[LAU] jack/oversampling

Jörn Nettingsmeier nettings at stackingdwarves.net
Sun Mar 16 18:50:57 UTC 2014


On 03/16/2014 05:45 PM, tim wrote:

> a) any non-linearity introduces harmonics, some non-linearities
> introduce an infinite amount of harmonics, which will cause foldover
> distortion. the large the sampling-rate, the lower the foldover.

ok, so you are trying to do weird synthesis that can produce 
non-bandlimited output? i can see how you might want to use high 
sampling rates there, but then again there will always be another 
processing step that causes yet higher harmonics - addressing that with 
high sample rates seems like a somewhat blunt approach that is bound to 
fail eventually.

> b) delay-lines have a higher precision at higher sampling-rates

that statement is definitely not correct. granted, if you only do delays 
with sample granularity (which has the big advantage of not requiring 
any computation), there is some benefit in using higher rates. but you 
can produce sub-sample delays with arbitrary precision easily. for IIR 
feedback, i sure see the point, but then the question becomes "why do 
you need to expose this to the outside world?" - just upsample in your 
processing application and leave the rest of the jack graph running at a 
sane rate.

> c) the tuning of digital filters is more precise at higher
> sampling-rates due to the frequency warping in the blt

i don't understand this. can you elaborate? what is "blt"?

> note on a:
> if your signal processor introduces the Nth harmonic, you have to
> upsample your signal by a factor of N. or apply a pre-filter on your
> signal by nyquist/N.

true. it's a funny and somewhat strange thought exercise for me to try 
and achieve the highest possible "fidelity" with brutal distortion 
algorithms  - obviously, since i don't work with distortion, i try to 
keep my signal chains as linear as possible.
but i can see how somebody well trained in distortion synthesis would 
want to eliminate aliasing artefacts, since those would conceivably 
interfere with systematic exploration of sounds based on prior 
experience, and make the sonic outcome even more erratic than it already 
is...

but in any case, there is no point in taking the internal higher 
sampling rates out into the real world, so the zita resampling approach 
might be your best bet.

> question for the reader: in order to completely prevent foldover
> distortion, how much do you have to upsample for a tanh waveshaper (a
> processor that introduces infinite harmonics)?

incidentally, just returned from musikmesse, and i've had my share of 
DXD/DSD loonies... if you want to go there, there is people who want to 
sell you 256-times oversampled single-bit delta sigma gear, and they 
will happily talk megahertz with you.
it would be a ton of fun to discuss with them the best way to handle a 
tanh waveshaper, and what new ultimate fidelity frontiers are required 
for the distortion synthesis crowd. just make sure you avoid the term 
distortion, call it "spectral enhancement processing" instead. >;->


best,


jörn



-- 
Jörn Nettingsmeier
Lortzingstr. 11, 45128 Essen, Tel. +49 177 7937487

Meister für Veranstaltungstechnik (Bühne/Studio)
Tonmeister VDT

http://stackingdwarves.net



More information about the Linux-audio-user mailing list