Precision Load Cell (Wheatstone Bridge) Simulator

[snarlingsnoyl]

First Post on EEVBlog, long time follower...

The goal of this project is to create a precision load cell simulator for testing/debugging load cell frontends/amplifiers/converters/etc. I am no expert in material science, precision resistors, low themal EMF design, etc; so I thought it best to share with everyone and hopefully get some good advice in return.

A load cell simulator is a device that produces a very small ratio-metric output simulating to the output of a strain-gage load cell. It can be used to provide a test signal to an instrument or device that is used to measure load cells.

Typically, the output is referenced to an excitation voltage and specified in terms of mV/V, or milli-volts output per volt of excitation. Most industry load cells are of the full-bridge type with a full scale output range around 2 mV/V; for example, if you excite the load cell with 10 VDC, then the full-scale output of the load cell would be +-20mV ( 2mV/V * 10V excitation).

The device will have a selection of outputs ranging from 0mV/V to 2mV/V in 0.2mV/V steps. Negative outputs will be accomplished via a polarity switch. My initial calculations show that this can be accomplished using a Wheatstone bridge and switching resistors in and out of series/parallel with each bridge leg in order to achieve the desired change in resistance.

Starting with a Wheatstone bridge:


We can then add parallel resistors across R1 & R4. However, in order to keep the total bridge resistance constant, it may be necessary to add resistors in series to R2 & R3. I've been double checking my math with a little LT Spice schematic:


The math (for a full bridge simulator) follows as such:
RB - Bridge Resistance:
     - Typically 350 ohms, also common are 120ohm, 1000ohm
RP - Leg Parallel Resistance:
     - This resistance is added in parallel to the active leg in order to lower its resistance by a calibrated amount
     - RP = RB * ~10 / Sensitivity
     - RP = 350 * 10 / 0.002 (or 2mV/V output) = ~1.75MegOhm
RS - Leg Series Resistance:
     - This resistance must be added to the opposite leg in order to compensate for the lower resistance caused by adding the parallel resistor
     - RS = RB²/(RB + RS)
     - RS = (350²)/(350+1750000) = ~80mOhm

It may be possible to add outputs or switches to reconfigure the simulator into half-bridge or quarter-bridge modes, but that could be a later improvement. Best to start simple.

Another possible improvement is for the simulator to work with both DC and AC excitation. For AC excitation, resistance need to be replaced with impedance. Fortunately, most industrial load cell systems using AC excitation typically only use 100-1000Hz carrier frequencies. This means that parasitic capacitance/inductance would need to be controlled, as well as balancing any resistor inductance, but should be reasonable easy to accomplish at such low frequencies.

As for parts, Caddock's low TC film resistors is the initial choice. However, I have to admit my experience with working with precision equipment is limited, and designing precision equipment is none. So things like temperature coefficients, parasitics, etc are on my radar, but I'm still trying to figure out how much impact on real world design and construction. This things are easy to

Final note, I lacking in the precision test equipment department. This project is going to require some precision resistance testing equipment. I've been looking through eBay and the like for a nice DMM/SMU. Anyone have any suggestions?

Be gentle

[RandallMcRee]

You might get more information if you post in Metrology. Everything I am about to say comes from there, so-to-speak.

First, the basics are nicely handled by reading Conrad's DIY Metrology Lab articles. The wheatstone bridge is the basics of diy metrology.
http://conradhoffman.com/mini_metro_lab.html

Attached is a photo of a basic setup using a Kelvin-Varley divider measuring the tempco of a Vishay s102k resistor. It looks crude, but lets examine it. The 720A is a 0.1ppm KVD, so quite good. The aluminum box in front is just used to connect the auxiliary wires needed to create the other half of the bridge. Not very visible is another s102k resistor serving as a reference. (This setup is diagrammed on page 78, figure 9 of Conrad's part 3). The usb stick measures temperature. The gray cable on the left goes to a Fluke 731b to supply 10v to the bridge. The output is measured with an HP3456a.

For a modern DMM folks seem to recommend the DMM6500 or DMM7510, but I've stuck to the tried and true as noted above.

I have not heard anything bad about the Caddock series. For higher precision I have good experiences with the AE z series. e.g.
https://www.newark.com/alpha-electronics/hcz10k000t/resistor-metal-foil-10k-0-01-radial/dp/96Y7830

In looking at what you have written, the major question is the switching. There is a reason that the my bridge is all hard-wired. Its to accurately fix all of the variables having to do with those pesky interconnections. If you need programmable switching--relays,  jfets, analog switches; that's a big step up in complexity and error sources.

Hopefully, this is helpful.

[MosherIV]

Hi.

I did not read everything in detail but it seem strang to me that you are trying to use a real bridge to simulate a strain gauge in a bridge.

If you want a simulator, all you need to know is the transfer function.
Knowing that, you know what output you should get for any input.
You then just build a system eg microcontroller with ADC and DAC. The micro then gives the simulated output voltage for a input voltage.

Or am I missing something?

[Kleinstein]

For the resistors I would consider the 350 / 120 and 1 K bridges as separate projects.
I would consider some thing like 350 Ohms + a number of around 1 M resistors in parallel  for all the resistors. So to keep the sum constant one could start with 2 x 1 M in parallel and add / remove the extra resistors where needed.  How large the extra resistors should be depends on the range, but there usually is a limit of some 0.1% of change for the usual DMS.  With 1 M resistors switch resistance should not be a major problem, it more like leakage limited. If needed the next step would be using something like 300 Ohm + 50 Ohm in parallel with a few 100 K resistors.

DMS bridges may use quite some power, so that self heating can be important.
At 350 Ohms I would consider wire wound resistors and possibly bulk metal foil if needed.
The larger resistors for signal steps are less critical and could likely be thin film types.

For some tests one could consider a real dummy type, that has the bridge part only for a much smaller voltage and just series resistors to get the overall resistance.  so more like 170 Ohms resistors at each end and than a bridge with 10 Ohms resistors. This version could get away with much less critical resistors and still get a very stable reading.

[jbb]

I think it's fair to say that if you want a continuously adjustable, high accuracy bridge simulator, you're going to have a difficult time.

Maybe you could split this into two sorts of tests: a lower precision adjustable type (maybe simple digital potentiometers would be sufficient) for debugging / exploring, and a number of discrete high-accuracy 'strain references' for precision testing?

[Kleinstein]

For testing strain bridge amplifiers, I don't see a need for a very fine adjustable bridge. More like a few (e.g. 5) well defined settings rather close to balance. Some tests might need a fast / smooth step switch - so more like electronic switching - this could be a little tricky.

So I think the circuit with extra resistors at all 4 terminals is a good idea. The extra resistors can reduce the required stability a lot (e.g. some 100 fold) - so no more need for very expensive resistors.  A limited number of settings could be generated by just a multi step switch and a string of low resistors (e.g. 1 Ohms). One may not even need to adjust both branches together, but could use separate switches, so that it could even be used for a linearity test or fine and coarse steps.

[snarlingsnoyl]

It always seemed like the community here was quick to respond to even the simple project, so thanks for all of the responses so far.

Why Not Digital (ADC->MCU->DAC)? - MosherIV
     - I hadn't really considered it before, but I suppose its possible.
     - Potentially a continuously adjustable range which may be useful in some testing/debugging situations.
     - The tricky part being low voltage/low noice DAC section for +-2mV at 0.1% (+-2uV), seems reasonable to use a normal DAC with a precision 1000:1 voltage divider, just have to deal with noise.
     - However, this would not likely not work with load cell amplifiers that measure excitation current to compensate for lead length and do not have dedicated sense lines.
          - It may be possible to simulate the proper excitation current with a DAC driven current sink.
          - Would have to watch power management closely for low excitation voltages.
          - Trickier to allow AC excitation, but not impossible.
     - Overall, this is a great idea. However, I personally would like to try and accomplish this with an analog solution first, then maybe try digital after that.

Next, lets narrow the scope of the project a little bit. Here is updated goals:

- Target 350ohm bridge load (Kleinstein), no need to be able to do all 3 at once. Should be easy to apply same principles to other loads
- Limit the number of steps to 5 (Kleinstein/jbb), 10 steps does seem like too many for an initial project, 5 discrete steps seems like plenty for its intended purpose.
- Target of precision is 0.1% across 5°C-45°C, not really precision but good enough to help spot issues in the field, may be able in do better in future revisions

The first post was a little short on construction details, so here's some more details
     - The original plan was to use a nice (low contact resistance/low thermal emf) multi-position switch. Most likely, it would be salvaged from an old decade box/transfer standard.
          - Both reed relays and other signal relays seem to have contact resistance specs (~100mOhm) that may make parts of this project more difficult.
          - Most multi-position switches have 10mOhm max initial contact resistance
          - Its expected that either solution would gain contact resistance over time, so it seems that starting out lower would lead to less error
          - Anyone have any information on contact resistance vs cycle count for multi-position switches or relays?
     - For each step, the switch would be adding a different set of resistors both in parallel or series with the bridge legs in order to achieve the desired change.
          - This is where the precision resistor stuff comes into play.
          - It would be easy enough to get resistances close using discrete precision resistances
          - Fine tuning can be accomplished with low TC trimmers
          - It also seems like the better approach would be to keep adding resistors in parallel/series such that the power (and heat) are more distrusted, instead of switching in and out
     - As mentioned, the resistors that are added in parallel are going to be less critical, so it should be easy enough to come up with proper combinations

Kleinstein also had a good idea worth exploring, split the bridge into a static section and a active section. For a 350 ohm half bridge to output 2 mv/V, the top and bottom resistance need to change (in opposite signs) by only 1.4 ohm. So it could be achieved using 340 ohm resistance of high precision, and then a set of higher tolerance resistors for each step. In order to remain in spec, the last 10 ohms of each leg would only need ~0.25% tolerance. It would be even better to split it at 345 ohm for the base and 5 ohm for the active, then the last 5 ohm would only need 0.5%.


This seems like a much better topology for this project. It also limits the number of expensive components.

My next steps are to spend a little time with some uncertainty analysis to determine the actual tolerances needed for my components over the temperature spec range. Then after that, onto laying out the construction details.

I am thinking of cross posting to Metrology but maybe later. I think as a project it fits better here, but when it comes time to do adjustments and calibrations, Metrology seems like a better place.

Thanks for everyone's replies so far, and please keep them coming!!!

[Kleinstein]

My verbal description of a possible circuit may not have been clear. Attached is a possible solution with 2 muti stop switches.
The resistor values are only rough order of magnitude and likely still too large steps.

[David Hess]

Load cell simulators do not work the way you have described because of the expense in switching or trimming the leg resistors accurately.  Sticking a switch or potentiometer in series with a precision bridge resistor is almost always a bad idea.

Instead, the four leg resistors are fixed and the variable output is provided in the same way that a bridge zero circuit is done; essentially a trimmed set of dividers are placed across the excitation which then drive one of the signal outputs through another high value resistor making it possible to have several full scale ranges with several settings each using only two rotary switches and none of them are in series with a low resistance path.  The bridge zero is trimmed in the normal matter.  This configuration also lends itself to electronic control.

We once made one where the bridge resistors were actual strain gauges mounted to a block of aluminum simply because those were the cheapest high precision resistors we had available.

I always find it surprising how little documentation is available online on this subject.  It is like everything is a trade secret or at least pre-internet which might be more accurate.

If it is not on the internet, then it does not exist - Jame Burke

[snarlingsnoyl]

My verbal description of a possible circuit may not have been clear. Attached is a possible solution with 2 muti stop switches.
The resistor values are only rough order of magnitude and likely still too large steps.

Your idea is very clear now. Its pretty obvious too, I'm a little ashamed I didn't see it before you had to draw it up for me. This seems to offer a lot of benefits over switching resistors in and out. I have updated my designs to reflect this concept for 0, 0.5, 1, 1.5, 2 mV/V on a 350 ohm bridge. A rough Monte-Carlo error analysis would suggest that 0.1% is achievable with 0.01% tolerance on 350 and 10 ohm resistors and 0.25% tolerance on the RP and RB resistors. These could be some Vishay Bulk Metal Foil potentiometers.


As for utilizing one half of the bridge for both +/- signals, it seems like this would leave the circuit vulnerable to input impedance of the measuring device. If it's not prohibitively expensive, it may be best to leave the it as a full bridge. Is this a valid concern or just ignorance?

Sticking a switch or potentiometer in series with a precision bridge resistor is almost always a bad idea.

This is an interesting comment. Is there any further explanation to the issues that arise from this. Of the top of my head, the biggest issues seem to be the contact resistance of the switch or the stability of a potentiometer (both material stability as well as mechanical stability). With the re-arranged circuit, the ~12k and ~250ohm resistor would likely need to be potentiometers and as such may be vulnerable to these problems.

Instead, the four leg resistors are fixed and the variable output is provided in the same way that a bridge zero circuit is done; essentially a trimmed set of dividers are placed across the excitation which then drive one of the signal outputs through another high value resistor making it possible to have several full scale ranges with several settings each using only two rotary switches and none of them are in series with a low resistance path.  The bridge zero is trimmed in the normal matter.

There are designs floating around similar to this. They don't always behave the same as a load cell would though in terms of input impedence. This design may be overkill, but the intent is to simulate as closely as possible to what's actually happening within the load cell.

I always find it surprising how little documentation is available online on this subject.  It is like everything is a trade secret or at least pre-internet which might be more accurate.

Would you be willing to illustrate your idea? This may become a helpful for those looking for information on this topic in the future.

[Kleinstein]

The circuit as shown last could also be used as a half bridge, though with 175 Ohms resistance as the 2 branches are in parallel.

For just getting the voltages near the center, it would really help to combine the upper und lower  350 Ohms resistors (R1+ R3, R2+R4) and add the extra output resistance to the signal sense lines.  The internal circuit is than quite different to a bridge, but the properties measured from the outside (especially with defined drive and sense) are very close.  The advantage is that this substitute is nearly balanced and stable by design. So even 1% resistors may be used to get a very stable bridge reading. The simple perfectly  (as good as the node is as a 4 wire zero) balanced case would be just 4 resistors of 175 Ohms joint at a hidden node. Even with just carbon resistors it would be more stable than a classical bridge with good (e.g. BMF) resistors.

The half bridge would need 2 really stable resistors and maybe a few in the center for steps.

[David Hess]

Sticking a switch or potentiometer in series with a precision bridge resistor is almost always a bad idea.

This is an interesting comment. Is there any further explanation to the issues that arise from this. Of the top of my head, the biggest issues seem to be the contact resistance of the switch or the stability of a potentiometer (both material stability as well as mechanical stability). With the re-arranged circuit, the ~12k and ~250ohm resistor would likely need to be potentiometers and as such may be vulnerable to these problems.

Variations in contact resistance are exactly the problem.  The strain gauges only change in resistance by a tiny amount and bridge circuits are used to make the most of this.

Instead, the four leg resistors are fixed and the variable output is provided in the same way that a bridge zero circuit is done; essentially a trimmed set of dividers are placed across the excitation which then drive one of the signal outputs through another high value resistor making it possible to have several full scale ranges with several settings each using only two rotary switches and none of them are in series with a low resistance path.  The bridge zero is trimmed in the normal matter.

There are designs floating around similar to this. They don't always behave the same as a load cell would though in terms of input impedence. This design may be overkill, but the intent is to simulate as closely as possible to what's actually happening within the load cell.

Load cell input resistance is not well controlled unless a shunt resistor is added because of any gain or modulus compensation resistors places in series with the excitation.  350 ohm bridges for instance often have a 40 ohm modulus compensation resistor, often in the form of a gauge, for 390 ohms.  Usually there is also a length of resistance wire in series with one element of the bridge for temperature compensation but its value is low so it can be ignored.  Various methods are used to trim the zero after all temperature compensation is done.

I always find it surprising how little documentation is available online on this subject.  It is like everything is a trade secret or at least pre-internet which might be more accurate.

Would you be willing to illustrate your idea? This may become a helpful for those looking for information on this topic in the future.

I thought I described it pretty well.  Use one rotary switch to select from a divider placed across the excitation and another rotary switch to select a higher value resistor which goes from the selected divider point to one of the signal output leads.

Note that strain gauge bridges really are precision devices.  Maintaining that level of precision in a finished load cell or a load cell simulator is not a trivial task.

A couple months ago I tried finding the schematics for the load cell simulators I am familiar with but there was nothing available online.

[snarlingsnoyl]

The internal circuit is than quite different to a bridge, but the properties measured from the outside (especially with defined drive and sense) are very close.

Forgive my ignorance, I just couldn't see it before but this make sense now after going through your design with input/output impedences.

Here is an updated schematic based on input from Kleinstein and David Hess:


Running through some more Monte Carlo simulations, here are the specs needed to achieve 0.05% total error across 0-40°C, with 0.1% long term stability:

NAME	RESISTANCE [ohm]	TOLERANCE	TEMPCO [ppm/°C]
RS	10	0.01%	2
R_BRIDGE	175	0.1%	50
R_OUT	175	0.5%	100
R_SET	1k	0.5%	100
R_SHUNT	1k	0.5%	100
R_ADJ	100 trimmer	5%	200

These spec for the RS resistors is a little misleading. These must match withing 0.01% and withing +/- 2ppm of tempco. After reading through RandallMcRee linked post (HIGHLY RECOMMENDED - http://conradhoffman.com/mini_metro_lab.html), it seems reasonable to find a matching set of 5 resistors from a large order 0f 0.5% 20ppm metal film resistors.

The RS resistors of 10 ohm will work for the 120 ohm simulator too. The key is getting the proper ratio of R_SET/S_SHUNT so that R_ADJ is just making minor adjustments. It looks like the 1000 ohm simulator would be too sensitive with RS of 10 ohm, so it would like need to be 100 ohm.

[EmmanuelFaure]

Output impedance problem in your last schematic : Output impedance = 175 ohm (R_out) in series with 2*175 ohm in parallel (R_bridge) = 262.5 ohm. To obtain the correct input impedance you have to use R_out = 262.5 ohm.

Another idea in attached file (Shown for output = 2mV/V. Every new setting requires one additional R_tuning resistor). Lower component count, input impedance is ok, output impedance is ok, the initial offset error is null, calibration can easily be done by tweaking the R_tuning value.

R1, R2, R3 can easily be implemented using a resistor network, doing so you'll get a perfectly tracking resistor ratio.

[gamalot]

I used to make a loadcell emulator using a Maxim 18-bit DAC to test the PCBA of the scale.

[snarlingsnoyl]

Output impedance problem in your last schematic

I made the same mistake when I was looking at Kleinsteins earlier post. In fact, the output impedence is just 2 * R_OUT. It goes Vsig+ -> R_OUT_2 -> R_OUT_1 -> Vsig-. Then for different settings, the output impedence rises by ~10-40.The R_BRIDGE resistors don't influence the output impedance at all.

Another idea in attached file

I believe this is the idea that David Hess was suggesting earlier. I do believe it is a valid design, definitely lower component count and easier to build.

The one difference that stands out immediately with this design is that the gain & linearity of the output will depend on the stability of every resistor in the network except for R5. With the current design, the linearity of the outputs is related only to the matching of the 4 RS resistors(which can be done very accurately, see Conrad Hoffman's articles); while the other resistors only affect the gain.

A similar analysis will answer this question, but I suspect that this design will require higher performance parts to achieve the same specs which may offset the advantage of the lower component count.

I used to make a loadcell emulator using a Maxim 18-bit DAC to test the PCBA of the scale.

This is certainly another valid idea and one that was suggested already. However, in order to scale this design into the more complex scenarios (like AC excitation), it may require significant complexity (high update rates, filtering for low noise, external power supplies). All of these issues can be overcome of course; however, it would require knowledge beyond my own at this point.

[EmmanuelFaure]

I made the same mistake when I was looking at Kleinsteins earlier post. In fact, the output impedence is just 2 * R_OUT. It goes Vsig+ -> R_OUT_2 -> R_OUT_1 -> Vsig-. Then for different settings, the output impedence rises by ~10-40.The R_BRIDGE resistors don't influence the output impedance at all.

Nope. Generally speaking the output impedance is measured between the output and the ground. What you're talking about is called the differential output impedance.

[Kleinstein]

For the DMS amplifier it is usually the differential output impedance that matters. This is what sets the noise due to current noise of the amplifier and offset due to bias currents. Also loading due to amplifier input resistance should depend on the differential resistance.

So the 4x175 Ohms version should be relatively close to a matched bridge, though not perfect in every aspect. The impedance to ground is slightly different from the real bridge.  For most tests I would consider it sufficient.

The tests that I would do on an DMS amplifier would be mainly roughly the zero point (to be sure it's reasonably in the center of the adjustment range) zero drift and a coarse scale factor check. Reading noise would be another factor -  LF noise and stability are mainly 2 ways to look at the same data. For dynamic use one might want to check the step function (this might need some electronic switching).  For the scale factor I would consider something like 1% accuracy sufficient, as calibration is often done together with the DMS / load cell anyway (e.g. using weights).

So the main requirement I would see is a really stable 0.  This is much easier with the star like configuration - up to the point of having an extra dummy (that could use normal 1%/100 ppm/K resistors) just for this test. If in doubt one could consider to switch between different resistors at the sense outputs. Usually the output impedance should not have a large effect.

[EmmanuelFaure]

For the DMS amplifier it is usually the differential output impedance that matters.

It depends : If the front end of the load cell condtionner is built around an instrumentation amplifier, the current noises at the inverting and non inverting input have 0 correlation. The offset induced by input offset current is less problematic : Although the 2 inputs do not share the same differential pair, they're made on the same chip and are often well matched, so the drifts caused by bias current flowing through the bridge compensate each others for any value of R_out.

[BrianHG]

I used to make a loadcell emulator using a Maxim 18-bit DAC to test the PCBA of the scale.

This is certainly another valid idea and one that was suggested already. However, in order to scale this design into the more complex scenarios (like AC excitation), it may require significant complexity (high update rates, filtering for low noise, external power supplies). All of these issues can be overcome of course; however, it would require knowledge beyond my own at this point.

Have you considered digital potentiometers.
Like 2(10 bit total swing, 1 on each side of the bridge) or 4(20 bit with the right dividing resistors) of these with added resistors outside them for the small narrow ranges you will be dealing with:
https://www.analog.com/en/ad5293
They will support up to +/-16v AC excitation which would be a problem with a normal DAC.

[snarlingsnoyl]

Hello All,

I've finally had some time to make a progress post for this project. I've jumped down the rabbit hole a little bit so this post may be slightly off topic and a bit long.

After my last post on here, I came to the conclusion that I would need to first build a couple of tools to help with matching resistance and tempco's of resistors. Since the performance of the simulator relies heavily on no less than 7 of the resistors, these would have to be either very high quality, or very well matched. So following through CHoffman's fantastic articles (http://conradhoffman.com/mini_metro_lab.html) I decided to build both a Null Detector and a measurement bridge. I know that I could have bought these devices, but this project is about learning and figuring all of this out, so why not try to build these too!

So for the null detector; following CHoffman's articles (Part 2). I shamefully copied the schematic and spun up some boards for a pre-fab enclosure. The construction consists of 2 boards; a pcba for the null detector amplifier and components, and a pcba for the front cover that also completes the enclosure's shielding with a ground pour. The null detector runs from a 9V battery inside the enclosure and has worked very well for me. The only deviation from CHoffman's original is the use of the LTC2050 which was a newer part with slightly better performance specs, cheaper and available in a SOT-5 package which helped the pcba layout.


And for the bridge, I built this one from some perf board and a switch to adjust the excitation from 9V, 18V, 27V or an adjustable linear regulator capable of 0.05V-7V. I followed CHoffman's designs for the brass clips from his Part 3 article and also tied in some binding posts for conveniently attaching measurements to. All of the binding posts are gold plated copper, and wired through with silver plated copper wire to reduce any potential thermocouple effects.


So with these tools in hand, I started the process of matching resistors for tolerance and tempco. I started out by trying to find some thermally stable resistors. This involved making (many) small purchases of generic %1 +-100ppm/°C resistors and then testing each order for a rough scale of tolerance and tempco. The setup changed slightly for the tempco testing by using some test leads to separate the resistor to be tested. Then a hot air rework tool was used to heat the resistor 100°C above ambient temperature and the change in output of the null detector was used to calculate the tempco.


There is only a set of 4 resistor in the design that need to be matched to both tolerance and tempco; everything else just needs to have a very low tempco. So after I had matched the resistor tempco, I began to match resistor tolerance using the test clips on the bridge. This involved a multi-stage process where each stage goes higher on the null detector scale.

I am not finished with this project however. I have had several setbacks and discoveries that have forced me to alter my designs. I will cover these in an upcoming post along with some interesting discoveries I have made about generic metal film resistor behavior. It will also cover the construction of the load cell amplifier where I encountered some more issues.

[Thinkster]

Any update on this project, I know it's been a few years...

[totaltech]

G'day,
See attached schematic. Using star arrangement for the bridge eliminates zero offsets. for 175R use 2 x s102j in parallel. For the shunt resistors use low tempco 0.1% resistors with a small value pot in series to trim each resistor/range.

[hozone]

Thank you for sharing all those information.
I would like to build a simulator like this one.
Find attached my LTspice simulation file and my schematic.
I can find 0.1 for all resistors (175 ore two 350 in parallel) and 0.05 +-10ppm for the 10 one at a reasonable price.
Do you think it will work?

[totaltech]

G'Day,
This may or may likely not work. A lot of signal conditioners expect to see the +/- signal levels at half the excitation level not swinging around 0V.

Cheers

Rob