Sample rates question

[thermionic]

is there anything to be gained sonically by bouncing to 24-bit 192k files rather than 44.1k?

I would tend to agree that upsampling for the sake of it is a great way to tax your cpu unnecessarily and hog your disc space.

However, standalone "upsampling" DACS have been around in hi-fi for a while, and, as stated, if you re-dither there could be a theoretical advantage if you want to get audiophool about it.

DCS have been selling their upsampling DACS to the hi-fi market for around 10 yrs now, and receiving awards in the process. From what I understand the DAC takes a bitstream from your 44.1 / 16bit player and upsamples / re-dithers to 192 (maybe even higher, long time since I checked, DCS ain't cheap).

Digital is not my forte, so take this with a pinch of salt.

Cheers,
Justin

 

[alk509]

[quote author="thermionic"]From what I understand the DAC takes a bitstream from your 44.1 / 16bit player and upsamples / re-dithers to 192 [/quote]

Dither has nothing to do with sample rate. To understand what happens when you up-sample from 44.1 to 192KHz, see below...

[quote author="JPrisus"]Imagine having a pizza cut into 8 slices. You could cut those 8 slices into 4 new slices each, for a total of 32 slices per pie.[/quote]

Not quite... Most - if not all - SR conversion algorithms don't just copy the samples, but take the relative level of contiguous samples and the rate of change of the signal around those samples and mathematically calculate interim values for the extra ~3.35 sub-samples. And because 44100 is not a multiple of 192000, even your original samples get re-calculated! It's like replacing every single pizza slice with "fake" slices that look like pizza, but are really computer-generated plastic mock-ups...

Now, no algorithm, no matter how advanced, can guess what was really happening to the original analog signal between two samples, after it has already been sampled. Up-sampling algo's can only try to "guess" what the in-between must have looked like, with no way of checking. And up-sampling does not necessarily let you use a gentler smoothing filter after D/A!

[quote author="JPrisus"]And Bouncing to Disk in PT is probably the worst thing you could do to a mix.[/quote]

This is true, but slightly inaccurate... The result of a digital addition (like with "regular" decimal addition) is always the same. In other words, a sample with a hypothetical value of 2, added to another sample with a hypothetical value of 2, will always result in a new sample with a value of 4. But read on...

Like Rochey said, the precision of the processor doing the adding does determine the precision of the resulting samples, and by extension, the quality of a digital mix. It is the Digidesign hardware, rather than ProTools, what determines the precision of mixed samples.

In other words, bouncing to disk in ProTools will give you the same result as bouncing to disk in, say, Logic, provided you use the same hardware to do the mixing. If you're using ProTools LE, which doesn't rely on on-board hardware DSP, the precision of your mixed down samples is determined by your computer's internal FPU.

And then there's some RME card that has something like 84-bit floating point precision!

Anyway, I'm getting dizzy...

Peace,
Al.

 

[Guest]

[quote="alk509] Not quite... Most - if not all - SR conversion algorithms don't just copy the samples, but take the relative level of contiguous samples and the rate of change of the signal around those samples and mathematically calculate interim values for the extra ~3.35 sub-samples. And because 44100 is not a multiple of 192000, [/quote]
Yes,
and then you must decompose it to
multipliing of fractions with small whole numbers.
and this represents as cascade of polyphase multirate filters.
And after it implement it effective.
total hell,
basic literature to creating this creatures is:
Fliege: Multirate digital signal processing.
Just read it.
xvlk

 

[JPrisus]

Thanks for the facts Al!

 

[jrmintz]

I wonder about things like if at 44.1 kHz the Nyquist limit is 22.05k and the brick wall filters start to roll off a few octaves below that, then if you upsample to 192k and the Nyquist limit is 96k and the filters roll off a few octaves below that, would you hear a couple of octaves of overtones that were recorded at 44.1 but filtered out on playback at 44.1? That is, if the kids will turn down the TV?

 

[JPrisus]

[quote author="jrmintz"]I wonder about things like if at 44.1 kHz the Nyquist limit is 22.05k and the brick wall filters start to roll off a few octaves below that, then if you upsample to 192k and the Nyquist limit is 96k and the filters roll off a few octaves below that, would you hear a couple of octaves of overtones that were recorded at 44.1 but filtered out on playback at 44.1? That is, if the kids will turn down the TV?[/quote]

No, because the theoretical information that you were missing at 44.1k playback was never captured to begin with, recording at 44.1k, so even at a higher playback SR, there's nothing to be heard above the 22.05k brickwall filter that was imposed on the source during initial capture.

To sum up everything that's been said in this thread, you can't hear what isn't there to begin with, no matter much you f*ck with it!

 

[Kev]

[quote author="jrmintz"].... in this particular case I'm the client. The buck stops here. [/quote]

ok
if you still need to work on the project and have more tracks to record ... like accoustic guitar or Lead and Group Vocals

:roll:
and have a great accoustic space to work in ...

and need some new inspiration to get the project cooking again ...

I say do it !
Up sample ... get a new converter and start singing again.

Perhaps you will find something and it may mean all new projects will be at the higher sample rates. Remember this is your baby and you have total control.

[quote author="JPrisus"].... I have yet to hear a client request a specific sample rate, but ya never know. Stranger things have certainly happened! [/quote]

It can happen ...
and one of the simple things to look for might be that a musical project is be destined for video and never for CD so the sample rate of choice may be 48 K.

 

[JPrisus]

[quote author="Kev"]

[quote author="JPrisus"].... I have yet to hear a client request a specific sample rate, but ya never know. Stranger things have certainly happened! [/quote]

It can happen ...
and one of the simple things to look for might be that a musical project is be destined for video and never for CD so the sample rate of choice may be 48 K.[/quote]

Good point! hadn't considered that, as it's never come up.

 

[jrmintz]

Thanks JP and Kev and everyone. I think this is fascinating stuff, and it's nice that we can discuss it here without being afraid to ask questions.

you can't hear what isn't there to begin with, no matter much you f*ck with it!

C'mon JP - haven't you had clients who heard good lyrics where there weren't any? Or heard great guitar solos that were never there? Hit songs that were definitely never there to begin with? Some people can definitely hear stuff the rest of us can't! Does anyone make high-rate wishful thinking to reality converters?

:green: :green:

:sam: :guinness:

We need a Jack Daniels emoticon.

 

[JPrisus]

[quote author="jrmintz"]

C'mon JP - haven't you had clients who heard good lyrics where there weren't any? Or heard great guitar solos that were never there? Hit songs that were definitely never there to begin with? Some people can definitely hear stuff the rest of us can't! Does anyone make high-rate wishful thinking to reality converters?
[/quote]

Hahaha, very true! My favorite is when they come in the next day and ask me why something sounds so much better/worse when I haven't touched a damn thing. I think it's dependent on the amount of alcohol consumed by said client the night before :sam:

 

[alk509]

[quote author="xvlk"]Fliege: Multirate digital signal processing.
Just read it.[/quote]

Errr... I have no clue who this Fliege fellow is, my DSP literature collection is limited to Vaidyanathan and a very few others... Now, what exactly is your point? Are you somehow saying that up-sampling does offer benefits over keeping the same SR throughout? Please do explain. (Feel free to reply privately if it's getting too OT).

[quote author="buttachunk"]the sample rate is almost exactly twice the highest frequency reproduced.
So, 44.1k will basically reproduce about 22khz.[/quote]

This is true in principle, but there's an antialias filter (a LPF with a very steep slope) before every A/D converter, and a smoothing filter after D/A conversion. For 44.1 audio, everything a little above ~20KHz is waaaay down there - not to mention severly phase distorted. For all practical purposes, most people just assume that the frequency response of a 44.1 system only goes up to 20KHz.

[quote author="buttachunk"]Thus, the advantage of 192k is more samples and higher frequencies which 44.1k data literally cannot contain.[/quote]

I don't know how true this is, since I can't hear **** above ~18K :sad:... I can, however, hear an improvement in quality between 44.1 and 88.2 or 96 or 192KHz audio, which I attribute to the gentler slope of the above mentioned antialias filter. Does anyone know if a 192KHz AA filter starts rolling off higher than at 44.1 or 48 or whatever? Or does it simply keep the extra bandwidth and hardcore-cuts stuff way the hell up there?

[quote author="buttachunk"]If the material was recorded at 44.1k and then internally raised to 192k, the number would contain repeats since there is no real way to interpolate the non-exsistent map data[/quote]

Actually this is not true: Every SR conversion algo I know of does exactly that: they take contiguous samples, find an average value between the two, figure out the rate of change of the other samples around this point and then they just make **** up!

For example, let's say we get a pseudo-exponentially-raising sample train with these values:

0 - 2 - 5 - 9

... and we want to upsample to twice the sample rate. Doubling up the values looks like this:

0-0-2-2-5-5-9-9

... which makes no sense and defeats the purpose of upsampling in the first place. So what they do is calculate the value of that extra sample so that the resulting sample train looks something like this:

0-1-2-3.6-5-7.7-9

I didn't do any real math here, but notice how they don't just average out the difference between consecutive samples, but look at a bunch of contiguous samples to see how the values are changing and come up with a better approximation of the original analog wave form.

It's smart, but it doesn't quite sound right - especially if you down-sample again to burn a 44.1 CD. This is the main reason why SR conversion (up or down, but specially up) sounds like ass to some people.

Damn, my head hurts...

Al.

 

[thermionic]

[quote author="thermionic"]From what I understand the DAC takes a bitstream from your 44.1 / 16bit player and upsamples / re-dithers to 192 [/quote]

[quote author="Alk509"] Dither has nothing to do with sample rate. To understand what happens when you up-sample from 44.1 to 192KHz, see below...
[/quote]

Forgive me if you thought I suggested sample rate and dither are not totally separate processes, my digital theory is not that bad...

A tutorial is available from this link for anyone reading this discussion that's unsure: http://www.rane.com/note137.html

(upsampling is reportedly voodoo)

Obviously you can't create information if it's not there in the first place, so there probably is a voodoo angle to it. Could upsampling create a "perceptual improvement" on account of finer quantisation resolution even though there is no actual new raw data? :?:

Justin

 

[alk509]

[quote author="buttachunk"]Al, I believe you are thinking of oversampling, not upsampling...[/quote]

No, I'm talking about upsampling.

[quote author="buttachunk"]there is no 5, 7.5, 10 in binary-- there is only 0101010101....[/quote]

Naturally, but I can't read binary so a decimal example is easier to understand. I didn't mean to imply that those decimal numbers were the actual digital representation in zeroes and ones; think of it instead as being voltage levels at each sample point...

[quote author="buttachunk"]oversampling is averaging, upsampling is scaling.[/quote]

Right on. But we're talking about SR conversion, not oversampling. Oversampling is a pretty straight-forward process.

[quote author="buttachunk"]you are right, SR conversion "sounds like ass"-- resampling sounds much better.[/quote]

What do you mean by "resampling"? Going out of the box and back in again??? Then I have to disagree with you... The extra D/A-A/D messes stuff up even worse than SR converting!

[quote author="buttachunk"]you can take an 8 bit 11khz sample of something, and you can upsample / SR rate covert to 24bit 192k, and it will still sound exactly the same (or worse depending on your algorithm)... the only difference is it will only take up much more hard drive space.[/quote]

Just to clarify... Upscaling scales the bit values from 8 to 24 by multiplying by a given factor (2^16). In other words, it just adds 16 zeros at the end of each 8 bit word. But upsampling (i.e., SR conversion) does depend on the algorithm used and invariably results in a different signal representation than you had before. It will always sound worse/different.

[quote author="thermionic"]Obviously you can't create information if it's not there in the first place[/quote]

Exactly. And this is just what SR conversion is all about, which is why it sounds... errrmm..... "different".

[quote author="thermionic"]Could upsampling create a "perceptual improvement" on account of finer quantisation resolution even though there is no actual new raw data?[/quote]

Upsampling has nothing to do with quantization resolution, upscaling does... I'm sure you're question refers to upscaling.

Upsampling = sample rate conversion.
Upscaling = bit depth increase.

There are situations where upsampling is inevitable (try importing a 44.1KHz file into a 96KHz PT session), and there are situations where upscaling is actually advantageous (the extra bits retain more of the low level detail that some DSP processes spit out).

The best alternative is to record at high bit depths to begin with, at the sample rate of your final medium (as long as you're staying inside the box).

Peace,
Al.

 

[TomWaterman]

Here's my humble take on it....

If the audio is tracked at 44.1k then there will be no improvement to the audio by sample rate conversion, especially not the stock PT algorithms....

I think the real improvement comes when running certain plugins at higher sample rates...the filter design in an FIR for example. Plugins such as these should in theory produce a smoother more accurate response when run at 96k etc.....some plugins upsample for this very reason and then sample back down before spitting your audio out of the backend...at the same rate as your session.

By upsampling, the plugins can move their anti-aliasing filters right out of the orignal audible bandwidth and among other things, create a benifit that we could possibly hear.....even for 44.1k recorded audio. The anti-aliasing filter design has a lot to do with this. If the anti-aliasing filters in the ADC and more importantly at this stage, in the plugins, create phase shifts, they could perhaps not sound as good when the session is at 44.1...

I would say that if you could upsample without damaging your audio, then the plugins and processing tools should actually create an improvement...even if your original audio was band-limited to around SR/2 (nyquist) BUT only when actually running the whole session at higher rates and mixing it again.

But again if your going to end up at16 bit 44.1kHz - the improvement might be negligible. I think its best to stick with factors of 2, so go 88.2k to 44.1k or 96k to 48k.....

I don't think there will be any improvement bouncing a master mix from 44.1k to 96k or so....

Cheers Tom

 

[krnt]

Hi.
This is my first post, because most of the subjects discussed here have gone way over my knowledge, but finally there is something I know something about.

So my question is that how about mastering? From what I've heard of some mastering houses use even 192kHz SR, which doesn't make any sense to me. Or have I understood something completely wrong?

--
Antti

 

[JPrisus]

If you mean a mastering house would upsample to 192k, you're very wrong, unless that particular mastering engineer has cotton in his/her ears.

Most likely, what you're saying is that any digital processing takes place at 192k, and that's very possible depending on the mastering engineer's taste. The original source would go through a DAC, then through the engineer's analog path first, then converted to said sample rate and processed digitally. Makes sense, kinda. Very specific to the mastering engineer more than anything else. Might wanna ask this over at Brad Blackwood's PSW forum. You'll get answers from real mastering engineers who do this every day.

 

[TomWaterman]

I think that if mastering guys process analogue then capture that at higher sampling rates before feeding their digital processors then they also hear an improvment of the digital processes at higher rates and IMO it must be a subtantial improvement because the problem here is going back to red-book standard again....you still have to downsample, which will throw away the information....

Cheers Tom

 

[ddt]

Tom W.

I think you hit the nail on the head there with your earlier post...

You've read Bob Katz's book, haven't you? :wink:

d.

 

[bluebird]

Whoops

 

[bluebird]

Al said this...I mean typed it.

What do you mean by "resampling"? Going out of the box and back in again??? Then I have to disagree with you... The extra D/A-A/D messes stuff up even worse than SR converting!

I have a couple questions about this.

I usually take individual tracks out of the box and use my outboard gear (all DIY!) to prosses the signal and record them back in. I record in a one room practice space (no control room) so its hard to use outboard gear on the way in. I use Samplitude in 32 bit floating mode and a Lynx II card. I record at 44.1. I feel this set up is not the best out there but at the high end of whats availible for digital recording.

Two things.

would it be better to use good plugins instead of outboard gear to avoid resampling? I've heard additive EQ with plug-ins is not the best idea.

And If I plan to do alot of "out of the box-back in to the box" recording should I use a higher sampling rate when recording the original material?