Mini Metrology Lab: some parts questions.

[Jebnor]

For those unfamiliar with this old project, here it is http://conradhoffman.com/mini_metro_lab.html
I believe the Author of said article is lurking around here somewhere too.   I've read the articles many times and find it quite interesting and enjoyable.

For no one in particular, the questions:

1) I'm planning on ordering the MAX6250ACPA+ as Vref. Cheaper and better specs (from digikey anyway.) Any particular problems with this chip?

2) For the K-V divider, What is the purpose of the continually decreasing values from decade to decade?  The maths tells me that I could do all decades with 10k resistors.  I seem to recall something, from somewhere,  that the reduction of value by decade had to do with output impedance or something.   I am planning on making a 6, possibly 7, decade box using the jumper method.  The decades I am planning are made of 10k, 10k, 10k, 4k99, 4k99, 1k and 499R.   Each stage will have a trimmer network made of 10k, 4k99,1k and 499R and 5K trimmers as appropriate (See attached picture below.)    The primary reason for considering the more limited quantity of values is cost, and things I already have in my junk box. (I once upon a time ordered 25 5k 10 turn trimmer pots for reasons I can't remember.)  Also, keeping the lowest value to 499R ensures that the connection resistance remains down in the 'noise'. I will be using 3 wire connectors I get from my hobby store for Servos.

3) Is the calibration service still offered by the author? 

I'll be rolling some of my own boards for these to fit the boxes I have chosen.  Pictures as I progress.

[Dave]

You could implement all decades with 10k resistors, but think about the currents.
The first decade will pull U/100kOhm, the second decade will see 1/10 of the voltage and have the same resistance, so one tenth of the current of the first decade. By the time you get to the fifth, sixth or even seventh decade, your current will be reduced so much, that pretty much any load (even a super-duper-low input bias current opamp) will create a huge error on the output voltage.

Also, bear in mind that you don't want to go too low on resistance for the less significant decades, as switch resistance will come into play.

[Andreas]

1) I'm planning on ordering the MAX6250ACPA+ as Vref. Cheaper and better specs (from digikey anyway.) Any particular problems with this chip?

The original cirquit uses a LT1027BCH in a hermetically tight package. Which is much more stable over time than any DIP8 plastic package which is more or less sensitive to humidity. Unfortunately the TO-99 package is no longer produced with the LT1027.

Datasheet specs have to be read carefully. Often the "good" specs are only valid for the best (military) package (not the cheapest) of the series. And unfortunately since ROHS these packages are mostly unobtanium.

So the MAX6250ACPA+ is a buried zener in plastic dip and plays in the same ball park as LT1027CN8 or AD586MNZ or MAX6350CP.

The pitfall of the cirquit is the resistor divider R4+R5 for setting the output voltage to 10V. Even if you select the resistors with TC the ageing of standard resistors will be much larger than the drift of the reference.
And real precision resistors will cost more than the reference itself.
So if you really want to have 10V as output voltage it is cheaper to use a 10V buried zener reference in a hermetically tight case.

With best regards

Andreas

[PA4TIM]

http://en.wikipedia.org/wiki/Kelvin%E2%80%93Varley_divider

A normal decade box is not a good idea. Your reference must see a high resistive constand low load. Your meter must see a constant resistance. .. A decade box will change impedance and the multimeter will form parralel resistance with the divider. The output of the KV will stay constant .

A KV is often used with a null-detector and standard cells. The KV was set at the value of the standard cell and the calibrator adjusted untill the null detector showed zero. In that case the calibrator was exact 1.1V,  10V or 100V and could be used for further tests. A Vref direct to a KV is possible if it is well buffered. I made a compound amp with a LT1052 chopper and LT1010 buffer (nummbers from the top of my head and I'm bad with that) it iis a voltage follower I use to buffer references before I connect some standards (I measure them with a 10 GOhm 7,5 digit to see if this is needed. My Fluke 731A surely needs it.
A decade is a waist of time and money for a reference.

If you use the last decade at 10, your 10 M multimeter is paralleled with 10 K. You can calculate the error. But now switch up to 9, so there is 90K parallel with 10M.
The KV will show the same resistance in every position.

Be carefull, the specs are often spread over a range. And the change you get one on the outerbounds is bigger as you think. They often sell several grades so carefull read the datasheets. Those letters often are important. The 1027 is available in A,B, C. I bought some C's at Farnell. I could not get a B. I then got a sample B from LT. I could not get the A version.

I have build a few standards. One with a LM399. That is now running 24/7 for a year. It drifted a lot at first. a few hundered uV. Then it jojo-ed plus/min 40uV a day (op 10V) but it does not drift any more. Changes are related to temperature and humidity. I monitored it a few weeks in the authum and noted the value together with rH% and tem in a graph. rH% seemed to give the biggest jumps. Tempco was good. It could jump 30 uV through tempchanges but that was over 5 or more degrees. so 6uV/C or 0.6ppm/C at 10V.  But to my suprise, one of my first references made with 4 x TL431 parallel and a OP277 opamp performs very well. But not close to the LM399.
But beside that, your meter will react to to the same things. And if the meter is worst as the standard it will react more. If the meter is 10ppm/C in the 10V rangeit wil change 100uV every degree. So when the standard changes 30 uV over 5 C, the meter will change 500 uV. And if the accuracy is 0.001% that is 100 uV at 10V, so 10 ppm.

So if your reference is 0.6 ppm/C and centre excact 10.000,000V at 20C your meter must read between 9.999,900V and 10.000,100V. But at 25 degrees  you should aid the tempco. and in our example +/500uV for the meter so the meter must read 9,999,400V and 10.000,600V. Then the 30uV from the reference. If you do not know the direction it will add that to the error budget.
These things can drive you (volt)nuts. I always warm the stuff up a day before I need it. Then compare everything to my standard cells to find out that if my roomtemp changes a few degrees I can start over again. So I must choose a day with constant temp, leave out my halogene lights (they heat up to much and cause EMC) and try to do a cal while watching hydro and thermometer. 

[klimm]

It might not be good manners to defrost such an old topic but  the title says it all and I think it would be ok to keep things related to this topic(title) under the same hood.

I need a cheap solution to keep my   31/2   digit voltmeters showing at least same values. Not  ICL7107 based,  they are built with discrete components and OAs from the 80s, so  I have a need to trim them from time to time. 

I already ordered  AD588 to build Jim Rowe's reference  when I find Conrad's Article about the mini lab. http://www.conradhoffman.com/mini_metro_lab.html  which seemingly would fit my needs
I am not an experienced hobbyist but  rather beginner so I dare to ask 2 questions about the schematic of the null detector which I intend to build.
1.Why one can't take advantage of the high impedance of the LTC1050 and need to get the input impedance so low as 200K when the Wheatstone bridge would love to see a high(non loading) impedance? I have in mind  the fact that for sorting out the 1% resistors of same value needed for the KVD (or other purposes) wouldn't bring the bridge out of balance in volts region and even if this would happen we have  the 914 diodes in place .
2. I read that enut11  was using  Conrad Hoffman's metrology chopper circuit feeding an analogue 100uA moving coil meter. How is this to be done? using a buffer? Is it  not inserting its own offset errors? I would prefer the moving coil instead of a DMM I don't have.
Thanks.

[exe]

For voltage reference, how much accuracy do you need?

I bought max6226 which comes in a ceramic package. It's the cheapest reference in a hermetic package I was able to find. It's only available for 2.5V out voltage. I also haven't tried it yet.

There is also lm399, which I tried on a breadboard in the following configuration: https://www.eevblog.com/forum/metrology/old-fashioned-zener-10v-reference/ . I think I had a decent result, meaning that the output it was not moving much (but I don't remember how good it was). I decided not to convert ~7.1V into 10V as this makes things more complicated and I don't need a precise voltage, I need a stable voltage track drift of my instruments.

I also have a bunch of voltage references from China, like these: https://www.eevblog.com/forum/metrology/cheap-ebay-ad584-voltage-references-my-experiences/ . One even came with a "cal-cert", which matched well with my K2000 (even though I suspect it has "unknown" calibration). I bough three more of these, but the first one, the one that looked old and rusty, has the smallest drift. Those a good-enough to calibrate 3.5digit DMMs .

[klimm]

Thank you exe
for trying to help, I partly am aware about what you're talking. Therefore I ordered the AD588 reference, it is 10V with buffering.

My questions are  about the nul detector from the mini metrology Lab project(down my initial  post)

As for the accuracy, the DMMs are rated to (0.2%FS +0.1% reading) for the 200mV, 2V, 20V and 200V ranges.  The way they are built request a good calibration on the 200mV and 2V ranges (The DC converter actually ) in order for me to attempt a calibration for the AC and R ranges which I even can't think of, yet.

[Mortymore]

...
I need a cheap solution to keep my   31/2   digit voltmeters showing at least same values. Not  ICL7107 based,  they are built with discrete components and OAs from the 80s, so  I have a need to trim them from time to time. 
...

Please check this album @Flickr and see if you find something that interests you.

https://www.flickr.com/photos/jcbarros71/albums/72157624135147315

There are photos with schematics. The construction of the voltage references is simple, and was done with samples from the manufacturers, so you have to check the prices if you are to buy the ICs.

[klimm]

Thanks Mortymore,
there are some nice picks showing practical approach. I will certainly grab a few photos from there.
I made a longer introduction with some details to avoid unnecessary questions but maybe my post was too long and the actual questions got lost (Questions 1. and 2. my initial post)
For the moment as the AD588s are on their way, I would like to build that null detector for sorting some resistors.

[Kleinstein]

In the null meter circuit the relatively low resistance to ground is not really needed. One may need a relatively low values for the higher voltage ranges, as high values resistors (e.g. > 1 M) are usually not that stable.
Ideally in a null-meter when adjusted to zero, there is no current flowing through the resistor. With 200K the circuit may be just closer to the all analog solution with a mirror - galvanometer.
So it is possible to build a null-meter with high input resistance, possible all the way to just the OP.

For the DMM adjustment: modern 3.5 digit meters got so cheap, that it is not really worth doing a calibration.
One can get relatively cheap ref. boards from China with measured (at least there is a good chance the values are real) voltages.

[klimm]

Kleinstein (For the DMM adjustment: modern 3.5 digit meters got so cheap, that it is not really worth doing a calibration)
Not the one in the attachment

About the null meter:
If choosing to supply the whole thing with +/- 9V which is possible according to data sheet could I  let apart the input divider which only offers an input resistance of 200k and connect the bridge to the RC network on pin3 of the OA?

[Kleinstein]

I would not increase the supply to the AZ OP very much. Leakage current and possibly also current spike may very well get higher than.  Getting away with the input divider is possible for a limited input range.

In principle one could have the low ranges without a resistor to ground and still use a divider for the higher ranges. So everything up to some 100 mV or maybe 1 V could be high impedance and only the higher ranges would use a divider (can be more then 200 K). It may need  switch with more contacts or a separate switch.

Just increasing the supply would not help much. The limiting factor in the given circuit is the 100 Ohms resistor to ground.  It is possibly to also switch away the 100 Ohms, though this may add some tiny bit of error from the switch, but this may be acceptable. The LT1050 / 1052 is not that low in noise, so one should get away with more than 100 Ohms too.

[klimm]

Kleinstein thank you so much for the detailed answer:

As I understand from the schematic the 100 ohms resistor together with the others(one by one for different ranges) switched paralel to C3,  set the gain as to get always 100mV FSD at the output. I don't think it is possible to let apart the 100 resistor.       
Am I wrong with this assumption?

[Kleinstein]

There are different ways to switch the amplification. The current circuit has a fixed 100 Ohms to ground and changes the other resistors. This has the advantage that there is no switch in the path from ground through the 100 Ohm to the negative OP input, so no problem with thermal EMF from the switch.

The other common version would use a single resistor string and switch the point where the inverting input is connected to. This system does permit lower gain, all the way to a gain of 1. An external DMM could than also use a 2 V range.

There still is a slight complication with the filter action of "C3". Usually one want less / no filter action at low gain and possibly some filtering for the very low voltages. This also depends on the type of meter (analog, DMM, ADC+µC+LCD) used. Some older DMMs can profit a lot from analog filtering (as the input is sampled for only some 25%). Modern DMMs with more continuous  input reading may not need the filtering and a custom µC solution could implement filtering in the digital domain.

[Conrad Hoffman]

Remember that those zero drift chopper amps have very little output capability. They can't drive a low total value feedback network. The resistor choice and f.s. output choice was a balance between sensible values and DMM range. My Fluke DMM has a 300 mV f.s. DC range, so 100 mV output made sense. I could have used two opamps but the idea was to keep it as simple as possible. As discussed above, input Z doesn't matter much. Some old mirror galvos had coils of a few hundred ohms or less. At zero volts (null) no current flows!

[klimm]

 I reached a point where I am going to build a Null detector. 
I keep in mind all the answers I already get above.
I would try two versions;
1.  built directly on the binding posts of   my Fairchild 7000A on the 100mV, floating range,  as it has a high impedance (about 1G ).
I will adapt the N.D. ranges to my situation.  The 7000A has a network power supply, not a battery.   
Could you explain what kind of problems could I encounter in this setup?  I am willing to add a buffer if that would help as I will anyway need it at version two;
 
2. Stand alone null detector battery powered. Output to a small chinese analogic, ±50uA class 2.5,  ammeter which still has a visible deflection at  1uA through it.
Would be  enough to use  a non inverting unity gain amplifier, to feed the analogic indicator,  say with an OP07C from PMI?


 

[Kleinstein]

A main reason for a dedicated null-meter is to have very good isolation from ground / mains.  In most cases the connection to a mains powered meter would be OK, but sometimes additional capacitance and possibly common mode hum may be a problem.

For a battery powered version the AZ OP can directly drive a 50 µA meter movement. There is no need for an extra buffer or OP. So one could get away with just 1 AZ OP for the whole circuit. The choice of OP is a balance between voltage noise, input bias and maybe supply current. How important the bias current is depends on the use, so there is not one best solution.

Mainly the old ones and some very low noise, but high bias types may need a relatively high current. So battery powered is a real option, as the current can be quite low (e.g.  200-500 µA range).

[klimm]

Thank you Kleinstein.
 I am aware about the need of isolation from ground /mains.  What I don't get is how this would translate in the behavior of the DMM/Detector amplifier. How one determines that such a problem occurs. Is this an offset voltage that actually would affect the null point?
I wish to try my version 1. described in the precedent post, dead bug with an IC socket to play with more than only the LTC1050 in order to see behavior differences as this  is meant to learn more about op amps. The ND assembly, battery included, will be shielded and connected to DMM's ground.

Concerning the version2. I wish to build a  permanent, dedicated null meter as I need it to use with a Wheatstone bridge for selecting resistors,  and  to null the Feussner Kompensator(FsK)  we were talking about  in another thread and which I was able to repair(only one 1K resistor interrupted at 5mm from one end)  Please see the picture.

A question about that often heard thing that when  a bridge or  potentiometer is in ballance there is no current flow...
 
Playing with the  FsK thing, I was able to get a null with the 7000A DMM at a resolution of 10uV.  I tried than, to see what deflection I could get with the bare 50uA  ammeter which has a resistance of about 1440R. I was not able to get the needle moving but only see the DMM getting out of balance due to the new load. 
While the null I've got was with a resolution of 10uV, I assume I had about 10uV at the microammeter, thus 10uV/1440R= 7nA through it so, too low to move the needle(at least1uA)  but still there. It actually would need a voltage of 1.44mV in order to see the needle moving.
How was than possible to use galvos with only  hundreds of ohms, as Conrad said, to ballance bridges for a reasonable resolution and accuracy? Were these galvanometers able to sense  nAmps?

[Conrad Hoffman]

No numbers at the moment, but those galvos suspended a small mirror on a very compliant suspension. The image of a curved scale was then observed with a small telescope, so it was far more sensitive than you might think. Some were even wall mounted and the optical lever arm could be across the room. If only they had had laser pointers back then!

[Kleinstein]

Imperfect isolation from mains can add some noise and hum and also some bias current. So both side may see a current of a few pA or so. The AC part of the current may be larger and depending on the DUT also create an offset if there are diodes or similar.
Most of the time mains powered is likely OK, but it is hard to know if there is an effect.

The old day galvanometers can be quite sensitive, especially if a mirror and some optics is used. Very sensitive however also made them slower. I rember using an old galvanometer directly connected to a thermopile. So maybe some 1 kOhms and maybe 100 µV, so some 0.1 µA as possible full scale range. Chances are one could see nA, though likely with more than 100 Ohm if internal resistance.

For the choice of Az OPs, there are not many that come in DIP - mainly the LTC1050 and ICL7650 which are not even that much different, the main point on the LTC1050 is having internal caps.  So you would likely need some adapter board for SO8 or Sot23-5/6.
A few possible candidates are:
LTC2054 for very low bias current, but a little more noise than the LTC1050.
max4238  low offset and a little lower noise than LTC1050
MCP6V66  low power and still lower noise, but more offset and bias, but not too bad
LMP2011  low noise and still reasonable low bias
AD8628   even lower noise, but more bias
OPA189   very low noise, but high bias , relatively high power

[klimm]

Conrad, Kleinstein,

Thank you both for your input. Amazing stuf those old galvos.
I'll most probably stick to Conrad's detector with LTC1050 if you say driving my 1440R uA would not be a problem but first a bit of getting wet with LF411, and some other Ops I have laying around. I even found an uncompensated Harris A2-5160 

[Kleinstein]

When using a mechanical meter movement, a nice trick is to have the meter movement in series with the feedback network. So the meter movement would be actually current controlled. This may need to recalculate the feddback network however.  For a null-meter the scale factor does not matter much, but the meters are usually with copper wire and thus have a TC of some -3500 ppm/K when controlled by voltage. 

For playing around on the breadborad the LM358 is nice. Fost OPs like > 5-10 MHz can get tricky on the bread board. 

[klimm]

I was reading about current driving in Hayes, Horowitz, learning the AoE. They have in chapter 7W2 the algorithm and rationale for the way one should choose an op amp for a milivoltmeter.
The thing is I want to keep Conrad's detector as is  and maybe use  a buffer amp to feed the uAmmeter but I still need some study as I don't have a ±100uV movement.

I once built a hi-Z voltmeter with CA3140 on a breadboard but was a bit dissapointing as I had unexpected movements of the needle.
I have some small metallic boxes I would use to shield any prototyp of the N.D.
Right,  I have some 358, 741, LF411...

[Conrad Hoffman]

When using a mechanical meter movement, a nice trick is to have the meter movement in series with the feedback network. So the meter movement would be actually current controlled. This may need to recalculate the feddback network however.  For a null-meter the scale factor does not matter much, but the meters are usually with copper wire and thus have a TC of some -3500 ppm/K when controlled by voltage. 

For playing around on the breadborad the LM358 is nice. Fost OPs like > 5-10 MHz can get tricky on the bread board.

Yes! When you put the meter movement in the feedback loop you can set the maximum current to protect the meter and it makes calibration easy. Surprised more people don't/didn't do this.

[klimm]

Thank you Conrad,

I think this is what you and Kleinstein are  talking about(attachment)
Transconductance amplifier as I remember ( I built a voltmeter with this). I am not confident yet about how one compute this to adapt to my situation but I am studying this right now.
As said, I want to use this as a next stage after the LTC1050. Would this approach be appropriate in your opinion?

[Kleinstein]

If one wants a voltage output (e.g. to connect an external DMM, e.g. for logging) and the analog meter ouput it makes sense to have an extra OP to drive the analog movement this way.
The resistor from ground to the meter set the gain. So the 10 Ohms shown in the link would give 1 mA for 10 mV input voltage. For a 100 µA movement and 100 mV full scale range one would need some 1 K for the scale factor.

Usually the meter movements can stand something like 500% overload. So the protection does not have to be very accurate.

Some movements need/want a parallel capacitor for damping.  A small cap in parallel (or directly from the OPs ouput to inverting input) may be needed for stability of the OPs FB loop.

A sufficient series resistor and than directly from the LTC1050 would also be possible - especially if the ouput full scale range is relatively large, like +-1 V.

[klimm]

Thank you Kleinstein,

I will try to come up with a schematic  adapted for my 50uA instrument and present here for evaluation.
It will take about 2 weeks as I have to go skiing a bit before season ending.