Kelvin Varley Divider [and Precision Voltage Source]

[Kleinstein]

Thermal EMF for the switches would add up for all stages, at least most of it (the average of the 2 switches of 1 decade). So low thermal EMF is still important for the lower stages.

[klimm]

amspire has a concept diagram in post#20 on the first page and I understand it is just a concept.
Specifying the LTC1050 was meant as a generic chopper/zero drift Op amp  in that schematic,  but  not to be used  with a 10V reference at input, right?
It was inviting at first to use the LTC1050 as I have them, but I read in the DS that  the max ratings are±9V  and surely I would not use the chopper at max ratings. Characteristics are given only for ±5V in the DS and I suppose this should be the nominal ±Vs to be used at.
 
What would you recommend to use as followers and summing amp  of the above mentioned concept for the following; 
 AD588 Vref (±3mV(or even 0.002% if I get it calibrated))   outputs max  10mA) followed by a decade of 10x10k or 10x100k (hesitate which to use)  mechanical switched 3 decades  to get a simple DC check unit for 3 1/2 to 4 1/2 DMMs.
I already built the reference and baked it for about 260 hours but  now of course I need  more voltages.

[Conrad Hoffman]

You might take a look at how Analogic/Data Precision did their voltage reference- http://w.ko4bb.com/getsimple/index.php?id=manuals&dir=06_Misc_Test_Equipment/Data_Precision

IMO, once get into summing buffered dividers, their method makes a lot of sense. I think they used an LM399, but the parts are not very exotic to get excellent performance.

[Kleinstein]

For the voltage followers there are several options, depending on the impedance (resistors values) and precision needed. The version with separate buffer for the decades may get away with lower grade buffers for the lower digits. 
The LTC1050 should be OK with a 10 V range (e.g. +12 V and maybe -0.6 V supply)

A few reasonable options are:
OPA202  (non AZ, relatively low bias and drift)
OPA140, OPA191  FET based OP with low offset (rel. expensive)

ICL7650,LTC1050  (AZ OP, some needs external capacitors, up to some 15 V)
LTC2057  (AZ OP up to some 30 V, but bias)
MCP6V51 (cheap AZ OP up to 45 V, but bias  and current noise higher than LTC2057  (useful up to some 30 K, but hardly with 100K)
            - the DS number on current noise is too good )
LTC2055, max4238, LMP2011, LTC2050,...   AZ ops with low bias, but only 5 V - these OPs would need some additional circuit for bootstrapping the supply. The classical way is driving the supply from the output, with some additional low pass filtering. The faster way uses a second amplifier from the input (can be faster and may be a little easier).

For high drive the BS supply version would need an additional in loop buffer for the current. For better precision the additional in loop buffer can also be good for a the other AZ OPs.

Usually a good concept avoids to many critical parts / buffers.

[klimm]

Thank you Conrad,
before asking I already read the whole topic at least twice and  was looking at 8200 and took notes. I will use some info from there but I don't want to use analog switches or Muxes as I would dive in, for me, unknown territory.  I would use some  rotary switches I have,  as they are metrological grade and work in decades not in heptal or octal system.

Kleinstein,
trying to be short, I think I was not very clear in my description. When talking about 3 decades I meant actually that after the one and only  resistor decade of say10X10K,  follow 3 rotary switches and 3 buffers(1 following each switch)  that deliver their signals to a summing amp. Going down the ranges from volts to tenths and hundredths of volt,  I would need at least same low offset and drift op amps  as for volts range, am I  right?

[Kleinstein]

In the circuit as in post 20 (summing amplifier with weights 1:0.1:0.01:..., the lower decades have less stringent requirement on the buffers, as there weight is lower. They still need very low input bias, if a common 1..10 chain is used.  So one may be able to use some good FET types (e.g. OPA145, OPA191) at least for the 3rd and lower decade. The 3rd decade has 1/100 the weight and thus the offset effect 100 times smaller. It would also still be a constant added to the output.

The common decade divider should be low impedance, so 10 x 1 K. With 10x10 K the demands on buffer input bias gets quite high !

[klimm]

Thank you for clarifying this.  I was only thinking at the lower voltages not taking weighting into account although I saw the  TL062s in the 8200.
In the D.P. 8200 they use 500 Ohm resistors  resulting a current demand from the Vref of 2mA. As my reference delivers max 10mA I think I am on the safe side with 10x1K and 1mA through the divider  but I can't  analyze  how this would affect if wanting  to hang more buffers to get  more resolution; there would be more current necessary to flow through the divider I suppose as energy won't come from nowhere even if Ib is really low, if using  say 5 Buffers? Sorry for bothering but I am by no means comfortable with  Op Amps

[Kleinstein]

The bias can be quite different between OPs:
The TL062 or similar are likely in the 10 pA range - so really low bias, but also poor DC specs.
With AZ OPs like LTC2057 / MCP6V51 expect a few 100 pA. So with 5 K impedance in the middle this could change things by a  may be 1 µV. Probably OK if not in the absolute volt nut range.
BJT base OPs may be a few nA  - and thus in the 10s of µVs.

One would probably not use the same type of buffer for all cases.
A possible combination may be ICL7650, LTC2057, MCP6V51 or maybe OPA191  for the 1st stage and output.
e.g. OPA191, OPA145 or maybe OPA171 for the 2nd digit.
e.g. OPA171 or TLC277 for the 3rd
e.g. TLC277, TLC272 or even lower grade for the 4th / 5 th
Especially for the lower grades there are many options.
For the later digits low bias is as important as for the first, but offset, noise and drift get less important.
It may help to have single supply capable OPs, so that one would only need a small negative supply, like -0.6 V.

[klimm]

I need to compare some 1K  wirewound resistor (flat wounded on 2mm epoxi-fiberglass PCB in the picture) in order to select them for the divider of 10X1k  .  They are about max 0.1% to each other according to my 4 1/2 DMM but resolution  is not enough. I think of using a Wheatstone bridge.
Now comes my problem;
The resistors are made of really thin wire,  0.1 to 0.2mm only, so very low power.  They were used into a Feussner Kompensator that was working at a current of 100uA. So I suppose that when measuring with the bridge I need to limit the current to  a maximum  of  200uA, 100uA for each branch, in order not to stress the resistors. Yes I would use a morse key to power the bridge for the time of the measurement  but still can't use a voltage like 9V
Any sugestions?

[Kleinstein]

One can always reduce the current to the bridge with an extra resistor in series to the whole bridge. So maybe start with some 1.5 V and than have some 5-10 K in series to limit the current. Another option would be to use a transformer with a low frequency AC and measure the bridge with AC. AC amplifiers are usually easier than DC and 1 K is low enough to work well with some 100 Hz. The transformer could also provide 1 side of the bridge.

Just looking for equal resistors one can have 3 fixed resistors and just compare the bridge reading for each each resistor.

The picture already looks like a good resistor decade or KV divider. So maybe do a little reverse engineering to see if it could be used largely as is.

[klimm]

Kleinstein,
in the picture you see a Feussner Kompensator.

I was thinking of a series resistor but I thought that I would see a lower voltage at the bridge input as the supply sees the resistor and the bridge as a divider. Is this wrong?

[Kleinstein]

The extra resistor and the bridge would form a divider - that is right. For the bridge the driving signal does not have to be very stable of exactly known, so the extra divider is not a problem. Of cause one would have a lower voltage at the bridge - can't change that is the current needs to be small.

The resistors don't look that small, so something like 1 or 1. V and 1 to 1.5 mA should be OK for the test. This is still only 1 - 2.3 mW, so not yet a real problem with self heating.

That original apparatus looks really nice and could also be used to set fine voltage steps - likely better than the circuit from post #20. The main part is a precision divider, just using more resistors than the KV circuit, but just as good, even with some slight advantages.

[klimm]

I will try to feed the decade with 10V and see if it heats at 1mA. Supposedly it would hold max 16mA as I can read on a decade box, I just opened and,  made with the same resistors.
this part added:==================
A short computation gives a max power P=0.016A*0.016A*1000R=0.256w. So, you should be right

I let here a link about bridge sensitivity that I found really helpful  https://electronics.stackexchange.com/questions/110588/sensitivity-of-wheatstone-bridge
End of added part=================
Yes, the Feussner kompensator is a neat thing. I opened it  to try getting the schematic as I don't have much info.  The lower range of switches  is an auxiliary decade to set the current in the auxiliary circuit.  One could calibrate 200mv and 2V ranges of 0.1 accuracy DMMs  in small steps, provided a Weston cell is present for calibration but  I don't have one. I was thinking of making a divider buffered by an OP07 on my Vref10V as the accuracy of the device is about 0.01 so far I know,   This one has some defective resistors I was said. Measuring the 1k decade I found one of them interrupted.  I am not determined to break it apart but can do some measurements to see how close resistors  are to one another and asses if I could use them.
Of course, best would be to be able to repair and use the thing as is.

[Kleinstein]

A broken resistor is a very real possibility.  On could very well be lucky and have only one broken resistor. Once broken the instrument my no be used very much.
A repair would be very well possible, if only a few bad resistors.