I'm not talking about existing implementations and the problems they have, I was asking more about plausible reasons to say better sound reproduction could be achieved with no filtering at 20 kHz - what audible benefits could come from that. It seems one claim is that the benefit is the lack of filter-produced ringing, which is assumed to be audible and detrimental (I've yet to investigate this), while my hypothesis for a potential benefit is the ultra-fast transient reproduction which requires ultrasonic frequencies (and this of course would require actual musical ultrasound captured from the instruments or added intentionally in mastering, not ultrasonic aliasing products from the sampling process itself).
-
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!
Why do NOS dacs sound different to oversampling designs?
- Thread starter Purité Audio
- Start date
solderdude
Grand Contributor
One can choose to leave it out or use 'slow' filters or filters that adhere to someone's line of thinking but in order to get the intended (and recorded) signal the reconstruction filter is essential.
For this there is 96/24/88.2/24, 192/24 and so on as well as DSDx(fill in any number) and DXD as a format.
Reproduce that on a filterless NOS DAC and this will sound fine.
There will be plenty harmonically related ultrasonics present and the non harmonically related 'garbage' is so high up there are no transducers that can make this audible. In some circumstances it may still come back as 'garbage' in the audible range though.
Most people, however, have more 44.1/16 content than hi-res. That's what my remarks are all about. Not about hires reproduced on filterless DACs.
There's no point in arguing with fundamentally flawed knowledge and a mind made up.
Be sure your speakers can accurately reproduce those 200 kHz transients.
BTW no audio ADC will accurately record 200 kHz so finding source material could be a challenge.
Be sure your speakers can accurately reproduce those 200 kHz transients.
BTW no audio ADC will accurately record 200 kHz so finding source material could be a challenge.
I'm merely giving this hypothesis the benefit of the doubt - far from having made up my mind. But thanks for the vote of confidence. 
Probably not even necessary - even if I go by the MQA studies/marketing, music already stops somewhere around 48k, so nothing higher than 96k sampling should really be necessary during recording or reproduction. (Then again music is always evolving - who knows what crazy sounds someone will think to include in their mix next year or next decade.)
Last edited:
If recorded at 96 kS/s then any signals over 48 kHz are images, i.e. signals not present in the original music. They are considered undesirable unless you are building a system using the images to generate higher-frequency band centers (done in some RF DAC systems). Or are an audiophile who somehow thinks images of the baseband signals add quality to the sound. They do not enhance the transients of the original sound.
It's a good thing you left out the word "you" from that sentence, because it's not at all clear who you're arguing with or who has said any of the things you're combating here. But hey, as long as you're having fun...
It's a good thing you left out the word "you" from that sentence, because it's not at all clear who you're arguing with or who has said any of the things you're combating here. But hey, as long as you're having fun...
Not actually directed at you or anyone, just general comments, though you are the one raising the points to debate. Could have said "one is". They do address the assertion you made that 200 kHz image signals from a DAC somehow improve the transient response. IME they do not, and in fact usually corrupt pulse integrity because the very HF images are not in phase with the baseband signals so they smear the pulses (edges). And then you said there was no music content over 48 kHz... You can actually measure frequency content from some things that exceed 48 kHz but I've never seen it proven that it is audible by humans. But at 96 kS/s from a DAC there can be no original signal content; Nyquist-Shannon holds. On the recording side such high frequencies would be aliased into the 0-48 kHz baseband if the anti-alias filter did not do its job.
Day seven in a row at work, sorry to be terse, not having fun. Back to SerDes testing (not passing jitter specs, arrrrghhhh...)
One other point of misunderstanding- the "5 us" thing is correct but absolutely unrelated to bandwidth or filtering.
I'm on my phone at the moment and can't link the Monty Montgomery videos, but they're posted in several places here, and if you want to get a wonderful short course in what all that means (and how wrong much audiophile lore is), it will be an enjoyable and eye-opening investment of 20 minutes or so.
I'm on my phone at the moment and can't link the Monty Montgomery videos, but they're posted in several places here, and if you want to get a wonderful short course in what all that means (and how wrong much audiophile lore is), it will be an enjoyable and eye-opening investment of 20 minutes or so.
If you mean the general D/A & A/D video, I did find it useful where it pointed out that post-filter ripples aren't really "added" by the filter, they're just part of the nature of a band-limited signal.
But on the question of timing it's not really an answer, as he only takes into account the case where you've already band-limited the signal before putting it through the ADC. In order to apply Milind Kunchur's 5-us finding to get higher fidelity you would of course not pre-bandlimit the signal anywhere near 20 kHz but would allow everything that real instruments and voices produce to be recorded and sampled. So in this case you would absolutely start losing the 5-us transients if you were sampling at a rate anywhere below 200 kHz.
Unfortunately, Kuncher completely misinterpreted the relationship between the interchannel timing and bandwidth. Don’t fall into the same error. Seek out and watch the video I mentioned, where there’s a great demonstration of this.
I did, and I found one video that doesn't directly address the issue. You're going to have to be more specific - like by posting a link - if you want me to land on some other exact video, I certainly can't read your mind.
(BTW, Kunchur's papers don't talk about interchannel timing but same-channel timing, where he finds the 5 us limit.)
*sigh* Already mentioned above - that part does not address the right issue, as it assumes CD-res-oriented pre-filtering, i.e. intentionally throwing out the 5-us transients before the digital realm is even reached. Of course the sampling rate discussion is moot at that point - you've ended it before it's begun. The real question is: if natural instruments produced harmonics up to 100 kHz (let's say), which would have transients as short as 5 us, and you actually wanted to record and reproduce those (not filter them out at the first step), how would you digitize them at 44.1 kSps? Last I checked, our common friend Harry Nyquist said you can't - you need 200 kSps.
Last edited:
I do not understand where you are going with this. NOS DACs, or any other DACs, do not create 200 kHz (or whatever) content that is not in the original recording.
You said this while postulating why NOS (and presumably "filterless") sound different/better:
No filtering at 20 kHz does not provide wider-bandwidth signals unless they were in the original recording, which would require sampling at >400 kS/s and then recreated by a >400 kS/s DAC. You seem to be saying audible benefits can be gained by not applying a 20 kHz low-pass filter but that is not true unless the recording (ADC) and playback (DAC) systems have higher sampling rates to begin with. Not filtering at 20 kHz for say a 44.1 or 48 kS/s recording only allows images of the baseband signal through. It does not provide wider-bandwidth transient information.
You said this while postulating why NOS (and presumably "filterless") sound different/better:
No filtering at 20 kHz does not provide wider-bandwidth signals unless they were in the original recording, which would require sampling at >400 kS/s and then recreated by a >400 kS/s DAC. You seem to be saying audible benefits can be gained by not applying a 20 kHz low-pass filter but that is not true unless the recording (ADC) and playback (DAC) systems have higher sampling rates to begin with. Not filtering at 20 kHz for say a 44.1 or 48 kS/s recording only allows images of the baseband signal through. It does not provide wider-bandwidth transient information.
The example may be at 44.1khz, but I don't see the theory of operation being inadmissible for any-sample-rate oriented prefiltering.
- Joined
- Feb 23, 2016
- Messages
- 20,791
- Likes
- 37,693
Yes, Harry is correct.*sigh* Already mentioned above - that part does not address the right issue, as it assumes CD-res-oriented pre-filtering, i.e. intentionally throwing out the 5-us transients before the digital realm is even reached. Of course the sampling rate discussion is moot at that point - you've ended it before it's begun. The real question is: if natural instruments produced harmonics up to 100 kHz (let's say), which would have transients as short as 5 us, and you actually wanted to record and reproduce those (not filter them out at the first step), how would you digitize them at 44.1 kSps? Last I checked, our common friend Harry Nyquist said you can't - you need 200 kSps.
If you think you've found a way to record everything up to 100 khz with a 44 khz sample rate by not filtering the output, then you are making a mistake.
Timing of 5 useconds is easy for 44.1 khz. Bandwidth of signals however isn't going to reach 100 khz waves. For that matter neither do your ears in any way, shape or form. So not recording those signals even though present, your hearing won't detect a difference either way.
Kunchur's research was misguided in several ways. You'll do yourself a favor to disregard it or learn enough about this so you understand why it was misguided. It is leading you astray vs what really happens.
It's not that the theory of operation is inadmissible, it's that it doesn't prevent 5-us transients from being digitized if you know how to do it (200 kSps).
None of you are stating clearly what it is that Kunchur is supposedly wrong about, so I still have no reason to think he's wrong. Apparently neither did those other physics professors who reviewed his papers and accepted them for publication in scientific journals. If you're going to contradict a published physics professor you better have some better argument than "Kunchur is wrong! Kunchur is wrong! Don't believe him! Don't believe him!".
What? You're not making any sense. You just said Nyquist was correct. Then 5 us at 44.1 kSps is not easy, it's impossible. You need 200 kSps.
- Joined
- Feb 23, 2016
- Messages
- 20,791
- Likes
- 37,693
Timing is very possible at 44 khz into the picosecond range. Even within a channel. You are confusing needing bandwidth directly related to timing accuracy. Two different things. An event can be timed much more precisely than the time between samples. Transients that require 200 khz bandwidth are a different matter.It's not that the theory of operation is inadmissible, it's that it doesn't prevent 5-us transients from being digitized if you know how to do it (200 kSps).
None of you are stating clearly what it is that Kunchur is supposedly wrong about, so I still have no reason to think he's wrong. Apparently neither did those other physics professors who reviewed his papers and accepted them for publication in scientific journals. If you're going to contradict a published physics professor you better have some better argument than "Kunchur is wrong! Kunchur is wrong! Don't believe him! Don't believe him!".
What? You're not making any sense. You just said Nyquist was correct. Then 5 us at 44.1 kSps is not easy, it's impossible. You need 200 kSps.
About Kunchur, one thing is he mistook JND's as the smallest level difference that could be heard as different. They aren't that. They are the smallest difference a person can hear and hear that there is a level difference. Smaller differences heard to be the same loudness, can sound like they have qualitatively different sound. I was appalled the review panel didn't catch that. A .25 db level difference in 7 khz square waves will be picked up blind at a high rate of reliability. He used squarewaves and analog filters with the thinking it would tell us if people perceive the ultrasonics. His analog filters altered the level of the fundamental 7 khz tone below the JND, but above the level it would be perceived as different blind. In fact if you look at the change in the fundamental caused by his analog filters it tracks almost perfectly. Around a .2 db difference it is detected, but not by all listeners. Above that it is detected by nearly everyone. And a little above that it is detected pretty much 100%. His test inadvertently was simply testing this, and the odd harmonics had nothing to do with it. It follows what someone knowledgeable with this would have predicted almost perfectly beforehand.
He also had the same misconception that timing accuracy between channels was limited by sample rate. Again, I'm appalled the review panel missed this. I think he later changed some of his wording on this. He then did other tests within one channel, and it was pretty much shown people were hearing IMD when one driver reproduced the main tone and harmonics. Separating that out and replicating the experiment gave null results. I think the fellow who replicated it was named Akashi. Or maybe I'm mis-remembering it.
There were other issues, but I've not read them in some time, and once it is clear the whole set of experiments by Kunchur were bungled, I've wasted no more time on it. You can confirm that what I'm saying is true easily enough.
Similar threads
