[sdiy] Dial-a-tempco

[Ian Fritz]

Hi all --

I had a chance this week to build and test the tempco-resistor corrector 
idea I posted a few weeks ago.

The basic idea is simple.  A perfect tempco resistor has a PTAT response, 
i.e., R = AT, where A is any constant and T is Kelvin temperature.  This 
gives a tempco of d(Ln R)/dT = 1/T = 3350 ppm/K at 25 deg C.

What if the 25 C coefficient is something other than 3350 ppm/K?  Well, 
it's a fundamental property of metals that their resistance is very nearly 
linear with temperature around ambient temperatures, so the tempco 
deviation has to be understood as a finite intercept at zero temperature, 
i.e., R = AT + B, where B is a constant.  So to turn an imperfect tempco 
into a perfect one just requires a simple circuit to cancel the B term.

I made a converter using this idea with a trimmer to zero out B, using two 
OPA2227 opamps and a CA3083 transistor array.  I separated the converter 
from the oscillator core so I could heat it alone and characterize just its 
own drift.  Since I had already measured the bare tempco's response to be 
about 3270 ppm, I knew in advance approximately how much correction to 
apply.  Starting from that value I was easily able to dial in the proper 
compensation.  Testing the octave from 4 kHz to 8 kHz from 25 C to 45 C I 
saw no observable frequency-ratio change.  So the tracking stays constant 
to the error of the frequency counter (call it 2 Hz) over a 20 C 
range.  This makes the scale-factor drift under 25 ppm/K, or about 1% of 
the uncompensated drift (2330 ppm/K in frequency).

For a sanity check, I figured out what value of B corresponded to the 
(measured) compensation voltage and got 50 Ohms.  This agrees pretty well 
with the value of B = 35 Ohms deduced from the direct tempco measurement.

There was also some linear frequency drift, which could be caused by offset 
voltages, resistor/power supply drift etc.  This, along with the drift of 
the core requires separate compensation.

As a final observation, I would note that the converter is reasonably 
quiet, in contrast to the active compensation schemes I have tried, which 
have extra noise because low-level circuitry must be used for the multipliers.

   Ian