K35 Filter

[Urs]

It seems that all of these implementations are linear? In that case I think the most straight forward and easiest to understand version is Andy's here:

https://cytomic.com/files/dsp/SkfLinear ... mised2.pdf

Hope you don't mind me asking Urs but you and your team have a done a stellar job of modelling this one in Diva:

Is it possible to get the 'screaming' aspect of this filter without non-linearities? If not, can you get this aspect with usual non-linearities (tanh etc) or is it something that really comes with an iterative method like Raphson-Newton?

You can get the screaming aspect with tanh(), but as this turns the equations into implicit territory, you need to do Newton-Raphson or some other kind of root finding algorithm or approximation.

Note that you'll want to have an original hardware around to get it right. There's some crazy shit going on that you do want to match, and you can match it by the right gain staging and asymmetry.

[juha_p]

There's this paper which presents an application of the port Hamiltonian formalism to the nonlinear simulation of the OTA-based Korg35 filter circuit and the Moog 4-pole ladder filter circuit :

Danish, Bilbao and Ducceschi - Applications of Port Hamiltonian Methods to Non-Iterative Stable Simulations of the KORG35 and MOOG 4-Pole VCF

September 2021
DOI:10.23919/DAFx51585.2021.9768301
Conference: DAFx20in21At: Vienna, Austria


Is this implementable as an alternative for filter in discussion?

[2DaT]

September 2021
DOI:10.23919/DAFx51585.2021.9768301
Conference: DAFx20in21At: Vienna, Austria


Is this implementable as an alternative for filter in discussion?

It's a known fact that a 2 pole filters with a single diode-like non-linearity can be solved implicitly with LambertW function (or similar non-elementary functions). However, it has some issues:

1) It doesn't solve aliasing. And aliasing in self oscillating filters can lead to instabilities even when you solve the equations implicitly.
2) Computation of LambertW is complicated. Approximation is possible, but it's still significantly more expensive than your typical exp/tanh/etc.
3)Mystran's method exists and it works.

[mystran]

3)Mystran's method exists and it works.

Arguably it works very poorly for some asymmetric cases (eg. a single diode), but in this case if you just want to model the whole feedback amplifier as a bulk linearity there's no issue.

[andy-cytomic]

For the record:

KORG35 = biased bjt epoxy job “chip” = MS10 and MS20 rev1 filters
KLM-307 = ota = MS20 rev2 daughter board filter

There is no KORG35 rev2 as indicated in the Port Hamiltonian paper.

https://www.timstinchcombe.co.uk/index.php?pge=korg