poor man's 1 : 100 resistance transfer standard

[Marcus_S]

Although still being in the metrology kindergarten I dare to present a little experiment.

Playing around with multimeters I liked to have a resistance transfer standard to transfer the 10 k and 100 k to 1 M and 10 M with moderate accuracy. Looking into the specifications of the multimeters (not very impressive in these ranges) and to my bank account I did not want to buy several SR-1010. But I thought about adapting the concept and building a low-accuracy poor-man's pocket version:

fig_1

The idea was to take leaded metal film resistors (1 M and 100 K, 1 %, TK 50, Vishay Beyschlag MBB), select them to be within 100 ppm of value with each other and assemble them in good thermal contact using a piece of aluminium. Using 25-pole sub-D-connectors as programming/shorting bars allows easy and fast switching from 0.1*R to 10*R.


How it was done:

fig_2

Taking 100 resistors, select them. Taking some some scrap aluminium and a milling machine gives the casing and will probably help to achieve somewhat homogeneous temperature and and electromagnetic shielding. A small hole in the bottom section of the casing allows the application of a temperature sensor.

Traces were cut into a piece of thin (1 mm) epoxy-pcb-material. The circuit board was glued into the casing. Copper wires were soldered to the traces to reduce the resistance of the traces. The leads of the selected resistors were bent carefully and the resistors were soldered onto the traces and to the sub-D-connector. Two contact pins of the connector were used in parallel to lower the contact resistance. Using a large area for soldering the resistor to the circuit board I tried to obtain a good heat-flow from the leads of the resistor into the casing:

fig_3
fig_4

To contact this module a "contact-box" was made. Two sub-D-connectors were "programmed" to give 0.1*R and the 10*R and connected to the 4 mm-sockets in parallel using the special very-ugly-wiring technique:

fig_5


The performance:

Playing around with a spreadsheet shows the error of the 0.1*R to 10*R ratio to be much lower than 1 ppm when using selected resistors with a tolerance within 100 ppm.

The max. error of the resistance due to self heating during measurement can be estimated to be 2,5 ppm (100 k-range, current of 100 µA, 100 µW per resistor worst case, assuming a R(thermal) of 500 K/W).

The isolation resistance was measured roughly (using an old Keithley 610C electrometer) to be approximately
75 GOhm (contact box - connector to connector),
25 GOhm (contact box with both modules installed - connectors to casing)
> 2 TOhm (modules - contacts to casing, surprisingly high).

The shocking 75 GOhm might add an error of approx. 140 ppm in the 10 M-range and approx. 14 ppm in the 10 M-range... One day I should modify the terminals... There must be some PTFE in the basement...

Finally I tested the device using a Tektronix DMM4050 (the only calibrated meter I have) in statistics mode:

Module  --  Range  --  T/°C   R (2 wire)  --  SD  --  N
10k/1M  --  10k  --  27.3  --  10.02457 k  --  1.90 m  --  128
10k/1M  --  1 M  --  27.5  --  1.002465 M  --  769 m  --  100

100k/10M  --  100 k  -- 27.4  --  100.0285 k  --  146 m     --  101
100k/10M  --  10 M  --  27.4  --  10.00291 M  --  73.9  --  100 (first measurement)
100k/10M  --  10 M  --  27.3  --  10.00290 M  --  145  --  100 (second measurement)
100k/10M  --  10 M  --  27.5  --  9.995030 M  --   ---   --  20   (cables twisted...)

Seems to work, seems to be good enough for the designated purpose. These measurements indicate an error of approx. 10 ppm for the 0.1 : 10 ratio. Comparing these results with the specifications of the DMM4050 (1 year):
1 M  --  100 ppm of rdg. + 10 ppm of range
10 M  --  400 ppm of rdg. + 10 ppm of range
shows no severe nonsense. But I cannot find the 140 ppm-error. I have to retry the measurements when it is a little cooler (30 °C today...).


And: the twisted cables seem to have an isolation resistance of around 12 GOhm dependending how they are twisted... Good for additional 800 ppm of error...


Best regards

Marcus

[zlymex]

Nicely done!
It's true that if the deviations of the resistors are all within 1%, the error of 1:100 ratio will be within 10ppm.
0.3% of deviation will result in less than 1ppm ratio error.
When I made my 100k hamon consisting of 11 Fluke 99.925k hermetic resistors, I adjusted the values to within 0.01% nominal so that the ratio error can be ignored safely for my measurement conditions.

[guenthert]

> Using 25-pole sub-D-connectors as programming/shorting bars allows easy and fast switching from 0.1*R to 10*R.
I like that idea. 

[Marcus_S]

Thank you very much for all the comments.

@ DiligentMinds.com
I like the idea to make a bridge circuit for resistance measurements but I have to think a lot about the design (to make such a device with my knowledge...). And currently I am busy with trying to adjust my other meters to my calibrated DMM4050. Therefore the bridge has to wait some time.

The short time stability I tried to achieve using the metal casing. And I decided to ignore drift due to humidity or aging for the duration of the measurement. But I agree - it would be better to use resistors with a lower t.c. (although I like the MBBs for very special reasons). As a first compromise I will check the t.c. of the assembly. A first quick test is very surprising, I have to repeat this more carefully. I will report back.

[JS]

> Using 25-pole sub-D-connectors as programming/shorting bars allows easy and fast switching from 0.1*R to 10*R.
I like that idea. 

I agree.  I also like the machined aluminium case, which will help ensure an isothermal environment for the resistors.  The only improvement I would make is to upgrade the quality of the resistors [which is easy to do up to 1M\$\Omega\$].

*** EDIT ***
The DB-25 might become an issue at the lower resistance values [say, <= 10K\$\Omega\$], and so for those, a lower resistance connection system will need to be designed.

This looks like fun, and I might try this.

  It seems he is doubling the connections, there are DB25 rated for <30m? connections, for the series it would add <30m? (one on each side, 2 in parallel). For the parallel would add <5.5m?, 5 connections in one side, 6 in the other, twice in parallel. For the 1ppm target is already too much for using the 10k resistors, as it would add 5ppm in the parallel mode.

  Using the second connection would allow for 4 terminal resistor approach in both, series an parallel modes. You could also use 9 resistors 3S and 3P and get the resistor value and use the 4 terminal approach here as well. One would be off the network but if you select just that one to be the closer to the average of the 10 you reduce that error. Using the 4 terminal connection it could probably go to quite low values and have reasonable results, probably not 1ppm but could still be useful.

  I've just got ten 10k? resistors (away from my lab and 1% are oddly expensive here) and they had an standard deviation of ~0.2% so if those numbers are ok 1ppm seems not a hard thing to get. I measured  all 10, got the average 9k996 (only 1LSB changed of one single resistor before or after soldering given enough time to settle). I used the worst DB25 I ever seen and only using single connection. The one off the 3P3S mode read 9k97 BTW. Luckily for me today is 20ºC in here, nice day to measure resistors.

AVG=     9k967
Series=  99k6
3P3S=    9k97
Parallel= 998?

  I'm using the oficial DMM to get my numbers, I've got it yesterday. I could get it to the university nest week and have a better look. I guess the extra ? from the parallel expected reading could be helped by the bad connections, which would make the DMM track for the LSB within the 3 ranges.

  As there still are 3 extra terminals in the DB25 a temp sensor could be added inside to have as a reference.

JS

[quarks]

Very nice, thanks for sharing