[sdiy] Makin' OTAs

[Ian Fritz]

Greetings --

After the recent talk about making discrete OTAs to replace the 
CA3080/3280's I decided to spend some time seeing how hard it would be to 
make a good OTA circuit.  My goal was to make an OTA/expo converter combo 
that would track at 1V/Oct within 0.2% or better over 6 octaves.  This 
turned out to be possible, but quite more difficult and expensive than I 
thought it would be.  The reason is that extremely good matching of both 
the mirror and amplifier pairs is needed to achieve this goal.

My circuit uses a 4-transistor circuit for the p-mirror, above a standard 
n-type differential pair amplifier, with the current output converted to a 
voltage using an accurate differential I-V op-amp converter.  The expo 
converter is a standard design.  I left the diff amp and the converter both 
referenced to ground, even though this has some drawbacks because of 
saturation effects.  All testing was done with a signal level of 12 mV at 
the amp input and a maximum Iabc of 1 mA.

This circuit works great if you use supertransistor pairs.  With a MAT03 
for the mirror and MAT02 for the amp the response was perfect on the first 
try.  But this is an expensive solution (around $20 for the 2 pairs.)

Replacing either transistor pair with an inexpensive monolithic pair pulled 
at random "out of the bag"  led to expo tracking errors of up to 
2%.  Clearly both pairs must be carefully selected for very accurate 
matching to get an accurate expo response.

The usual matching procedure of driving collector current through the 
transistors and writing down their Vbe's is not good enough here, unless 
you have something like a 6-digit DVM.  I figured out a couple of ways to 
set up bridge-type circuits to measure either (1) the Vbe difference with 
the same Ic's or (2) the Ic ratio at the same Vbe's.  These methods are 
roughly equivalent, with a 1 mV Vbe difference corresponding to a 4% 
current ratio error.  With care and patience I was able to measure Vbe 
differences down to a few uV's.

For the p-type mirror transistors I measured 20 2SA798 pairs. These cost 
$.50 each on sale from ACME (usually $1).   The results for the Ic errors were:

7 of 20 under 0.2%
14 of 20 under 0.4%
18 of 20 under 0.8%.

For the n-type diff amp transistors I selected through 20 CA3083 Q1/Q2 
pairs.   You can still get these excellent arrays for under $1 if you look 
around.  The results, now in terms of Vbe error, were:

8 of 20 under 50 uV
12 of 20 under 100 uV
18 of 20 under 200 uV.

Interestingly, both of these sets binned out about the same.  To get 
MAT-type performance requires Vbe matching to about the 50 uV level (or 
0.2% in Ic).  This means that you have to de-select about 2/3 of the pairs, 
so the cost for a set of two good pairs comes out to about $6.  Better than 
using the MAT's, but a lot of work.  If you use CA3280's, then you still 
have to de-select some of the devices, maybe 1/3 (based on limited 
measurements), so that turns out to be a comparable price.  (Isn't market 
economics amazing?)

With a 2SA798 pair matched to 0.08% in Ic ratio and a CA3083 pair matched 
to 15 uV in Vbe, I was able to achieve my goal of <0.2% tracking accuracy 
in the overall OTA/converter system.  This level of accuracy is only needed 
for critical applications such as tri-core VCOs, accurate hi-Q filters, 
precise ring modulators, and the VC chaos generators I have been 
developing.  For less critical applications the matching specs can be 
relaxed a bit.  But with offset voltages under a few mV over the entire 
Iabc range, non-thumping VCAs should also be possible with the closely 
matched pairs.

   Ian