[sdiy] Tempco adjuster idea

[René Schmitz]

Hi Ian,

> So maybe I could save one amp.  I'll think about that method some more.

Maybe one could even save that extra opamp, if the divider can be used. 
I have the feeling that the NTC could also take care of reducing the 
nonlinearity of that divider arrangement.

> Yes, and the tempco of carbon film increases with the R value, also.  
> But I could measure one from a batch and assume the rest are the same.

That is the part which I would want to avoid. :-)

> But what would you recommend for a negative tempco component?  
> Thermistors are really awfully nonlinear.

Awfully nonlinear is what we'd want. What ideally would be required 
would be a device with R~1/T.

We don't have that, but the e(-T) law of a single thermistor can be 
fitted to approximate the 1/T dependancy within 0.5% for a temperature 
range of 0 to 50°C.
(THat is for the method where the NTC does the whole job.)

Since the correction term for the Pt-tempco is much smaller than what we 
need when the thermistor does the job alone, I guess the overall impact 
of the nonlinearity should be vanishingly small. Roughly estimated, the 
nonlinearity makes <0.5% for 3300ppm, so it should be arround 0.015% if 
we merely generate a ~100ppm correction term, as everything is 
approximately scaled down linearly. That would be the error of the 
correction term alone then, which when propagated into the total gain vs 
temperature error of the whole NTC/PTC setup will diminish further.

A more mathematical approach with a proper application of the error 
propagation law would be in order however.

Btw, there is a link to a comparision table on my tempco tutorial for 
the NTC, PT-resistor and the KTY, which Martin brought into the 
discussion. All compared to the ideal dependancy. The PT was for a 
divider arrangement, the KTY was also used in the divider with 1340ohms 
series, and the NTC network was the one at the tutorial page.
As one can see the KTY isn't so bad in a temperature range of 20 to 40°.

> All true. But if the residual drift after fixing the scale factor is 
> nearly linear it can be measured and a single compensator can be used to 
> take care of all these sources at once.  This is what I did with my 
> previous VCO's.  I used a combination of metal film and carbon film 
> resistors to make the oscillator trip point slightly temperature 
> dependent.  As you say, measuring the drift accurately is much harder 
> than actually fixing it.

Sure, all residual linear drift can be compensated (either by your 
method or by brute force summing a PTAT source into the expo inputs.)

> Yep.  But this isn't an impossible task.  Good laboratory 
> instrumentation (Keithley, HP, etc.) routinely makes it into the 50 
> ppm/K regime, and so can we!

I think so too!

Cheers,
  René

-- 
uzs159 at uni-bonn.de
http://www.uni-bonn.de/~uzs159