Filterless DACs question

[Digital Enigma]

NOS filterless DACs allow in a load of imaging and distortions, measure like a bag of nails and look completely broken on paper. So why is it people say these types of DACs sound pleasant and smooth? Shouldnt the high amounts of imaging cause buzzing and harsh tones in the audible spectrum? I can hear up to 14khz are these artefacts happening above there?

 

[theveterans]

People don't listen to illegal signals to induce those ugly looking sine waves

 

[Digital Enigma]

If I’m understanding this correctly then, with actual music (old or modern) the ADC filter prevents aliasing and then when played back on filterless DAC there is no imaging?

Imaging only shows on test tones to prove if a DAC is filterless?

 

[theveterans]

If I’m understanding this correctly then, with actual music (old or modern) the ADC filter prevents aliasing and then when played back on filterless DAC there is no imaging?

Imaging only shows on test tones to prove if a DAC is filterless?

ADC does not prevent aliasing and that's why DAW is always sampled at 88.2 or 96KHz environment. With reconstruction, a filterless DAC is implementing a zero order old approach and you get imaging as you get near nyquist

 

[Digital Enigma]

you get imaging as you get near nyquist

But not on actual music?

 

[theveterans]

But not on actual music?

It does, but the harmonics (that makes up the timbre) in the actual music is has more amplitude than the image so you basically don't hear the imaging harmonics as much due to auditory masking

 

[danadam]

So why is it people say these types of DACs sound pleasant and smooth?

When you play audio sampled at 44.1k on such DAC, the zero-order-hold reconstruction of the signal will result in increasing attenuation of higher frequencies, up to 3 dB at 20 kHz. If that's not compensated, and I'm guessing such DACs won't do that, then it may sound pleasant and smooth to some people.

Shouldnt the high amounts of imaging cause buzzing and harsh tones in the audible spectrum?

Depends. It may, if the components downstream (amp, speakers) are prone to causing intermodulation distortion.

If I’m understanding this correctly then, with actual music (old or modern) the ADC filter prevents aliasing and then when played back on filterless DAC there is no imaging?

No, ADC filters have nothing to do with DAC images.

You can upsample (properly) the audio before sending to such DAC and this will move the generated images further in the frequency spectrum and it will attenuate them more and this, I think, may reduce probability of triggering IMD. This will also remove reduce the attenuation in the audible band mentioned above. You basically do OS (oversampling) yourself instead of DAC.

 

[Digital Enigma]

When you play audio sampled at 44.1k on such DAC, the zero-order-hold reconstruction of the signal will result in increasing attenuation of higher frequencies, up to 3 dB at 20 kHz. If that's not compensated, and I'm guessing such DACs won't do that, then it may sound pleasant and smooth to some people.

If most people can’t hear above 15khz then this wouldn’t be audible surely.

 

[gregorio]

If most people can’t hear above 15khz then this wouldn’t be audible surely.

It’s a very shallow roll-off starting around 2kHz. Depending on the content you’re playing, it can be audible under DBT conditions. I passed a DBT with a NOS DAC many years ago (and so have many others), it wasn’t particularly difficult, although I doubt I’d be able to do so these days.

If I’m understanding this correctly then, with actual music (old or modern) the ADC filter prevents aliasing and then when played back on filterless DAC there is no imaging?

The ADC filter does indeed prevent aliasing, however, “images” when converting back to analogue are not related to aliasing, as danadam stated. Aliasing is when frequencies in the input signal above the Nyquist frequency are mirrored down, into the in-band freqs (“in-band” meaning the freqs below the Nyquist frequency). When converting back to analogue “Imaging” is the reverse of aliasing, the in-band freqs are mirrored up, above the Nyquist frequency. In both cases, the solution is a filter that removes freqs above the Nyquist frequency: An anti-alias filter in an ADC that removes frequencies from the input signal above the Nyquist frequency AND an anti-image (or reconstruction) filter in the DAC that removes images from the output signal above the Nyquist frequency. Both filters are required.

I’m not sure how easy to understand my above explanation is, but hopefully you now understand that you will have images above the Nyquist point in the DAC’s output, even though the digital data input to the DAC has no content above the Nyquist frequency. As danadam also mentioned, unless these images are filtered out during the DAC process, there’s a good chance they will cause IMD downstream (in the amp and/or speakers). So the reason “people say these types of DACs sound pleasant and smooth” is one or a combination of: A. They *might* be able to actually sense the very subtle high freq roll-off, B. They may be getting audible levels of IMD and somehow think that’s “pleasant” or C. Misleading marketing has led them to believe NOS DACs “sound pleasant and smoother”, so that’s what their biased perception experiences/“hears”.

ADC does not prevent aliasing and that's why DAW is always sampled at 88.2 or 96KHz environment. With reconstruction, a filterless DAC is implementing a zero order old approach and you get imaging as you get near nyquist

None of the assertions in these two sentences are correct. An ADC does indeed prevent aliasing. DAWs are quite rarely used with a sample rate of 88.2 or 96kHz, by far the most common sample rate used is 48kHz, followed by 44.1kHz. And lastly, you do NOT “get imaging as you get near Nyquist”, EVERYTHING below Nyquist (1Hz to 22.05kHz in the case of 44.1k sample rate) will be mirrored above Nyquist and indeed not just once but at multiples of the sample rate.

It does, but the harmonics (that makes up the timbre) in the actual music is has more amplitude than the image so you basically don't hear the imaging harmonics as much due to auditory masking

Again, no. You will not hear the images due to the fact they start above the Nyquist freq and are therefore outside the range of human hearing, nothing to do with auditory masking (which is the phenomena of frequencies within the audible band being rendered inaudible).

G

 

[castleofargh]

Objectively, filterless is bad. That's it. We have clear rules for how digital audio can have good fidelity, and no reconstruction filter goes against those rules. We can argue all day about what we like or dislike, but IMO, rejecting the very means of fidelity is a little broken. And here, we're not even talking about the pleasing harmonics of certain tube amps, we're talking about types of distortions that are not known to be pleasing(if noticed!!!). I've never read an argument in favor of NOS filterless that didn't seem to misunderstand how digital signals work.

NOS isn't strictly against the rules of digital audio, at least in theory we could have very good NOS DACs. But it makes the work so very much harder at the filter level when trying to handle 44kHz music. In practice, the result tends to be a mix of distos and treble roll off, but it's up to the designer to decide to overshoot in one direction, or the other or to stick to the middle of the road. So the actual impact could be fairly different, depending on the philosophical priorities of the designer. That would also tend to mess more with phase, so maybe someone somewhere justifies not filtering a NOS DAC with that excuse? IDK.

Subjectively, It would also depend on your system, hearing, and what you find objectionable or not. My go-to examples for that are background hiss from some amps, and vinyl. When I go nuts hearing any amount of audible hiss, almost nobody in my entourage seems to care or even notice unless I point it out first. Different strokes for different folks.
Chances are that your amplifier was either designed to handle ultrasonic content fairly high without issue, or has its own low pass filter to deal with what would cause instability/disto. Then your transducer probably rolls off significantly(maybe before even reaching ultrasounds), or at least with headphones it probably has those comb like ups and down at high frequencies that still attenuate a good deal of them, lowering the global ultrasonic energy. And obviously, you're not going to hear frequencies you're not hearing(even if that's a high concept for audiophiles).
It's hard to generalize how much worse the sound can be, headphones already have various and often significant amounts of distortions(we only ever see THD measured, but it's not all there is).
So there are various circumstances potentially helping to mitigate the mistake of not following the typical way of handling signal reconstruction. Everything is a game of relative amplitude and audibility/personal taste/sensitivity.

 

[2leftears]

Have the proponents of NOS DACs ever come up with any hypotheses as to why a NOS DAC sounds 'better'? Or in other words, what do they think the drawbacks are of oversampling?

Is it worth pointing out that if a NOS DAC modification is made to the built-in DAC of a CD player, in some filter/DAC topologies (e.g. the TDA1541/SAA7220 combo) the ability for error concealment for imperfect/damaged CDs is also compromised?

 

[71 dB]

Pretty much everything has been answered and explained already.
I'll add a graph of the gain of zero-order-hold signal reconstruction at 44.1 kHz sampling rate:

Have the proponents of NOS DACs ever come up with any hypotheses as to why a NOS DAC sounds 'better'? Or in other words, what do they think the drawbacks are of oversampling?

I believe their claim is the more the signal is processed, the less "musical" it becomes.

 

[Digital Enigma]

Have the proponents of NOS DACs ever come up with any hypotheses as to why a NOS DAC sounds 'better'? Or in other words, what do they think the drawbacks are of oversampling?

From what I’ve read the idea is with large amounts noise reduction, error correction, negative feedback and an impulse with ringing is what makes it sound artificial.

They live with distortions for no processing and perfect impulse response, they say leads to natural sound. This one is the lesser of two evils I guess.

I like to be open minded about this stuff and stay in the middle. Example here, if you take a photo on an iPhone it does a lot of processing to clean up the image and it is no longer a raw photo but if you zoom in a bit it has this painted and smeared artificial look, I would of actually preferred original the less sharpened picture with more noise because it looks more real.

Now this easily visible to your eye but the question is does the same apply to audio? Is this processing audible to the ear in the same way?

 

[gregorio]

Now this easily visible to your eye but the question is does the same apply to audio? Is this processing audible to the ear in the same way?

“This processing” is not audible to the ear and that’s because the processing you mentioned is not occurring. There is no negative feedback and there are no impulses in music. Very occasionally there are waveforms within music which do cause ringing but it’s a rare phenomena, of far lower magnitude than the ringing from impulses and nearly all of it is above 20kHz, so when/if it does occur, it’s definitely inaudible. In addition, there is no noise reduction in the process, let alone large amounts and there is no error correction. Maybe you are referring to noise-shaped dithering? If so, then the whole point is to linearise the output so there are no distortions in the audible band. In effect, quantisation error is converted to noise and that created noise is placed in the very high and ultrasonic range where it cannot be heard but it’s very low level anyway with 16bit.

Your analogy is incorrect, the iPhone image processing is considerable and is supposed to be visible, the whole point of it is to make the images look better. With digital audio conversion there is very little processing and the whole point of it is to be inaudible (invisible). There is no “processing to clean up the image” in digital audio conversion, no equivalent of sharpening, noise reduction or other image enhancements being applied and the audio output is the equivalent of a RAW image.

G

 

[theveterans]

From what I’ve read the idea is with large amounts noise reduction, error correction, negative feedback and an impulse with ringing is what makes it sound artificial.

They live with distortions for no processing and perfect impulse response, they say leads to natural sound. This one is the lesser of two evils I guess.

I like to be open minded about this stuff and stay in the middle. Example here, if you take a photo on an iPhone it does a lot of processing to clean up the image and it is no longer a raw photo but if you zoom in a bit it has this painted and smeared artificial look, I would of actually preferred original the less sharpened picture with more noise because it looks more real.

Now this easily visible to your eye but the question is does the same apply to audio? Is this processing audible to the ear in the same way?

It's a load of audio BS IMHO. For my subjective taste, I vastly prefer the sonics out of proprietary filtering such as in the case of Mola Mola Tambaqui or Playback Designs MPD-8AI over NOS