# Shunt's thermal EMF (Seebeck effect) vs Kelvin connection?

While reading this Vishay document about shunts, I saw that temperature mismatch between the shunt's terminals can introduce a "parasitic" voltage drop which will in turn affects the current measurement accuracy.

https://www.vishay.com/docs/49159/_power-metal-strip-shunts-current-shunts_pl0005-1801.pdf

Typically, shunts provide kelvin connection pins, so the voltage is measured directly across the resistive element and not across the whole shunt. With that in mind, I don't see how the EMF can affect the measurement, since this "parasitic" voltage won't appear at the kelvin connection level, but beyond that (At terminal level).

Does this mean that whenever an EMF figure is specified, the voltage is not probed directly across the element or is this applicable to all shunts and I did misinterpret something?

The Seebeck effect is entirely due to dissimilar metals in contact. If the Kelvin connections are connected directly to the sensing element, then a Seebeck voltage will appear on the Kelvin connections if the two connections are at a different temperature. Current is not required.

I don't see how the EMF can affect the measurement, since this "parasitic" voltage won't appear at the kelvin connection level,

So, from a measurement perspective, the Vishay diagrams are missing the Seebeck junctions that I have shown in my diagram below as VT3 and VT4. Cu means copper, Mx means the metal used in the sensing element.

Edit from comment by Ben Voigt: While the Vishay diagrams demonstrate the Seebeck effect, the voltage measured by a Kelvin connection must include the Seebeck voltage.

The Kelvin connection cannot eleminate the thermocouple effect. However that the connections are made, maintaining the connections at the same temperature is a must to reduce Seebeck errors.

simulate this circuit – Schematic created using CircuitLab

• I don't think the Vishay diagram is "missing" Seebeck junctions.... I just don't think the "voltmeter" symbols on the Vishay diagram represent Kelvin connection, or in fact any physical connection whatsoever. They are ideal symbolic (theoretical) meters, introduced to show the change in potential resulting from thermocouple effect. The ideal meters don't have lead resistance, chemical composition, calibration error, thermal noise... Jun 27, 2023 at 20:21
• OK, I take your point @BenVoigt . "My perspective" is from a measurement point of view . I have adjusted my answer Jun 27, 2023 at 20:36
• typos: cannection should be connection, thermocoiple should be thermocouple Jun 27, 2023 at 20:38
• Interesting! Your diagram clarifies the issue I had with Vishay's, perfectly! So even if we probe the resistive element directly, there is still a bit of EMF to consider, although to a lesser extent since there is no power dissipation in the kelvin connected circuit and the element is typically small compared to the whole shunt, so the temperature differential is smaller. Very helpful explanation @RussellH. Thank you Jun 28, 2023 at 7:59

Consider this type or similar type of shunts:

The main resistive element is still manganin alloy (manganin has virtually zero thermal coefficient so brings a rocksteady stable resistance), but the shunt has no Kelvin connection whatsoever. The best you can do is to take the measurement traces from the PCB footprint i.e. create your own so-called Kelvin connection:

Image src

Now the thermal EMF phenomenon mentioned in the article applies to this situation. Since you won't be able to take the measurements across the resistive element in any case because of the imperfections, you can expect a thermal EMF i.e. voltage difference which is caused by possibly unequally heated/cooled terminals.

For shunts with true Kelvin connections there's nothing to worry about.

• Suppose the "high current" is a dissipating source off to the left, causing a thermal gradient in the PCB (higher temp to left, lower temp to the right). It'd seem to me that thermal EMF's influence your Kelvin connections unequally. Jun 27, 2023 at 16:04
• @Rohat, Yes, for the shunt you proposed, I understand. Please consider this one as an example: vishay.com/docs/30135/wsms5515.pdf. Although the Kelvin connection seems to be directly on the resistive element, they still provide a thermal EMF figure. I don't think anyone would be interested in using the other pin further down the busbar. Would you be worried about EMF here? Jun 27, 2023 at 16:17

There will be dissimilar metals in most shunts because the copper connects directly to the resistance material (which will be a low-temperature-coefficient alloy such as Manganin or perhaps Constantan). Copper-Constantan is a standard thermocouple (type T) and has competitive output with other commercial thermocouple types (40 or 50 uV per degree C, if memory serves). Manganin is a higher performance material (lower tempco). There is one junction in series with each lead so you can probably assume an error of roughly 40-50uV per degree C of temperature difference. It can be better specified if you know the exact resistance material alloy and look up the behavior vs. copper.

It would be possible to (mostly) eliminate thermocouple EMFs by attaching wires that were made of the same resistance wire material and making the junctions to copper close together and far away from any source of heat (so they are at the same temperature), but that is seldom done.

Failing that, the best approach is likely to maintain symmetry (for example the heavy current-carrying wires are long and of the same gauge and there are no asymmetries that would cause a temperature difference (eg. mounting is not such that the shunt is hanging over the edge of a panel etc.) And, of course, avoid any external sources of heat as much as possible, particularly if they would affect one end of the shunt more than the other.

Keep in mind that the EMF errors will result in an offset approximately proportional to the temperature differential, whereas the self-heating (and, likely, the temperature differential due to the self-heating) is proportional to the current squared. Depending on what you are measuring (for example, a low current between pulses of much higher current) the effect may be more or less objectionable.