Understanding different circuit grounds. How should I connect a shield can??

[Mighty Burger]

Hi all,

I'm designing a PCB and had a question regarding grounding.
Here are a couple pretty pictures of both sides. It's only a two-layer board.



If you're curious what this is for, this is one of two PCB's I plan on using in a Nixie Tube clock. This one has the high voltage power supply and the microcontroller.

Right now I went with something simple - just one giant GND net. Made a big copper pour on the bottom layer and hooked all the circuit GND's up to it.

I asked a coworker to take a peek. He has lots of experience as a professional EE. He recommended that I change how I connect the shield can to the ground plane. Rather than just stitching it right to the ground plane with a bunch of vias like I am now, he said I should only connect it to the ground plane at one point through a 1M resistor, potentially in parallel with a capacitor. (Not sure what value cap.)

I wonder if you guys agree, or would know why it would be better to connect it that way, rather than connecting it directly to the ground plane? Maybe it's so the shield can doesn't become a return path for signals..? Having separate circuit grounds has always been a bit of a puzzler for me.

I'd assume the same wisdom would apply to those mounting holes (i.e. they should all be connected together, and connected to the circuit GND through a 1M resistor). Those mounting holes are indeed connected to the circuit GND right now, it's just that KiCad thinks they need thermal reliefs for some reason.


For context - the shield can here covers the 12V to 170V converter. The only purpose of the can is to stop a bunch of spurious signals emitting from it and interfering with other devices and other parts of the circuit. When I experimented on a breadboard, it was putting out some pretty sick interference: https://www.eevblog.com/forum/beginners/help-designing-emi-filter-for-12v-170v-nixie-power-supply/ I'm not concerned very much about the converter's susceptibility to interference.


Thank you!

[bob91343]

I would suggest connecting the shield directly to a single ground point.  This prevents it from becoming an antenna, encloses the sensitive areas, but still allows no current flow through it.  Once you have current flow through it by grouding it in more than one place, it becomes a path for ground loops, a bad thing.

[ataradov]

Here is one of the most comprehensive videos on this topic. It is long, but worth the watch, IMO.



There is basically no simple answer. I would keep the stitching, but isolate the ground under the can from the rest of the circuit. That voltage converter should have well defined entry and exit lines. Then you can deal with EMC if there are any issues.

[David Hess]

I doubt you need such shielding unless you need the whole thing to be EMI quiet, but in the general case for shielding RF, the seam between the can and ground should be continuous with a minimum or no gaps; otherwise a slot antenna is formed.

In your case do whatever is easiest; the RF from the boost converter will not care how the shield is connected, but do connect it.

A shield would only be connected at one point in a different type of application, like for blocking leakage or electrostatic coupling.  Sometimes a shield is connected to a single point through a parallel RC network to provide limited RF shielding, while preventing the shield from becoming a shorting hazard.

[Mighty Burger]

Thanks for the advice guys. Although it's always a little scary to get conflicting information from knowledgeable people.

From what it sounds like, different applications call for different ways of dealing with grounding a shield. With my application, just trying to keep the voltage converter from spewing out EMI, it sounds like it would probably work fine either way.

I would probably think it best to keep the stitching, connect at a single point, and maintain the GND plane over the voltage converter's inputs and outputs. However it seems like this would be problematic with a 2-layer PCB and the shield can I'm using, which has no cutouts for traces to pass through. This is why I routed the traces on the bottom side of the PCB when it crosses over the shield can. Otherwise the shield might cut through the soldermask and short with either the power trace or the GND net distinct from that of the shield.

I'd like to get rid of the possibility of ground loops by connecting it at a single point but I'm not sure what the best way to do that is. Really not sure where to go from here.

[Mighty Burger]

On a related note, I realized that the mounting holes have a potential to create a ground loop. So I put them on their own net and made a kind of ring around the board to connect all of them, but it doesn't go all the way around to avoid creating a loop antenna. I'll connect it to the main ground plane via a 1M or 100K resistor. Probably overkill for this project especially since I'm not really dealing with analog signals but it's probably good practice I think?


Although.. now that I think about it.. I could probably just not even connect those mounting holes to GND whatsoever.. that would probably be simpler.

Haven't updated the shield yet simply because I'm not sure what to do there, see my above post.

[ataradov]

Ultimately, it does not matter that much what you do. Without simulation it is very hard to guess how things will behave in real life. You can do your best, and if first iteration causes issues, then you treat it as a prototype and fix them on the second iteration once you had a chance to do the measurements.

I would isolate mounting holes in case the screws are exposed on the outside of the enclosure and there is a safety risk. There is no need to connect them to each other or even make them plated. But again, this is such an esoteric thing that it does not matter what you do.

[Mighty Burger]

Ultimately, it does not matter that much what you do. Without simulation it is very hard to guess how things will behave in real life. You can do your best, and if first iteration causes issues, then you treat it as a prototype and fix them on the second iteration once you had a chance to do the measurements.

I would isolate mounting holes in case the screws are exposed on the outside of the enclosure and there is a safety risk. There is no need to connect them to each other or even make them plated. But again, this is such an esoteric thing that it does not matter what you do.

Thank you. I would have gone with your suggestion in the previous post, it's just that having a 2-layer PCB and a shield without slots for traces to come in makes the implementation difficult. I hope you don't take that as me disregarding your advice.
I'll certainly have to see how it performs once I assemble it. It's no wonder some companies are willing to spend so much money on simulation software packages for this kind of stuff. I went into this thinking there'd be one easy, concrete answer but it seems like I've opened up a whole can of worms here. Though I think my current application is not critical and the solution I go with is likely going to be good enough.

Regarding the mounting holes, they are indeed exposed to the outside world. My original thinking is that it would be best to connect it to GND to avoid ever having 170V on the outer metal bits, in which case there would be a 170V potential difference between the bottom plate and the shell of the barrel socket, ouch!! Would be better to have it short to GND and have the fuse blow than the user shocked. (Which means my 1M resistor thing nullifies that safety aspect, oops.) Maybe just to keep things simple it is best to have the mounting holes non-plated and not connected to anything like you said. Not sure. Still going back and forth

Sorry if I'm rambling a lot. I'm learning