Sample Rates Redux

[noisewreck]

As I mentioned in this thread there are instances when higher sampling rates would be useful. Specifically this was in relation to the internal DSP processes that occur in soft-synths and was not dependent and was regardless of AD/DA process, external I/O or any coversion (such as subsequent SRC and bit depth reduction/dithering).

So, this is what I have done:

Created a simple Reaktor instrument that simply outputs Reaktor's built-in primary-level Sawtooth oscillator's output. I then setup a simple sequence in Cubase going from C7 to D#8 in 1/8 note steps. Then exported this sequence to audio twice. One was done at 44.1kHz/32bit float, the other was at 88.2kHz/32bit float. I then dithered both files down to 16 bits, using Cubase's built-in UV22hr dithering, and in the case of 88.2kHz files it was also sampled down to 44.1kHz (using Cubase's built-in feature by saving to 44.1kHz file).

To remove any bias on your part I have renamed files as A.wav and B.wav.

You can download the files here:
A.wav
B.wav

Once you click on the links above, you'll be directed to the download page for the respective file. Click on the "Download" button. It will direct you to a "countdown" page and then will make the download link available to you. You will need to download the files to your computer (I am sorry for that, that's the way 4shared.com operates on free accounts, and I don't feel like paying for it just for this experiment )

Once you have listened to the files, we can start discussing the "whats", "whys" and "hmmms"

Let the games and flames begin!

 

[WhiteStrat]

I thought the drums in "B.wav" were more punchy and forward.

 

[miroslav]

I agree...even the guitars sounded fuller in the B.wav

 

[miroslav]

Aren't they artifacts?

I guess you're a Brit...

Those square waves sound just like the ringing in my ears.

 

[noisewreck]

Well,

I was hopeful that we'd be discussing WHY one of the examples exhibits more aliasing than the other, but I can't say I am surprised that there has been no discussion in that regard whatsoever. It is easy to have idle, half-assed, half-misinformed and wishful conversation about the merits of higher sample rates, analog headroom and all that nonsense, but it's too much work to actually download samples, compare them, listen to them carefully and then offer some insightful comments on the said experiment.

I was hoping this would turn into an educational experience for some of us (including myself), but I guess it isn't meant to be.

Perhaps I didn't construct this in a meaningful way, so if you guys have suggestions on how I can provide a better way of testing that would fuel actual discussion about higher sample rates as far as DSP is concerned, please let me know.

 

[noisewreck]

Aren't they artifacts?

I guess you're a Brit...

Those square waves sound just like the ringing in my ears.

Both spellings are valid. No I am not a Brit, although I am not a US born citizen either, although that has nothing to do with why I spelled it as "artefact".

Although both artifact and artefact are the same thing, I like to put a distiction between the two. For me "artifact" is an archeological find. "Artefact" on the other hand are "unwanted" side-effects of some process. In this particular example, aliasing is an unwanted artefact of generation of a sawtooth wave through DSP process where the process generates more higher frequencies than the available sample rate can handle.

 

[WhiteStrat]

Well,

I was hopeful that we'd be discussing WHY one of the examples exhibits more aliasing than the other, but I can't say I am surprised that there has been no discussion in that regard whatsoever. It is easy to have idle, half-assed, half-misinformed and wishful conversation about the merits of higher sample rates, analog headroom and all that nonsense, but it's too much work to actually download samples, compare them, listen to them carefully and then offer some insightful comments on the said experiment.

I was hoping this would turn into an educational experience for some of us (including myself), but I guess it isn't meant to be.

Perhaps I didn't construct this in a meaningful way, so if you guys have suggestions on how I can provide a better way of testing that would fuel actual discussion about higher sample rates as far as DSP is concerned, please let me know.

Sorry for pissin' on your parade. I downloaded the wav's at the office and couldn't have a good listen. Couldn't tell much of anything in the headphones I keep at my desk. I'm gonna be home tomorrow, so I'll fire up the samples in some good speakers. I'm curious as to what I'm going to find--if my ears find anything, that is.

 

[Greg_L]

Well,

I was hopeful that we'd be discussing WHY one of the examples exhibits more aliasing than the other, but I can't say I am surprised that there has been no discussion in that regard whatsoever. It is easy to have idle, half-assed, half-misinformed and wishful conversation about the merits of higher sample rates, analog headroom and all that nonsense, but it's too much work to actually download samples, compare them, listen to them carefully and then offer some insightful comments on the said experiment.

I was hoping this would turn into an educational experience for some of us (including myself), but I guess it isn't meant to be.

Perhaps I didn't construct this in a meaningful way, so if you guys have suggestions on how I can provide a better way of testing that would fuel actual discussion about higher sample rates as far as DSP is concerned, please let me know.

Lol. Calm down.

 

[Greg_L]

I tried DL'ing, but both files get stuck halfway through the DL, so fuck this shit.

 

[SouthSIDE Glen]

Perhaps I didn't construct this in a meaningful way, so if you guys have suggestions on how I can provide a better way of testing that would fuel actual discussion about higher sample rates as far as DSP is concerned, please let me know.

While disappointing, I wouldn't consider the results of this thread thus far unexpected.

This happens all the time in the MP3 clinic and the M/M forum and elsewhere; the average response to posting audio is not a critical analysis of the engineering, but rather "I like this" or "I don't like that" or "I'd do this different".

So let me try taking things in another direction and ask, George, what is the mechanism behind this? I'm not sure I understand just how Reaktor is generating it's sound, and as to why the DAW's sample rate should make the difference. Is this difference the fault of the sample rate, the DAW software or Reaktor?

I'm not questioning the effect, I just honestly don't understand it's origin.

G.

 

[miroslav]

Honestly, it's not easy/pleasant listing to those high-pitched sawthoot waves, so the possibility of any focused LISTENING didn't work for me.
Basically...they are very annoying to listen to with a critical ear.

That said, I believe the B.wav is the one exhibiting more aliasing, but then, I wouldn't exactly say that it sounds bad or good relative to A.wav...'cuz like I said, sawthoot waves are annoying to listen to begin with!
The B.wav sounded a bit fuller and the A.wav sounded cleaner but thinner.

So there's my "constructive" $0.02...thought I'm not sure it provides the feedback you were after…?

You know…you were kidding around and kinda' bustin chops in a couple of other threads...
…I didn't think a little humor was going to upset you.

 

[noisewreck]

You know…you were kidding around and kinda' bustin chops in a couple of other threads...
…I didn't think a little humor was going to upset you.

I know where you are coming from. And you are right about me kidding around and busting chops in other threads. But at the same time you have to admit intermixed with all that "having fun" or as the Brits say "taking the piss", I also offer quite a bit of insightful information, or at least attempt to

Plus, people bust my chops in the Cave all the time and I have no problem taking it with humor

Had the comments about guitars and other sillyness been interspersed with some actual discussion of the topic at hand, I would likely join in the sillyness and go along with it.

I agree with you that those high-pitched tones can be pretty irritating and I am sorry about that. It's just that it's easier to hear aliasing issues at these higher pitches than the lower ones, which is why I chose these particular notes.

 

[noisewreck]

So let me try taking things in another direction and ask, George, what is the mechanism behind this? I'm not sure I understand just how Reaktor is generating it's sound, and as to why the DAW's sample rate should make the difference. Is this difference the fault of the sample rate, the DAW software or Reaktor?

I'm not questioning the effect, I just honestly don't understand it's origin.

G.

The aliasing occurs within the Sawtooth oscillator itself.

If you create a mathematically correct Sawtooth or any oscillator that is rich in harmonics (thus any oscillator other than Sine wave generators), you are going to have aliasing issues at any sampling rate (yes, even at 192kHz and higher), at least at some higher frequencies. This is because these oscillators, especially Sawtooth and Square wave generators theoretically consist of unlimited amount of partials, which if left unchecked will attempt to go beyond the NyQuist frequency and cause aliasing since there is no way to accurately represent frequencies beyond the NyQuist frequency for a given sample rate. Now, at higher sampling rates this aliasing occurs at higher frequencies, which means the partials that get affected are higher up in the harmonic series. This in turn minimizes the audible effects of aliasing since the higher partials have less energy (amplitude).

This is why, if you listen to the sample I provided, you can hear the notes sound more consistent in timbre in one example compared to the other where a couple of notes sound obviously "off".

Again, it is important to note that when digitally generating mathematically perfect audio oscillators they will alias at any sampling frequency. The aliasing artefacts will be less audible at higher sampling frequencies, because the frequencies that are getting aliased are at lower amplitudes.

The only way around this issue is to veer away from mathematically correct representations of such oscillators, and develop their logic in such a away as to disallow the generation of any partials/harmonics that would go beyound the NyQuist frequency, i.e. using band-limited oscillators. In fact most modern soft-synths, especially the "Virtual Analogs" use this method to take that "digital edge" off of their sound.

This is why I was stating that this type of aliasing is irrespective of the AD/DA process, and in fact it is irrespective of the DAW too. In fact in the case of Reaktor, it is irrespective of Reaktor as well, since it is entirely possible to create band-limited oscillators using Reaktor's lower level modules. In fact at least one person has done exactly this and has contributed his oscillator designs to the Reaktor User Library.

While I understand all this when it comes to oscillators, this is where it starts getting somewhat murky for me:

1. EQ. Often-times when applying hi-shelf EQ on sounds, they tend to become irritating rather than just bright. Conceivably an EQ shouldn't be able to "add" frequencies that aren't already there, but only boost or cut what's already in the audio, so I would like to understand the mechanics of "air" vs. "thin irritation" better.

2. Compressors. By nature, compressors are non-linear processors, in some respects not THAT dissimilar from distortion/waveshaping devices. So, theoretically these CAN add further frequencies (and in fact do for example when using extremely fast attack/decay settings). I wonder if some sort of band-limiting is employed in such devices.

3. Distortion. Which developer pays attention to this and who doesn't? Especially when it comes to analog modelled distortion devices (Guitar Rig, Amplitube) this should be important.

So, if we take these into account, I would think that band-limiting will be necessary with all DSP processes, and especially be important for non-linear functions. Assuming that some developers make some trade-offs in their algorithms, for example band-limiting vs. CPU efficiency, would higher sampling rates then become important when working strictly ITB, irrespective of AD/DA process?

I don't know the answer to this as far as most circumstances are concerned. From synthesis point of view, it MAY be beneficial, but general processing?

 

[noisewreck]

BTW, from some of the comments I've gathered that some of you might need some help in identifying exactly what aliasing sounds like. Rather than go into description, I will post a sample of heavily aliased stuff later on this evening. And Miroslav, I'll try to do it at lower frequencies so it doesn't sound so irritating, although can't say I can take all irritation out of it, since the sound of aliasing itself, in general, is not a pleasant one

 

[miroslav]

OK...post some more...but of the two you posted...which is which???

Wadda you gonna do...keep us in suspense?

And mind you...I only listend to them on my desktop computer speakers, so they may not reveal everything...
...based on that, I described what I heard.

 

[SouthSIDE Glen]

Well, not only are sawtooths about the most irritating of the basic oscillator types, but those are some butt-ugly sawtooths coming out of that software. They're almost as irritating to look at as they are to listen to . Nah, I'm just having some fun; as you know the majority of my experience with oscillators was back in the ARP analog heydays. Those sawtooths were ugly enough , especially when viewed on a cheapo Eico 2Mhz oscilliscope From the local Heathkit outlet. Back then "PC" pretty much only meant "poo chunks" . And I also understand how difficult it must be to get a decent sawtooth wave out of bandwidth-limited digital.

And yeah, I believe (though we are at the edge of my understanding here, so it's kinda wobbly) that pretty much everything digitally generated has to be bandwidth limited; Nyquest demands it, so I think it all hinges pretty much upon liming the bandwidth. I should have picked up on that before I asked the question, I guess I just didn't pick up on the idea that Reaktor generated at different sample rates itself. I should have, that's sounds pretty obvious now; I guess I still have just too much of the old analog thinking ingrained in me when it comes to waveform synthesis.

Oh, and BTW, I believe it's "A" that has the heaviest aliasing artifacts, and therefore, based upon your description, the lower original sample rate.

G.

 

[noisewreck]

And yeah, I believe (though we are at the edge of my understanding here, so it's kinda wobbly) that pretty much everything digitally generated has to be bandwidth limited; Nyquest demands it, so I think it all hinges pretty much upon liming the bandwidth.

Oh, and BTW, I believe it's "A" that has the heaviest aliasing artifacts, and therefore, based upon your description, the lower original sample rate.

G.

So, having said that, then the question becomes how well band-limiting is implemented in other processing software such as EQs, Compressors, etc. The reason I ask this is because I have noticed that when I put a frequency analyzer after some some processors (for example UAD 1176LN compressor), there is a noticable bump in the sub-30Hz area, increasing all the way down to whatever the system can handle/show (10Hz or so). In my experience, this low frequency bump is a direct result of aliasing, although I may be wrong, and this bump may attributable to other factors that I am unaware of.

And yes, example A.wav was the one recorded at 44.1kHz.

 

[SouthSIDE Glen]

So, having said that, then the question becomes how well band-limiting is implemented in other processing software such as EQs, Compressors, etc. The reason I ask this is because I have noticed that when I put a frequency analyzer after some some processors (for example UAD 1176LN compressor), there is a noticable bump in the sub-30Hz area, increasing all the way down to whatever the system can handle/show (10Hz or so). In my experience, this low frequency bump is a direct result of aliasing, although I may be wrong, and this bump may attributable to other factors that I am unaware of.

I know enough to understand your question, but nowhere near enough to venture an answer, to be honest. It's an interesting question, though.

And yes, example A.wav was the one recorded at 44.1kHz.

Do I get a kewpie doll? .

@miroslav; when you're not absolutely sure - which is easy when it's an irregular wave for which you haven't heard the un-aliased source - your eyes can help :

G.

 

[Bguzaldo]

Woo Hoo! I don't know what's going on but my ears did not lie to me! And I listened on laptop speakers. I'm having a hard time following with all the gizmo talk, but I'm trying seeing as this is an area that somewhat concerns me. I've even noticed this effect in some of my older digital recordings.

So does a higher sample rate cut down on these artefacts?

Thanks for doin this,
-Barrett

 

[noisewreck]

I tried DL'ing, but both files get stuck halfway through the DL, so fuck this shit.

That's OK, it's not supposed to work in TX