0.5 uOhm shunt current sensing practical resolution?

[max_torque]

Hey all,

I'm after some info and experience on sensing currents across fairly low resistance shunts.  If i had a hald microohm shunt, what is a resonably practicable sensing resolution and noise without going ridiculous pcb layout and componentry?  If i wanted to sense 1000amps max, how low a current would i expect to be able to resolve repeatably?  (offset would be handled by individual calibration, so tempco and basic resolution / noise are my concerns)

At say 1 amp, id get half a microvolt of voltage drop across the shunt, so i'd need a gain or at least 250 and probably 500 to get this to a reasonable magnitude for an ADC to sample?

Any one got any direct experience here?

[Conrad Hoffman]

No direct experience but what you have is a dynamic range problem, not much different than many others. The impedance is low so noise isn't much of an issue, but you'll have trouble at the low end as with any other microvolt level system. Is there any way to divide the problem by switching a low value shunt in and out, in parallel with a slightly higher value shunt? Is there enough power/space/ventilation available that you could use a tapped shunt? Or, don't use a shunt at all, but a current transformer or possibly Hall effect sensor? IMO, trying to do this in one shot with a single shunt is going to be suboptimal.

[joeqsmith]

You really don't provide enough details about what you are attempting to do.  Normally for higher higher DC currents where I am concerned about the losses,  I will use a LEM sensor.  For AC, some sort of transformer.    A while back I looked at one of the TI 32-bit ADCs.  The part offers several features including an internal PGA. I stuck it into a thermal chamber with roughly 0.2C error along with a shielded box.   I doubt even this would be good enough for what you want but maybe it will provide you some insight.       

https://www.ti.com/product/ADS1262

[joeqsmith]

Showing a 100uOhm 500A shunt along with a LEM sensor.   Way outside of what you are asking, still it may provide some perspective.   
 

[Jay_Diddy_B]

Hi,

I think that it is un-realistic to have a 0.5u shunt.

I found a 50u 1000A shunt from Ohmite SH8536F1K0APEP.

These are large:



A shunt like this will have 50W of dissipation at full-scale.

You will be mounting the PCB on the shunt, not the shunt on the PCB.

Vishay has a similar part 25u   WSBS8536L0250JK80




This would be 25W dissipation at 1kA

and 25uV at 1A You have reasonable chance of reading 25uV with reasonable accuracy.

Consider that a square of 1oz copper on PCB has a resistance of 500u  . You do not want run 1000A through the PCB, even if you could 10oz copper.


Jay_Diddy_B

[mzzj]

Really depends on the shunt and how isothermal you manage to keep the shunt. Half microvolt is easy on short-term and in stable indoor environment. 10 years without possibility to adjust the offset and over automotive temperature range would be more like 20 microvolts.

Fluxgate current sensors give you pretty easily 1:1000000 dynamic range but these are not cheap. Plus they are quite power-hungry

[Jay_Diddy_B]

Hi,
Here is a link to a commercial product:

https://riedon.com/media/pdf/SSA.pdf


Digikey:    
696-SSA-1000-ND




Jay_Diddy_B

[max_torque]

I’m trying to use a rather non optimum shunt that is actually a copper through pcb top hat.  The pcb sits directly on the terminals of a lithium cell and current flows to that cell through copper top hats that are bolted to the cell tabs

This means my shunt is short ie only the thickness of the pcb itself and because it also has to carry the mechanical loads of attaching bus bars to the cell it has to be generously dimensioned in terms of area and hence becomes pretty low resistance.  I can’t practically use a LEM or similar all in one type sensor but I could perhaps do something clever with sort form of Hall effect type sensor?

[max_torque]

To make that clearer:

The pcb has large plated through holes of 11mm ID into those holes is soldered a copper top hat which has a 7mm hole up the middle.  The top hat sits on the terminal tab of a large lithium cell and conducts current up and to a laminated planar bus bar asst that sits on the opposing side of the pcb.

On each side of the pcb a track samples the voltage drop across the top hat so the shunt is just 1.6 mm long.  It is low resistance because the cross sectional area must carry the mechanical clamp loads and the fastener itself is conductive

The aim is to be able to sample cell terminal voltage and current directly on the tabs of the host cell

[PartialDischarge]

The TC of copper is so bad that you can forget about the rest of the electronic accuracy issues

[max_torque]

We are measuring the temp of the shunt so can correct for temperature effects and the temperature of the top hat is actually dominated by the thermal inertia and thermal impedance of the large planar bus bar assy to which it is very firmly bolted and by the thermal coupling to the cell below it too

[PartialDischarge]

We are measuring the temp of the shunt so can correct for temperature effects and the temperature of the top hat is actually dominated by the thermal inertia and thermal impedance of the large planar bus bar assy to which it is very firmly bolted and by the thermal coupling to the cell below it too

That is like saying that you're developing a car with square tyres, but the suspension system is being reengineered to ease on those square edges hitting the road.

But if you are really like copper for its low cost for example, and are using "correction" methods there is another simpler solution: Split the current passing by a cable or busbar in 2 branches, lets say 95% goes thru one branch unmeasured, and the other 5% goes thru other branch with a real less expensive low TC shunt that you can measure. In practice you'll need to determine how much % goes to each branch

[mzzj]

We are measuring the temp of the shunt so can correct for temperature effects and the temperature of the top hat is actually dominated by the thermal inertia and thermal impedance of the large planar bus bar assy to which it is very firmly bolted and by the thermal coupling to the cell below it too

That is like saying that you're developing a car with square tyres, but the suspension system is being reengineered to ease on those square edges hitting the road.

But if you are really like copper for its low cost for example, and are using "correction" methods there is another simpler solution: Split the current passing by a cable or busbar in 2 branches, lets say 95% goes thru one branch unmeasured, and the other 5% goes thru other branch with a real less expensive low TC shunt that you can measure. In practice you'll need to determine how much % goes to each branch

how would you keep the balance at 95% when other branch still has large tempco? 

Copper tempco is not that big problem if OP just wants resolution instead of absolute accuracy.
Quite often 1% accuracy is enough good but you need big range. Hall sensors suck on both accuracy and range compared to proper shunts.

[PartialDischarge]

We are measuring the temp of the shunt so can correct for temperature effects and the temperature of the top hat is actually dominated by the thermal inertia and thermal impedance of the large planar bus bar assy to which it is very firmly bolted and by the thermal coupling to the cell below it too

That is like saying that you're developing a car with square tyres, but the suspension system is being reengineered to ease on those square edges hitting the road.

But if you are really like copper for its low cost for example, and are using "correction" methods there is another simpler solution: Split the current passing by a cable or busbar in 2 branches, lets say 95% goes thru one branch unmeasured, and the other 5% goes thru other branch with a real less expensive low TC shunt that you can measure. In practice you'll need to determine how much % goes to each branch

how would you keep the balance at 95% when other branch still has large tempco? 

Exactly, I did not invent something extraordiary and I would not do it this way, still would need corrections. But I still think it would be a more sane approach than the initial one

[max_torque]

In an ideal world I would have the package space to implement a proper shunt or LEM solution but unfortunately I don’t work in the ideal world!  I have an awkward and small package space that precludes nice options hence me trying to work out what can be sensed in the space available which is Around 2mm thickness….

We also have an external series high precision IVT-S current shunt that gives us effectively an
external DC current reading but I need a higher bandwidth local current value but this can be correlated and corrected in SW against the external sensor under close to dc conditions

[Marco]

I think you're going to need to experiment.

In theory a 1 nV/rthz amplifier should be able to get you to a couple kHz for 1A resolution purely from a Johnson noise perspective, but how much temperature differential could you expect across the "shunt" when there's 1000A going through?

[coppice]

I’m trying to use a rather non optimum shunt that is actually a copper through pcb top hat.

The temperature coefficient of copper is 0.393% per degree C.It might be good enough for a shunt for some kind of rough and ready current sensing, but its useless for any serious measurement. When you build a low resistance shunt with a metal that has a reasonable temperature coefficient you still need to get every last scrap of copper out of the measurement path. Its usually necessary to use a 4 terminal shunt, with the sense terminals coming from the low coefficient metal, well before the current path transitions to copper.

I don't think you said what you are measuring. If its DC, controlling the offset in the amps and ADC is usually a challenge when trying to achieve high accuracy from a small signal. If you are measuring AC you can kill the offset easily. 0.5microohms is really low. People usually use at least 10 to 20 micro-ohms for 1000A. With careful design you can achieve good accuracy like that. The I2R issue is tough. At 100A you can use 200 micro-ohms, get a decent sized signal, and only dissipate 2W. Above 100A shunt dissipation starts to be a real problem. Really high current shunts typically have a large heat sink, and people accept 10' of watt of loss, because its a tiny part of the total power.

[max_torque]

I'd like to be able to "see" 1000A, but frankly, above around 400A i have no accuracy or resolution targets. 1000A is peak, instantaneous current that occurs rarely and is of a magnitude that all i need to know is that it's happening, and not how much of it is happening if that makes sense?

Surely the temp co is irrelevant as to the absolute shunt impedance, ie a change of 0.4% per degree is a change of 0.4% per degree irrespective of your nominal shunt R, ie it effects a 100ohm shunt exactly as much as a 100uOhm shunt?

A low absolute shunt impedance only affects gain, noise and offsets from your analogue front end surely?  On advantage of a very low impedance seems to actually be that self heating is going to be low (I^2R)?  ie for any given physcial shunt a 10 times lower nominal impedance means 10 times less heating in the shunt?

[Marco]

To me it seems temperature differences between both sides of the PCB (the copper pipe is a pretty good heat conductor too, but it has to be close to perfect for Seebeck effect at these kinds of minute voltages) and thermal expansion of the PCB and copper pipe stressing the solder joint with unpredictable results are going to be the limiting factor.

PS. you probably don't want to connect the amplifier directly to a via when bringing the signal from the other side, run a thin trace surrounded by a ground pour for a bit. It doesn't take a lot of temperature difference to crowd out a couple 100 nV.

[coppice]

Surely the temp co is irrelevant as to the absolute shunt impedance, ie a change of 0.4% per degree is a change of 0.4% per degree irrespective of your nominal shunt R, ie it effects a 100ohm shunt exactly as much as a 100uOhm shunt?

For a pure copper shunt this is true.  However, consider putting a piece of low coeff material between 2 copper leads. If the resistance of the low coeff material is 1k ohms, the crazy coefficient of a few milli-ohms of copper doesn't affect things too much. When you have 10 milli-ohms of low coefficient material a small amount of copper really starts to affect things. When you get to 100 or 200 micro-ohm shunts, even small lengths of copper can be a substantial part of the total resistance. Then, you REALLY need to tap from the low coefficient material for sensing.

[JohnG]

There are a number of sensors that are sensitive to DC magnetic field. Allegro makes a bunch of hall sensors with various compensations built in for temp, etc. There are some newer ones that use a tunneling magnetoresistive (TMR) sensor. Some companies that make TMR sensors for contactless sensing are Crocus (https://crocus-technology.com/products/contactless-sensors/), which is owned by Murata, and TDK (https://www.tdk.com/en/featured_stories/entry_012.html).

You would need to figure out a good place to mount them near the conductor and figure out some gain experimentally or otherwise.

These sensors aren't the cheapest, but I think it will be really expensive to measure current with any accuracy and dynamic range using a hunk of copper with 5e-7 ohm resistance.

John

[David Hess]

At say 1 amp, id get half a microvolt of voltage drop across the shunt, so i'd need a gain or at least 250 and probably 500 to get this to a reasonable magnitude for an ADC to sample?

Any one got any direct experience here?

The input referred noise of the amplifier determines the usable resolution, but even if the shunt is perfect, there are also errors from thermocouple junctions between the shunt and amplifier which will be particularly tenacious because of heating of the current shunt.  Excluding the thermocouple problem, resolution down to less than a microvolt is feasible with proper component selection and design.

[max_torque]

After some new 3d design work, and some calculations, it looks like i can get the "shunt" resistance up to around 4 to 5 uOhm's, which is a much better figure i think, and not using copper is going to help, and after doing the thermal calc's it looks like we don't need too (because the average current is much lower at only around 250A). I'm also happy to use a PGA in the system somewhere and have a "high range" mode where i accept a marked reduction in resolution in order to be able to not go overrange.  The total energy used during these short high current periods is small, so offset and absolute errors here will not affect the energy sumation nearly as much as the much longer periods spent at lower currents

I'm also looking at non contact bus bar sensing using a Allegro hall effect sensor as a fall back, but i'd like to avoid this due to the additional mechanical design and build complexity it brings. I'll probably put on a seperate FFC header to run out to this sensor board, giving me options as to where it sits and if to include it or not.

[Marco]

It doesn't necessarily help with thermocouple effect, more electrical resistance will likely mean more thermal resistance too so they could just scale together. Nor necessarily with effects of differential thermal expansion of the "shunt" and the PCB/solder joint, which could also scale along.

It helps with noise, but noise wasn't a hugely limiting factor to begin with.