0
\$\begingroup\$

I've run into a connection between two PCBs, and it's not giving me a warm fuzzy feeling. The master is a carrier board, and the second board is a standalone sensor that communicates over SPI. This is more of a thought experiment for me to form a better conceptual understanding of EMI issues. I understand there is a lot more at play here, signal rise time, frequency of the SPI communication, proper line impedance and termination, but I want to focus on cable setup.

Original Image

In my mind, this is a problem because power and ground are no longer tightly coupled, creating a large loop area that is ripe for high-frequency EMI noise. similarly, the SPI bus forms a large loop area and is also susceptible to EMI noise PWR coupling

replacing the power for a ground line poses the same issue as above, with the added problem of a large "ground loop" Ground loop

Best case Without changing the design too much, I would imagine the best option would be to remove power or ground from the SPI cable. The issue of a large loop area still stands, but the noise is localized to the SPI instead of the entire system.

Ideally, I would think bring the power across with the communication lines and using a properly shielded cable would be best.

\$\endgroup\$
0
2
\$\begingroup\$

There MUST be at least one ground in every connector. It is not a bad idea to put ground on either side of the clock. If it is not ribbon cable you can even twist CLK and GND together if it is not too much trouble. For ribbon cable twisting is not needed.

The reason for all this is to provide a relatively low impedance and consistent impedance from CLK to GND all along the cable length. If you are going to have a problem with too-high radiated emissions, it will be due to CLK above all else. If your CLK signal connector does not have a GND wire in it, you will have a larger loop for harmonics of CLK, and greater coupling into adjacent signals (MISO, MOSI, CS).

Likewise, when it comes to noise coupling in, having a ground wire adjacent to signal minimizes the ability of the noise to couple in (similar to having a differential pair).

In extreme cases, you can run one ground for each signal wire (one for CLK, one for MOSI, one for MISO). This improves signal integrity and reduces EMI. However it is kind of extreme. And if you are going to allocate that many wires it may actually make more sense to use LVDS signalling (if timing allows). Any ground wires in the signal connector do not need to be sized for DC load. They are just signal wires.

\$\endgroup\$
1
\$\begingroup\$

A lot depends upon the distance between the master and the sensor. But given what you've shown, I would go with option 3 but with a shield around your data cable, like this:

enter image description here

The shield gets grounded to the chassis/enclosure on both ends. The GND in the data cable helps to minimize loop area for signal currents.

You don't need to shield the power cables, though many people do. But you should twist the PWR and GND in that cable.

\$\endgroup\$
4
  • 1
    \$\begingroup\$ There would be no shield in this situation, it would be header cables. \$\endgroup\$
    – Lpaulson
    Dec 13 '20 at 0:52
  • 1
    \$\begingroup\$ I would swap locations between GND and CLK so that CLK is not immediately adjacent to MOSI. It is best to keep CLK away from other signals in a cable if possible. You want CLK to couple to GND and not to other signals. \$\endgroup\$
    – mkeith
    Dec 13 '20 at 1:57
  • 2
    \$\begingroup\$ @mkeith - Valid point. I was looking at OP's diagram more like a schematic than a signal-order-in-the-cable. In fact if you really want to go the extreme, use a ribbon cable for the signals with an alternating signal-gnd configuration. \$\endgroup\$
    – SteveSh
    Dec 13 '20 at 11:53
  • \$\begingroup\$ My comment was for the OP mainly, I guess. I agree with you, @SteveSh. \$\endgroup\$
    – mkeith
    Dec 14 '20 at 21:40
1
\$\begingroup\$

I actually don't think the distant PWR line is that big a deal because local decoupling caps can handle that the high frequency currents so high frequency currents don't flow in loops over the distant PWR wire, or either PWR wire. If you have high frequency currents flowing through the bottom PWR wire that will give you issues regardless.

It's the high frequency currents of the SPI lines that have to travel the big loop that don't get the benefit of decoupling caps. To that end, I believe having a two GND wires is the more conservative option.

Ground loops are only a problem if they cause a problem, if that makes any sense. If they did cause a problem, all things being equal, the GND wire I would actually remove would be the one dedicated to the separate power connector and keep the GND line next to the SPI wires.

\$\endgroup\$
1
  • \$\begingroup\$ That seems like a pretty reasonable assessment \$\endgroup\$
    – Lpaulson
    Dec 13 '20 at 1:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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