Descret OTA was "that Japanese site"

[terry michaels]

Message text written by INTERNET:jhaible at metronet.de
>Are you speaking about the 4-transistor-OTA ? The gm cell of the SSM2040 ?
(not the 2044)
The limit of the output swing is *the* secret of the 2040 sounding
different
that a 3080-based filter. The 2040 has npn darlington buffers (similar to
the
LM13700 buffers, though David Rossum spoke of some trick to increase the 
input impedance). So a voltage swing that is limited to approx 0V at the
OTA
output would be limited to approx. -1.2V at the buffer output. That's why
the
SSM2040 data sheet tells us to keep the level below 1v peak. (10:1 input
divider, and x10 output amplifier in a 10V amplitude application)
What the data sheet does *not* tell, and what probably wasn't intended
by Rossum and his partner either, is that the real fun starts when you
overdrive
the filter so much that the first OTA *output* is clipping. (The input's
amount
of overdrive is frequency dependent because of the feedback loop, but for
the
output things are different.)
Then you have asymmetrical clipping -> even numbered harmonics.
The following stages will filter these harmonics with 18dB / oct, i.e. you
have a dominant 2nd harmonic left.
That was the static case. Dynamically, the feedback loops will correct for
the
remaining offset voltage (caused by asymmetrical clipping), and will change

the amount of distortion in the time after the overdrive occurs.
Haven't figured it out completely yet, but that's the coarse direction at
least.
BTW, one might try some experiments with SSM2024 in that direction. It
also has a simple gm cell without additional current mirrors. (Is that why
the 2024 has such a pleasant sound ??) But the positive swing is limited,
too.
I gues that's because they tried to keep the effect of the Early voltage on
the
symmetry low. So the unsymmetrical clipping effect would not be as
prominent as with the 2040.

JH. 
<

Hi Juergen:

I took a look at your SSM2040 clone, and found it very interesting. The gm
cell as used by ARP is the same circuit.  What ARP did different is run the
output of the gm cell into the summing node of an op-amp, instead of a
darlington buffer.  The limited output swing of the gm cell no longer is an
issue, as the output of the op-amp is where the output swing occurs.  Now
your voltage swing is limited by the op-amp.  If you replace the feedback
resistor of the op-amp with a capacitor, creating an integrator, you then
get a current controlled first order low pass function.  Again, the output
voltage swing is limited only by the op-amp being used.

I am puzzled why the SSM2040 was designed with this inherent limitation on
voltage swing.  Instead of trying to fix it by keeping the input level down
with a 10:1 divider, all you need do is use the op-amp integrator approach
as described above.  All voltage swings are nearly rail to rail.

Unless, of course, the asymmetrical clipping was intentional.  

Terry Michaels