3
\$\begingroup\$

I am struggling to get my head around Kelvin PCB routing for shunt resistors.

  1. First how do I tell if a shunt resistor on a PCB is routed using the Kelvin method?

Below are three images:

THE TWO RED RECTANGLES ARE COPPER POURS

THE TWO THIN TRACES ARE GOING TO THE SENSOR IC

PLEASE SEE QUESTIONS AT THE END

FIGURE 1

enter image description here

FIGURE 2

enter image description here

FIGURE 3

enter image description here

  1. Which one of the three figures is regarded as using the Kelvin method and why?

  2. For the Figures that are not Kelvin routed, why is it?

\$\endgroup\$
3
  • 1
    \$\begingroup\$ A big factor is where the current enters and where the current leaves the power connections. \$\endgroup\$
    – Andy aka
    Apr 18 at 20:06
  • 1
    \$\begingroup\$ It will also matter a lot whether it is a 1 milliohm or 100 milliohm shunt and whether you are aiming for 1% or 0.1% accuracy. The copper pours will have resistance of somewhere around 0.1 milliohms, depending on actual dimensions. \$\endgroup\$
    – jpa
    Apr 19 at 16:34
  • 1
    \$\begingroup\$ Good AD app note on optimizing pads for kelvin measurement of non-kelvin packages. Good LTC app note on combining these for parallel systems. \$\endgroup\$
    – Mark Omo
    Apr 19 at 18:55

3 Answers 3

8
\$\begingroup\$

The main goal of a Kelvin connection is that potential drops in the wires delivering the current to the resistor don't get measured as part of the resistor drop.

That means that all of your examples are more-or-less representative of Kelvin connections, although the first one is probably the best example, which will have the least error due to feed-line resistance.

It should also be said that the additional resistor devices shown between the two contact pads are likely to cause some errors in the measurement, as you can only make a Kelvin connection to one of the three devices, not all of them at the same time. It would be better to use one bigger (higher-power) device than three devices in parallel if you want to be able to make a Kelvin connection.

An even better example would use a resistor with additional sense pads, to avoid measuring drops over the contact resistance (the solder joint between the board and the SMT device) contributing to the measured drop.

\$\endgroup\$
3
  • \$\begingroup\$ Can you elaborate on this "That means that all of your examples are more-or-less representative of Kelvin connections, although the first one is probably the best example, which will have the least error due to feed-line resistance." How do I "see" the feed line resistance, can you draw it for each Figure? \$\endgroup\$
    – JoeyB
    Apr 20 at 10:20
  • \$\begingroup\$ @JoeyB, a worst-case "non-Kelvin" connection would be if you hooked up the voltmeter inches or meters or kilometers away from the sense resistor. For example, at the terminals of the source producing the current. \$\endgroup\$
    – The Photon
    Apr 20 at 17:34
  • \$\begingroup\$ Well, that answers everything! Why do people or documentation not differentiate between those two cases as easily as you did? \$\endgroup\$
    – JoeyB
    Apr 21 at 8:39
4
\$\begingroup\$

I would dare say neither; this option seems to have been overlooked:

enter image description here

But this still isn't actually much better.

The reason is, a Kelvin connection is only defined for the device it's calibrated for.

A lone resistor will do well enough, but an array of them cannot compensate for the resistance of the copper polygon connecting them, and so a true Kelvin connection cannot be made this way.

But "true" is a matter of degree.

Most responsible, is to at least approximately calculate the spreading resistance between resistors, and determine if that is an acceptable error given the requirements of the system. The resistance may simply be low enough that it is negligible, and then any of these four methods will suffice. Conversely, if the shunt resistors are small enough, or precise enough, the copper spreading resistance will exceed that variance and reduce precision of the system.

The primary concern, in any case, is that 1. initial (design) value cannot be as well controlled, because PCB resistance varies relatively widely (copper thickness and trace width both have fairly modest tolerances), and 2. copper's temperature coefficient is relatively large and positive, so the total resistance changes with temperature (and since temperature varies with current flow, load- and history-dependent error can be introduced this way). And because of (1), (2) cannot simply be compensated by choosing slightly-NTC resistors (even if they were commonly available).

It may be that, the sum of these variances falls well within requirements, and the solution is trivial. Since you don't give an example application and requirements, I won't jump to conclusions whether one or the other is acceptable for your present concern.

Alternately, choose a single four-terminal component with explicit -- and manufacturer calibrated -- Kelvin resistance. These are often relatively expensive parts (particularly when precision values are made with bulk metal foil, even moreso in power ratings), but when you need that level of precision, it's a price well paid.

There's also the option to connect sense traces to each resistor (inside the pads as above, or to its Kelvin terminals if provided), use independent diff amps, and sum their results. This incurs a lot of ICs, and their attendant errors, but it could be justifiable in some situations.

\$\endgroup\$
3
  • \$\begingroup\$ The PTC response of typical resistors and copper seems like it should act towards balancing the currents among the three resistors. Though I agree it will not be perfect. \$\endgroup\$
    – jpa
    Apr 19 at 16:35
  • 2
    \$\begingroup\$ Instead of separate sense amplifiers I remember an article from Linear Technology doing the summing with resistors, it seemed pretty sane: Proper Kelvin Sensing with Multiple Sense Resistors \$\endgroup\$
    – pipe
    Apr 19 at 17:25
  • \$\begingroup\$ @jpa In this case, balance isn't the issue; likely the mismatch is well below any thermal limits. The problem is that it's changing at all, whereas we generally want stable values for sensing. For example, avoiding adding a local temp sensor plus multi-point calibration with respect to it. \$\endgroup\$ Apr 19 at 22:20
0
\$\begingroup\$

In a Kelvin (4-wire) measurement two test points are connected to each side of the measurement. One point acts as the “force” that supplies the current and the second as the “sense” which is used to perform the voltage measurement. The place at which the force and sense paths meet is called a Kevin point.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.