[sdiy] OTA performance (was SSM chip reissue)

[Tom Wiltshire]

On 25 Apr 2017, at 11:57, Neil Johnson <neil.johnson71 at gmail.com> wrote:

> Tom Wiltshire wrote:
>> In this case, we *know* that the 2164's matching is good enough, because we've tried it. It can howl.
> 
> Except you were the one going on about larger differences:
> 
>> What's the smallest decibel difference people can detect? About 3dB, isn't it? Quite a bit more than 0.25dB.
>> 
>> To be honest, if it were 1dB out by -60dB, who's really going to notice? 2dB out at -80dB? And that's a lot worse than the spec'd figures suggest is likely.
> 
> 0.25dB is about 3% error.  1dB is about 12% error - an SVF would have
> quite a poor Q with that kind of gain error.
> 
> And quite honestly, making a circuit howl is easy - circuits will
> quite happily do that all by themselves, even when you don't want them
> to.  Making it howl nicely, or precisely, is much harder.

I can still see a lot of useful applications for the chip even if the gain errors were as bad as I suggested, and I was exaggerating. Not everything *requires* carefully matched gains, although there will be a few things that do. 
More numbers in the datasheet would be very useful, and some idea of what sort of spread is hidden behind that word "typical" would be very useful too.

You raise an interesting possibility though - it could be that variations in gain matching at large attenuations are one reason why it's harder to make 2164-based SVFs resonate at the low end than at the high end. Previously when this has been discussed, phase lag has been proposed as the likely culprit. But if the gains match well at 0dB, pretty well at -20dB, and not so well by the time you're down at -80dB, then you'd expect resonance to drop off at lower frequencies.

Incidentally, Oberheim managed to make an SVF with CA3080s in the OB-X, and they're not matched at all (unless they matched them?). The filter is quite highly thought of. Ultimately, it's not always about technical perfection.

Tom