[sdiy] Oakley COTA filter (was: help finding roland filter schematics?)

[David G. Dixon]

Tony,

Thanks for your response.  I hope you didn't take anything I said as
criticism, as I think your COTA filter is the best sounding thing I've ever
heard.  I only hope mine will sound half as good when I'm done with it.
Comments below...

> The op-amp input summing stage of the COTA module has a gain of 0.3 thus
> preventing any major overload to occur within the cascade itself.

I guess I misread your COTA document.  I thought it said the input summing
stage had a gain of 0.5.

>  > Doesn't this imply that the outgoing minus 24dB signal at high cutoff
> would be +/-8V for an incoming signal of +/-5V?
> 
> Yes, at low resonances there is a overall gain in the cascade itself.
> However, in conjunction with the input stage gain loss, and some make up
> gain at the output stage, the overall gain of the module itself is less
> than unity. ***

At 0.3 * 1.47 * 3, the overall gain should be... let me punch the numbers
into my calculator here... 0.95!  My original calc gave 1.59.
 
>  > It would also imply that the cutoff control would effective act as a
> volume control at high cutoff.
> 
> I don't follow you there. For a sinewave at a given frequency any low
> pass filter can be seen to act as a volume control. At high cut-off the
> cascade can be seen as a linear system, so long as we don't overdrive
> each element too much. As such any gain could be considered lumped into
> a single element at any point along that chain, including the output.

What I meant by that is that (assuming my gain values were correct, which
they in fact were not) as the cutoff frequency increases, and the output
signals from each stage approach (what would normally be) unity gain with
respect to the input signal, in your filter the gain would actually increase
by a fair margin because of the inbuilt gains of each stage.  However, as
you say, this happens with every filter.
 
> What is interesting though about this is that clearly the signal
> headroom before soft clipping occurs does get smaller as you work
> through the COTA's cascade. Does this affect the sound? Quite probably -
> but the actual 'depth' of this affect is very much dependant on the
> signal input to the module.
> 
> My ears told me they preferred the sound of this set up compared to a
> straight unity gain cascade. But ears have been known to be deceived.

My approach is to put a resonance OTA on each stage.  The resonance gains
will be apportioned across the stages at a ratio of roughly 10:3:3:3.  This
is how I will ensure unity gain at zero resonance, and constant amplitude at
self-oscillation.  My simulations suggest that I will be able to avoid
clipping at self-oscillation by carefully tweaking the resonance, but I am
investigating the use of automatic gain control on the first stage for this
purpose (albeit so far with limited success -- I'm not happy with the degree
of THD imposed by the linearizing diodes in the AGC circuit from the LM13700
datasheet, although its definitely better than back-to-back zeners!).

> As a matter of an aside - what should the gain of any filter module be?
> I usually make it less than one - usually because at resonance the
> signal level goes up to clipping with a 5V signal input.

Resonance amps at each stage eliminate this, again if the resonance is
tweaked carefully.