[sdiy] expo accuracy? or integrator accuracy?, or both?

[Magnus Danielson]

From: Grant Richter <grichter at asapnet.net>
Subject: Re: [sdiy] expo accuracy? or integrator accuracy?, or both?
Date: Mon, 10 Feb 2003 02:25:47 -0600

Dear Grant,

> With all due respect to Ian, Magnus, Rene et al.
> > 
> > While I agree with you, I just want you to remember that one of the things we
> > where discussing was an alternative to TEMPCO resistors and alternative
> > solutions to that. That is one of the things which I think is fashinating and
> > the experiments that Tim and Scott have done is encouraging to say the least.
> > But onces you've got an oscillator which allows itself to be built from not so
> > rare components and doesn't need constant tuning to be reasnoble, then looking
> > at means of modulation, timbral aspects, etc. all come in, I agree, but that
> 
> As my understanding goes, each component is specified with a tempco, but
> then there is a tolerance on the tempco. For example polycarbonate caps with
> a +150 ppm +/-50 ppm or between 100 and 200 ppm.
> 
> Now someone pointed out that the silicon will have a tempco tolerance and is
> only centered on 3300 ppm.

That's not quite how it works. In semiconductors you have two tempco
responses, that of the exponential property, one of the Is reference current
(which actually is a quite messy expression) and to some degree also the
bulk resistance. The bulk resistance influence is something which is reduced
by simply minimizing the bulk resistance parameter (one of the selling points
for the MAT-02). The Is dependence is removed in the expo-pair setups we
have, and the exponential property is due to the q/kT multiplicative for which
the voltage over a NP-junction depends. This is actually due to the quantum
mechanics in action.

So, the 3300 ppm/C is not a constant with per device fluctuation (given that
the classic percaussions have been taken) but rather just a linearized
estimate which is as such temperature dependent. Using 3300 ppm/C tempco
resistors have been enought to get into the right neighborhood and
sufficiently accurate tuning.

> So each unique collection of components will have a unique composite ppm
> drift.
> 
> So if you develop a stable but temperature variable multiplier for the 1
> volt to ~18mv conversion, wouldn't you still have to do a zero and span
> calibration using an actual temperature chamber to calibrate it? Mucho work!

No, not really. The proposed scheme efficiently removes the major component
of the expo-pair temperature dependence with sufficient accuracy and no
specific trimming is needed for temperature tracking. The trimmings needed do
IMHO not depend on ultrastable or ultraknown temperatures but is more of the
classic offset/scale type and should be possible to do quite quickly.

> I have also noticed that the 1 watt copper wirewound tempcos (KRL) exhibit a
> better short term stability, than the same circuit with an 1/8 watt film
> tempco (because the 1 watt has a much larger thermal mass?). I don't think
> you WANT to change the scale factor very fast. I get the impression that
> adds instability from air movement.

The thermical system is always important to properly consider, and the really
stable double-oven setups do real modeling on this.

> It seems these exquisite design enhancements may have reached the law of
> decreasing returns.

It is there, and the point is certainly there that you run into other
problems, but I kindly disagree with part of your reasoning. I think that the
work done in doing a propper q/kT compensation is thrilling, but that what is
now needed is reduction in size. For instance, the use of dual OTAs is not
necessary and I have advocated for the diffrential current drive of a single
linearized OTA. It would be interesting of this linearized drive could be
neatly combined with the kT/q sensing.

The early measurements have been encouraging and the final doom of the design
complexity needed have not been laid down. Given the preliminary results I
think that with wise choice of other components, the oscillator should be
good enought for most uses. Naturally will other aspects like reset-time
compensation etc. remain to be done, but thankfully it's an orthogonal issue.

I do agree that the current designs look a little too big. Also, important
aspects like thermal interconnection have not been explicitly addressed but is
assumed due to the nature of the field. I can see the reason for worry, but as
I said I kindly disagree with part of your reasoning.

Cheers,
Magnus