Reducing aliasing of waveshaping using convolution

DSP related issues, mathematics, processing and techniques
User avatar
Spogg
Posts: 3368
Joined: Thu Nov 20, 2014 4:24 pm
Location: Birmingham, England
Contact:

Re: Reducing aliasing of waveshaping using convolution

Post by Spogg »

OMG Martin! Is there no limit to your talent?

I didn’t know about your tracks but the one you linked to above is simply lovely. I will listen to the others very soon.

I think you’re a very private and modest person but I’d love to know a bit more about you.

Thanks for everything you do.

Spogg
francoisreme
Posts: 29
Joined: Wed Feb 15, 2017 4:01 pm

Re: Reducing aliasing of waveshaping using convolution

Post by francoisreme »

Hi Martin,

Thanks for your work, did you also tried the other kernels, I was wondering if there is any chance that "big" kernels could resolve analyticaly (min phase band limited impulse approximated by low order polynomials or bspline) ?

I recently read a paper about cascading rectangular convolutions kernel to get higher order kernel alias supression. (like convolve on 2 or 3 samples rather than 1). This paper is called ANTIDERIVATIVE ANTIALIASING, LAGRANGE INTERPOLATION AND SPECTRAL FLATNESS.[url]https://www.researchgate.net/publication/321785604_Antiderivative_antialiasing_lagrange_interpolation_and_spectral_flatness
[/url]

Hope it can give you ideas !
User avatar
martinvicanek
Posts: 1334
Joined: Sat Jun 22, 2013 8:28 pm

Re: Reducing aliasing of waveshaping using convolution

Post by martinvicanek »

Thanks for the nice words and the beer, Spogg! ;)

Francoisreme, thanks for the reference. Yes, you can go to higher order but why would you? The math becomes unwieldy and you gain very little if anything. At least with the guitar amp application in mind, there is no reason to seek for spectral flatnaess because the real thing is anything but spectrally flat - there is pretty much nothing above 5 kHz, you see. In fact the AA waveshaper will not sound good unless you actually add another lowpass after it!

I think that the box shaped kernel marks a sweet spot in terms of effort vs. effect. Personally I would stop there.
francoisreme
Posts: 29
Joined: Wed Feb 15, 2017 4:01 pm

Re: Reducing aliasing of waveshaping using convolution

Post by francoisreme »

Hi Martin.

for audio plugin dev it's convenient to have saturation spectraly flat a low volume and adding only harmonics. But maybe basic correction with a few poles-zeros can do the trick !

I never thougth about this zeros reflections pattern around nyquist frequency as you proposed in your previous post.
about higher order kernels, I was talking about that only to get higher aliasing reduction.
Post Reply