Discrete SSM2040 clone built

[Joachim Verghese]

I'm bringing back the 2040 discussion again, I hope it's not too boring.

Last night I experimented with a circuit which I believe is quite close
to being a discrete version of the SSM2040, and I think Juergen would
agree.

The gain cells are based on those described in Dave Rossum's first
patent, i.e. a PNP mirror pair on top of an NPN differential pair. I
used regular 2N3904 and 2N3906 devices whose Vbe's I measured and
grouped into loosely matched (5 mV) pairs. The batch of devices I had
wasn't large enough to permit tighter matching.

I used BC517 low power darlingtons (current gain of 30,000) for the
buffers, and a 10k resistor from the output emitter to VEE. The
"external" resistors were the same as those in the 2040 data sheet,
i.e. 10k and 220 ohms.

The exponential converter is a standard NPN/PNP pair with sourcing
output and operating near ground level. This feeds a low-side NPN current
mirror whose single output is divided among the four gm-stages.

How does it sound? Well, my initial impression was that it sounded
almost identical to the SSM2024-based filter I built a few days back.
(Maybe I've been listening too much to filter sweeps lately).
Very smooth and warm, just as Juergen has told us. The *big* difference
comes when you overload the input. Definitely asymmetrical clipping,
very nice sounding harmonic modulation when swept.

The response to different resonance settings was similar to that of
the 2024 filter. I could get what sounded like fairly high Q without
oscillation, but I had to tweak the resonance control very carefully,
so as to prevent the filter from suddenly jumping into oscillation.

Too bad I didn't have a real SSM2040 for comparison.

Control rejection was far from ideal, probably a result of insufficient
matching of transistor pairs. I may build an array-based version
sometime soon, and this should help a little, since you get better
thermal coupling as well.

-joachim