DIY 20A 0,1R Current shunt, Metrology grade(?)

[Micke]

Hmm... easy to go down the rabbit hole...
It started I bought a defective Datron 4600 (Transconductance amplifier 11A DC-20kHz) to be accompanied to my Datron 4705 (bought with broken 100V and 1000V range).
The symptom with the 4600 was that after like 2h it turned off and lit the red "PSU" LED.
After changing all electrolytic capacitors in the switched mode PSU to new Low-ESR 105C A-brand, the 4600 has worked fine. Opening the PSU was difficult, box is all welded (amazing design), only access is from one of the short sides.
But having the 4600 fixed, problem arised with adjustment... using normal Amp-meters are not accurate enough. First thought of buying used FLUKE Y5020 0.1R 20A, but having seen teardown pictures here on the forum https://www.eevblog.com/forum/buysellwanted/ice-teas-score-thread/msg3439280/#msg3439280 I was not impressed, no hermetic resistors, it looked like a very expensive heating fan...
I then thought of buying a used Burster 1282-0.1R, but it is only specified at 50Hz, I E-mailed Burster and asked for additional data at different frequencies, but it was unfortunately not available. https://www.burster.de/fileadmin/user_upload/redaktion/Documents/Products/Data-Sheets/Section_1/1282_EN.pdf

So what to do...
Why not buy 10pcs 1R Hermetic Julie Research and put in parallel?
https://www.ebay.com/itm/202568048586
The specs are amazing, 1R +/-0.005% 6W TC +/-2.5ppm/C   
Datasheet on Xdevs: https://xdevs.com/doc/JRL/JRL1/CH48T4HK.pdf
It is quite expensive buying 10 pieces, but less expensive than used Fluke Y5020 or Burster 1282.
Said and done, ordered 10 pieces from USA.
Bought 60mm aluminum tube (thickness 3mm) from a car modding site that was just perfect fitting the resistors on.
Found a very nice grey cube steel box (Hammond 1415D) 152.4x152.4x154.23mm
In the back I put a 80mm 12V fan, and a PCB with LM317 for adjusting the fan speed.

But how to connect Sense?

I thought of 3 options, but decided to go for option 2, using sense connection on ONE resistor.
Did an experiment with option 1 (using Sense at the connection points of the resistors), but the result was an absolute disaster, sensed voltage was increasing constantly... Copper has terrible TC, but I thought that since the leads are quite short, and the temperature rise should not be that high, it would be OK... but not!
Next problem was how to arrange the connection with the resistors, I then found that old coins with diameter 27mm and high copper content was perfect, so sacrified and grinded down two 71 year old Swedish 5-Öre coins that were in bad shape from the beginning.
https://en.ucoin.net/coin/sweden-5-ore-1910-1950/?tid=22822

Next I will send the shunt for calibration   
Will be interesting to see the AC performance, I am hoping for good performance up to 20kHz...

[KT88]

For frequencies higher than a few Hz this paper may be interesting:
https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=24169
Probably already a bit on the academic side but for 20kHz worth reading...

Cheers

Andreas

[TiN]

Did you actually measure tempco of paralleled setup. I've tried to combine multiple of these JRLs into higher power 1R device, but due to copper leads TCR the resulting tempco was much worse than any single resistor.

I didn't figure out way to improve it yet, but did log results on xDevs draft.

[1audio]

Lots of good work and probably a very accurate .1 Ohm resistor at DC. At high audio frequencies it may depart from the target accuracy. The resistors have inductance; I measure 150 nH at 100 KHz and 50 nH at 1 KHz on my sample but I would not take those numbers as absolutes. The overall construction is more important and you want the current flow to be as coaxial as possible. Maybe drill a hole in one coin and wire with coax going through the coin to the other side.

What you are really looking for is a Fluke A40A current shunt. There are a few equivalents but all are quite rare. If you were nearby you could borrow mine (which sees virtually zero use). The A40B is closer to what you have built. https://us.flukecal.com/products/electrical-calibration/electrical-standards/a40b-series-precision-dc-and-ac-current-shunts 

I suspect that TCR issue with the copper is also an issue on the precision decade resistors.  How did they overcome that limitation? Did they? https://www.ietlabs.com/esi-tegam-db62-decade-box.html  says 20 ppm up to 10 Ohms.

[KT88]

As long as the device is used as a current shunt, kelvin sensing would mitigate (most of) the problems related to copper leads. The Kelvin connections can be summed with some low Ohm resistors (it looks like Fluke did just that with the A40B series).
Once one tries to put multiple shunts in series to increase R it gets difficult.
Thermal EMF is a different story though...

[MaxTesla]

Hello guys,

very interesing project, and not an easy topic at all. It gets quite complex for frequencies abouve 100Hz.

So first of all at my job we work with very high pulse currents (like 300kA peak) and we mainly use shunts to measure it.
I am aware 20Amps are not quite that, but the same bassic rules apply. If you want to calibrate with a shunt and high frequencies i totally agree that you will need a fully coaxial setup.
Typiclly for high power applications this is done by an outer tube of copper and an smaler inner tube of pure maganin. The connection for the voltage measurement is done by a smaler wire right through the center of the manganin tube. As a said the setup should be as coaxial as possible, so position of the voltage measureing port matters. I have attached the frequency charactarisics of a typical HP shunt...nominal resistance is 40uOhm.

The example shunt is up to 140kA, 1 Impulse in 10 min to limit self heating. I have attached a researche paper discussion the frequency response a little bit more in detail.

This is more or less a secondary topic, manly the self heating of your system will be an issue. So .1 Ohm and 20A is 40W of power loss distributed on 10 reistores, this is still a lot of power to dissipate and leads even with oil cooling to a significant self heating of the resistors.

I would go as fare as to say something like the IT 60-S Ultrastab from lem is a better choice if you like to have the high signal level:

https://www.lem.com/en/it-60s-ultrastab

Basic accuracy is excellent and it can do DC as well. The current to measure drops with this one 600:1, so you can use a much smaler, mach stabler resistor to get your voltage out, less self heating and so on.

We use the bigger brother ITZ 10000-SB for calibration purposes in our lab. Attached is the characteristic of it. I am not saying these are generally better then a shunt, but in this order of magnitude easier to implement maybe. Obiviosly you can get a shunt characterized and use a dynamic scaling as well (thats what we do with our ITZ) to get the precision needed, but personaly i like the elegance of a zero flux current transducer

Hope this gave you some inside

[Micke]

For frequencies higher than a few Hz this paper may be interesting:
https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=24169
Probably already a bit on the academic side but for 20kHz worth reading...

Thanks! I have downloaded the paper, will read it in the weekend! 

[Micke]

Lots of good work and probably a very accurate .1 Ohm resistor at DC. At high audio frequencies it may depart from the target accuracy. The resistors have inductance; I measure 150 nH at 100 KHz and 50 nH at 1 KHz on my sample but I would not take those numbers as absolutes. The overall construction is more important and you want the current flow to be as coaxial as possible. Maybe drill a hole in one coin and wire with coax going through the coin to the other side.

What you are really looking for is a Fluke A40A current shunt. There are a few equivalents but all are quite rare. If you were nearby you could borrow mine (which sees virtually zero use). The A40B is closer to what you have built. https://us.flukecal.com/products/electrical-calibration/electrical-standards/a40b-series-precision-dc-and-ac-current-shunts 

I suspect that TCR issue with the copper is also an issue on the precision decade resistors.  How did they overcome that limitation? Did they? https://www.ietlabs.com/esi-tegam-db62-decade-box.html  says 20 ppm up to 10 Ohms.

Hmm, good idea with coaxial cable through one coin, did not think of that! If I build another I will build like this!
At least I had the coaxial design in mind when placing the resistors, and I kept wires to minimal length.
Actually I DO have a Fluke A40A-10A    But thanks for the offer!    measured it to 43.17mOhm 
Would be interesting to see a teardown of a A40A!

But I built this anyway because:
1) Useful with higher current rating
2) My DIY uses hermetic resistors, unsure of the long time stability of the A40A
3) It is fun building things 

Now in the weekend I will use same cables (coaxial cables for current and sense), and compare A40A and the DIY 0.1R from DC to 20kHz.
Moment of truth! 

[Micke]

Did you actually measure tempco of paralleled setup. I've tried to combine multiple of these JRLs into higher power 1R device, but due to copper leads TCR the resulting tempco was much worse than any single resistor.

I didn't figure out way to improve it yet, but did log results on xDevs draft.

Have not measured tempco yet, but will try do that!
But I have a "feeling" it is rather good, just finished the shunt like 2 days ago, have not done any serious measurements yet, but when starting with a cold shunt and putting 10A through it, the voltage across does not change many ppm (recall like 30ppm? Will check again!) even after shunt getting warm. Of course if I was really unlucky, the tempco of the 4600 and the shunt was of opposing direction, but quite unlikely 
Can imagine there are lot of pitfalls when doing series/parallel like you tried, not obvious!
I was surprised how really bad the result was when I measured Sense on the coins (very close to center), I knew it was a bad idea due to high TC of copper, but this bad? 

[Micke]

As long as the device is used as a current shunt, kelvin sensing would mitigate (most of) the problems related to copper leads. The Kelvin connections can be summed with some low Ohm resistors (it looks like Fluke did just that with the A40B series).
Once one tries to put multiple shunts in series to increase R it gets difficult.
Thermal EMF is a different story though...

I was thinking of putting low ohm resistors on all sense connections (Option 3 in picture), but discarded that idea due to risk of impacting AC performance. Yes, I also think putting resistors in series is more difficult than parallel, higher inductance, bigger "loop area" etc.

[Micke]

Hello guys,

very interesing project, and not an easy topic at all. It gets quite complex for frequencies abouve 100Hz.

... [Removed text]

Hope this gave you some inside

Interesting insight MaxTesla! Thanks!
Wow, currents up to 300kA, that is something!
It is strange it is so difficult making good shunts for AC, one would think that audio frequencies up to like 20kHz would not be that difficult, but as you say, it is!
The 140kA shunt you posted looks really massive, how heavy and big is it? looks like 5-10kg!
Have read before on the forum about LEM Ultrastab, high performance, they do have some advantages!
Yepp, in my case the dissipation is 40W at 20A, so I decided having a fan is mandatory! 
Right now the DIY shunt is under some kind of "burn in" I am running 15-20A through it to stabilize, maybe unnecessary, but thought it might help.

[MaxTesla]

Hey Micke,

more like 35 Kilos and really massiv

Here is another german company who does shunts fyi : https://www.hilo-test.de/products/product-categories/shunts/item/178-wsm-en
For the higher currents at least yes, there you have all kinds of magic effects

Maybe your TC problem is only a secondary effect...did you check wether the current is shared equally? maybe one resistor is taking more and therefor getting much warmer?

BR,

Max

[Kleinstein]

I don't see why the extra resistors should effect the AC performance. The sense side sees relatively low current and is less critical with prasitic inductance. It is more the inductive coupling that can be a problem. So the sense wires should get geometrically close together relatively symmeric. No need for a coax, twisted wires should be enough.

[quarks]

very nice, thanks for sharing

[jfphp]

HOLT HCS1. 3 years ago Mr DeWitt was still working and was of great help with the cal tables of my shunt box (he was more than 90 years old !)

[Micke]

Did some comparison today between FLUKE A40A-10A and DIY, only one shunt at a time, and used same cables.
Used coaxial cables, RG213 for current and RG58 for voltage.
The shunts are quite consistent up to 1kHz, then things start to happen...
Hope the spreadsheet is understandable! 

[Micke]

HOLT HCS1. 3 years ago Mr DeWitt was still working and was of great help with the cal tables of my shunt box (he was more than 90 years old !)

Really nice shunt kit you have!
The specs are really really good 
Wow, Mr DeWitt was still active despite his age, impressive! I hope to be that too...

There is a kit on Ebay, but unfortunately above my budget...

[HighVoltage]

Very nice and interesting project.
Thanks for sharing this.

[Andrew Morton]

Did you consider running the enclosure for the resistors at more than room temperature, say 35 °C, as the calibrated point? Like an OCXO.

[Micke]

Did you consider running the enclosure for the resistors at more than room temperature, say 35 °C, as the calibrated point? Like an OCXO.

No, I have not considered keeping the temperature constant at higher temperature, the temperature coefficient of the resistors is specified 2,5ppm/C which is low enough for me 
But not bad idea though, performance would be even better!

I did an unofficial calibration at a calibration lab in September last year, including comparison with my Fluke A40A shunt.
The Fluke A40A had quite much flatter frequency response than my DIY... 

[DavidKo]

When I have seen the HCS design the bifilar winding come to my mind, that you can reduce the inductance by packing 2 resistors in series close together to reduce the inductance of its wires. Pity I have not done a pictures from my repair of Metra L110 resistance decade. There have been the resistors done on PCB as traces (probably not from Cu, decade had issues with switches, which was a design flaw) in bifilar configuration, but parallel traces or wires can be used for connection of known resistors to reduce the effect of inductance at the first step. Second level is to evaluate its impedance of such a line.

Micke: Have you had measured the frequency characteristic of single resistor? Maybe it is given by its internal construction and you cannot change it. At the end can be simpler to go with 1 or 10 mOhm shunt from cheap non inductive resistors with known TCR. You will have less heat to care about. Second reason why to use 1mOhm is, that you will lose 2V@20Amps on your shunt. As you can see from Agilent, Keysight 34330a 1mV/A  (15A continuous) you can do 15 Amps shunt completely enclosed in epoxy. From my point of view the 10x0,1Ohm non-inductive resistors can be a sweet spot for your application. You will have only 4W and 0.2V of Burden voltage. Power loss is small so I do not expect necessity of cooling in case you will use big resistors.

Edit: corrected the wrong calculation

[Micke]

DavidKo: No, I have not unfortunately not measured the  frequency characteristic of a single resistor, should have done that of course...
But I do suspect though the performance of single resistor is much better! My sense connection is taken from one of the resistors, might cause problems as well.
Yepp, the power dissipation it rather high on my shunt...  but at 10A it is "only" 10W    massive cooling area and fan helps!
Why I chose 10x1R was that I found them at reasonable price and they were hermetically sealed and have low TC, I was mainly after long term stability. If I had found lower resistance with similar performance at reasonable cost I would have gone more your route   

[HighVoltage]

If you need a LEM Ultrastab for measurements, let me know. I bought a few over the years and have too many laying around.

What source do you use for the AC current?

[DavidKo]

Micke: than the only way is to try to cancel out the inductance as much as possible. Use bigger connection surface to shorten the legs (if possible rotate them 90° and solder them to bigger plate). Parallel twisted cable to the middle of the "resistor" split the cable to the sides.

[lugaw]

Is resistance welding better for this kind of metrology grade instruments?