Does DSD sound better than PCM?

[JJB70]

I'm completely lost now, brain overload. I'm listening to my CDs of the Beethoven piano concertos recorded by Gilels and Szell and they are magnificent. At the risk of sounding like a rampant subjectivist that's what really matters to me.

 

[March Audio]

OK, so now phase response doesn't matter either. In fact, nothing seems to matter.

I didnt say that. I asked you to show some data regarding the audibility of phase error.

 

[svart-hvitt]

I'm completely lost now, brain overload. I'm listening to my CDs of the Beethoven piano concertos recorded by Gilels and Szell and they are magnificent. At the risk of sounding like a rampant subjectivist that's what really matters to me.

I guess this is what happens when you call another engineer’s technicals «faulty» without taking the time to back up the accusation with evidence.

 

[March Audio]

I'm completely lost now, brain overload. I'm listening to my CDs of the Beethoven piano concertos recorded by Gilels and Szell and they are magnificent. At the risk of sounding like a rampant subjectivist that's what really matters to me.

It is indeed.

There are several issues discussed here.

one is the problem of alias signals getting back into the audio band. A problem at the start of the AA filter. In simple terms the actual audio signals (real music not test tones) in this region are not high enough to cause problems in the audible band. The alias ends up in the noise floor.

Second is the fact that dacs put out ultrasonic noise which has the potential to cause intermodulation in a subsequent amplifier which can end up in the audio band. Again with real music (not test tones) the noise is not significant - unless your amplifier has no input filter in which case it wil probably suffer problems from other sources such as RF.

 

[March Audio]

I guess this is what happens when you call another engineer’s technicals «faulty» without taking the time to back up the accusation with evidence.

I have done so.

I said real music energy (not test tones) in the transition band isnt high. I demonstrated so. The level its at puts in band aliases in the noise floor.

Archimagos data which I can replicate shows ultrasonic noise with real music (not test tones) is not an issue.

 

[svart-hvitt]

I have done so.

I said real music energy (not test tones) in the transition band isnt high. I demonstrated so. The level its at puts aliases in the noise floor.

Archimagos data which I can replicate shows ultrasonic noise with real music (not test tones) is not an issue.

Why don’t you make 16/44 DACs? All else is waste, isn’t it?

 

[Miska]

So even if you look at the highest signal of -36dB, the alias are buried in the noise floor. Where is the problem?

That highest signal consists partially of alias... With halfband filter, rejection at Nyquist is pretty much non-existent. So the band between 20 kHz and 22.05 kHz is mixture of base-band signal and alias of frequencies above it (22.05 - 24.1 kHz). In my castanets example -20 dB level hits this 22.05 - 24 .1 kHz band.

Broadcast radio starts at 148kHz

Hmmh, what radio starts there?

But yeah, there are radio transmitters even in ELF range for communicating with submerged submarines:
https://en.wikipedia.org/wiki/Extremely_low_frequency

 

[Miska]

I said real music energy (not test tones) in the transition band isnt high. I demonstrated so. The level its at puts in band aliases in the noise floor.

You didn't do any exhaustive search even, you quickly picked something you found from your own collection. That proves really nothing.

And that doesn't have anything to do with in band aliases.

I didnt say that. I asked you to show some data regarding the audibility of phase error.

And I'll ask you to show some data about non-audibility. I assume perfect audibility unless proven in statistically realiable way otherwise. And it must be statistically reliable against entire population on earth.

@Blumlein 88 data showed a reduction of -85dB before you hit the audible band (20kHz).

His data showed one case for one ADC and then you extrapolate from that all-encompassing generalization that it covers all the cases. And you think that audible band starts exactly at 20000 Hz, so 20500 Hz is already totally inaudible.

 

[DonH56]

This thread is a mess. I miss Thomas.

Without touching the rest, there are a number of radio bands in the LF region (ELF is much lower), many used for navigation systems (e.g. LORAN-C, differential GPS, etc.) plus many (perhaps most?) radio clocks use that band. There are a number of AM stations in the 150~300 kHz range or so in Europe and Asia.

At this point the rest of it is just throwing barbs back and forth.

 

[Sal1950]

At this point the rest of it is just throwing barbs back and forth.

Agreed, Going back to the OP,
If DSD sounds better (or different) than PCM, it's because the source master was different or the DAC is poorly designed.

 

[Miska]

Some upper limit of hearing figures for a very small sample set of 32 people:
https://asa.scitation.org/doi/10.1121/1.2761883

Now for fun you can factor in potential boost of about 20 dB from metal dome tweeter resonance somewhere between 20 kHz and 25 kHz.


But anyway, I perplexed to see the gap between subjectivist who say "you cannot measure what I hear" and objectivists saying "you cannot hear what I measure". Now I understand better what kind gap there really is!


And we are still talking about some of the more basic stuff. We haven't even touched modulator properties and performance yet. I guess most here have not heard ESS presentation about their modulator design and subjective listening tests about the sound.

Really, huh!

 

[solderdude]

Thanks for answering some of my questions and avoiding others

I measure also higher, but there's usually nothing because the analog reconstruction filter cuts out. Typical analog filter is 2nd order with fc=100kHz so it has rolled off dirty output by 5 MHz already.
Taking one of the dirties devices I have (in that respect), Focusrite Forte. Almost all the the junk is in first 10 MHz band.

yes for differential mode, no for common mode.

Without starting to write my own story here now, here's one example from other people:
https://www.diyaudio.com/forums/class-d/321632-hypex-ncore-nc400-input-anti-alias-filter.html

Class D isn't what most people use and is very obvious that it needs an input filter, likewise ADCs.

Q1: What I was wondering about is do you have any indication how big this issue is and how relevant it is to music reproduction ?
That is with normal (headphone) amps on the market.
Q2: Examples of analog amps that are known to exhibit this behavior ?

You mention:

So any intermodulation products they may create in later stages (like power amp) are also fully correlated. So if you hear it sounds like artificial hardness.

Q3: Is this your experience ?
Q4: Can you clearly correlate this to ultrasonics being present ?
Q5: What amplifier / transducers are you using to listen for this (so I can avoid that one) as you clearly state they MAY create it I assume you own one to test.

To these questions you answered:

One of the things you can simulate...

Really ?
You can simulate the sound of artificial hardness and perception ?
You can simulate amplifiers and transducers and how they are perceived by owners ?

Why not multi-tone? Images are correlated multi-tone because they reflect music spectra around multiples of sampling rate.

Because it would be easier to spot, with multitones it becomes a mess where it is difficult to see if something is caused by the HF.

Where does this -60 dB come from? It is not necessarily even that much down.

Talking about 1MHz which most likely is down to at least -60dB but -50dB would also be fine.

If it can be measured - it matters. If it cannot be measured, it still matters, but measurement resolution needs to be improved...

Strange thesis. If something measured is so far below audibility why would it matter for audio ?

I don't like to make positive assumptions about amp behavior, rather negative worst-case scenarios. So I aim for clean DAC output that represents proper reconstruction instead of "half-assed" one.

That's sound thinking.
In how many cases (not class-D) would you say > 50kHz is a real life problem ?

Yes, my own listening tests. Not blind. When I have two algorithms at hand, I have no reason to prefer A over B. I have my favorite algorithms based on listening. Some other people prefer other ones, that's completely OK for me. I offer options to choose from.

Options are always good to have. Do such in my designs as well.
Understood on the listening test .. sighted it is.

No you may not.

LOL, then I won't disclose my age and hearing limits either...

I don't know why what I hear matters to anybody else except me in first place. I usually don't talk much about my listening experiences unless specifically asked. Sometimes if people ask what are my favorite settings, then I tell my current favorite.

Oh .. I agree, was just wondering if your endeavors originated from things you heard and wanted to correlate or that you measured something and went from there.

Now what I don't understand is why this particular thing has caused this kind of shit storm towards me. I was just sharing my point of view about this detail based on spending 20+ years designing digital filters.

Much appreciated.. I don't dwell on other fora (except my own) and thought it was an interesting opportunity to pick your brain.

Don't forget .. there are other folks with more than 20 years of experience in audio here as well.
I think you very well realize that not all 'experts' have similar thoughts.
It's O.K. to ask and answer questions.
Although it does appear to be the Spanish inquisition here now and then and folks can be quite opinionated, as you are as well it seems.

 

[Blumlein 88]

Arse!

I knew some clever dick would spoil things Im aam revising my post to 0.5 ohms.

Which model is it? Are you sure its 0.75 ohms?

Thats not actually a speaker, thats a link of wire. That will cause a world of pain for amps.

It an Aura. Which is pretty much the M3 in better build quality. And yes depending on the settings on the interface it is around .75 ohm at the upper frequencies. Here is a plot from long ago of the A1, but it is pretty much the same as mine in impedance.

 

[Miska]

Class D isn't what most people use and is very obvious that it needs an input filter, like wise ADCs.

It is quite common these days, Hypex modules, B&O modules (BelCanto, etc), NAD's amps, etc. Active speakers...

This is going just the way I predicted, sigh.

You can simulate the sound of artificial hardness and perception ?

I can simulate different kinds of errors, that's just DSP as well.

Strange thesis. If something measured is so far below audibility why would it matter for audio ?

I've said many times before, I don't draw any "below audibility" lines. It is more straightforward to handle things that way. I'll just leave it there.

Don't forget .. there are other folks with more than 20 years of experience in audio here as well.
I think you very well realize that not all 'experts' have similar thoughts.

Sure. Just sometimes surprised about the attitude that nothing matters, even if it is objectively measurable and fixable.

And maybe a little bit surprised that I see so few measurement results from other people than Amir and few others.

 

[Blumlein 88]

Sure, I'm a bit confused about the response though.

I first explained how I think particular choice of ESS digital filter in his DAC fixes certain common problem from source material used in the test tracks created by @Blumlein 88 . I'm kind of surprised of the response. Based on the response he could of course also go back to his chip vendor (ESS) and complain that they in first place even offer any filter options at all because none of that matter. Instead of complaining about that to me... I have my opinion about the matter and he is not going to change that anyway.

But this thread has gone massively off-topic at least. The only relevance to OT is that with DSD ADC you don't need digital decimation filters at all so this issue being discussed now doesn't exist.

This goes back somewhat to the test listening files. I didn't reveal it because it is a dead tell in a sound editor. The first generation has already cut off those transition band frequencies. The test file in this respect sounds like listening to the March DAC. Which is using filtering more or less as it should be. You then say the DAC has fixed the original file, but it would do this simply by listening to that DAC. If someone is listening to my files with another DAC that leaves the transition band there or aliases it further then it is a worse signal and would be a way to differentiate by sound the reference vs the test file. With the assumption such things do make a difference in sound. I don't see how this is a complaint. You seem to have made an issue of what some DACs do because the one I used doesn't do that. I still feel like I'm misunderstanding you somehow.

 

[Blumlein 88]

Some upper limit of hearing figures for a very small sample set of 32 people:
https://asa.scitation.org/doi/10.1121/1.2761883

Now for fun you can factor in potential boost of about 20 dB from metal dome tweeter resonance somewhere between 20 kHz and 25 kHz.


But anyway, I perplexed to see the gap between subjectivist who say "you cannot measure what I hear" and objectivists saying "you cannot hear what I measure". Now I understand better what kind gap there really is!


And we are still talking about some of the more basic stuff. We haven't even touched modulator properties and performance yet. I guess most here have not heard ESS presentation about their modulator design and subjective listening tests about the sound.

Really, huh!

Nope haven't heard of ESS doing this. Can you point us to where it is?

 

[solderdude]

Yes, class-D is going to be even bigger for obvious reasons.
All it takes for the manufacturers is to use a good input filter and there should not be much of a problem.

Perhaps exposing behavior of different amps (as Amir seems to be attempting) is more constructive as one can avoid crappy gear.
Maybe you could persuade Amir to measure the output of DACs well into the MHZ range as well as amplifiers.

I don't think there is the attitude here that nothing matters.
I do believe there are quite a few knowledgeable folks here that have come to understand (through experience) what matters to them and in general in audio and what might not be.
They DO draw a line and defend it.
It is a choice not to draw one though.

When things are fixable I am all for it (in fact is what most of my 'work' is about).
Some things don't need fixing though and work fine as they are.
They may not measure that great though.

Not many folks have AP gear.
There are other measurement folks here as well.
Some have AP too or other equipment.
I don't.

 

[Miska]

The first generation has already cut off those transition band frequencies. The test file in this respect sounds like listening to the March DAC. Which is using filtering more or less as it should be. You then say the DAC has fixed the original file, but it would do this simply by listening to that DAC.

Yes, that was my point.

With the assumption such things do make a difference in sound. I don't see how this is a complaint. You seem to have made an issue of what some DACs do because the one I used doesn't do that. I still feel like I'm misunderstanding you somehow.

No, for some people it seems to be an issue that I said the DAC in question cleaned that part due to choice of the particular ESS filter. I didn't complain anything, it was an observation from the file and DAC data. First I was wondering which side it was, DA or AD until you told the DAC which pointed to the roll-off curve which immediately described it was DA side. Then I compared that to plots in ESS datasheet to figure out which of the filters it likely was.

Some/many other DACs don't do that, so listening to the DA+AD loop file may actually sound relatively better in some respects through such DACs even though it has gone through the extra conversion loop and degraded in other ways.

Maybe people got upset because they know my software gives choice of oversampling filters that do similar cleanup. Or something, I just don't know.

 

[Blumlein 88]

Yes, that was my point.



No, for some people it seems to be an issue that I said the DAC in question cleaned that part due to choice of the particular ESS filter. I didn't complain anything, it was an observation from the file and DAC data. First I was wondering which side it was, DA or AD until you told the DAC which pointed to the roll-off curve which immediately described it was DA side. Then I compared that to plots in ESS datasheet to figure out which of the filters it likely was.

Some/many other DACs don't do that, so listening to the DA+AD loop file may actually sound relatively better in some respects through such DACs even though it has gone through the extra conversion loop and degraded in other ways.

Maybe people got upset because they know my software gives choice of oversampling filters that do similar cleanup. Or something, I just don't know.

Well, I'm not upset with you. I respect you and sometimes learn from you. I was having some problem understanding what you were ultimately getting at in your posts. I can be hard headed even when wrong of course. I just hear so much BS from people with complaints about nothing in the end. You are not among those people even if I disagree with where you draw a line or don't.

The 8th generation thing was an attempt to show what that sounds like. I'm surprised it is as good as it is. And other than the filtering I don't think anything else about it is improved. Distortion, response, noise, jitter are all worse. Yet it doesn't end up sounding like garbage the way some think it would. I'm not for adding distortion, noise, aliasing, imaging, jitter or any of that stuff. But some of those are worried over when they don't rise to the level to make a difference to most people (or probably any people). It is a probability about some aspects as you never get to test everyone in the world for a result.

 

[Miska]

All it takes for the manufacturers is to use a good input filter and there should not be much of a problem.

Now the trick is design of such filter... My point was to clean up the DAC output so much that you don't need to rely on amps having a near-brickwall filter in their input. So not pushing anything on shoulders of the next stage, which probably has enough design challenges of it's own.

Maybe you could persuade Amir to measure the output of DACs well into the MHZ range as well as amplifiers.

I just do that for myself. Others can decide what they want to do and how.

Some things don't need fixing though and work fine as they are.
They may not measure that great though.

Now that is the part I find a bit confusing statement on this site...