes DSD measure like PCM.