Fractal Talk

[JiveTurkey]

 

[Orvillain]

Whether someone can hear aliasing or not is complicated. It's a psychoacoustic thing.

It depends on the exact nature of the aliasing. When the Kemper first came out the aliasing was very noticeable because the aliased tones occurred in the midrange when playing higher notes. If the aliases are near the distortion products they are masked by a phenomenon known as perceptual masking. This is the same phenomenon that is the basis for MP3 encoding. Humans can't hear quiet tones that are close in frequency to loud tones.

As with all engineering problems it's a trade-off. To reduce aliasing you can either increase the sample rate (via oversampling, for example) or use Antiderivative Anti-Aliasing (ADAA). Both of these approaches increase the amount of CPU required. Simply oversampling is very effective but doubles the amount of CPU cycles.

ADAA can achieve aliasing reduction but it's no free lunch either. The problem is the antiderivative is typically a more complex function, sometimes much more complex (tanh(x) becomes log(cosh(x))or example). For example if the function is, say, x^2, the anti-derivative is 1/3*x^3. This requires two extra multiplies. You might as well simply double the sample rate. Then you have to take do a difference and if the difference is near zero you risk generating a result that exceeds the numerical range so you have to add conditionals which slow things down even more. ADAA also doesn't work for nonlinear systems with memory. Nearly all our algorithms employ memory so we don't use ADAA.

Wouldn't doubling the sample-rate require upsampling filters and downsampling filters, and have more CPU impact than ADAA? Wouldn't SIMD help with the CPU impact of of ADAA as well?

 

[Dom Mcsweeney]

Yup, he's a wingnut. I call him a "sociomediapath".

Aliases:
Jean Bloomberg
Yvan Bloomberg
Yvan Apothekar
Stainless Steel Frets
Shattered Square
Elon Mask
Popstar
and maybe Dada.

oh man.... and the rest

some bonus funny ones:

Francois NeuralNet
MXR ROCKMANN - (was allegedly messaging a few industry types, pretending he was the official MXR brand)
Mark Nembrini
RichSwissGuy

....and really quite worrying and disturbing:

last year I caught him pretending to be some young innocent girl from Canada =(

he's definitely a "sociomediapath", but I have firm reasons and evidence to back it up, that he's a lot worse than that

 

[GTR37]

Yup, he's a wingnut. I call him a "sociomediapath".

Aliases:
Jean Bloomberg
Yvan Bloomberg
Yvan Apothekar
Stainless Steel Frets
Shattered Square
Elon Mask
Popstar
and maybe Dada.

Wasn’t Shattered Square at the peak of Mt Stupid a few years back ?

 

[yek]

Right but if they fixed the issue with the amp block aliasing, what changes in the Drive block needed to be made that results in less aliasing?

There are other functions in the Fractal ecosystem that provide a choice between normal and higher oversampling. Such as the preamp in the Cab block, and the amp modeling itself. There's another one, I think, but I can't recall it right now. Anyway, normal suffices, but it simply is an option for the OCD crowd to go for ultra.

 

[FractalAudio]

Wouldn't doubling the sample-rate require upsampling filters and downsampling filters, and have more CPU impact than ADAA? Wouldn't SIMD help with the CPU impact of of ADAA as well?

We already upsample and downsample. You have to upsample and downsample at least 2x when distorting, otherwise the aliasing is objectionable. For heavy distortion at least 4x is desirable.

SIMD may or may not help, probably not. Even if it did it would also help in the non-ADAA case so it's not a gain.

Regardless, ADAA doesn't work in our algorithms because almost all our algorithms use memory (which is why they sound good). Analog circuits have memory, even if there aren't explicit reactive components in the network. For example, a tube triode with just resistors connected to its terminals still exhibits memory effects due to parasitic capacitances (Miller effect).

 

[blandified]

We already upsample and downsample. You have to upsample and downsample at least 2x when distorting, otherwise the aliasing is objectionable. For heavy distortion at least 4x is desirable.

SIMD may or may not help, probably not. Even if it did it would also help in the non-ADAA case so it's not a gain.

Regardless, ADAA doesn't work in our algorithms because almost all our algorithms use memory (which is why they sound good). Analog circuits have memory, even if there aren't explicit reactive components in the network. For example, a tube triode with just resistors connected to its terminals still exhibits memory effects due to parasitic capacitances (Miller effect).

Now you got me thinking... do you consider physical setup/CCA layout in your models?

For example, trace geometry in a CCA... maybe a 10mil trace over a GND plane with some dielectric height will have some sort of capacitance. Something in the order of pF, but still.

I have been doing a lot of layout and signal integrity/power integrity for work using Hyperlynx and we have to consider all that fun stuff.

 

[FracKing]

Now you got me thinking... do you consider physical setup/CCA layout in your models?

IIRC, he explained how they DO and how, for example, the tube socket material and the like makes a difference or has effects.

Found the post:

F

Post in thread 'The technical modelling thread'

Jan 1, 2026

You rarely, if ever, need sin or cos for amp modeling. More common trig functions encountered are tanh and atan. I personally am not a fan of LUTs but use them when necessary. I prefer polynomials when possible.

There are many different techniques for nonlinear modeling (which is the crux of the issue). If the circuit is linear you can simply use linear digital filters.

There are nodal techniques: K-method, DK-method, modified nodal analysis (MNA), etc. SPICE is an example of MNA. Wave Digital Filters (WDFs) are another technique and are a type of scattering method.

Regardless of...

• FractalAudio

• 
• 
•