On a related subject, a recent project I've been working on, started out with half a dozen grounds
on one board alone! Worse still, it was two layer routed, no planes/pours; and worst of all, it was just a switching power supply, no reason at all to split grounds! Needless to say, it was an EMI nightmare beyond a few MHz. I talked them down to a 4-layer ground plane design with only three ground domains. Connections between domains are made either with common mode chokes, when the connections are between galvanically linked domains; or with isolators and optional 'Y' capacitors, for galvanically-isolated domains. The results will be coming in soon, hopefully with significant reduction in emissions.
I told them, that one board had more grounds than any project I had worked on up to that point,
including multi-board designs linked by cables, counting each board's local ground separately. That fact didn't seem to impress them very much though...
If you don't know what you're doing, tie everything together, as often as possible, as widely as possible -- on a ground plane.
I'd expand that to "if you don't know or if you think you know what you're doing". Otherwise nothing to add.
There's a lot of use cases that may require isolation here. Most of these cases require "real" isolation, like T3sl4co1l said. Only a few may work without, but most approaches I've seen to separate Earth and circuit GND without real isolation failed in some way.
Yes, quite; to omit the Dunning-Kreuger effect would be negligent.
There is something very dangerous, I think, about discussing alternate grounding designs at all. It's an important design tool, on the rare occasion that it turns out to be needed; but the number of articles discussing it, is absolutely blown out of proportion to the number of applications that actually need it.
So, reading on these topics, can give the novice the impression that these things are quirky and cool and useful, far more often than they are -- without understanding the huge scope required to make such a judgement. Most especially, those that "know just enough to be dangerous", or that think they know everything that needs to be considered.
Life is weird, on the far side of the D-K curve. I can't honestly give an answer with certitude, because I know not only that
everything is uncertain, but that my knowledge is itself uncertain, to varying degrees. Mind, the uncertainty associated with core, well-practiced topics (like circuit theory, or E&M), is very small indeed. But it would be foolish of me to state otherwise, especially when the information feeding those tools is itself uncertain. And equally so, topics are around the edge of my knowledge are much less certain, and consciously so. (Mind, there is the know-it-all's disclaimer: I can infer or extrapolate on those topics, and sound awfully scientific and certain about it in the process, but really I'm just grasping at straws, constructing hypotheses about it -- which may well be obviously wrong to someone with a deeper knowledge of those subjects!)
The other grounding topics that come to mind, for those curious -- slotted grounds, trace routing (path of least impedance), star grounding*, and similarly for supplies: local (usually analog) supplies, gaps between planes, etc. Shielding of cables (protip: always ground both ends, to do otherwise is approximately equivalent to not shielding at all). Ground loop (especially in audio and metrology).
*There are two kinds of star grounding, in fact. One is the somewhat naive kind, which is done to minimize errors in precision metrology instrumentation -- it's only applicable to low frequencies, where the inductance of those return paths can be ignored. It falls over, in many magnificent ways, at high frequencies. Which may be mere kHz for precision AC, or low MHz for outright instability in the circuit.
The other kind is nuanced by the consideration of fields -- namely, following the path of least impedance back from each subcircuit, so that signals between subcircuits pick up as little error as possible due to induced voltage drops on each "spoke". Here, the topology resembles a central hub with spokes. All power, signals and ground are routed from each subcircuit back to the hub. I suppose "hub" is a better fit than "star". The important part is, signals get shielded by their grounds, so they do not pick up induced noise between subcircuits, whereas in a naive star-ground design, signals may end up routed between subcircuits by straight line connections, where they pick up all the awful induced voltage between spokes. From another perspective: it violates each ground domain by routing a trace away from local ground; this is why you must never route traces over a ground split/slot.
...
I hope I'm not lecturing too much and that it comes off as self-inflating. I mean, I do like talking about high level topics, but I want others to learn these things. The world needs more, and better, engineers! I suppose a lot of "thinking about thinking" can seem redundant, or confusing? Critical thinking is key, though. Once you can critique anything, and learn about anything, the world is yours.
Tim
Home
Help
Search
Login
Register
