31 Aug
2014
31 Aug
'14
8 p.m.
On Sun, Aug 31, 2014 at 07:42:48PM +0000, Fons Adriaensen wrote:
Ah, the fons-filter. Extremely stable and a nice acid-y squelch. Since a tanh function
tends to unity at either end, I can't see how that *would* go unstable. You'd
need to have greater than unity gain for it to really go off.
Fons - ever considered turning your analytical skill to the MS20 and Steiner Synthacon
diode-chain Sallen-Key filter?
--
Gordonjcp MM0YEQ
On Sun, Aug 31, 2014 at 08:13:01PM +0100, W.Boeke
wrote:
Correct. Even the "18dB/octave" Roland TB303 ladder filter has four stages.
Why isn't it 24dB/octave? Because they're more honest about the real-world
performance than Moog :-D
Once you go hanging weirdass source and load impedances off your ladder the response can
vary wildly. Tim Stinchcombe has an incredible analysis which shows all sorts of odd
bumps and peaks in what you'd typically expect to be flat passband.
Sorry Fons, if you google "moog filter
schematic" you will
see 3 stages, where a stage is a capacitor in series with 2
transistor emitters (which have a diode V/I characteristic).
All versions I've every seen have four stages, and with three
the feedback wouldn't have the right phase. If you find a
schematic with three stages please post a link.
Each stage is a capacitor in parallel with the series connection
of two emitter impedances, and driven by a current source, the
collector currents of the previous stage. About that complicated code: I saw implentations
with a lot of
code lines, and preventing oscillations originating from the
non-linear tanh functions needed a lot of care.
I analysed this thing to dead around ten years ago, when I wrote
the MCP plugins. There is no instability problem resulting from
the non-linearity. The main problem with a straightforward digital
implementation is that it becomes a bad approximation at higher
frequencies, this requires some attention to get right. The easy
solution is to run it at a higher sample rate.