Why can I hear a difference between 32bit 384khz and 24bit 44.1khz?

[Lambda]

Glad to see the upsampler was the issue

It is noted and interesting however that the OP prefers the sound of up-sampling first with possible errors occurring... it is not uncommon for coloration to be pleasing, e.g. vinyl, tube amps etc. I suppose.

That's why i think its the DACs up sampler that adds unpleasantness coloration.

 

[JSmith]

That's why i think its the DACs up sampler that adds unpleasantness coloration.

Maybe... but from what has been posted the last few I'd suggest not.

There is this I guess too, not perfect though;

https://www.audiosciencereview.com/...s-of-topping-d10-dac.2470/page-67#post-157302

Who knows, but one would imagine a decent DAC is going to do things properly re interpolation filter.



JSmith

 

[voodooless]

That's why i think its the DACs up sampler that adds unpleasantness coloration.

Maybe... but from what has been posted the last few I'd suggest not.

No, the D10 is not at fault here. It doesn't even do upsampling, so can be ruled out on that basis alone. By now it's quite clear that the Windows mixer is at fault.

 

[Lambda]

D10 is not at fault here. It doesn't even do upsampling,

Of cause it does. how do you think its output filter works? analog?
No it has an Digital filter and this filter is of cause working at higher frequency
https://www.audiosciencereview.com/...s-of-topping-d10-dac.2470/page-67#post-157302

Who knows, but one would imagine a decent DAC is going to do things properly re interpolation filter.

Not perfect but perfect is impossible (in real time)

Practical filters have non-flat frequency or phase response in the pass band and incomplete suppression of the signal elsewhere. The ideal sinc waveform has an infinite response to a signal, in both the positive and negative time directions, which is impossible to perform in real time – as it would require infinite delay.

This actually looks not to bad but of cause it's having the pre-ringing.

http://archimago.blogspot.com/2018/01/musings-more-fun-with-digital-filters.html
Some don't like this digital Linear phase filters and prefer "NOS dacs" without pre ringing and minimum phase.

Apple seems to use minimum phase up sampling by default.

 

[JSmith]

It doesn't even do upsampling

Doesn't it upsample to 384Hz or some multiple of the input, then low pass filter using a high-performance FIR?



JSmith

 

[Veri]

Doesn't it upsample to 384Hz or some multiple of the input, then low pass filter using a high-performance FIR?

Most ESS chips use ASRC, followed by oversampling. Unless ASRC register is turned off in which case it'll only filter.

 

[voodooless]

Doesn't it upsample to 384Hz or some multiple of the input, then low pass filter using a high-performance FIR?

Yes, but that is oversampling. This is fixed and cannot be influenced from the outside (except for filter selection). Besides, it's synchronous.

Most ESS chips use ASRC, followed by oversampling. Unless ASRC register is turned off in which case it'll only filter.

The D10 does not have an ESS chip. And even if it had, you cannot influence the ASRC process of the ESS chips in the way you do in a normal upsampler. It's bound to the master clock and is meant to bring the audio data in the stable clock domain of the master clock, and in the process getting rid of some more jitter.

In short: there is no way to tell these chips to up upsample to 375 kHz.

 

[JSmith]

Most ESS chips use ASRC

True... but then there is the weird ESS hump and this;

https://www.audiosciencereview.com/...own-ess-es9038pro-dac-thermal-analysis.10348/

Also don't most ESS chips employ ARSC by default?

Apologies I didn't realise the D10 uses the ESS ES9018K2M.

Maybe it is the D10 then the OP doesn't like.

Yes, but that is oversampling.

No, it's zero stuffing... and the D10 seems to use an ESS anyway.



JSmith

 

[escape2]

The D10 does not have an ESS chip.

Really? What chip does it have then?

 

[Lambda]

Yes, but that is oversampling.

Whats the difference between Oversampling and Upsampling.
I think one is a special case of the other. both generate new/more samples at a higher rate.

This is fixed and cannot be influenced from the outside

If you do it in software it can be influenced?

 

[Tks]

This is one:

View attachment 116045

But you can check out other releases from 2L

I love this group and this album.

 

[Veri]

The D10 does not have an ESS chip.

Well if you're going to argue about things at least get your facts straight D10, D10s and upcoming D10 balanced are all ESS 9038Q2M based. And you're right that you can't upsample to a fixed kHz rate, but the ASRC will do an interpolation before the oversampling filter (the latter which uses a really high rate). You can configure both of these via register parameters but most of the ESS config is behind NDA anyway. So who knows, exactly, the manufacturers with experience do..

 

[voodooless]

Really? What chip does it have then?

Hmm, Google mist be crazy because all I came up with was AK4490. My bad

Changes nothing though.

No, it's zero stuffing... and the D10 seems to use an ESS anyway.

Well yes, that is the basis of oversampling: you leave the original samples, leave the rest zero, and use the low pass filter to then reconstruct the original waveform.

Whats the difference between Oversampling and Upsampling.
I think one is a special case of the other. both generate new/more samples at a higher rate.

Well, sort of. Usually oversampling is synchronous. But there is more to it. For instance, some AKM's have 256 oversampling, but the digital filter "only" operates at 8x. The rest is used in the modulator, which makes it even more convoluted.

If you do it in software it can be influenced?

Not the one in the hardware obviously. That one is active regardless of what you input in the DAC (yes, you can disable the ASRC of the ESS, but there really is no point in most cases, and most devices don't let you anyway). A software ASRC can obviously be influenced in all kinds of ways depending on the implementation. See sox documentation on the sheer amount of options available.

You can configure both of these via register parameters but most of the ESS config is behind NDA anyway. So who knows, exactly, the manufacturers with experience do..

Yes, you can do some things there. I've had a look at the NDA'd documents of the older chips once upon a time But you really cannot compare this to what's been done in the windows mixer, which by now is the obvious source of all the issues. I really don't know why we're still arguing about the DAC. It's clearly not the problem, regardless of the chip used.

 

[Lambda]

Well yes, that is the basis of oversampling: you leave the original samples, leave the rest zero, and use the low pass filter to then reconstruct the original waveform.

Ok looks like it is only semantics and there is no real disagreement here.
i would say zero-stuffing is a verry basic form of upsampling/oversampling.
And since its followed by a low pass filter (normally) its not leaving the original samples as they are.

Not the one in the hardware obviously.

Most DACs can't disable there hardware oversampling and low pass filter.
But if you send the DAC 44,1kHz its low pass filter will operate at ~20kHz close to hearing limit and its low pass filter will add artifacts to the very high frequency content of the 44,1kHz sample data.
For example Pre-ringing, phase distortion, imaging/aliasing or even inter modulation.
It is very questionable if humans can here a difference from this artifacts but many people claim they do.
https://archimago.blogspot.com/2015/04/internet-blind-test-linear-vs-minimum.html
https://sci-hub.se/http://www.aes.org/e-lib/browse.cfm?elib=17497


If you instead up sample and lowpass filter for example with sox from 44,1 to 192khz you can chose your own low pass filter.

The new signal is now 192kHz but it's max frequency content is about ~22kHz
the DAC will also oversample this 192kHz to ~375 kHz but the low pass filter will operate at ~90kHz.
Since there is almost no 90kHz content the DACs low pass filter is effectively not doing a thing and is not adding artifacts.

This way its the sox filter adding artifacts and not the DACs filter.

Don't know if this is better or worse but it can be different.
If the difference can be heard is an other question.

 

[JSmith]

oversampling

Upsampling is usually associated with ASRC used in ESS chips and ARSC is followed by the oversampling digital reconstruction filter.

Oversampling is generally Synchronous Sample Rate Conversion (SSRC).

Anyway, what was your point again apart from saying firstly there was no upsampling, then it didn't have ESS and then it did? I was saying all along it likely wasn't the D10 as are you I believe.



JSmith

 

[voodooless]

It is very questionable if humans can here a difference from this artifacts but many people claim they do.

Yes, so we can rule that out here..

Anyway, what was your point again apart from saying firstly there was no upsampling, then it didn't have ESS and then it did? I was saying all along it likely wasn't the D10 as are you I believe.

Yes, it's getting very confusing . The main contention point is post #61, where @Lambda blames the DAC, and you find that to be a possibility in #62. It's just not! Not in this particular setup anyway. You guys might have a very blunt occam's razor

 

[KSTR]

Not perfect but perfect is impossible (in real time)

Practical filters have non-flat frequency or phase response in the pass band and incomplete suppression of the signal elsewhere. The ideal sinc waveform has an infinite response to a signal, in both the positive and negative time directions, which is impossible to perform in real time – as it would require infinite delay.

At 44.1kHz, full sinc filtering/resampling intrinsic delay is about 1 second for 16bit precision and ~10 seconds for 24bit precision, so it is doable in (very delayed) realtime but probably most players might be moaning about plugins that need to (pre-)buffer that huge amount data.

 

[Lambda]

Yes, so we can rule that out here..

From the filter low pass filter alone there should be no obvious difference.
Like "the bass is deeper and tighter..."
can rule that out here

There might be a difference but its supper minuscule and only effecting high frequency and maybe stereo image.
We can't rule this out because there is AES paper claiming it has effect in blind tests.

A possible explanation for would be a volume difference possibly caused by the up sampler.

where @Lambda blames the DAC, and you find that to be a possibility in #62. It's just not! Not in this particular setup anyway.

The DAC is changing the sound in a different way by up sampling and filtering as windows up sampler. why is this not possible?
remember OP claims Up sampled with windows sounds better.

On top 44,1kHz
"Badly" up sampled with windows to 192kHz huge inter sample distortion
Third the windows 192kHz version over sampled to 384kHz
And the last one 44,1kHz over sampled to 384kHz without clipping (hopefully this is what would happens in the dac?)

But the last version is abut 2dB lower
"good" over/up sampling maybe reduce volume and therefore "is has less bass" or what ever they claim to hear.

 

[voodooless]

There might be a difference but its supper minuscule and only effecting high frequency and maybe stereo image.
We can't rule this out because there is AES paper claiming it has effect in blind tests.

Sure, there are probably situations where it could be audible, but not with casual listening like happened here.

A possible explanation for would be a volume difference possibly caused by the up sampler.

For a software version yes, not so regarding the hardware version of the D10. Why would the volume be different depending on the input sample rate? That doesn't make any sense. Besides, you cannot set any sample rate on the D10... That only invalidates the whole premise. It is very clear that the changes were made somewhere in a software setting.

The DAC is changing the sound in a different way by up sampling and filtering as windows up sampler. why is this not possible?

The point is that DAC is irrelevant in the whole debate. It's a constant, we know the properties of the filters in those things, and they are properly designed for all sample rates. It's very unlikely that you will hear any differences regardless of sample rate you input (for > 44.1 kHz) if your source is properly sampled and volume matched.

 

[Lambda]

Why would the volume be different depending on the input sample rate? That doesn't make any sense.

If you up sample first ignoring inter sample clipping and then oversample in the DAC volume can be higher.
If up/oversampling is completly done in the DAC and if the DAC avoids inter sample clipping it must be lower volume output.

look at the last two waveforms.

It's a constant, we know the properties of the filters in those things, and they are properly designed for all sample rates.

For a different sample rate it uses a different filter. so it is not constant.