[sdiy] Switched cap filters

[Tim Stinchcombe]

> I'm curious as to what the list thinks about switched 
> capacitor filters in 
> general and more specifically the Nat'l Semi LMF100.

>From comments posted here over the years, my initial reaction is that if you
stick with the examples on the datasheets, you will probably be
disappointed. However some years ago I played with an MF10: I cannot say the
results were 'stunning' or anything, but I thought it didn't sound that bad;
the downside is that you have to work quite hard at it, so it really turns
into a 'can this be done' kind of exercise (the answer is 'yes'), but what
you end up with is hardly that practical, in the sense that there are much
better and easier ways to achieve the same thing using non-switched
capacitor means.

This is what I did: I cascaded both sections of the MF10, but re-calculated
the resistor values so that it sort of 'emulated' the perennial four-pole
favourite, 1/((s+1)^4+k) (a la Moog ladder, CEM3320, SSM2040 etc.) by making
each section roughly 1/(s+1)^2=1/(s^2+2s+1). I also added a feedback loop
with an OTA in it for voltage-controlled resonance. It didn't sound too bad,
and would oscillate quite nicely too.

As for the clock frequency issue, again there is some latitude here: the
cut-off frequency (for the MF10 at least) is not merely the clock frequency
divided by 100 - there is a ratio of resistors in there too, so this factor
can be made higher to push the clock higher into the inaudible range at
lower cut-off frequencies. The resistor values I had suggested I finally
settled on about 400:1, so a clock of 16kHz gave a cut-off of 40Hz. But
again there is a downside - this now means there are some pretty large
resistance values in the circuit, and this increases the DC offsets, to the
point where there is no room left for any signal swing! It should be
possible to trim-out some of these offsets, but I didn't get as far as
trying (and again this just further increases the circuit complexity).

The other problem was how to clock the thing, especially since it is
desirable to have something approaching a 1V/octave response for the cut-off
frequency. I was using the VCO in a 4046 PLL chip: as I was running the
whole circuit of +/-5V split supplies, the CV for the 4046 has only a few
volts to play with across the whole control range, which at the time seemed
to imply a super-accurate expo converter was going to be needed. However
since then I have learned that it is possible to control the 4046
current-wise through pin 12, and that seems to suggest better possibilities
- a simple test showed it was easy to get from 33kHz to in excess of 600kHz
from the 4046, ample for a decent spread of cut-off frequencies. Also
running the circuit off 0 - 12V might help, and biasing the MF10 to be
mid-rail. But again this implies yet more added circuit complexity.

At some point in the future I intend to return to this to see just how
plausible it all is, but as I said at the start, that is really all it will
be, rather than a practical proposition!

Tim
__________________________________________________________
Tim Stinchcombe 

Cheltenham, Glos, UK
email: tim102 at tstinchcombe.freeserve.co.uk