Fluke 895A 19.999k precision resistor pairs

[trobbins]

A friend has a pulled Kelvin-Varley board assembly from a Fluke 895A Differential Voltmeter.  The board had 12 pairs of 19.999k +/-0.01% 1W Fluke 192583 resistors that the manual describes as factory matched for resistance, tolerance and tempco.  Each pair is in series with 10 ohm trimmer as a 40k section.  The board now has 6 pairs left.

The pairs are comprised of one resistor with FHIP0.5 marking, and the other with FLIN0.5 marking.  Similar to photo in #37 (https://www.eevblog.com/forum/metrology/best-way-for-decade-resistor/msg1028398/#msg1028398).

I don't have the instrumentation to resolve their resistance to better than 19.999k, let alone their tempco.  As a form of confidence, my HP3597A does measure each of the 12 remaining individual resistors with values varying from 19.999k to 20.001k, with one down at 19.997k.

Although a single resistor could be used as a stand-alone reference (ie. removed from pcb and remounted in an enclosure), it would seem wiser to re-use them as pairs, given their original factory matching may still provide some tangible benefit even given age. 

Would the original factory matching (and the different markings) indicate more credence to a pair being tempco matched (ie. one exhibiting negative and the other positive tempco) in the region of nominal lab temp, rather than resistance matched (given there is a 10 ohm trimmer in series with each pair)?

My RLC meter can't resolve to better than 10 ohm for each resistor, but does indicate a noticeable self inductance and a need for some shunt capacitance to normalise impedance if they were used for low frequency (<10kHz) measurement setups.

[alm]

I would indeed say they are resistors with opposing temperature coefficients. Search the forum for discussion about the Fluke 5450A or 742A. I'm pretty sure this topic came up in relation to one of them. So keeping them in pairs makes sense.

The differential voltmeter is a dc only device, so the designers didn't take ac performance into account.

[razvan784]

Yes, they are certainly tempco-matched (P 0.5 with N 0.5 should ideally cancel). The resistance values would simply be within the design tolerance, in order to keep the trimmer value low, I see no reason to match the values within the pair to better than nominal tolerance.
Edit: The most probable reason for this approach is that it's very difficult to build a 0.0 ppm/K tempco resistor. You can target 0 and get something like a -3 ... +3 ppm/K distribution. You can select a few "perfect" resistors from that batch, but you'd probably throw out >95% of the total. Or, you can label every resistor with its measured tempco and then select opposing tempcos to get an almost perfect cancellation.

[Dr. Frank]

yes, these resistors are at first matched pairwise for opposing T.C., so definitely keep all those pairs together, with their individual trimming resistors.
Then they are aligned /trimmed for exactly 40k , I.e. 2x19k999 + trim pot + fixed resistor, or similar trimming scheme.
As these are probably the resistors within the 1st decade of the Kelvin Varley Divider, this exact trimming gives a very good linearity, to around 1ppm, or so.
Make your decision what you want to do with those resistors. if you have 4 pairs available, you could build either 4x 10k reference resistors by paralleling two 19k999 plus trim, or one super stable 10k by paralleling 4 pairs, which are in series each, and adding a very fine trimming network, to resolve < 1ppm
Frank
P.S. I published a description about the FLUKE matching method in the thread 332B/AF in the slaughterhouse, or similar title, and/or in my article about the 5450A resistor standard

[trobbins]

Thanks for the directions and enlightenment - that makes it easier to search for and read through other relevant comments/threads. 

The post (https://www.eevblog.com/forum/metrology/fluke-742a-10k-and-hp3458a/msg1078377/#msg1078377) clarifies the original resistors I have are +0.5ppm/K and -0.5ppm/K batched, although I'm not quite sure if the FHI and FLI markings are tempco or resistance batch related.

As the board is a friends, I will try and get him to at least relocate one pair into a separate enclosure with a triangle of 4mm sockets to allow easy measurement of either 10k or 40k from the matched pair, and two 20k options from low ppm parts.  I can't measure them to any level of confidence yet so there is no incentive to include trimming at this time, or to choose one pair over another.

And it was pleasant to get some insight on tempco variation, so there is half a chance these parts have not been stressed.  I'll just try and construct a reference using the many examples and comments in the various threads.