Aliasing in synths. How to prevent it?

[aciddose]

No it will be easy to test. Try my "clipper" plugin from the Xhip site (effects) and play a sine with oversampling disabled as I did in the example above with the spectrum graphs and .wav clip. You could hear that easy, right?

If you can't easily tell the difference, don't bother. I'm never talking about small differences that you need to have a hard time to identify when I'm talking about aliasing.

I'm always talking about really horrible dissonance and noise that bugs me so much I'm almost ready to jump out of my skin.

[nyarfautca]

Hello Guys. First post but long time reader.
I've been thinking of this sample rate thing for a very long time and now here is my big dilemma, hoping for some help:

First of all I make instrumental music in the box, no HW instruments, no recording stuff, YET. Let's say I do agree that higher sample rate is better for SW instruments and (IMHO) even more for effects. No problemo at all when I use just AUs or VSTis. I start to worry when I'm using audio files from sample CDs or sound banks which is in 44.1khz originally. Someone on another forum suggested using the final sample rate in your projects in this case due to the fact more conversion in the end = more chance fore unwanted artifacts.

Do you guys know (who use higher sample rate in case of the final product 44.1khz) what is the best solution?

1.) If you place audio in the project, first sell I convert all my audio to (let's say) 48 or 96, set the project to 48 or 96 (or 88.2 now it doesn't matter - however one say it's better to double), so work in the project with the higher SR and when I finished, convert everything down after the rendering work?

2.) Or for instance I set the project to higher SR but don't want to convert all my sample content. Instead I might just load them into samplers like Kontakt, or Ableton Live in rewire (to Logic or Cubase) and then Kontakt maybe has some inbuilt oversampling feature?

I don't know how others do this. Lemme tell you sg. I'm a big fan of mau5 the dead the godfather. This is really offline but to me his music is the most relevant in terms of sound so obviously I try to follow his techniques and advises. He has a live stream where you can catch some useful stuff. Once he mentioned in a tweet that it's so worth to work in higher SR and it was an Ableton or Bitwig (donremember) project set to 192khz.
But I know he records his modulars and all the HW things. So maybe he use it just for the recording or maybe not. Maybe He works with that during the project. But I know he uses samples as well from his own Xfer bank sometimes from Vengeance what I own and I know they are 44.1khz. And most of the times he loads the samples into Ableton's sampler. This is.. for sure.

I would like to know how he set up sample rate. Haha I know I'm not gonna get this answer here. But If you guys know any professional how does this please explain me. Thanksss

[aciddose]

Like I've said, it depends upon the particularities of the situation you're working in.

You must decide on a case-by-case basis based upon evidence whether one choice is better than another.

It isn't possible to make a general rule and follow it without thinking about it and investing the effort to make the right decision.

Sorry!

[nyarfautca]

Yeah I see. But I was asking about the audio files. 44.1khz sample content. There is a common rutin for sure. Upsample your sampels and use them like that and then after rendering convert back. Or leave them on 44.1 and load them into samplers in the project. I am just guessing some samplers will upsample them automatically. Any practical thoughts..

The way I do it these days: leaving the project on 44.1 (cause this will be the final SR so don't have to deal with conversion) but using just synth amd effects which can do internal ovrsampling...

[kx77free]

Hi,

We can do VA with the main processes oversampled with a good sounding at 44.1k but with a mix using complex patches or a lot of rich pads with reverb, it's better to increase the DAW FS when you record your mix .
Even if the VA use oversampled processes that will be always better to use the DAW FS upper than 44.1k or 48k.

Because each modulation could generate aliasing, which depends of the signals used and theirs frequencies.
For example a saw LFO (no BLT) to 200 hz generates aliasing, not if it's a sine LFO.
In DSP audio to do a simple ring modulator, we use a simple multiplication, x*y.
Each multiplied signal is potentially like a VCA based ring modulator. Generally the aliasing generated on complex patch is very low, not really audible but with a lot of modulations...
The other thing is that the oversampled signal quality is deteriorated by the filters used to eliminate the artefacts when the up or down frequency of the internal FS. It's like a SD movie to HD screen, you search always the focus!
If you want to do HD movies you will not do your rushes in SD!

You must think about that:
44.1k/10k=4 -> 4 samples for 10khz!
So the high frequencies quality at 44.1k depends more of your brain and of the DAC of your audio card than any thing.

The big actual problem with the audio DSP quality is that the musicians use computers optimized for the video not for the audio (the chipsets of the laptops are very crappy).
Your audio card manages only its internal mix with its DSP, all is done with the CPU/FPU.
If the video had the same technical evolution than the sound, we look movies with size of 320*240 in 2015 on our HD screens.
An oversampled VA use the same CPU use than if it is not oversampled in a DAW at 96k.
This is why the devs prefers to propose you good VA at 44.1k with a low CPU use with generally only the main processes which are oversampled.
So you have the choice to increase the DAW FS if you use a lot of VA or FX in your mix. And do not forget to avoid to clip the signal of each source of modulation.

Xavier

[EnGee]

Thank you Xavier for your input.

I have a tutorial about mixing. The tutor is a pro mixing engineer and advises to use 24bit/44.1 khz for mixing. I believe this is good enough.

Now, for recording? This is where things are getting confusing! Is recording audio at higher sample rate than 44.1 khz essential? What about acoustic instruments, vocals and external synths recording vs Software instruments recording? What the A/D converter's sample rate? Is there any oversampling in the converter?

The modulation is a good point also. Using a lot of crazy sounds and modulations can lead to aliasing?

Then, the one million dollar question! Is the aliasing is the most important issue in determining the quality of the sound?

[jupiter8]

I have a tutorial about mixing. The tutor is a pro mixing engineer and advises to use 24bit/44.1 khz for mixing. I believe this is good enough.

Now, for recording? This is where things are getting confusing! Is recording audio at higher sample rate than 44.1 khz essential?

I'd say it's the other way around. Higher samplerates are more important for mixing than recording. For recording 44.1 is good enough. It can be for mixing too,it depends.

What the A/D converter's sample rate? Is there any oversampling in the converter?

There is. All modern converters sample in the GHz range and downsample internally with a digital filter.

Then, the one million dollar question! Is the aliasing is the most important issue in determining the quality of the sound?

Hard to say. I think aliasing sounds absolutely horrible but if it's THE most important thing is hard to say. How much aliasing compared to what ?

Like I've said, it depends upon the particularities of the situation you're working in.

You must decide on a case-by-case basis based upon evidence whether one choice is better than another.

It isn't possible to make a general rule and follow it without thinking about it and investing the effort to make the right decision.

Sorry!

^this

[EnGee]

I have a tutorial about mixing. The tutor is a pro mixing engineer and advises to use 24bit/44.1 khz for mixing. I believe this is good enough.

Now, for recording? This is where things are getting confusing! Is recording audio at higher sample rate than 44.1 khz essential?

I'd say it's the other way around. Higher samplerates are more important for mixing than recording. For recording 44.1 is good enough. It can be for mixing too,it depends.

Thanks, do you have any links for further reading please?

Then, the one million dollar question! Is the aliasing is the most important issue in determining the quality of the sound?

Hard to say. I think aliasing sounds absolutely horrible but if it's THE most important thing is hard to say. How much aliasing compared to what ?

Well, I meant is it the most difficult problem for the developer to solve? Considering the oversampling requires a lot of cpu power.

[aciddose]

Depends upon the specifics.

As far as oscillators are concerned efficient solutions for anti-aliasing filters are long solved.

When it comes to over-sampling, this is the least efficient method available. Say you could create a maximally efficient filter that had zero cost.

Now when you over-sample by 2x it still costs you twice as much because you had to compute twice as many results.

It can sometimes be "less than 2x" because the original cost included some overhead, but when we're accounting only for the true cost without overhead it is inevitable, doing twice the work requires twice the effort.

Lifting twice the weight or lifting a weight twice the height requires twice the energy.

[kx77free]

Hi

I have a tutorial about mixing. The tutor is a pro mixing engineer and advises to use 24bit/44.1 khz for mixing. I believe this is good enough.

Sorry but the sujet is:
Aliasing in synths. How to prevent it?

The majority of mixing engineers do not understand something about the internal DSP audio processes!!

I done DSP researches since 10 years and after a lot of discussions with some sound engineers, my conclusion is that they ignore an important focus about that: all audio DSP processes generate aliasing. Even if they use the 44.1k FS, theirs professional machines or plugins use in intern a lot of DSP ic or processes which use upper FS than their Pro Tools project.

Some examples:
Auto-oscillating resonant filters like Moog or other: FS *2 min , ideal *4 or *8.
Ring mod: FS *2 min with BLT signals, ideal *8 or *16 for RM diode sim.
Distortion and limiters: FS *2 min, ideal *8.
Compression: FS *2 min, ideal *4, *16 for hard clipping.
Sync oscillators: FS *8 min, ideal *16 or *32.
That depends of the DSP processes, with BLT lookup tables there are some complex processes which avoid the oversampling but the sync produce less high harmonics than a true analog synths. Some BLT oscillators produce good sync but they do not use lookup table method to generate BLT waveforms.
Other modulations like multiplication: FS*2 is the frequencies are very high or no BLT, ideal *4 or *8.
If your VA use for example an oversampled filter with its intern FS to *4, at 88.2k the process works at *8.

You cannot use rules for recording about the VA synths mixing.
When you record a track with a microphone the hardware filters cut the frequencies upper than FS/2 and effectively they are not aliasing but if you modify the track with DSP audio processes you generate potentially some aliasing, it's the rule... It's not an opinion.

Please, done a mix with fat modular patches with several VA with a reverb on 44.1k and after the same mix to 88.2k, you must listen a difference in the mix, the sound appears less digital rigorous and aggressive in the high frequencies, more aerial and smoothness.

And do not forget to verify the setting of your audio card if its FS outputs or inputs setting are 44.1k for 88.2k and 48k for 96k, because some hardware audio cards which can record the DAW work flow directly produce aliasing when the FS of the outputs and inputs use a bad FS conversion setting.

Xavier

[EnGee]

Depends upon the specifics.

As far as oscillators are concerned efficient solutions for anti-aliasing filters are long solved.

When it comes to over-sampling, this is the least efficient method available. Say you could create a maximally efficient filter that had zero cost.

Now when you over-sample by 2x it still costs you twice as much because you had to compute twice as many results.

It can sometimes be "less than 2x" because the original cost included some overhead, but when we're accounting only for the true cost without overhead it is inevitable, doing twice the work requires twice the effort.

Lifting twice the weight or lifting a weight twice the height requires twice the energy.

Thank you aciddose It is more clear now

[EnGee]

Hi

I have a tutorial about mixing. The tutor is a pro mixing engineer and advises to use 24bit/44.1 khz for mixing. I believe this is good enough.

Sorry but the sujet is:
Aliasing in synths. How to prevent it?

Yes, you are right The thing is little bit difficult to have so many stages and things with sample rate. So, I better stay focused on the synths only.

The majority of mixing engineers do not understand something about the internal DSP audio processes!!

I done DSP researches since 10 years and after a lot of discussions with some sound engineers, my conclusion is that they ignore an important focus about that: all audio DSP processes generate aliasing and that even if they use the 44.1k FS, theirs professional machines or plugins use in intern a lot of DSP ic which use upper FS than their Pro Tools.

Some examples:
Auto-oscillating resonant filters like Moog or other: FS *2 min , ideal *4 or *8.
Ring mod: FS *2 min with BLT signals, ideal *8 or *16 for RM diode sim.
Distortion and limiters: FS *2 min, ideal *8.
Compression: FS *2 min, ideal *4, *16 for hard clipping.
Sync oscillators: FS *8 min, ideal *16 or *32.
That depends of the DSP processes, with BLT lookup tables there are some complex processes which avoid the oversampling but the sync produce less high harmonics than a true analog synths. Some BLT oscillators produce good sync but they do not use lookup table method to generate BLT waveforms.
Other modulations like multiplication: FS*2 is the frequencies are very high or no BLT, ideal *4 or *8.
If your VA use for example uses an oversampled filter with its intern FS is *4, at 88.2k the process works at *8.

You cannot use rules for recording about the synths mixing.
When you record a track with a microphone the hardware filters cut the frequencies upper than FS/2 and effectively they are not aliasing but if you modify the track with DSP audio processes you generate potentially some aliasing, it's the rule... It's not an opinion.

Please, done a mix with fat modular patches with several VA with a reverb on 44.1k and after the same mix to 88.2k, you must listen a difference in the mix, the sound appears less digital rigorous and aggressive in the high frequencies, more aerial and smoothless.

And do not forget to verify the setting of your audio card if its FS outputs or inputs setting are 44.1k for 88.2k and 48k for 96k, because some hardware audio cards which can record the DAW work flow directly produce aliasing when the FS of the outputs and inputs use a bad FS conversion setting.

Xavier

Thank you for the explanation. I will re read what you wrote tomorrow morning and try to grasp it

[aciddose]

Sync oscillators: FS *8 min, ideal *16 or *32.

Are you talking about naive?

It is possible to use a method like FIR filters to fully anti-alias sync without any over-sampling.

"Ideal" also is a bit tricky. To produce the same result as a "very low-quality" (5 samples long) FIR filter requires *32 and depends upon a "very high-quality" interpolating filter.

This should be easy to understand once you understand that the convolution is the same as infinite times oversampling, so the FIR filter acts directly as an interpolating filter with infinite resolution of input.

[kx77free]

Hi,

It is possible to use a method like FIR filters to fully anti-alias sync without any over-sampling.

For my oscillators I use band limited look up tables.
I do not use FIR to reduce aliasing for oscillator sync but a complex mix of the waveform with itself.
When the slave waveform is broken by the master wave Sync, the slave waveform restarts but it is mixed with a very small fade (few samples following its frequency) with itself.
And because on the upper frequencies the length of the fade is too long than the waveform period, the slave waveform is mixed progressively to identical band limited waveform with the same frequency but with no Sync.
By this way the oscillator CPU use is increased of 100% only and that works fine without oversampling.
It's a compromise, the harmonics created by the Sync are progressively decreased around the high frequencies but it's not really important because there are only few harmonics in this case.

[Ichad.c]

If possible I record (analog) @96kHz (to save a master copy) and then mix @48kHz (software + hardware), never had much to complain about. There is really no reason to use 44.1kHz for mixing - anti-aliasing filters need space to work, and the CPU usage / disk space difference for 44.1kHz vs 48kHz is negligible. Also - I find aliasing in effects more annoying than in synths, probaly more due to the complex interacting with intermodulation distortion.

(RANT) -> Different quality settings for real-time vs offline rendering is one of the greatest evils in this world! I want the sound what I hear when I mix - nothing more nothing less - real-time export all the way!

Cheers
Andrew