Mini Metrology Lab Build (update)

[Kleinstein]

The TCR matching applies to a single package. Using separate packages can work but matching is not guarantied. However there are also chips with 10 and 12  resistors in one package. Anyway one still needs reasonable matching between the decades. So the total TCR is still important, not only matching.

If at all self heating is a problem for the first decade, the second decade sees only 1/5 the voltage and half the current, so 1/10 the power. It also depends on the resistance chosen.

For trimming it is not such a good idea to use low value pots, as the wiper contact is one point that can change over time. For the combined TC there is no principle difference in using a series or parallel path. The parallel way is usually easier, unless the resistor value is very large.

The more usual way would be a 1 K main resistor in parallel with a 100 K resistor (of reasonable quality, e.g. <= 100 ppm/K) and pot of e.g. 10 K in series - this gives about an 0.1% trimming range. depending on the precision needed / given the trimming range might need to be smaller for the first decade. In this example the TC of the 100 K resistor contributes with about 1/100 and that of the pot with about 1/2000 of its value, so they are far less critical. 

With parallel trimming the resistor value is reduced - so one will need an adjustable resistor shunting the second decade as well (should be something like 200 K + 20 K pot for the example above).

With 1 K resistors in decade 2, one would be at 5 K resistors for decade 1. This would be more like using two 10 K resistors in parallel.

[Dbldutch]

OK JS and Kleinstein, I humbly stand corrected! 

I obviously made a mistake with my PPM calculation. I was influenced by the trim construction that Fluke used in the 720 on the first decades.
OK, another little step closer, thank you for taking the time to point this out.

Kleinstein, or anybody, I could only find one more candidate, the Vishay OSOP series with 10 resistors that seem to have the right specifications. Could you please point me in the right direction for a suitable resistor network with a low TC that packs 12 resistors with adequate precision?

After thinking some more about the practicality of resistor matching for the second decade, it dawned on me that it is quite impossible, or as a minimum really messy, to combine resistors from networks in other decades. With a matching requirement of +/-0,037%, that is not easy to do in any case. If a PCB layout is ever constructed, it will be much easier to bite the bullet and spend another 10-15 Euro's on trimpots and also make the second decade with trimmers. The third decade needs a matching of +/-0.37% and that can be met by using 0.1% individual resistors or a network with that precision, or also use trimmers.

How that will turn out is largely depending on the available resistor values and precision for the networks. The AORN, NOMCA and the OSOP series look good, but the first two are limited in the available resistor values, and pack only 4, 7 or 8 resistors. The OSOP supports 1K, 2K and 10K values with 10 resistors in a package.

For the first decade, I have found with my current KVD prototype, that the resistors should match the 10K value per resistor pretty well. Unless I'm wrong with that, this means that the first decade needs to have a small resistor of 100 Ohm in series with the 10K, in order to adjust it around the 10K value with a shunt of a 1M resistor and a 25K trimpot, following JD's and Kleinstein's suggestions. With the trimpot at 50%, the result is 10K0, and there is a +/- 1.2 Ohm trim possible. That span will most likely be too much for a single turn pot in order to get a match within +/-0,0037% because for 10K the span is only +/-0,37 Ohm. Values can still change to a tighter span, but that depends on the specifications of the available networks.

Thank you all in advance for taking the time to contribute!

[Kleinstein]

You only need the series resistor if you need to get the adjusted value to 10 K exactly, so that to resistance of the whole KDV is at 100 K.  Just for matching to the second decade one can add an adjustable resistor in parallel to the second decade. In this case the total resistance of the KVD might end up at 99 K or so, but this is usually not such a big problem.

With the resistor networks it's not that practical anymore to do individual matching. One could still use adjustment for the whole decade resistance and rely on matching inside the decade.

An important factor for the mechanical form are the switches. Also don't forget about the extra contacts needed to do the adjustment in circuit. With the 10 K resistors one also has to have an eye on isolation. So a printed board might not be that good here - without a board these SMT resistor array may not be that practical anymore. For the lower resistance and less precision lower decades a board might be OK.

[Dbldutch]

Good points! Thank you.

After hemming and hawing, I decided to forget about the use of resister arrays. At first they looked really good because of the resistor-to-resistor low ppm values, but because you can't select them on value, the whole price advantage goes out the window.

So, I guess that if you really would like to have a reduced TC effect on the KVD, you need to go for low PPM/C resistors and because they are too expensive to buy in bulk and then do the precision matching, you have to use individual shunts.

Looking at the TC effects on the first 3 decades, the PPM budget of the decade must be pretty low to stay within the +/- 0,0037% (or 37 PPM/C) matching requirement for the first decade. (BTW, where is this 0,0037% number coming from?) The second is +/- 0,037%, or 370 PPM, and the third is still +/- 0,37%. If you select 5 PPM/C resistors, my hope is that you can reach some reasonable stability in your hobby environment. Provided you can stay within a few degrees temperature variation between calibration and then taking the measurements.

In any case, I found 10K2 5 PPM/C 0,1% resistors (PTF5610K200BZEK)  so they are easy to shunt towards 10K0 for less than 2 Euro's and shunting them with a 510K resistor and a 19 turn 5K trimmer (T93YB502KT20) may actually work. For the second decade, I found a 2K1 resistor (PTF562K1000BYEK) 10 PPM/C 0,1% and if shunted with a 42K2 and a 19 turn 1K trimmer you can trim just above the 2K0 value, so a decade shunt can be used to trim the decade to decade trimming. The third decade can be created with a 1K resistor (PTF561K0000BXR6) with 15 PPM/C and 0,1%. With this precision you may not need an individual shunt, otherwise you can use one with 100K and a 19 turn 10K trimmer. A decade shunt will do the rest, same as with the other decades, but they will only need 1% or less precision to stay within the matching requirement.

For my second KVD prototype, I need to purchase some of these parts and do some experimenting myself, while learning to deal with this TC circus, which I am pretty new at.

[JS]

Kleinstein, why wouldn't you trim the resistors within a decade using a network?

  The resistors are in series and adding the trim circuit is just the same as in the schematic before. If the matching from factory isn't good enough the trim could be added, and the better the matching of the original network the lower the effect of TC of the trim circuit.

  The value of 37ppm is there for a very good reason, the 250µV instead of 270,27µV for the desired 40ppm in the design. IIRC, the 40ppm are there to try to get 20ppm of absolute accuracy. As a comment 10ppm where archived at the end of that project, but not assured by design. That design to that level of confidence is to be able to build it with common metal film resistors, if you are going for a more expensive option with pretrimmed arrays with very good tracking or 0.1 resistors with 5ppm/C you should probably aim higher, thus better matching of the first decades are needed. Also I'd try to get higher power rating for R3 to prevent self heating. If you are aiming for 20ppm you are probably good enough with the original design.

  Using TH resistors I don't know if insulation is still a problem, probably not as long as you keep the PCB clean. Even then, to avoid insulation errors and still have an easy way of building the trim you could use small PCBs mounted over stands and each PCB would receive one end of the 2 resistors (R1 and R3), have the trim pot and connect to the other end of R1 and R3 of the next stage. The alternative is using guard traces from a dummy divider in parallel with the main one guarding each node of the primary from the respective node of the dummy divider. As I said, with big TH resistors, you could even add cut outs, probably not needed of extra efforts. With SIP TH networks the guarding could be done.

JS