[sdiy] Potted expo converters

[Magnus Danielson]

From: Peter Grenader <peter at buzzclick-music.com>
Subject: [sdiy] Potted expo converters
Date: Tue, 30 Dec 2003 06:24:16 -0800
Message-ID: <BC16CA89.FD35%peter at buzzclick-music.com>

> Ladies Tangents,

Peter Buzzington,

> I remember one day I got all excited at this brainstorm I got that you could
> (sort of) duplicate the functionality of the ua726 by using the remaining
> trannies in a 3046 as a self heater, only to find it had been done a billion
> times before when I posted the idea on this list.
> 
> d'OHI!
> 
> So, here's another lightbulb that appeared over my head that I'd thought I'd
> bounce off you all  that I'm also willing to bet ain't original, either...
> 
> I remember when I was an engineering tech two billion years ago, we used to
> glue thermocouples to the windings of 1/2 horsepower electric motors using
> thermally conductive epoxy for UL lock-rotor tests. the material seemed to
> be was astonishingly heat conductive. But...we were dealing with
> thermocouples which could detect a match being lit in the next room (not
> really of course, but you know what I mean).
> 
> Along the same lines, has anyone tried potting the tempco/expo converter
> pair in conductive epoxy to A) improve heat transfer from one another and B)
> aid in isolation from ambient conditions?
> 
> Just curious.

Well... yes!

If you use separate transistors, it have been done for quite some time and in
many different fields. For instance, I've seen many cases where the TO-72s have
been put "face-to-face" and with a copper-clamp around them for both heat-
conduction and just plain mechanical clamp. dBX used a small aluminium case on
top of the pair for their famous VCAs.

When you use MAT-02s, LM394s or what have you, the transistors sit on the same
silicon-chip, which as such binds the transistors together. However, when you
do that you make 4 transistors in a square and bind them together in a diagonal
fashion, so that the (1,1) and (2,2) transistors forms one transistor and the
(1,2) and (2,1) transistors forms the other transistors. This makes the
compound transistors form in a diffrential pair situation a better thermical
stability from thermical gradients, i.e. a temperature difference between the
sides of the transistor setup. The compensation is not perfect, but nocks off
the first-grade difference.

Now, if we now succseeded having the transistors of the diff-pair at the same
temperature, where' all set? No!
The temperature difference between the transistors helps to null-out the "Is"
differences, but doesn't null-out the temperature differences of the exponents
q/(kT) term. There is two traditional approaches to this offset, either
external compensation of the gain difference, usually done with the Tellabs Q81
temperature-dependent resistance which *almost* nulls out the temperature
dependence in the scale. The other approach is to lock the temperature of the
transistors through a oven-design and then make a static compensation of the
gain constant (trimmable ofcourse, but not under electrical/temperature
control).

Those being alert have seen the discussions and experiments (ending up very
succsessful) of a tempco-resistorless compensation scheme, by the use of the
same temperature-dependence, namely that of a semiconductor. This is just a new
approach to the external compensation, but on a theoretical point of view
reaching for a perfect compensation.

The external compensation tricks however all depend on the sensing of the
temperature of the NP-junctions of the transistors in the diff-pair, and the
missmatch there will cause missmatching. Therefore it is important to have the
tempco-resistor (or whatever sense-method in use) thermically ties to the
NP-junctions of the diff-pair transistors.

For instance, in one version of the MiniMoog, the tempco-resistors where
sitting fairly far away from the expo-transistors, causing the compensation not
to track the correct temperature. Moving the actual resistors over to the
diff-pair, tieing it thermically to the diff-pair and then wires over to the
original solderingpoint made the same electrical design much more stable. I.e.
the lesson is that it's not all in the schematics.

There is one additional lesson to learn. The expo-pair transistors and whatever
else you have thermically tied to them is not "dead" components, they are quite
active (or they would be useless) and this means they themselfs dissapate heat.
The outgoing transistor dissapates a very large range of heat energy, since it
depends on the current it currently is transmitting. So, in order for correct
compensation, it is hard to keep the same temperature of the components
individually, but tight thermical coupling between them is necessary to keep
them close in temperature.

> I know a few manufacturers, Steiner-Parker included,  used to place all
> three parts in a little metal box filled with thermal compound.  The
> alternative I'm mentioning here seems much less messy.
> 
> Considering we're dealing with comparatively low voltages, I assume there
> wouldn't be a lot of difference, but if it meant an improvement in stability
> even a little may be worth it.

I don't think you have to go to such extremes. Tying things together
thermically is what you want. However, I would spend a few moments caring about
the output collector off the diff-pair. You don't want to create a too low-
impedance path to anything else. So you might creating a new problem when
trying to solve another. Don't do that, it could spoil your low-freq tracking
if you care about huge ranges of frequency-tracking. It's more about watching
out and avoid a few selections than anything else.

> I tell you what - it's a lot easier than
> scanning the globe for 726s all the damn time.

Indeed.

> anyone?
> 
> anyone??

How's bad?

Cheers,
Magnus