24/96khz

[SWAN808]

can someone correct me if Im wrong - but the way Ive come to understand it was that the process of recording audio at 44.1 and playing it back - is different from processing audio at 44.1 once inside the box...?

So there are 2 issues - recording and playback

and DSP processing at a particular sample rate

So my own personal understanding at the moment is that for recording and playback - 44.1 is generally fine - because of the arguments made by the NQ theroum.
However from what Ive heard and gathered from developer comments on the topic (eg useful input from Sean Costello on this very htread)...moving aliasing generated by processing (compression, saturation etc) out of the audible range - by running at 96k - can be beneficial. There is also the issue of some software instruments which may produce aliasing depending on how they have been coded. This can also be moved into the inaudible range.

Oversampling is an option. However it seems in theory it might be better to avoid realtime oversampling (for its own pitfalls) and just run the whole session at 96 if able-then downsample offline using HQ SRC. Id love to hear a test of oversampling vs 96k but not seen anyone do it.

Thats where Im at with it at the moment - but open to movement...

[eidenk]

But then why not running the DAW at 96khz with less or no need to use oversampling for those effects and thus save CPU and avoid as much as possible any kind of possible degradation of the sound due to multiple resamplings all over the processing chains? There is nothing wrong with that reasoning or is there?

Is that a joke question?

Seriously?

Yes at 96khz oversampling is mostly not needed but:

You do realize that if you use plug at 96khz with no oversampling it will use same amount of cpu as it is using when you are driving it at 48khz + 2x oversampling ?!?!! So there isn't any saving of CPU or anything like that..

I agree that when you run a 2x oversampled plugin at 48khz it should more or less consume the same amount as running it non oversampled at 96khz for processing the audio but then you've forgotten that CPU cycles are also needed to upsample in and downsample out which is what you save.

[Dean Aka Nekro]

Since Dan Lavry makes some of what are IMveryhumbleHO the best AD and DA converters, His papers on the matter are good reading and gives his reasoning why he does not build any that operate above 24-bit/96kHz:

http://www.lavryengineering.com/lavry-white-papers/

Covers pretty much everything except DSP processing at lower and/or higher sampling rates. Bar some spelling mistakes they are easy to digest and he doesn't mince his words much

All the best to all as always

Dean

[loopdon]

Since Dan Lavry makes some of what are IMveryhumbleHO the best AD and DA converters, His papers on the matter are good reading and gives his reasoning why he does not build any that operate above 24-bit/96kHz:

http://www.lavryengineering.com/lavry-white-papers/

Covers pretty much everything except DSP processing at lower and/or higher sampling rates. Bar some spelling mistakes they are easy to digest and he doesn't mince his words much

All the best to all as always

Dean

Dean, thanks!

[BertKoor]

can someone correct me if Im wrong - but the way Ive come to understand it was that the process of recording audio at 44.1 and playing it back - is different from processing audio at 44.1 once inside the box...?

The way you describe it, I see not a single problem or difference there. Probably I don't fully understand what you wrote...

However from what Ive heard and gathered from developer comments on the topic (eg useful input from Sean Costello on this very htread)...moving aliasing generated by processing (compression, saturation etc) out of the audible range - by running at 96k - can be beneficial. There is also the issue of some software instruments which may produce aliasing depending on how they have been coded. This can also be moved into the inaudible range.

Correct, and that's exactly why some plugins do oversampling themselves or offer a high-quality mode which can be implemented by internal oversampling.

Listening tests were conducted in times far gone by, see the link on one of the first pages. The result was inconclusive: half the people liked the gritty aliasing sound

[dreamkeeper]

can someone correct me if Im wrong - but the way Ive come to understand it was that the process of recording audio at 44.1 and playing it back - is different from processing audio at 44.1 once inside the box...?

So there are 2 issues - recording and playback

and DSP processing at a particular sample rate

So my own personal understanding at the moment is that for recording and playback - 44.1 is generally fine - because of the arguments made by the NQ theroum.
However from what Ive heard and gathered from developer comments on the topic (eg useful input from Sean Costello on this very htread)...moving aliasing generated by processing (compression, saturation etc) out of the audible range - by running at 96k - can be beneficial. There is also the issue of some software instruments which may produce aliasing depending on how they have been coded. This can also be moved into the inaudible range.

Yes, that's how I understand it as well. And in every debate I've seen so far, many people don't distinguish between recording and processing. That, along with misconceptions about Nyquist and the basics of digital audio in general always lead to confusion. Oh, and many people seem to not understand what aliasing is either.

As an aside, when advocating recording at higher rates, people tend to forget that there are not many (affordable) mics that pick up content above 20kHz.

Oversampling is an option. However it seems in theory it might be better to avoid realtime oversampling (for its own pitfalls) and just run the whole session at 96 if able-then downsample offline using HQ SRC. Id love to hear a test of oversampling vs 96k but not seen anyone do it.

There was some discussion about that in the thread a year ago. See the quote below and the following posts by Urs:

http://www.kvraudio.com/forum/viewtopic ... 31#4698431

The problem is that the added harmonics in a signal chain add to each other. That is, a nonlinear process will also add harmonics to the already added ones from the previous process. So each time you double the samplerate all you get is one more plugin in the signal chain to end up with the same level of artifacts.

The best way to keep aliasing low is to bandlimit each single process in the chain. Thus deploying steep filters inbetween. 'tis the only way.

I think, the same problems may occur if those high frequencies are in the recording already.

[dreamkeeper]

can someone correct me if Im wrong - but the way Ive come to understand it was that the process of recording audio at 44.1 and playing it back - is different from processing audio at 44.1 once inside the box...?

The way you describe it, I see not a single problem or difference there. Probably I don't fully understand what you wrote...

That's pretty simple: it means, when it comes to possible benefits of higher samplerates, that there is a difference between the AD/DA conversion (record/playback) and the processing in between (in the box).

[BertKoor]

OK, capice. Thanks for the "translation".

[Dean Aka Nekro]

Since Dan Lavry makes some of what are IMveryhumbleHO the best AD and DA converters, His papers on the matter are good reading and gives his reasoning why he does not build any that operate above 24-bit/96kHz:

http://www.lavryengineering.com/lavry-white-papers/

Covers pretty much everything except DSP processing at lower and/or higher sampling rates. Bar some spelling mistakes they are easy to digest and he doesn't mince his words much

All the best to all as always

Dean

Dean, thanks!

Anytime/happy to help/be of some use now and again loopdon

Now if only I could just afford a set of Lavry AD and DA conveters for my home setup Not going to happen anytime soon!

[whyterabbyt]

However from what Ive heard and gathered from developer comments on the topic (eg useful input from Sean Costello on this very htread)...moving aliasing generated by processing (compression, saturation etc) out of the audible range - by running at 96k - can be beneficial. There is also the issue of some software instruments which may produce aliasing depending on how they have been coded. This can also be moved into the inaudible range.

Now, I might just cock this up here, but Im sure doubling the sample rate doesnt 'move aliasing' out of the audible range; it moves the threshold at which aliasing occurs up an octave.

But aliasing artefacts are a sort of 'foldover'; frequencies which exceed the Nyquist limit get 'reflected' back into the the audible range. That still happens whatever your Nyquist frequency is. At a sample rate of 44Khz, they get 'reflected' back into the 0-20kHz range, and all of them wind up the audible range. At a sample rate of 96Khz, they get 'reflected' back into the 0-40Khz range, and 'half of them' wind up in the audible range. But you still potentially have them across your entire audible range.

(Consider that the harmonic content of a 15kHz square wave being PWM'd exceeds 44Khz by the 4th partial...)

Maybe that's why oversampling on the more 'modelled' stuff goes up much higher than just doubling; you see up to x16 and such... because a 'mere' doubling of the sample rate isnt actually enough for those cases. And which is why I suspect that in many more cases than people realise the issue of downsampling filters using 'extra' CPU or causing phase issues etc isnt something which is 'bad' at 44.1Kz and 'doesnt happen' at 96Khz. 96KhzKhz 'saves you' one doubling, out of 3 or 4...

(Its been a long day so Im a bit dottled, and if anyone wants to pull this apart, feel free.)

[jupiter8]

Yes, in D/A converter the waveform is recreated. But consider that you do ALL your complex/multitude DSP and mixing with the low precision "6 sample" example instead of higher precision (unless DSP uses oversampling of course).
At D/A stage you can't recover the benefits of using higher precision in the first place.

It does not matter. Higher sample rate isn't higher precision. It might seem counterintuitive but all a higher sample rate does is capture higher frequencies.
The Sampling Theorem clearly states that the waveform can be 100% accurately recreated. A higher sample rate will not result in a more accurate waveform whether it's been processed or not. You're just wrong but in your defense it is the most common misunderstanding about digital audio.

[dreamkeeper]

However from what Ive heard and gathered from developer comments on the topic (eg useful input from Sean Costello on this very htread)...moving aliasing generated by processing (compression, saturation etc) out of the audible range - by running at 96k - can be beneficial. There is also the issue of some software instruments which may produce aliasing depending on how they have been coded. This can also be moved into the inaudible range.

Now, I might just cock this up here, but Im sure doubling the sample rate doesnt 'move aliasing' out of the audible range; it moves the threshold at which aliasing occurs up an octave.

A tad picky... but correct.

But aliasing artefacts are a sort of 'foldover'; frequencies which exceed the Nyquist limit get 'reflected' back into the the audible range. That still happens whatever your Nyquist frequency is. At a sample rate of 44Khz, they get 'reflected' back into the 0-20kHz range, and all of them wind up the audible range. At a sample rate of 96Khz, they get 'reflected' back into the 0-40Khz range, and 'half of them' wind up in the audible range. But you still potentially have them across your entire audible range.

Apropos picky: that would be 22 and 48 respectively.

[whyterabbyt]

However from what Ive heard and gathered from developer comments on the topic (eg useful input from Sean Costello on this very htread)...moving aliasing generated by processing (compression, saturation etc) out of the audible range - by running at 96k - can be beneficial. There is also the issue of some software instruments which may produce aliasing depending on how they have been coded. This can also be moved into the inaudible range.

Now, I might just cock this up here, but Im sure doubling the sample rate doesnt 'move aliasing' out of the audible range; it moves the threshold at which aliasing occurs up an octave.

A tad picky... but correct.

But aliasing artefacts are a sort of 'foldover'; frequencies which exceed the Nyquist limit get 'reflected' back into the the audible range. That still happens whatever your Nyquist frequency is. At a sample rate of 44Khz, they get 'reflected' back into the 0-20kHz range, and all of them wind up the audible range. At a sample rate of 96Khz, they get 'reflected' back into the 0-40Khz range, and 'half of them' wind up in the audible range. But you still potentially have them across your entire audible range.

Apropos picky: that would be 22 and 48 respectively.

Yup, 'twere crude aproximations only.

(Although Im sure Ive read that although Nyquist states you need to use a sample rate of twice the maximum frequency you want to represent, in reality you're actually better working with a sample rate of 2.2* the maximum frequency you want, because of, hmm, downsampling filter slopes or something? I plead dottled, again. Hence 44.1Khz. That may be apocryphal though, Ive never entirely been sure. )

[S0lo]

But isn't all the range from 20Khz and above are completely inaudible. The human ear simply can't notice it. All 96khz is doing is capturing an extra range from 22khz to 32Khz which is all practically silence to the ear.

IMO, benefits of higher sample rates doesn't come solely from increased frequency range. There is also increased precision for DSP and summing process in audible range. For example, with 44.1K sample rate you can have less than 6 samples available to form one cycle of 8000 Hz sine wave. With 96K you have 12 samples.

But according to Nyquist theorem (if I understand it well), you only need those 6 samples to recreate the original waveform EXACTLY as it was, no loss.

Exactly. Nonsense argument.

Yes, in D/A converter the waveform is recreated. But consider that you do ALL your complex/multitude DSP and mixing with the low precision "6 sample" example instead of higher precision (unless DSP uses oversampling of course).
At D/A stage you can't recover the benefits of using higher precision in the first place.

hmm, so your saying for example, Sampling two different signals at lower rate (44.1Khz), mixing them (addition) together digitally then sending the sum to the output will result in a different signal than if this exact process was done at higher rate. (I mean different bellow 22Khz)

If thats what you meant, then let me rephrase this. If we mix two different signals totally in Analog domain, and then we mix the same two signals in digital domain (i.e. sample at 44.1Khz (ADC), add (no clipping), reconstruct (DAC), output). Then this will result in two sum signals that are different bellow 22Khz.

If thats what you meant, then I'm not sure, but I doubt. Unless clipping happens at some stage in the digital process, I think you will get the same final waveform if you look bellow 22Khz. And even if clipping happens, it will not be caused by the lower sampling rate, and it will not be fixed by a higher sampling rate. The two added samples are above the peak value, weather you sample more or less. Notice here that I'm assuming ideal hardware as proposed by Nyquist, I'm neglecting circuit error which is always there and has nothing to do with the sampling rate.

If I assume your right, then doing mixing digitally for like 10000 signals or more (Without clipping) should finally sound very different than the 10000 signals mixed in analog, because the small differences would add up, right?

[dreamkeeper]

Yup, 'twere crude aproximations only.

Yeah, figured that...

(Although Im sure Ive read that although Nyquist states you need to use a sample rate of twice the maximum frequency you want to represent, in reality you're actually better working with a sample rate of 2.2* the maximum frequency you want, because of, hmm, downsampling filter slopes or something? I plead dottled, again. Hence 44.1Khz. That may be apocryphal though, Ive never entirely been sure. )

Interesting question. I'm not an expert, but I'd be surprised if the required margin for clean filtering would depend on the samplerate. IOW, if 2kHz are enough at 44.1kHz, it should be enough at 96kHz (or whatever) too? Maybe a dev could clear that up?