matched pairs/quads

[Martin Czech]

Beeing a triangle oscillator disciple I need a PNP matched pair for
the expo converter. (Any suggestions how to circumvent this? Please,
no mirrors). I also need them for all OTA filters, but in this case I
could live with less accurate results.  At least I think so.

(What about amplitude drop problems for a high Q tracking bandpass filter?
B=f/Q, allowed error is thus +-B/2 = +-f/2Q, thus frequency error: 1/2Q,
this will be 5% for Q=10, and only 2.5% for Q=20.  Huu, this requires
quite accurate computation as well, but allows about 10 times more error
than in the audio vco case.)

So far I toyed arrround with CA3096, they are not cheap, and crap, but
available (I guess lateral process option in a npn process).

Next thing is MAT03 and THAT120 and 2SA1349.

Let's see the spec (typical, @ 25C, values made comparable if possible):

                         THAT120             MAT03F           CA3096    2SA1349

Current gain @ 1mA       75                  165              47        200 min
Current gain matching    5% (1mA)            0.5% (100uA)     ?         10% max
Rbe                      2 Ohm               0.3Ohm                     ?
Offset voltage           500uV               40uV             150uV     10mV max
Offset current           70nA                6nA              2nA       ?
Noise voltage density    0.75nV/SQRT(Hz)     0.7nV/SQRT(Hz)   ?
Collector Base Leak      25pA                50pA             60pA      100nA max

So for expo conformance the MAT03 is clearly the winner. High current
gain and lowest base resistance. And better matching.  But very expensive
(I think they really measure each MAT, then put them into different bins,
thus they can guarantee max values, thus expensive).

Next comes THAT and finally 2SA1349 and 3096, the last having very wide spec limits.

Easy. No, not easy at all:

We all know that temperature has a strong influence. The MAT has only 2
transistors, so on chip servo self heating will not be possible. However,
the environment temperature could be regulated (potting, or potting
and tempco).

The transistors will also heat due to the expo current. We can estimate a
maximum transistor Uce 15V and a maximum current of 1mA, thus 15mW. The
MAT comes in a TO78 metal can.  I don't know how many K/W this package
may have. I guess 25K/W, so we have a temperature difference of 0.37K
for maximum expo current modulation.  For a reference current of 100uA
this should give about 0.1% additional error.  This is only true for
the final asymptote value.

If we apply a voltage step to the expo converter, the excess heat
(I don't know how many K peak) will first make the frequency to be
a bit flat (in the normal configuration), then after equilibrium the
frequency rises to the final value. I don't know how fast this will go,
I guess 1s.  Another reason for irregularities in analog VCO circuits,
thus "rich" sound.

The quad packages don't offer any advantage here, because the on chip
regulation can only heat, but not cool down. This would mean that
at least one side of a pulse response would suffer from the
same expo current heating problem.

The same applies for tempco versions also.

So I think that the THAT option does not offer any technical advantage
over the MAT type. MAT03 is not as hard to get but quite expensive ($15),
which I don't care about. And if I would go for a cheap part I'd use some
"dual" PNP as in the previous postings, or maybe even the CA3096.

Unfortunately there is no data sheet available for the SSM types,
but they seem to be out of spec MAT types (red bin). Do they also come
in sealed metal cans? If not, then long time stability is usually by
far worse due to mechanical stress inside the plastic molded package,
temperature dependend of course.


m.c.






----- End Included Message -----