• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Does DSD sound better than PCM?

Status
Not open for further replies.

March Audio

Master Contributor
Audio Company
Joined
Mar 1, 2016
Messages
6,378
Likes
9,317
Location
Albany Western Australia
Are we discussing music or audio?

Sound reproduction or music reproduction?

:)

Can you expand on your differentiation there?

My comment was referring to a real musical performance recording as opposed to say a test recording where you have just recorded that instrument to close to 0dB.
 
Last edited:

svart-hvitt

Major Contributor
Joined
Aug 31, 2017
Messages
2,375
Likes
1,253
Can you expand on your differentiation there?

My comment was referring to a real musical recording as opposed to say a test recording where you have just recorded that instrument to close to 0dB.

I think @Miska discusses digital perfection while you discuss pragmatic reproduction of music.

So you may talk past each other?
 

Zek

Major Contributor
Joined
Feb 17, 2018
Messages
1,582
Likes
2,146
In the "DSDIFF/DSF Settings" uncheck the "Direct SDM" box to allow DSD-to-DSD processing. Then in main window, select "SDM (DSD)" as output format. In settings, set for example "44.1k x256" or "44.1k x512" as rate limit and make sure "Auto rate family" in unchecked.

Cover art embedded in files is shown automatically. External cover art is shown when playing from the library, in which case JPEG or PNG format files with "cover" or "folder" in the file name are picked up.

For further questions about my software, please send me email, start a new thread here, or post at the HQPlayer thread at AS: https://audiophilestyle.com/forums/topic/19715-hq-player/
Just to avoid derailing this thread too much from the original topic. Thanks! :)
Thanks a lot.:)
 

svart-hvitt

Major Contributor
Joined
Aug 31, 2017
Messages
2,375
Likes
1,253
Granted, but there are a few things discussed here that even on a theoretical level don't add up.

If things don’t add up, that’s more interesting!

But I don’t think @Miska bothers about psychoacoustics (which is part of audio science), only theoretical information processing.

:)
 

March Audio

Master Contributor
Audio Company
Joined
Mar 1, 2016
Messages
6,378
Likes
9,317
Location
Albany Western Australia
If things don’t add up, that’s more interesting!

But I don’t think @Miska bothers about psychoacoustics (which is part of audio science), only theoretical information processing.

:)
Well its not really psycho acoustics to be unconcerned about low level images in the transition band which are above the audible hearing range.
 

Miska

Addicted to Fun and Learning
Audio Company
Joined
Feb 20, 2019
Messages
615
Likes
448
But as Blumleins earlier numbers show an alias could be 85dB down by the time you hit the audio band. So where is the problem?

Because it is extra crap that shouldn't be there. And it is way too close to "audio band", even if defined tightly, which is anyway quite loose definition. I consider at least first 100 kHz as audio band.

-85 dB down is 0.006% of distortion. If people here see a DAC to have harmonic distortion products at -85 dB many will complain, why aliasing at -85 dB wouldn't be?

My comment was referring to a real musical performance recording as opposed to say a test recording where you have just recorded that instrument to close to 0dB.

You can record it even at -60 dB, it doesn't change anything because it is about relative levels between different frequencies and not about absolute level.

Well its not really psycho acoustics to be unconcerned about low level images in the transition band which are above the audible hearing range.

Of course it is fine for you to choose what you are concerned about. I'm concerned about everything that is deviation from perfectness. This is one of the things that can be technically avoided, so I would like to hear the reason why not be concerned about something that can be fixed.

But hearing doesn't have a brickwall filter at 20 kHz, some people hear past 20 kHz. Especially young females.

In addition, aliased tones there, which usually in music coincide transient leading edges generate exactly same kind of IMD products as the usual 19+20k IMD test tone. So it is new kind of TIM, related to digital audio and doesn't exist in analog domain. TIM, which in it's original form seem to have been somehow forgotten from measurement arsenal...
 

svart-hvitt

Major Contributor
Joined
Aug 31, 2017
Messages
2,375
Likes
1,253
@Miska, out of curiosity: If you look at the development of DAC chips and other digital architecture (for example FPGAs), at what time (how many years from now) can HQ Player be plugged in as part of firmware to fix for perfection 95 percent of the "problems" of DACs?
 

Miska

Addicted to Fun and Learning
Audio Company
Joined
Feb 20, 2019
Messages
615
Likes
448
@Miska, out of curiosity: If you look at the development of DAC chips and other digital architecture (for example FPGAs), at what time (how many years from now) can HQ Player be plugged in as part of firmware to fix for perfection 95 percent of the "problems" of DACs?

Already now... I would say it largely depends on how much CPU performance and software one is ready to put inside a DAC. So far maximum I've got out of a small embedded passive cooled CPU has been DSD256. This is without any external heatsinks, only a small on-chip heatsink. But even CPUs like T-series Core (35W TDP) is quite easy to cool in a passive way.

As more DACs become networked, it is more in line with those capabilities as well.

Sure, with some heavy filters for extra finesse and things like digital room correction for DSD you need a lot more CPU/GPU power, but that is then more question of what kind of feature set one wants. OTOH, one can also ask if it is necessary to actually put these inside the DAC or if it is better to have a two box solution, like many manufacturers have had for a while an external upsampler device (and for example Chord just jumped on that train too).

Mostly the limiting factor is device BOM and how it matches the final wanted price point. It likely won't happen soon for the sub 1000€ devices, for those I see using an already existing computer as the vehicle is way to go. If one plays music using a computer, instead of CPU sitting idle near 0% load, it can instead do something useful...
 

pjug

Major Contributor
Forum Donor
Joined
Feb 2, 2019
Messages
1,775
Likes
1,561
If there is IM in the normal audio band we can measure it. Without commenting on the likely hood of this, even though its technically feasible, do we have any measurements that correlate and support the proposition?

Another view is don't use an amp that is wide open to RF frequencies.

Hi - I have been following this part of the discussion regarding the high frequency DAC output and whether it affects downstream equipment. It does seem like it would be unlikely to affect Class A, Class BB amplifiers. But maybe with Class D there is an issue. For example, see here where it seems like a possible problem when sent to ncore: https://www.diyaudio.com/forums/class-d/321632-hypex-ncore-nc400-input-anti-alias-filter.html

@March Audio since you are making an amplifier with ncore, I am wondering if you are adding LP filtering to your input to guard against this. Or do you think the ncore input is OK as supplied by hypex?
 

March Audio

Master Contributor
Audio Company
Joined
Mar 1, 2016
Messages
6,378
Likes
9,317
Location
Albany Western Australia
Because it is extra crap that shouldn't be there. And it is way too close to "audio band", even if defined tightly, which is anyway quite loose definition. I consider at least first 100 kHz as audio band.

-85 dB down is 0.006% of distortion. If people here see a DAC to have harmonic distortion products at -85 dB many will complain, why aliasing at -85 dB wouldn't be?



You can record it even at -60 dB, it doesn't change anything because it is about relative levels between different frequencies and not about absolute level.



Of course it is fine for you to choose what you are concerned about. I'm concerned about everything that is deviation from perfectness. This is one of the things that can be technically avoided, so I would like to hear the reason why not be concerned about something that can be fixed.

But hearing doesn't have a brickwall filter at 20 kHz, some people hear past 20 kHz. Especially young females.

In addition, aliased tones there, which usually in music coincide transient leading edges generate exactly same kind of IMD products as the usual 19+20k IMD test tone. So it is new kind of TIM, related to digital audio and doesn't exist in analog domain. TIM, which in it's original form seem to have been somehow forgotten from measurement arsenal...


So you think images above the audible band, which is 20 kHz and not 100 kHz, at - 85dB are of audible consequence?

Do you have any evidence of that, or is this just a theoretical desktop exercise?

Sorry, Im not following you with the relative bit. If an image in the transition band is at - 85dB then it's at - 85dB. Pretty much inaudible even if it were in the audible range, but it is above the audible range. What you are forgetting is that numbet was relative to full scale signal, which you simply will not have in the transition band.

I'm evaluating what's of concern based on engineering and acoustic knowledge, it's not arbitrary.

Very few people can hear above 20kHz.

If you look at good amplifiers a full scale 19+20khz imd test may create in band tones below - 110dB. Transition band images at significantly lower levels will produce IMD products a lower levels again. You shouldn't conflate amplifier performance with this filter stuff.
 
Last edited:

March Audio

Master Contributor
Audio Company
Joined
Mar 1, 2016
Messages
6,378
Likes
9,317
Location
Albany Western Australia
[
Hi - I have been following this part of the discussion regarding the high frequency DAC output and whether it affects downstream equipment. It does seem like it would be unlikely to affect Class A, Class BB amplifiers. But maybe with Class D there is an issue. For example, see here where it seems like a possible problem when sent to ncore: https://www.diyaudio.com/forums/class-d/321632-hypex-ncore-nc400-input-anti-alias-filter.html

@March Audio since you are making an amplifier with ncore, I am wondering if you are adding LP filtering to your input to guard against this. Or do you think the ncore input is OK as supplied by hypex?

The modules in my current amps aren't the NC400. The testing I have performed (I did a similar thing to that example) hasn't shown issues with RF input or with dac outputs. I'm sure if you try hard enough you might find a dac with 21mV of RF output in the wrong place as the test in that example :)

As I mentioned in an earlier post any amp should be appropriately filtered. IM is a risk
 
Last edited:

Miska

Addicted to Fun and Learning
Audio Company
Joined
Feb 20, 2019
Messages
615
Likes
448
So you think images above the audible band, which is 20 kHz and not 100 kHz, at - 85dB are of audible consequence?

It is quite far from touted perfect reconstruction and easily within measurement limits. Once you have cleaned all the measurable aspects you still have all the other aspects left which are not revealed with couple of simple measurements. Such as modulator behavior under complex circumstances. Just to get the very basics right before going further.

I'm evaluating what's of concern based on engineering and acoustic knowledge, it's not arbitrary.

Me too. I fix all the obvious and non-obvious problems I can fix, without going to assumptions/claims about audibility which are just bad excuse from fixing obvious technical problems. I spend years developing fixes for such and very happy doing so. I was doing this for ten years just for myself before I realized that someone else could just possibly maybe find it useful/interesting. Now it's over 20 years old project...

Sorry, Im not following you with the relative bit. If an image in the transition band is at - 85dB then it's at - 85dB. Pretty much inaudible even if it were in the audible range, but it is above the audible range. What you are forgetting is that numbet was relative to full scale signal, which you simply will not have in the transition band.

If difference between lower frequencies and the 22.8 kHz tone is -20 dB, it doesn't matter if you record at 0 dB or -60, but the relative level between lower frequencies and 22.8 kHz is still the same 20 dB. If you record at -60 just turn up the volume in headphone amp and there we go.

If you look at good amplifiers a full scale 19+20khz imd test may create in band tones below - 110dB. Transition band images at significantly lower levels will produce IMD products a lower levels again. You shouldn't conflate amplifier performance with this filter stuff.

World is full of all kinds of amplifiers. These transition band (20 - 22.05 kHz) images are pretty much around -3 dB at 22.05 kHz. But you'll have much more problems with the images around 352.8 kHz.

How do DSD advocates handle DSP based crossovers, room correction, and EQ without converting to PCM?

DSD is a marketing term. I treat all under generic SDM category, 2-level or some other number of levels. (at the moment I can deal from 2-level output to 256-level output)

I do that all the time. By processing at native rates with remodulators built to perform such tasks. My approach is to have two separate DSP engines, one for PCM outputs and another one for SDM outputs. Sure, processing data at something like 24 MHz SDM sampling rate takes a little bit more oomph than at mere 44.1k PCM. That's were modern GPUs are good help. And maintaining two engines is double work, but that's not a problem.
 

Miska

Addicted to Fun and Learning
Audio Company
Joined
Feb 20, 2019
Messages
615
Likes
448
Hi - I have been following this part of the discussion regarding the high frequency DAC output and whether it affects downstream equipment. It does seem like it would be unlikely to affect Class A, Class BB amplifiers. But maybe with Class D there is an issue. For example, see here where it seems like a possible problem when sent to ncore: https://www.diyaudio.com/forums/class-d/321632-hypex-ncore-nc400-input-anti-alias-filter.html

It affects also other types of amps, for example class AB. But you really need to measure their performance at least from 0 - 1 MHz bandwidth to know. There are huge differences in distortion vs frequency behavior.

So my take is that starting with clean output makes sure that you don't have surprises later on in the chain. IF there are still effects due to some output properties, I think it is good to realize what kind of effects those are! That would go under good engineering practice?

I am wondering if you are adding LP filtering to your input to guard against this

Now this gets into to the other topic we were discussing which is overall system phase/transient response...
 

March Audio

Master Contributor
Audio Company
Joined
Mar 1, 2016
Messages
6,378
Likes
9,317
Location
Albany Western Australia
It is quite far from touted perfect reconstruction and easily within measurement limits. Once you have cleaned all the measurable aspects you still have all the other aspects left which are not revealed with couple of simple measurements. Such as modulator behavior under complex circumstances. Just to get the very basics right before going further.



Me too. I fix all the obvious and non-obvious problems I can fix, without going to assumptions/claims about audibility which are just bad excuse from fixing obvious technical problems. I spend years developing fixes for such and very happy doing so. I was doing this for ten years just for myself before I realized that someone else could just possibly maybe find it useful/interesting. Now it's over 20 years old project...



If difference between lower frequencies and the 22.8 kHz tone is -20 dB, it doesn't matter if you record at 0 dB or -60, but the relative level between lower frequencies and 22.8 kHz is still the same 20 dB. If you record at -60 just turn up the volume in headphone amp and there we go.



World is full of all kinds of amplifiers. These transition band (20 - 22.05 kHz) images are pretty much around -3 dB at 22.05 kHz. But you'll have much more problems with the images around 352.8 kHz.



DSD is a marketing term. I treat all under generic SDM category, 2-level or some other number of levels. (at the moment I can deal from 2-level output to 256-level output)

I do that all the time. By processing at native rates with remodulators built to perform such tasks. My approach is to have two separate DSP engines, one for PCM outputs and another one for SDM outputs. Sure, processing data at something like 24 MHz SDM sampling rate takes a little bit more oomph than at mere 44.1k PCM. That's were modern GPUs are good help. And maintaining two engines is double work, but that's not a problem.


Well you neatly avoided the question there. Do you think signals above the audible range of 20kHz at - 85dB are audible?
We can measure things that are of no consequence and way below our ability to hear.

If we accept your premise regarding differential, we have to accept that, for example, a spurious signal at - 120dB is as significant as a spurious signal at - 20dB. This quite obviously is not the case.
Your premise makes no sense, especially as you are still talking about signals that are in an area above our hearing range. In the real world you will not have high level signals in the transition band. Spuria will be significantly below that level.

If you ignore the real world then what you are doing is somewhat arbitrary.
 

March Audio

Master Contributor
Audio Company
Joined
Mar 1, 2016
Messages
6,378
Likes
9,317
Location
Albany Western Australia
I

World is full of all kinds of amplifiers. These transition band (20 - 22.05 kHz) images are pretty much around -3 dB at 22.05 kHz. But you'll have much more problems with the images around 352.8 kHz.

.

No, you are conflating amplifier performance with other issues. If the amp has good IM performance, it's a non issue.

If it has IM problems with signals in the transition band then it has issues with IM in the audio band. It's a shit amp - period.
 

Miska

Addicted to Fun and Learning
Audio Company
Joined
Feb 20, 2019
Messages
615
Likes
448
Well you neatly avoided the question there. Do you think signals above the audible range of 20kHz at - 85dB are audible?
We can measure things that are of no consequence and way below our ability to hear.

Given for example said 22.8 kHz, and it's aliasing product at 21.3 kHz, it is not down by -85 dB in this case. And yes I consider 21.3 kHz likely audible for notable portion of population on earth (at least more than 1 out of entire population).

But again, I don't evaluate audibility of N within population X people. That is much harder evaluation to do than just fixing the problem, instead of speculating about audibility. More, evaluating audibility of some objectively found problem is trying to find excuse for not fixing an obvious problem.

If you ignore the real world then what you are doing is somewhat arbitrary.

How am I ignoring the real world? If I can fix something I can find, objectively, as fault in real world, I will fix it.

Can you describe why you are against fixing some obvious problems?

For me, this is non-issue. It is one of the many things I've fixed already and I move on to many of the other issues to be fixed. No need to make a lot of posts about one particular fix, which is among the easier ones.

No, you are conflating amplifier performance with other issues. If the amp has good IM performance, it's a non issue.

In real world (pun intended) not everybody has a perfect amplifier... How about playing nice with those with less than perfect ones too?

Otherwise we very quickly move to talk about amplifiers and then I start hearing statements "if a DAC has good quality it doesn't output ultrasonic images". Just IMO, you cannot push problem of component A to shoulders of component B. Every component has enough of design challenges in it's own domain.

If it has IM problems with signals in the transition band then it has issues with IM in the audio band. It's a shit amp - period.

I don't see it that way. Usually 2+3 kHz twin tone is quite different from 19+20 kHz win tone, both have the same 1 kHz difference.


But really, I don't know why I'm having this conversation. It is off-topic, except for the nice detail that with properly done DSD content you don't have digital decimation filters (so whatever we discuss doesn't exist)... You can make you own decisions about what is important to you and I make mine. When I can objectively measure differences, I'm usually fine talking about those, from mathematical perspective. If you want to tell people what they can and cannot hear, it's your decision. Personally, based on my experience, I don't do that. People have surprised me enough many times.
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,522
Likes
37,050
Nothing should be ever mixed to 0 dB, but there are periods in recordings (some Spanish music (Flamenco) and such) where castanets play solo. What matters are the relative levels. And what stops me from publishing my recordings of castanets playing? I would be upset if someone considers my musical creation unimportant to reproduce properly.

You can also find similar effects in periods where someone drives ADC into clipping. That shouldn't be done either, but it still happens. And recordings shouldn't contain digital clipping, but most RedBook content does, thanks to loudness wars.

I never ever make assumptions about what "musical recording" contains. There are so many different types of recordings in the world. From trash metal to Taiko drummers and sounds of nature. Roger Waters' Amused to Death album contains a very nice scene of fighter plane flying towards you and over you and dropping a bomb that explodes. Regarding earlier comment about MQA, that is one of the mistakes they make. In their presentations they present typical spectra of classical music, like that would be all there is.

So playback gear better behave properly in all conditions. That is important to me, other people may have different priorities.

I agree with you in principal, but there is the constraint of what is possible. Reducing those constraints is good. But for instance if we took your approach to amplifier distortion you'd be unhappy until there is none. There are at least as many ways amps can be tripped up as can ADC and DAC. But if you apply their use well distortion is low enough not to be heard with good amps, and any pathological conditions have been made so rare they aren't much of a concern.

We need people like you trying to reach perfection, but the phrase don't throw the baby out with the bathwater comes to mind.
 

svart-hvitt

Major Contributor
Joined
Aug 31, 2017
Messages
2,375
Likes
1,253
Well you neatly avoided the question there. Do you think signals above the audible range of 20kHz at - 85dB are audible?
We can measure things that are of no consequence and way below our ability to hear.

If we accept your premise regarding differential, we have to accept that, for example, a spurious signal at - 120dB is as significant as a spurious signal at - 20dB. This quite obviously is not the case.
Your premise makes no sense, especially as you are still talking about signals that are in an area above our hearing range. In the real world you will not have high level signals in the transition band. Spuria will be significantly below that level.

If you ignore the real world then what you are doing is somewhat arbitrary.

We have producers of DACs that promise distortion at -150 dB: https://www.mola-mola.nl/dac.php

At a price of €10,000...

Is this real-world problem solving?

So what’s the issue if you can fix «problems», even if they’re inaudible, in the digital chain for €150?

I would have problems with a product like HQ Player if it didn’t do what it says. Or worse; does harm where there was no problem.

Are you suggesting HQ Player doesn’t do what it claims, or that the software does harm?
 
Status
Not open for further replies.
Top Bottom