[sdiy] Tempco adjuster idea

[Magnus Danielson]

From: Ian Fritz <ijfritz at earthlink.net>
Subject: Re: [sdiy] Tempco adjuster idea
Date: Sun, 27 Apr 2003 09:12:15 -0600

> Hi Magnus --

Hi Ian,

> At 05:47 AM 4/27/2003, Magnus Danielson wrote:
> >
> >Things like bulk resistance can be handled separately by fortunately much 
> >smaller
> >alterations to the curcuit. However, the bulk resistance compensation 
> >usually seen
> >out there only handles part of the bulk resistance problem, as I have 
> >pointed out
> >before.
> 
> Correct, but it is pretty close as long as you use a decent transistor. 
> Also, remember that Rene came up with a clever improvement to the usual Rbe 
> scheme that should take care of your objections.

Which of them? I've lost track really...

My point was that there is two rBE to compensate, and they run under sufficiently
different conditions since the current through the two rBEs is not nearly the same.
On the reference transistor, the steady state current is almost perfectly the same
all the time. On the output transistor it varies over a huge range.

I had some proposals for how to remedy that as I recall it. I think it came out of
refinements of Rene's curcuit, but I don't recall it exactly now.

> >So, handling the temperature compensation which comes from the fundamental 
> >expression
> >kT (which is a quantum mechanic energylevel) is still a valid study.
> 
> Actually, the kT comes from classical statistical mechanics 
> (thermodynamics), but thanks for the support!  :-)

Yes, it does... and to show why they invented the quantum mechanics, since it prooved
to be part of quantum mechanic behaviour, where statistical mechanics is mearly an
extention to that field (even if the areas where originally invested the other way
around).

> > > The TKR is defined as 1/Ro*(R(T)-Ro)/(T-To), which is why you always
> > > need to know what Ro and To is. This is different from the above
> > > differential definition except around To and also different from
> > > (dR/dT) except for linear temperature dependence.
> >
> >True linear curve may not exist really, but it may be accurate enought in 
> >the range
> >we are discussing.
> 
> That's what I want to get at ultimately.  With the correct tempco at room 
> temperature how do the remaining nonlinearities and drift stack up vs. 
> whatever errors you get with active compensation (like the spectacular 
> Patchell result)?

That takes carefull measurements. I have two good DMMs, but no means to measure
temperature very accurately. It seems nobody really seems to give out the details on
that. Any proposals?

Ease of calibration is my concern. (Ice-water and boiling water in my lab doesn't
*really* feel like a good idea.)

> >However, the tempco-less discussions we had before seemed very interesting 
> >and with
> >the amount of hardware used to compensate the curve of tempcos we are not 
> >too far off
> >in hardware to start with.
> 
> Agreed. And I know Jim is working on simplifying his design. Now if I could 
> just understand how it works!

Ah! Actually, I looked at Jim's design and considered an alternative design. Jim used
up a complete OTA for the inverse properties, where as I worked on how to do things
properly with the use of linearizing diodes. I put it away, but I felt quite confident
that reductions of Jim's curcuit could be done.

The whole trick where to do a "bandgap" temperature reference and multiply the
incomming CV with that reference, that was the idea. This multiplication naturally
needs to be very linear... and you end up having some problems in another corner.
However, such things can be solved. Once you know that general plan, it's the
execusion of the plan which should form the next layer of confusion.

> > > No, you still have to fight both coefficients. The offsets introduced
> > > by B can be canceled by trimming at a single temperature. Then when
> > > you find that the compensation fades as you go away from that
> > > temperature, that means that A is incorrect. You can trim A towards
> > > zero rather easily with passive components, but not the other way
> > > around. BTW, Conrad has a bunch of 3900ppm/K temp sensors from Heraeus
> > > that are almost reasonably priced.
> >
> >Why do I start to consider ovenized oscillators as a fair deal?
> 
> I have no idea.  :-)
> 
> Are you talking about putting the whole circuit in an oven, or just the 
> usual heat-the-converter-transistor business?

Whatever parts that could seriously damage tuning properties by temperature
sensitivity, naturally. Naturally you'd like to keep as little in an oven as you can,
since the bigger curcuits the harder to control and heat up.

The comparator seems to be forgotten all the time BTW. I've never seen an analys of
its sensitivity from external sources (temperature included) into the frequency of the
oscillator.

> >My point. Ovenized oscillators is a known trick of the trade and when done 
> >properly
> >you can acheive *really* stable temperature, enought to make that ~ 3300 
> >ppm/C dampen
> >down to way below other sources of drift. This is done for crystal 
> >oscillators on a
> >regular basis.
> 
> It could be the ultimate solution, although a bit inelegant.  One thing to 
> watch for would be whether the high temperature degrades any component 
> performance. You might want to get rid of all the FET-input op amps, for 
> example.

So you are saying that I should not heat things to 1500 K unless I take out the
FET-opamps? ;O)

Cheers,
Magnus