How can a non-oversampling DAC sound good?

[manisandher]

Following on from my DAC comparison thread (http://www.computeraudiophile.com/f6-dac-digital-analog-conversion/digital-analogue-converter-test-old-school-vs-new-kid-31354/), I simply don’t understand how the Altmann Attraction DAC, a non-oversampling and filterless DAC, can sound as accurate as it does. On listening to its output, it manages to reproduce a very convincing facsimile of the original file (see aforementioned thread for details).

 

Here are spectra of some test tones fed to the Altmann DAC and also to a Chord 2Qute for comparison:

 

2Qute 2500Hz:

 

Altmann 2500Hz:

 

2Qute 15kHz:

 

Altmann 15kHz:

 

The aliasing is clearly visible on the Altmann spectra, and yet it sounds more accurate than the 2Qute. I know there are some serious filter designers here (Miska and PeterSt spring immediately to mind) and a bunch of other very knowledgeable people too. Could you help me understand how this could be? Is the total elimination of pre- and post-ringing in the Altmann a more important factor than the introduction of some aliasing?

 

Mani.

[Jud]

Following on from my DAC comparison thread (http://www.computeraudiophile.com/f6-dac-digital-analog-conversion/digital-analogue-converter-test-old-school-vs-new-kid-31354/), I simply don’t understand how the Altmann Attraction DAC, a non-oversampling and filterless DAC, can sound as accurate as it does. On listening to its output, it manages to reproduce a very convincing facsimile of the original file (see aforementioned thread for details).

Here are spectra of some test tones fed to the Altmann DAC and also to a Chord 2Qute for comparison:

 

2Qute 2500Hz:

[ATTACH=CONFIG]32815[/ATTACH]

 

Altmann 2500Hz:

[ATTACH=CONFIG]32816[/ATTACH]

 

2Qute 15kHz:

[ATTACH=CONFIG]32817[/ATTACH]

 

Altmann 15kHz:

[ATTACH=CONFIG]32818[/ATTACH]

 

The aliasing is clearly visible on the Altmann spectra, and yet it sounds more accurate than the 2Qute. I know there are some serious filter designers here (Miska and PeterSt spring immediately to mind) and a bunch of other very knowledgeable people too. Could you help me understand how this could be? Is the total elimination of pre- and post-ringing in the Altmann a more important factor than the introduction of some aliasing?

 

Mani.

@lmitche and I just happen to have been guests at a "mad scientist's lair" last weekend and heard a *very* interesting presentation of a filterless DAC. (Not in production, all discrete parts, the internal cabling you wouldn't believe....) It was feeding speakers well capable of 50KHz response, through a custom designed amp.

 

The immediacy was captivating, showing me what we're missing with the necessarily at least slightly more "polite" filters in the software or hardware we're accustomed to. Percussion hits, instrumental and vocal attacks, had an absolute "you are there" feeling, and the timing of all the transients seemed to me as if it helped with localization.

 

However, for any extended period, it was teeth-gritting. When we listened through the filtering of one of the software players on the market, everyone's feet started tapping and we were really enjoying ourselves for the first time all day. This is something I think the "mad scientist" and his lovely wife were loath to admit, even to themselves, but their feet were tapping too.

 

So yes, I think we are missing some dimension of reality that is very much emphasized by filterless DACs. The results are often sonically spectacular. But I don't want to listen to "sonic spectaculars," I want to listen to music. With the state of the art today, I am literally happier listening to music that *doesn't* have that last ounce of reality in the time dimension at the cost of teeth-gritting distortion.

[manisandher]

Hey Jud, thanks for sharing. It's exactly the sort of description I've heard many people give of non-oversampling DACs - really impressive for a while, but then fatiguing.

 

I should have called this thread, "How can non-oversampling DACs sound accurate?", rather than 'good'. The Altmann sounds accurate and true to the source. How do I know this? Well, take the three files from the other thread:

 

Original 16/44.1 file: https://drive.google.com/open?id=0B0...0F4SXJRZmZ0Unc

 

2Qute analogue output: https://drive.google.com/open?id=0B0...XNBMFZZVmlYSE0 (1.5dB down on original)

 

Altmann analogue output: https://drive.google.com/open?id=0B0PU5LO5jVjfNExvRVJWUjVHSDg (1.5dB down on original)

 

I've replayed these files on many different systems now and the outcome is totally consistent and repeatable: the Altmann sounds truer to the original file than the 2Qute, and is without any hint of fatigue. I know it's a pain downloading and spending time comparing these things, but the files are there for anyone to verify my claim for themselves.

 

Mani.

[Jud]

Hey Jud, thanks for sharing. It's exactly the sort of description I've heard many people give of non-oversampling DACs - really impressive for a while, but then fatiguing.

 

I should have called this thread, "How can non-oversampling DACs sound accurate?", rather than 'good'. The Altmann sounds accurate and true to the source. How do I know this? Well, take the three files from the other thread:

 

Original 16/44.1 file: https://drive.google.com/open?id=0B0...0F4SXJRZmZ0Unc

 

2Qute analogue output: https://drive.google.com/open?id=0B0...XNBMFZZVmlYSE0 (1.5dB down on original)

 

Altmann analogue output: https://drive.google.com/open?id=0B0PU5LO5jVjfNExvRVJWUjVHSDg (1.5dB down on original)

 

I've replayed these files on many different systems now and the outcome is totally consistent and repeatable: the Altmann sounds truer to the original file than the 2Qute, and is without any hint of fatigue. I know it's a pain downloading and spending time comparing these things, but the files are there for anyone to verify my claim for themselves.

 

Mani.

 

Unfortunately I wouldn't be able to listen to those files properly, Mani, lacking an NOS DAC.

 

 

Sent from my iPhone using Computer Audiophile

[manisandher]

I've played the files back on a bunch of regular DACs and the outcome is exactly the same - the Altmann sounds remarkably similar to the original file. If you have the time and inclination, feel free to give it a go, though I know these things can be a bit of a drag.

 

Mani.

[Jud]

I've played the files back on a bunch of regular DACs and the outcome is exactly the same - the Altmann sounds remarkably similar to the original file. If you have the time and inclination, feel free to give it a go, though I know these things can be a bit of a drag.

 

Mani.

 

The thing is that the filtering in my software or DAC will likely take at least some of the edge off the unfiltered version. Even if I adjusted Audirvana's upsampling to zero slope, there'd still be the SDM from the software or DAC.

 

 

Sent from my iPhone using Computer Audiophile

[YashN]

Could you help me understand how this could be? Is the total elimination of pre- and post-ringing in the Altmann a more important factor than the introduction of some aliasing?

You should definitely check the other REW measurements results tabs for more info, like Waterfall, Distortion and Impulse Response, etc... These could show additional differences.

[ted_b]

I've played the files back on a bunch of regular DACs and the outcome is exactly the same - the Altmann sounds remarkably similar to the original file. If you have the time and inclination, feel free to give it a go, though I know these things can be a bit of a drag.

 

Mani.

 

No offense Mani, but how do you know what "the original file" sounds like? Aren't you playing it through something that affects it somehow, or were you there when it was recorded?

[Indydan]

Here is a great paper by Metrum on NOS DACs, that explains why they measure badly:

 

http://www.metrum-acoustics.com/Design%20Philosophy%20Metrum%20Acoustics.pdf

 

It seems our own hearing system acts as a filter. So what is measured at the DACs output, is not what we perceive once we hear the music through our ears. That is how I understand it.

 

 

"Because our hearing naturally functions as a strong filter, our brains tend to interpret the signal from the NOSDAC as if it has passed through a FIR-filter. This is due to the limited bandwidth of our hearing."

[manisandher]

... how do you know what "the original file" sounds like? Aren't you playing it through something that affects it somehow...

 

I don't think it's necessary to know what the original file sounds like, in absolute terms, in order to determine which DAC under test is the most accurate. Let me explain my rationale.

 

I take a 16/44.1 track - the 'original file'. I play it through each DAC and capture the analogue output as I do so. I then play the original file and the captured files back to determine which captured file sounds closest to the original file.

 

Of course, no playback system is 100% perfect. However, whatever affect the playback system might have on the original file, it'll have on the captured files too. Provided the replay system is half decent, you should be able to determine which captured file sounds closest to the original file. Of course, the more transparent the replay system, the easier this is to do. I've compared the files on a number of different systems and can hear the differences consistently and repeatably in all cases. The method seems to work.

 

So in answer to your question, no I don't know what the original file sounds like in absolute terms, but I don't need to for this test.

Mani.

[manisandher]

The thing is that the filtering in my software or DAC will likely take at least some of the edge off the unfiltered version. Even if I adjusted Audirvana's upsampling to zero slope, there'd still be the SDM from the software or DAC.

 

It's interesting... As I mentioned, I've used a number of playback systems to compare the files. Irrespective of the type of playback DAC, NOS or SDM, the differences between the 2Qute and the Altmann are easy to hear in all cases.

 

But I think I know what might actually be going on now... and you're on the right tracks (see later).

 

Mani.

[manisandher]

You should definitely check the other REW measurements results tabs for more info, like Waterfall, Distortion and Impulse Response, etc... These could show additional differences.

 

Yep, no doubt there's more than meets the eye going on. Will look into these other areas...

 

Mani.

[manisandher]

The thing is that the filtering in my software or DAC will likely take at least some of the edge off the unfiltered version.

 

It seems our own hearing system acts as a filter.

 

Aha! You're both onto it I think, but have got the filtering at the wrong part of the chain.

 

I've spent a lot of time over the last few years assessing various ADCs. The one I've settled on is the Prism AD-124 - it's simply the best ADC I've come across. The reason I have a Tascam DA-3000 isn't for it's ADC (it's mediocre at best) - it's to capture the digital output from the Prism. Unfortunately, the Prism is only 44.1/48kHz capable, so I recorded the outputs of the DACs under test at 16/44.1. However, I recorded the test tones shown in the opening post with the Tascam, at 24/192.

 

The test tones clearly show the expected distortion from the Altmann. However, this distortion most likely doesn't even exist in the captured output of the Altmann. Why? Well because the Prism, working at only 44.1kHz, would have invoked its anti-alias filter and taken out any content above 20kHz or so. This would explain why on replay, the Altmann capture doesn't sound as you'd expect a non-oversampling DAC to sound, because it's already been filtered.

 

Make sense?

 

Mani.

[audiventory]

I simply don’t understand how the Altmann Attraction DAC, a non-oversampling and filterless DAC, can sound as accurate as it does.

 

Mani,

 

What does you mean as «filterless» - no-digital or no-analog filter or no-both?

 

 

Could you help me understand how this could be? Is the total elimination of pre- and post-ringing in the Altmann a more important factor than the introduction of some aliasing?

 

Ringing almost is not matter comparing aliases in my opinion.

[manisandher]

What does you mean as «filterless» - no-digital or no-analog filter or no-both?

The 15kHz plot suggests the Altmann has neither a digital or analogue filter.

 

Ringing almost is not matter comparing aliases in my opinion.

Oh sure, they're different things. However, having no anti-imaging filter in the DAC would eliminate any ringing. The compromise though is that it will introduce imaging. And I was wondering which of the two is the most detrimental to the sound, ringing or imaging?

 

I suppose we're back to accuracy in frequency domain vs. accuracy in temporal domain! I suppose most anti-imaging filters in DACs go for some sort of compromise.

 

Mani.

[audiventory]

Oh sure, they're different things. However, having no anti-imaging filter in the DAC would eliminate any ringing. The compromise though is that it will introduce imaging. And I was wondering which of the two is the most detrimental to the sound, ringing or imaging?

 

Where we use ears for estimation, we get subjectivity. One man seems that -70 dB noise is OK. Other man dislike -90 dB even.

 

At ideal hardware absence of analog filter is not matter by aliases.

 

These aliases placed at ultrasound and we can’t hear it.

 

But for real hardware non-linear distortions cause generating of audible products by ultrasound components.

 

Level of the audible products depend on power of ultrasound components.

 

Also ultrasound components consume part of dynamic range of analog part.

 

Ringing looks so scary for abstract square test signal.

More difference between next samples - more ringing energy.

In real musical signal almost absent big differences between next samples.

[manisandher]

But for real hardware non-linear distortions cause generating of audible products by ultrasound components.

Level of the audible products depend on power of ultrasound components.

Also ultrasound components consume part of dynamic range of analog part.

 

These distortions, even though ultrasonic, must have some effect on the sound for the reasons you cite. Here's a linear scale plot of the Altmann at 15kHz:

 

 

The imaging around 44.1 kHz is clearly seen.

 

Mani.

[manisandher]

"Because our hearing naturally functions as a strong filter, our brains tend to interpret the signal from the NOSDAC as if it has passed through a FIR-filter. This is due to the limited bandwidth of our hearing."

 

This, I don't understand at all. Surely the filtering needs to be before the digital to analogue conversion takes place to be effective. As audiventory points out, a non-oversampling DAC will be sending a whole bunch of ultrasonic grunge to the electronics downstream (its own output stage included).

 

Anyone want to chime in on this?

 

Mani.

[audiventory]

These distortions, even though ultrasonic, must have some effect on the sound for the reasons you cite. Here's a linear scale plot of the Altmann at 15kHz:

 

[ATTACH=CONFIG]32837[/ATTACH]

 

The imaging around 44.1 kHz is clearly seen.

 

Mani.

 

Pure digital spectrum is symmetrical.

 

See here (upper part of 2nd picture from top - source signal audio spectrum):

How Convert Sample Rate. Oversampling

 

At your spectrum we see half of full digital band (0 ... [sample rate]).

 

"Hill" at upper part of the spectrum looks like DSD modulation noise. But I can't understand how it appear.

 

At my sight, picture looks like a system: PCM audio data (88 kHz sample rate, probably) -> DSD modulator. And we observe the output of the DSD demodulator.

 

Non-filtered PCM spectrum is symmetrical along frequency [sample rate]/2 like by link that I reffered above.

 

At your specrum poducts along 44 kHz havn't same levels. So probably there applied a low frequenct filter.

 

Real picture may be changed by measurement tool.

[manisandher]

I should have been more careful with terminology in my original post. The plots I posted show clear DAC 'imaging' and not ADC 'aliasing'. So we should expect symmetry only around 44.1kHz (due to imaging in the Altmann filterless DAC) and not around 96kHz (due to any aliasing in the Tascam ADC).

 

At your spectrum we see half of full digital band (0 ... [sample rate]).

 

"Hill" at upper part of the spectrum looks like DSD modulation noise. But I can't understand how it appear.

 

Yes, it's due to noise shaping in the Tascam. The Tascam uses a PCM4202 1-bit ADC chip, but I set it to PCM 24/192.

At your specrum poducts along 44 kHz haven't same levels. So probably there applied a low frequenct filter.

 

Yes for sure, the Tascam would have applied an effective anti-alias filter before ADC.

 

Mani.

[Jud]

Aha! You're both onto it I think, but have got the filtering at the wrong part of the chain.

 

I've spent a lot of time over the last few years assessing various ADCs. The one I've settled on is the Prism AD-124 - it's simply the best ADC I've come across. The reason I have a Tascam DA-3000 isn't for it's ADC (it's mediocre at best) - it's to capture the digital output from the Prism. Unfortunately, the Prism is only 44.1/48kHz capable, so I recorded the outputs of the DACs under test at 16/44.1. However, I recorded the test tones shown in the opening post with the Tascam, at 24/192.

 

The test tones clearly show the expected distortion from the Altmann. However, this distortion most likely doesn't even exist in the captured output of the Altmann. Why? Well because the Prism, working at only 44.1kHz, would have invoked its anti-alias filter and taken out any content above 20kHz or so. This would explain why on replay, the Altmann capture doesn't sound as you'd expect a non-oversampling DAC to sound, because it's already been filtered.

 

Make sense?

 

Mani.

 

I would very much like to see plots with 16/44.1 input like the ones you did above for 24/192 input, if you have the time and interest.

 

 

Sent from my iPhone using Computer Audiophile

[manisandher]

I would very much like to see plots with 16/44.1 input like the ones you did above for 24/192 input, if you have the time and interest.

 

Yep, happy to do this - I'm interested in knowing too. The biggest issue will be that the plots will only extend to 22.05 kHz and not 96kHz. For the Altmann, I would expect exactly this (plot stopping at 22.05 kHz):

 

 

There should be no new aliasing effects below 22.05 kHz, as the Prism's anti-alias filter will have eliminated this. However, the Altmann's imaging effects around 44.1kHz will still be there, but totally cut off above 22.05 Khz. Below this, all the imaging effects should remain, as in the original plot.

 

We'll see...

 

Mani.

[manisandher]

I would very much like to see plots with 16/44.1 input like the ones you did above for 24/192 input, if you have the time and interest.

Here you go...

 

Original 24/192 plot through Tascam (Altmann DAC with 15kHz tone):

 

 

My expected result, Altmann through Prism ADC (with anti-aliasing filter):

 

 

Actual result, Altmann through Prism ADC (44.1 kHz file upsampled offline to 192 kHz to normalize scale):

 

No wonder the Altmann capture doesn't sound like that of a NOS DAC - all of it's 'NOSness' above 22.05 kHz has been taken out by the Prism's anti-aliasing filter!

 

Mani.

[Fokus]

However, having no anti-imaging filter in the DAC would eliminate any ringing.

 

No.

 

That is a common fallacy. But no, a non-filtered DAC does not eliminate ringing at all.

 

Think about how the recording has been made. Or try it yourself. You seem to have the tools.

[manisandher]

No.

 

That is a common fallacy. But no, a non-filtered DAC does not eliminate ringing at all.

 

Wow, really? I thought pre- and post-ringing was purely down to the use of a low pass anti-imaging filter in the DAC. My assumption was, no filter, no ringing. And this seemed to be the only (debatable) advantage of a filterless NOS DAC. But I'm happy to learn and stand corrected.

 

Think about how the recording has been made. Or try it yourself. You seem to have the tools.

I'm willing to give it a try. How would you suggest I go about this?

 

Mani.