[sdiy] Polyphonic temp compensation in VCF's!

[Tom Wiltshire]

On 23 May 2011, at 23:37, karl dalen wrote:

> -- Den mån 2011-05-23 skrev Tom Wiltshire <tom at electricdruid.net>:
> 
>> Many of the synths you mention that *don't* have tempcos
>> for the filter will use an autotune routine instead. So if
>> you think the filter FM sounds 'out of tune', you can just
>> bang the autotune button - at whatever temperature.
> 
> The autotune in those machines as i have understod them
> are probably multi pass all parameter tuning method, not
> continuous background single parameter periodical tuning
> method (compare with A6) which would make sense. If that
> actually works? Anyhow, in many cases are slow on the
> 8Mhz Z80 of the day.
> 
> So slam the auto button dont seem like a viable method
> to me, well if it would they should have not put in
> tempcos in the first place i reasoning.
> 
> OBmx are a resonable new design has no tempcos but FFM
> it would have been interesting to se how auto tuning
> works on the 2 discrete filters.
> 
> KD

Most filter trimmers give you "range" and "offset". E.g. they assume that the filter CV response is linear and allow you to adjust it so that its 1V/Oct. Most autotune routines sample several points across the range (octaves, for example) and then assume the response is linear between those points. So the autotune typically gives a better result than the trimmers, since it can compensate for at least some non-linearity in the filter response. Better routines or more modern code could assume a more complicated polynomial instead of a straight line and get a further improvement.

I hear you about the autotune routines being slow. But the processor isn't always at fault. If you're counting pulses from a oscillating filter fed through a comparator, and testing the bottom octave (say 20Hz) then counting for a least a couple of seconds is necessary for reasonable accuracy - still only 40 pulses. Multiply those couple of seconds up by several filters and several pairs of oscillators and you have a routine that takes half a minute to run, whatever processor you use. You could use better techniques to measure frequency (like timing the length of the period), but that's the simple way that some old synths use. So it may be the algorithm at fault.

To be honest, I think the autotune solution is perfectly acceptable. Rarely does the temperature you're operating a synth in vary enormously within a few minutes, so if you make sure the synth is at the ambient temp and switch on and allow it to warm up before you autotune, you'll get a result which is pretty consistent between sessions.

I once took my Polysix around to friends house on a cold winter's day. Although it was only outside for five or ten minutes or so when I arrived the tuning was unplayably bad. And there's no autotune on a polysix...

T.