Kelvin sensing for current measurement

[anvoice]

I'd like to measure a current of up to about 30A through a sense resistor. I understand that Kelvin sensing is the preferred method for this as it eliminates the voltage drop across the conductors in the high current path. What I don't understand is why 4 terminal resistors are developed specifically for that purpose. In essence, a 2 terminal SMD resistor where the voltage sense pads come straight off the main pads seems to do exactly that, as evidenced by the picture in the operating principle for Kelvin sensing from Wikipedia.
That would mean I can get away with using a 2-terminal resistor and get the same advantages as a 4-terminal one provided I connect them properly. This is important because I just can't find a 4-terminal SMD resistor of about 5mOhm (I think I don't want to go much lower for reasonably accurate measurements) with the right power rating. Am I missing something?

[rx8pilot]

The 4 terminal devices are generally for repeatable precision by defining the exact location of the sense.

I use regular two terminal devices all the time and run sense traces to the inside of the device.

[anvoice]

Not sure why 4 discrete terminals would be more repeatable than what is essentially 4 terminals using traces from pads?
What this means though is that there is functionally no difference if done right if I understand correctly?

[rx8pilot]

I cannot think of a functional improvement other than guiding the designer toward a proper Kelvin connection.

[coppice]

You can't find a 4 terminal resistor in stock, or you can even find a part number for one? Most serious resistors makers offer 4 terminal parts, both surface mount and through hole mount. Not every distributor stocks them, so you might need to look around. Since 30A through 5 milliohms means 4.5W of dissipation in the resistor, a suitable resistor is rather large, and usually of through hole design. These days people usually use a lower value resistor for 30A, keep the dissipation down, and use an SMD part.

If you use a 2 terminal SMD resistor you'll probably need something a bit lower value than 5 milliohms, because of the heat. If you follow the layout guidelines you can find in http://www.analog.com/en/analog-dialogue/articles/optimize-high-current-sensing-accuracy.html you can get pretty good results. However, the complete isolation of a 4 terminal resistor normally has the edge, especially in keeping the temperature coefficient under control.

Trying to use a very low value 2 terminal through hole resistor for current sensing gives really poor results, as you just can't keep copper out of the sensed path. The 0.4%/C temperature coefficient of even a small amount of copper can really mess with the accuracy.

[coppice]

The 4 terminal devices are generally for repeatable precision by defining the exact location of the sense.

I use regular two terminal devices all the time and run sense traces to the inside of the device.

Don't use that layout. It gives a horrible temperature coefficient, because you haven't isolated the copper in the sense path from the copper in the load path. Use the kind of layout you'll find in http://www.analog.com/en/analog-dialogue/articles/optimize-high-current-sensing-accuracy.html

[rx8pilot]

Don't use that layout. It gives a horrible temperature coefficient, because you haven't isolated the copper in the sense path from the copper in the load path. Use the kind of layout you'll find in http://www.analog.com/en/analog-dialogue/articles/optimize-high-current-sensing-accuracy.html

Excellent app note for precision layouts.....

[ejeffrey]

There are at least 2 ways you can do this:

1) Use a standard SMD footprint for your device, and run two traces to each pad.  This seems to be what anovice is suggesting.  This is simple and has no assembly problems even on small resistors, but has the problem that the solder joint is included in the resistance.  There is also a voltage gradient across the pad as the current moves towards the current trace, so the value you get depends exactly on how the sense lead comes in.  This is fine for basic use, but is really inferior to any true 4 terminal connection.

2) Use two completely separate pads, and solder the resistor terminal to both of them.  This is much better because you get the solder joint (mostly) out of the equation, but still has the same problem that there is a voltage drop across the terminal, and you are sensing the voltage at a single point.  If the terminal has a significant fraction of the total resistance, your effective resistance depends on where the sense lead is connected.  Take a look at that analog devices link that coppice posted -- the exact design of the pads make a pretty big difference.  If the datasheet doesn't specify a 4 terminal layout (which cheap 2 terminal resistors don't normally do), then you can do your best, but you can't guarantee you get the best results. 

In addition, the terminal material may not be the same as the resistive element, and therefore may not have the same temperature coefficient or aging performance as the main resistive element.

It all comes down to what level of accuracy and repeatability you need.  A 4-terminal device lets the manufacturer control exactly where the sense leads are connected and make sure that it meets the datasheet performance when used that way. 

[anvoice]

You can't find a 4 terminal resistor in stock, or you can even find a part number for one?

I can't find a 5mOhm, 4 terminal SMD resistor rated for at least 4.5W. I found one of the through-hole variety though. I'd like reasonable measurement accuracy so I didn't want to go much lower than 5mOhms, but that will depend on what I can find of course.

If you use a 2 terminal SMD resistor you'll probably need something a bit lower value than 5 milliohms, because of the heat. If you follow the layout guidelines you can find in http://www.analog.com/en/analog-dialogue/articles/optimize-high-current-sensing-accuracy.html you can get pretty good results. However, the complete isolation of a 4 terminal resistor normally has the edge, especially in keeping the temperature coefficient under control.

Thanks for the link, very useful. Problem with using a 4-terminal is that I will then either need to go through-hole, or settle for using a smaller resistance value 4-terminal device. Will my measurements with a true 4-terminal 1mOhm resistor really be better than with a 2-terminal 5mOhm one, properly arranged?

There are at least 2 ways you can do this:

1) Use a standard SMD footprint for your device, and run two traces to each pad.  This seems to be what anovice is suggesting.  This is simple and has no assembly problems even on small resistors, but has the problem that the solder joint is included in the resistance.  There is also a voltage gradient across the pad as the current moves towards the current trace, so the value you get depends exactly on how the sense lead comes in.  This is fine for basic use, but is really inferior to any true 4 terminal connection.

2) Use two completely separate pads, and solder the resistor terminal to both of them.  This is much better because you get the solder joint (mostly) out of the equation, but still has the same problem that there is a voltage drop across the terminal, and you are sensing the voltage at a single point.  If the terminal has a significant fraction of the total resistance, your effective resistance depends on where the sense lead is connected.  Take a look at that analog devices link that coppice posted -- the exact design of the pads make a pretty big difference.  If the datasheet doesn't specify a 4 terminal layout (which cheap 2 terminal resistors don't normally do), then you can do your best, but you can't guarantee you get the best results. 

In addition, the terminal material may not be the same as the resistive element, and therefore may not have the same temperature coefficient or aging performance as the main resistive element.

It all comes down to what level of accuracy and repeatability you need.  A 4-terminal device lets the manufacturer control exactly where the sense leads are connected and make sure that it meets the datasheet performance when used that way. 

Won't pretend that I have all of that grasped just yet (I'm working on it), but seems like very valid points. So the recommendation is to go with a 4-terminal through-hole, as that is the only 4-terminal resistor I can find meeting the power dissipation requirement? I'd like good accuracy for data logging and possibly torque control.

[rx8pilot]

For 30A, consider a lower value and larger package.

Short and misplld from my mobile......

[ejeffrey]

. So the recommendation is to go with a 4-terminal through-hole, as that is the only 4-terminal resistor I can find meeting the power dissipation requirement? I'd like good accuracy for data logging and possibly torque control.

It just depends on what accuracy you need/want.  Is 5% good enough?  1%?  0.1%?  I tried to do a rough estimate.  For a 5 mohm shunt in a 5W package, 2 terminal devices should easily be good for 1% with good layout.  If you were using a 500 micro-ohm resistor in a 1 W package, that would be more difficult to reach 1%, but 5% might be fine.

[anvoice]

It just depends on what accuracy you need/want.  Is 5% good enough?  1%?  0.1%?  I tried to do a rough estimate.  For a 5 mohm shunt in a 5W package, 2 terminal devices should easily be good for 1% with good layout.  If you were using a 500 micro-ohm resistor in a 1 W package, that would be more difficult to reach 1%, but 5% might be fine.

I feel that 1% should be a reasonable starting point. The resistors are pretty massive (I'm looking at this: https://www.mouser.com/ProductDetail/66-SLN5TTED5L00D), but I should be able to fit them on if I try hard enough. The other choice would be 1mOhm in a 1W, true 4 terminal package (https://www.mouser.com/ProductDetail/71-WSKW06121L000FEA). My intuition says the 5 mOhms will be more accurate despite possible advantages of the 4 terminal resistor, for what that's worth. I'll start designing around the 5mOhm ones for now.
Thanks for your guidance.

[coppice]

I feel that 1% should be a reasonable starting point. The resistors are pretty massive (I'm looking at this: https://www.mouser.com/ProductDetail/66-SLN5TTED5L00D), but I should be able to fit them on if I try hard enough. The other choice would be 1mOhm in a 1W, true 4 terminal package (https://www.mouser.com/ProductDetail/71-WSKW06121L000FEA). My intuition says the 5 mOhms will be more accurate despite possible advantages of the 4 terminal resistor, for what that's worth. I'll start designing around the 5mOhm ones for now.
Thanks for your guidance.

Intuition? So you're not trying to actually engineer something.

What are you trying to measure? Mains power or something with a wider bandwidth? For mains power, most solutions designed to accept a signal from a shunt are designed to accept a much smaller signal than you will get from 5 milliohms at 30A. People don't want to have to get rid of 4.5W from a small spot, as it requires so much space for adequate cooling. Cooling isn't just about avoiding excessive temperatures. Its also about accuracy, because of the temperature coefficient of the shunt. "I should be able to fit them on if I try hard enough" makes it sound like you are only allowing space for the shunt, and not thinking through the surrounding space needed to keep the shunt nice and cool. Most 0.1% accuracy wide dynamic range measurement solutions, like those in mains energy metering devices, are designed for the sort of signal you will get from a 0.5milliohm shunt at 30A, and can sustain their 0.1% performance over a 1000:1 current range. In most uses the key parameter is the CMRR of the differential input to the electronics. If that's high enough, you can work with a really small signal from the shunt.

[David Hess]

Even if geometry was not a problem, the temperature coefficient of resistance of the copper, solder joint, and part termination would spoil the accuracy compared to a 4-wire resistor.  In less critical applications, I try to run the Kelvin connections directly opposite to the high current connections instead of at 90 degrees.

[anvoice]

Intuition? So you're not trying to actually engineer something.

What are you trying to measure? Mains power or something with a wider bandwidth? For mains power, most solutions designed to accept a signal from a shunt are designed to accept a much smaller signal than you will get from 5 milliohms at 30A. People don't want to have to get rid of 4.5W from a small spot, as it requires so much space for adequate cooling. Cooling isn't just about avoiding excessive temperatures. Its also about accuracy, because of the temperature coefficient of the shunt. "I should be able to fit them on if I try hard enough" makes it sound like you are only allowing space for the shunt, and not thinking through the surrounding space needed to keep the shunt nice and cool. Most 0.1% accuracy wide dynamic range measurement solutions, like those in mains energy metering devices, are designed for the sort of signal you will get from a 0.5milliohm shunt at 30A, and can sustain their 0.1% performance over a 1000:1 current range. In most uses the key parameter is the CMRR of the differential input to the electronics. If that's high enough, you can work with a really small signal from the shunt.

Let me clarify: I'm frantically reading about all the aspects of the design I'm making, but because my education is pretty far from electronics, it's taking a while. Hence I have to occasionally rely on intuition as a placeholder for knowledge I plan to attain. I really do try.

This design is for a brushless DC motor controller, to be run at 12V. I don't know how that will compare to measuring mains, but I wanted a safety margin as far as accuracy goes. I'm gladly taking advice, including critical advice.

Heat will be produced by the motor as well as the MOSFETs driving it. I don't know exactly how much yet because I don't know the motor's internal losses, but it may well rival the 4.5W of the shunt resistors. I don't think I want active cooling as that will complicate the system and increase noise, rather I plan to have an adequate aluminum enclosure to drive the heat off.

Even if geometry was not a problem, the temperature coefficient of resistance of the copper, solder joint, and part termination would spoil the accuracy compared to a 4-wire resistor.  In less critical applications, try to run the Kelvin connections directly opposite to the high current connections instead of at 90 degrees.

I have nothing against 4-terminal resistors, the only reason I was considering the 2-pads is because they're easier to find in the right power dissipation rating with a higher resistance. From what I'm reading though, it may be difficult to tell exactly whether 2-terminal 5mOhms in a 5W package will be better than a 4-terminal 1 mOhm in a 1W package without testing it. Unless of course one could quantify the temperature coefficient of resistance of the copper, solder joint, and part termination easily.

[rx8pilot]

This design is for a brushless DC motor controller, to be run at 12V. I don't know how that will compare to measuring mains, but I wanted a safety margin as far as accuracy goes. I'm gladly taking advice, including critical advice.

Precision should not be too serious for that application. A typical center tap layout with a 2 terminal resistor should be adequate. While the layout example I posted earlier will not pass in any metrology application - it is rather typical for motor controls, SMPS, etc. where ultimate accuracy is not required.

[anvoice]

Precision should not be too serious for that application. A typical center tap layout with a 2 terminal resistor should be adequate. While the layout example I posted earlier will not pass in any metrology application - it is rather typical for motor controls, SMPS, etc. where ultimate accuracy is not required.

What would you classify as acceptable for motor control? 1% or higher? There are 3 use cases for measuring current that I see, 1) overcurrent protection, where a crude measurement will do, 2) data logging, where 1% seems acceptable, and 3) torque control, which may or may not be important at all depending on motor application.

[Jwillis]

This might help to understand the difference.http://www.rcdcomponents.com/rcd/press/R-31%20Two-Terminal%20vs.%20Four-Terminal%20Resistors.pdf . And may help in your choices.

[anvoice]

This might help to understand the difference.http://www.rcdcomponents.com/rcd/press/R-31%20Two-Terminal%20vs.%20Four-Terminal%20Resistors.pdf . And may help in your choices.

Thank you, that is exactly what I'm looking for.
One thing I don't understand: I can certainly see how spacing is important for a through-hole resistor, but what about SMD? There the spacings would be equivalent to the spacing between PCB pads, is that not the case? Also, how does the 4-lead greatly minimize the thermal instability? The document briefly mentions that at the end but offers no explanation.

[coppice]

It does tell you. It's the horrible temperature coefficient of the  leads. Something you avoid with an SMD part. I don't know exactly what the leads are made from, but most resistors use some kind of copper alloy. They want the leads to be

stiffer than pure copper, but the copper content will ensure the temperature coefficient is pretty bad.

[anvoice]

Ok, so the 4-terminal devices are made of a special allow that allows their temperature coefficient to be controlled better. That does make sense.

I still don't understand why one can't make the same design with the same type of alloy using a 2-terminal resistor soldered onto 4 pads, but I do notice that the 4-terminal resistor from Vishay does have an extremely low temperature coefficient (20 ppm), lower than any 2-terminal part I can find.

[coppice]

Ok, so the 4-terminal devices are made of a special allow that allows their temperature coefficient to be controlled better. That does make sense.

I still don't understand why one can't make the same design with the same type of alloy using a 2-terminal resistor soldered onto 4 pads, but I do notice that the 4-terminal resistor from Vishay does have an extremely low temperature coefficient (20 ppm), lower than any 2-terminal part I can find.

No. A 4-terminal resistor gets the copper of the load path out of the sensing path, so it's temperature coefficient doesn't affect measurements.

[anvoice]

No. A 4-terminal resistor gets the copper of the load path out of the sensing path, so it's temperature coefficient doesn't affect measurements.

My understanding is that since the 4-terminal resistor is a monolithic device, it will still get hot where the sense connection is. Thus introducing some temperature coefficient at least. Perhaps it just doesn't get as hot, which is why its apparent temperature coefficient is lower.

[ogden]

My understanding is that since the 4-terminal resistor is a monolithic device, it will still get hot where the sense connection is. Thus introducing some temperature coefficient at least. Perhaps it just doesn't get as hot, which is why its apparent temperature coefficient is lower.

High current supposedly shall not be running through sense path. For high impedance input of ADC or opamp it does not matter - sense path resistance changes by < 0.1 Ohms or not.

[coppice]

No. A 4-terminal resistor gets the copper of the load path out of the sensing path, so it's temperature coefficient doesn't affect measurements.

My understanding is that since the 4-terminal resistor is a monolithic device, it will still get hot where the sense connection is. Thus introducing some temperature coefficient at least. Perhaps it just doesn't get as hot, which is why its apparent temperature coefficient is lower.

It seems you fail to understand what a 4 terminal shunt is all about.

With the sense points of a 4 terminal shunt being right on the resistive element, they are not sensitive to the temperature dependent voltage drops along the copper based leads carrying the load current. Those leads have just as bad a temperature coefficient as a 2 terminal shunt, but it doesn't affect the sensed voltage drop across the resistive element. That is the reason 4 terminal devices are manufactured.