Best No oversampling dac to buy??

[Blumlein 88]

Perhaps I am mentally conflating aliasing in the ADC versus DAC...evinced further by my total mental inability to now at this point remember why the heck there is a lowpass filter...
(Somehow I was thinking that if there was not ultrasonics really getting into the recording in the first place then lack of a lowpass filter like in an NOS design doesn't really matter much).

Maybe learn about aliasing (an ADC thing) and imaging (a DAC thing). Then it might make sense you need low pass filters on both ends. Or better yet, watch this cool video which has been cited here like hundreds of times. By Monty Montgomery. Watch a few times, digest what it says, ask questions after you have and you'll learn so much about digital audio.

 

[srkbear]

What I meant was transient attack
At sampling rate of 192 kHz, this moves the Nyquist frequency to 96 kHz and that helps a lot.

Where are you getting this information from? Obviously it’s not consistent with the accepted science on how digital audio works, as you can see from the reactions you’re receiving here. You just joined a week ago—are you here to learn? That’s why I joined. You seem to be ignoring the feedback others are offering you to put you on the right track, which is why the reactions are getting increasingly comical and derisive.

Instead of persisting in defending your point, if you adjust your strategy to show curiosity and ask for help, I think you’ll enjoy the site and find its members very generous and knowledgeable. Your current approach is going splat. Try another, it’s worth it! Peace…

 

[dc99]

Where are you getting this information from? Obviously it’s not consistent with the accepted science on how digital audio works, as you can see from the reactions you’re receiving here. You just joined a week ago—are you here to learn? That’s why I joined. You seem to be ignoring the feedback others are offering you to put you on the right track, which is why the reactions are getting increasingly comical and derisive.

Instead of persisting in defending your point, if you adjust your strategy to show curiosity and ask for help, I think you’ll enjoy the site and find its members very generous and knowledgeable. Your current approach is going splat. Try another, it’s worth it! Peace…

maybe you can’t, but I can definitely hear the reconstruction filter on the delta-sigma type DAC destroying sound quality. If you don’t believe me, listen to some great piano sonatas, and hear the piano hammers hitting hard on the strings, specially on the high notes.

 

[srkbear]

maybe you can’t, but I can definitely hear the reconstruction filter on the delta-sigma type DAC destroying sound quality. If you don’t believe me, listen to some great piano sonatas, and hear the piano hammers hitting hard on the strings, specially on the high notes.

Oh right! Because I’ve never listened to piano sonatas before. Thank goodness you’re here!

I tried.

 

[Prana Ferox]

Gotta love a resurrected zombie thread.

The thread had moved on from the perceptible AudioScienceReview passband but something caused it to ring back audibly

 

[AnalogSteph]

Wonky digital filters with concerning levels of periodic passband ripple (and associated pre-echo if FIR) do exist but they are quite rare these days. You might find them in older oddball onboard audio chips and the like, the IDT job in my Dell Latitude E6330 is such a case (+/- 0.1 dB). The issue can be mitigated by using software upsampling and a higher DAC sample rate then (the SoX resampler DSP for Foobar and 192 kHz in this case; things are virtually back to normal by 96 kHz already). This moves the pre-echo closer in time so it's more easily masked.

This should be a complete non-issue in any DAC worth its salt. Your average Realtek audio codec even has lower periodic ripple than some dedicated chips, and none of those exhibit concerning levels either, even if up to +/-0.005 dB still is a tad higher than I (or Julian Dunn) would really like to see. I guess designers are also concerned about latency implications.

Potential issues with intersample-overs are avoided easily enough these days, too (leave a few dB of headroom in the playback chain / bring down hot tracks with volume normalization).

Yes, you can get away with only analog filtering when running a NOS DAC at quad speed. It still needs to be fairly steep though. You've got about 2.8 octaves (tops) for some decent attenuation, so unless you pull the Bessel card and pop a notch filter in there, that's roughly a 4th-order lowpass. Still much better than the monstrosities they had to use back in the NOS days, mind you (I remember seeing something like an 11th or 13th-order Bessel lowpass filter being mentioned, yikes). That being said, a lot of the DACs used for NOS are not exactly state of the art performance-wise.

 

[voodooless]

What I meant was transient attack

You do realize that “transients” in bass aren’t really transients? Bass is low frequency, low frequencies are slow changes. Even the overtones aren’t that fast. Bass is not a square wave. We have $ 139 DACs reconstructing multI-tones like this:

This would look exactly the same at 44.1 kHz sampling. There is zero difference.

maybe you can’t, but I can definitely hear the reconstruction filter on the delta-sigma type DAC destroying sound quality

Compared to what?

 

[Head_Unit]

Then it might make sense you need low pass filters on both ends.

Well if there was zero actual sound energy to be filtered, it wouldn't matter. The reconstruction filter is to prevent aliasing, where energy above the cutoff frequency would get folded back down into audio frequencies making the sound yucky.

My question should be rephrased as "IS THERE enough energy in actual recording situations that it would matter if you removed that reconstruction filter and its allegedly deleterious effects?" I've never seen a measurement of such. If someone could measure and post data it would be very cool!* How to do so accurately...???

At say 96 kHz with a 48 kHz cutoff, how much energy above 48 kHz is even in the studio? That I have no idea. But recording microphones, my recollection of their specs is they die before 48k anyway. And and, even if there was energy up to 96k a dim corner of my time-to-go-to-bed brain says those frequencies would alias down into the ultrasonic range anyway and not be audible.

*some searching turned up https://hal.science/jpa-00230587/document whose data would seem to indicate very low ultrasonics but that is a significant extrapolation, not measured.
This https://www.mdpi.com/2624-599X/1/4/48 suggests recording in public places *might* run into problems...mmm...

 

[Head_Unit]

What I meant was transient attack

You do realize that “transients” in bass aren’t really transients? Bass is low frequency, low frequencies are slow changes.

Ah this puts me in mind of my second interview fresh out of college for a loudspeaker design position. Got into a bit of an argument about "bass transient response" with the hiring manager, who said exactly the second quote. I replied in a bit of an eye-winking way "Weeelllll, yes, yes, BUT YOU KNOW WHAT I MEAN." He laughed and said "OK, yeah."
And hired me
I think the quality that gets labeled as "fast" or "transient" probably has more to do with midbass response of the system in the room, and also possibly the post-ringing of bass notes. Because hey all bass notes may be "slow" but some systems DO sound "tighter" than others...

 

[voodooless]

I think the quality that gets labeled as "fast" or "transient" probably has more to do with midbass response of the system in the room, and also possibly the post-ringing of bass notes. Because hey all bass notes may be "slow" but some systems DO sound "tighter" than others...

That may be, but none of this has anything to do with reconstruction filters.

 

[Blumlein 88]

Well if there was zero actual sound energy to be filtered, it wouldn't matter. The reconstruction filter is to prevent aliasing, where energy above the cutoff frequency would get folded back down into audio frequencies making the sound yucky.

My question should be rephrased as "IS THERE enough energy in actual recording situations that it would matter if you removed that reconstruction filter and its allegedly deleterious effects?" I've never seen a measurement of such. If someone could measure and post data it would be very cool!* How to do so accurately...???

At say 96 kHz with a 48 kHz cutoff, how much energy above 48 kHz is even in the studio? That I have no idea. But recording microphones, my recollection of their specs is they die before 48k anyway. And and, even if there was energy up to 96k a dim corner of my time-to-go-to-bed brain says those frequencies would alias down into the ultrasonic range anyway and not be audible.

*some searching turned up https://hal.science/jpa-00230587/document whose data would seem to indicate very low ultrasonics but that is a significant extrapolation, not measured.
This https://www.mdpi.com/2624-599X/1/4/48 suggests recording in public places *might* run into problems...mmm...

The reconstruction filter is on the DAC. It is to prevent imaging. Nothing on the output folds back down into the audible range.

An anti-aliasing filter is on the ADC to prevent aliasing obviously. Here yes, ultrasonics can fold back into the audible range.

Imaging is when you get mirror imaged output above half the sample rate. For instance your bass frequencies would show up just below 48 khz with 96 khz sampling without a reconstruction or anti-imaging filter. Maybe it is a problem, maybe not. Why build a DAC this way as I don't see any benefit?

On the ADC end when recording you are risking high frequencies that would alias down into the audible range. One more time, why risk this issue when I see no benefit?

Probably would get away without a filter with a high sample rate on the DAC better than on the ADC. But to no benefit. People make filterless DACs and certainly your ears, and probably your speakers are an effective enough filter for the most part. It is a broken design however. Or as I like to call them broken as designed.

 

[Hayabusa]

While I agree that this is most likely the best NOS option, it still sucks in NOS-mode (by design):

That's around 3 dB down by 15 kHz, 5 dB by 20 kHz [source / review].

@KehaDNb If you really like NOS DACs, I suspect that this is the reason. Get any other competent oversampling DAC, dial in that frequency response by reducing treble by ~3 dB at 15 kHz and then compare it to your current setup. Preferably volume-matched and blind. Tell us about your results.

its more 3dB down at 20Khz.. don't over estimate
Also these curves are for 48KHz..

 

[IAtaman]

Well if there was zero actual sound energy to be filtered, it wouldn't matter. The reconstruction filter is to prevent aliasing, where energy above the cutoff frequency would get folded back down into audio frequencies making the sound yucky.

My question should be rephrased as "IS THERE enough energy in actual recording situations that it would matter if you removed that reconstruction filter and its allegedly deleterious effects?" I've never seen a measurement of such. If someone could measure and post data it would be very cool!* How to do so accurately...???

At say 96 kHz with a 48 kHz cutoff, how much energy above 48 kHz is even in the studio? That I have no idea. But recording microphones, my recollection of their specs is they die before 48k anyway. And and, even if there was energy up to 96k a dim corner of my time-to-go-to-bed brain says those frequencies would alias down into the ultrasonic range anyway and not be audible.

*some searching turned up https://hal.science/jpa-00230587/document whose data would seem to indicate very low ultrasonics but that is a significant extrapolation, not measured.
This https://www.mdpi.com/2624-599X/1/4/48 suggests recording in public places *might* run into problems...mmm...

If you look at the samples that is supposed to represent, say a 5KHz tone, you can only guarantee that samples represent 5KHz if the signal is band limited. If it is not, it might be samples from a 25KHz signal as well. So even if the HF is not in the original recording, oversampling DAC would produce it anyway, wouldn't it?

 

[Killingbeans]

maybe you can’t, but I can definitely hear the reconstruction filter on the delta-sigma type DAC destroying sound quality. If you don’t believe me, listen to some great piano sonatas, and hear the piano hammers hitting hard on the strings, specially on the high notes.

Assuming it's not placebo, are you absolutely sure you aren't just hearing differences in roll-off?

 

[srkbear]

The reconstruction filter is on the DAC. It is to prevent imaging. Nothing on the output folds back down into the audible range.

An anti-aliasing filter is on the ADC to prevent aliasing obviously. Here yes, ultrasonics can fold back into the audible range.

Imaging is when you get mirror imaged output above half the sample rate. For instance your bass frequencies would show up just below 48 khz with 96 khz sampling without a reconstruction or anti-imaging filter. Maybe it is a problem, maybe not. Why build a DAC this way as I don't see any benefit?

On the ADC end when recording you are risking high frequencies that would alias down into the audible range. One more time, why risk this issue when I see no benefit?

Probably would get away without a filter with a high sample rate on the DAC better than on the ADC. But to no benefit. People make filterless DACs and certainly your ears, and probably your speakers are an effective enough filter for the most part. It is a broken design however. Or as I like to call them broken as designed.

Yes I think he’s mistaking the ADC anti-aliasing filter with the post reconstruction filters. And thus assuming that upsampling causes aliasing, instead of preventing it.

 

[srkbear]

Assuming it's no placebo, are you absolutely sure you aren't just hearing differences in roll-off?

When you have the science around backwards out of the gate, cognitive bias seems unavoidable. I don’t think there’s hope of convincing him until he surrenders to the basics personally.

 

[oleg87]

What I meant was transient attack
At sampling rate of 192 kHz, this moves the Nyquist frequency to 96 kHz and that helps a lot.

I know its temping to look at pre-ringing in an impulse response and think it must do something similar to transients in music, but that is not the case. Your piano, like all band-limited signals, will come out of an oversampling DAC quite unmolested.

 

[dc99]

What I’m trying to say is that when you feed a Hi-Res signal (96/192) into a DAC, IME, it’s better to choose a NOS filter, to get the best sound, and better transient attack. (The Nyquist freq for 192 is 96, way above our human hearing range, I doubt anyone can hear such microdynamics).
I agree with you, when listening to 44.1/48kHz source, and AK or ESS upsamples your signal, then the linear-phase, sharp roll-off filter is technically the most accurate. But this is not what I am talking about.
My point, as more and more music are available in Hi-Res, I think NOS DAC will become popular again.

 

[IAtaman]

What I’m trying to say is that when you feed a Hi-Res signal (96/192) into a DAC, IME, it’s better to choose a NOS filter, to get the best sound, and better transient attack. (The Nyquist freq for 192 is 96, way above our human hearing range, I doubt anyone can hear such microdynamics).
I agree with you, when listening to 44.1/48kHz source, and AK or ESS upsamples your signal, then the linear-phase, sharp roll-off filter is technically the most accurate. But this is not what I am talking about.
My point, as more and more music are available in Hi-Res, I think NOS DAC will become popular again.

The only advantage higher sampling rate offers is that it makes it possible to accurately sample and recreate higher frequencies. If you agree that you can not hear frequencies >20KHz, what advantage do you think 192KHz sampling rate might offer?

 

[voodooless]

What I’m trying to say is that when you feed a Hi-Res signal (96/192) into a DAC, IME, it’s better to choose a NOS filter, to get the best sound, and better transient attack

No, NOS, or sample-and-hold is broken by default. It does not give you better reproduction of the original, nor does it have better transients. See @oleg87 here. A higher sample rate will only lessen the issues, not cure them.