Modelers and Aliasing

[James Freeman]

Jays test signal emulates harmonically rich signal in the frequency range that electric guitar produces going into the modeler, while the 10-20 khz signal will make you see and hear aliasing better it is not a realistic signal and will not mask the reflected aliasing overtones.

Another thing that is overlooked here is everyone has different ears and audio systems, the effectiveness of masking will be different for different people.

This thread taught me that I need to test using both methods to get a clearer picture, the 10-20khz sweep will tell me about the modeler's anti-aliasing filter implementation AND the harmonically rich sweep at guitar range will tell me if I can actually hear aliasing when it is being masked by a realistic input signal.

 

[Orvillain]

Here's a comparison of the Aliasing performance between ToneX software vs ToneX Pedal vs Axe-Fx III using @jay mitchell's aliasing test signal.

The ToneX software plugin and pedal are running the JCM800 capture by SDS and the Axe-Fx III is running the JCM800 model matching the settings visible in SDS's video (which puts in the neighborhood gain and tone wise).

First ToneX Software plugin running at 48KHz, then ToneX Pedal, then Axe-Fx III


I don't hear any aliasing in the Axe-Fx III case.
The ToneX Pedal seems to be running at greater than 48KHz, since there's less aliasing (to my ears) than the software plugin running at 48KHz.

Yep, I hear the aliasing in the ToneX clips, not in the AxeIII clips.

 

[Mongillo19]

Here's a comparison of the Aliasing performance between ToneX software vs ToneX Pedal vs Axe-Fx III using @jay mitchell's aliasing test signal.

The ToneX software plugin and pedal are running the JCM800 capture by SDS and the Axe-Fx III is running the JCM800 model matching the settings visible in SDS's video (which puts in the neighborhood gain and tone wise).

First ToneX Software plugin running at 48KHz, then ToneX Pedal, then Axe-Fx III


I don't hear any aliasing in the Axe-Fx III case.
The ToneX Pedal seems to be running at greater than 48KHz, since there's less aliasing (to my ears) than the software plugin running at 48KHz.

Hey I can actually hear that! Pacman sound is a great way to describe it lol I only hear it as the pitch become higher, which checks out?

 

[jay mitchell]

So far, the clips using my alias test file reveal either no or relatively low levels of audible aliased frequencies. The modelers I first used the test with ca. 2007 generated very obvious descending tones. The worst that have been shared in this thread so far are worlds better.

 

[MirrorProfiles]

not really too invested in this discussion, but I just happened to be checking out a few different emulations of OCD pedals and thought I'd have a quick look in PluginDoctor.

Helix and Amplitube looked very similar. STL looked nothing like either of them (more like what a tubescreamer looks like). Neural DSP have a model in the Soldano plugin. Was KIND of like Helix and Amplitube but also a bit unique. The Fractal model was also quite different to the other versions, but I was surprised how much more it had aliasing than any of the others I tested. It was quite visible at even low signal levels, so it was quite surprising to me and made me think I'm doing something wrong.

Anyone able to compare the response of some OCD plugins against Fractal (and ideally the real pedal too)?

 

[jay mitchell]

not really too invested in this discussion, but I just happened to be checking out a few different emulations of OCD pedals and thought I'd have a quick look in PluginDoctor.

What did you see? How did you distinguish aliasing from other nonlinear artifacts?

 

[MirrorProfiles]

What did you see? How did you distinguish aliasing from other nonlinear artifacts?

sweeping a sine wave up and down in plugin doctor

and running a sinesweep through and recording it

 

[northern_fox]

Possibly of interest - https://www.experimentalscene.com/software/antialias/
Obviously this doesn't *fix/prevent* any aliasing, but claims to attempt to filter out the artifacts. Disclaimer: I have absolutely no idea if this is valid or not.

 

[James Freeman]

Thanks.
How can it possibly undo aliasing that's already there?
Also, VST2 32bit is so outdated I don't think any modern plugin offers that format?

 

[yeky83]

Possibly of interest - https://www.experimentalscene.com/software/antialias/
Obviously this doesn't *fix/prevent* any aliasing, but claims to attempt to filter out the artifacts. Disclaimer: I have absolutely no idea if this is valid or not.

Seems like a dumb plugin, it's essentially a low-pass filter... it won't do anything to mitigate aliasing.

 

[BenIfin]

Possibly of interest - https://www.experimentalscene.com/software/antialias/
Obviously this doesn't *fix/prevent* any aliasing, but claims to attempt to filter out the artifacts. Disclaimer: I have absolutely no idea if this is valid or not.

Thanks.
How can it possibly undo aliasing that's already there?
Also, VST2 32bit is so outdated I don't think any modern plugin offers that format?

Hey Guys !

Interesting find NF :)

I just d/l it and installed and ran it in Cakewalk Bandlab. I recorded a Tonex Track of Crunch guitar and then applied it as a VST.

I don't know how its *supposed* to work but it *seemed* to "dull-off" the high frequencies ..... so I gather its a very detailed kind of low-pass eq as yeky83 noted. As as James and Y83 wrote, I cant see how it can de-alias a plugin that is not already de-aliased in it core engine code (?)

Maybe (?) the trick is to run an Audio Track and have it as the 1st VST and then have [say] Tonex VST or NAM VST after it .... but even then it cant possibly (?) get into the Plugin engine and de-alias whats hard-coded in there (?)

Really interesting find though !

Thanks :)

Ben

 

[northern_fox]

Hey Guys !

Interesting find NF :)

I just d/l it and installed and ran it in Cakewalk Bandlab. I recorded a Tonex Track of Crunch guitar and then applied it as a VST.

I don't know how its *supposed* to work but it *seemed* to "dull-off" the high frequencies ..... so I gather its a very detailed kind of low-pass eq as yeky83 noted. As as James and Y83 wrote, I cant see how it can de-alias a plugin that is not already de-aliased in it core engine code (?)

Maybe (?) the trick is to run an Audio Track and have it as the 1st VST and then have [say] Tonex VST or NAM VST after it .... but even then it cant possibly (?) get into the Plugin engine and de-alias whats hard-coded in there (?)

Really interesting find though !

Thanks :)

Ben

It's not meant to de-alias, I think it's meant to literally remove/reduce the offending frequencies by band - so instead of a low pass filter, maybe a bunch of very small cuts? It's a slightly destructive operation for sure, but maybe still helpful.

 

[MirrorProfiles]

I think if you’re working at a higher sample rate, and there is no anti aliasing filter on a plugin, and you’re running a few one after the other in series, it may have some benefit.

Pretty sure it’s just a filter though, and no idea how good it actually is for this purpose. They aren’t created equal.

 

[yeky83]

There’s no specific “offending frequencies” to cut since aliasing noise is broadband. And an "anti aliasing filter" that oversamples and filters by itself does nothing after aliasing noise has effectively been printed on the track already. This plugin is perhaps useful if you have a DAC with a piss poor reconstruction filter...? Otherwise it's pretty silly.

 

[pipelineaudio]

We used to say stuff like this about the impossibility of removing reverb, or tuning individual notes in a chord. Never say never

 

[MirrorProfiles]

has anyone compared results when using a pedal and amp?

For instance:

- NAM (or ToneX) Pedal capture into a NAM amp capture
- algorithm based pedal capture into NAM/ToneX amp capture
- real pedal into NAM/ToneX capture

 

[Raaphorst]

I see IR is measured sometimes for aliasing but an IR cannot produce aliasing. It can use the frequencies which are in the IR table. Frequencies which fall outside that table are simply not affected in any way.

Or even shorter: an IR is lineair.

 

[northern_fox]

I see IR is measured sometimes for aliasing but an IR cannot produce aliasing. It can use the frequencies which are in the IR table. Frequencies which fall outside that table are simply not affected in any way.

Or even shorter: an IR is lineair.

2 explanations:

1 - the IR loader itself, possibly
2 - I suspect a lot of people "hear" sounds that they just attribute to aliasing that aren't actually that

 

[norminal]

At what frequency range does aliasing typically occur?

 

[AlbertA]

At what frequency range does aliasing typically occur?

Any frequencies above half the sample rate will be aliased (see Nyquist-Shannon sample theorem).

For example a digital system with a sampling frequency (Fs) of 48KHz, any frequencies above 24KHz (Fs/2) will mirror/fold/alias in the range of 0-24Khz.

Say your input signal contained a 40KHz frequency, which is 16KHz above Fs/2 (24 KHz) - the alias would occur at 24KHz-16Khz = 8KHz.

That's why in analog-to-digital conversion, your input signal is filtered to limit its bandwith to Fs/2.

Now, if you are already in the digital domain and you do non-linear operations, this can introduce frequencies that go above Fs/2.
To avoid aliasing in this case, you have to increase the internal sample frequency or your signal (upsample or oversample), then do the operation, filter the signal to the original required BW and then downsample back to the system sample rate.

This is why you'll see modelers mention things like 16x oversampling, 8x oversampling, etc.

For exampe, let's say a system sampled an 18Khz sine signal and you squared it. The squared operation would result in a signal with a 36Khz. frequency in this case. Without oversampling, a 36Khz frequency is 12Khz above Fs/2, so that means it will alias/fold as a frequency 12KHz below Fs/2, in this case, 12KHz.

Here's a cool animation that shows this in action:

Aliasing - Wikipedia

en.wikipedia.org

So in this particular case, 2x oversampling would be more than enough to avoid aliasing, since 2x oversampling would mean you could generate frequencies from 0-48KHz (24Khz x 2).


Now imagine an amp modeler, which is taking a complex guitar input signal and performing non-linear operations on it, that introduce a lot of harmonics of the original frequencies of the input signal. 16x oversampling on a 48Khz system, means that amp modeling operation can produce frequencies from 0-384Khz, which it can then properly filter from 0-24Khz, before any aliasing even occurs.