[sdiy] uA726 transistor array

[Magnus Danielson]

From: harrybissell <harrybissell at prodigy.net>
Subject: Re: [sdiy] uA726 transistor array
Date: Sun, 13 Oct 2002 00:57:40 -0400

> The uA726 is also extraordinarily stingly with heater
> current... and therefore does not heat the whole synth
> at the same time (making everything ELSE drift).
> 
> I tried some experiments using an LM399 with the
> heatshield removed... and epoxied to a matched pair.
> It did work... but I'd say not as well as a diff pair and
> tempco...

Your milage may vary...

The trouble is how well the heater, the sensor and the transistors sit
thermically. You also must care about how these are oriented in relation to
the path of heat flow and the thermically resistance between the heater and
the other elements. In precission oscillator research they to nifty things like
Spice and FEM analysis of the thermical and electrical path. In a fairly recent
article in IEEE UFFC they had a double-oven crystal oscillator where this have
been done. The uA726 works fairly well due to the tight connection between all
these parts since they share the same chip. Also forgotten is the capacitive
heat property of materials in conjunction to the heat transfer properties.

If you have a too large bulk of material, it will make the heating lag and if
you have a too small bulk you must react quickly to changed properties.
Double-ovens excell over single-ovens, but for the properties we are discussing
I doubt that a well designed single-oven of the expo-pair could be made much
better without also handling the rest of the design in terms of thermal
stability.

Jörgen Bergfors reported to me in a private email-session that he noted that
the LM311 comparator where a source of thermally induced frequency shift. This
is logical since the comparator is used to detect the phase-ramp against some
stable level. The comparator core is another diffrential transistor-pair, so
we should not be supprised.

I haven't spent quality time on trying to design and analyse how a good
ovenized diff-pair should be done, but I have alot of chrystal oven material
to get inspiration from. I even think I have a FEM analysis program around here
somewhere.

> An experiment with better quad transistors like the
> MAT-04... or those.. I mean THAT transistor arrays
> might be better than the 3046....

I've been thinking about MAT-04's too. What I really would like to see is a
MAT-02 (which is a MAT-04 but with pair-wise wireing of transistors to better
handle temperature gradients, a chip-trick being a bit hard to do otherwise)
but with a handfull of extra transistors so that a distributed heater and
multiple senses can be applied. 

I haven't looked at the THAT stuff for ages. Anyone care to toss me an URL?

Cheers,
Magnus