EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: TimNJ on October 05, 2018, 05:20:19 pm
-
Hi all,
I'm working on a (fairly) high density power supply design. The initial prototype was a 2-layer PCB, and it seems that radiated EMI might be tough to meet. We have been doing lots of investigative work into the sources of EMI and have gotten down the levels considerably from where they were. But again, still borderline.
So, I'm thinking about changing to a 4-layer board, or at least trying it out in the lab. In the 2-layer layout, we have a bit of a pseudo-star ground scheme. The idea is to keep high di/dt currents from mixing in with control circuitry return currents, to keep current loops as small as possible, and to reduce injecting noise onto critical nodes.
My question is: If we put a ground plane under the existing layout and tie everything together, we lose some of the isolated return paths we accomplished on 2 layers. Now the return path inductance will be much lower, but I still feel uneasy about connecting everything to a plane. (This is the first layout of my EE career, by the way!)
Is there any merit into not connecting high di/dt (power train) components to the internal ground planes? That is, connect control circuitry returns together on the ground plane, and to the bulk cap, but let the high di/dt currents flow through the top and bottom layer conductors (as they did on 2 layers).
Thanks in advance.
-
There are a lot of _single_ layer switch mode supplies out there.
I suggest you find an EMC lab and ask them to do a ‘debug’ radiated emissions test on your 2 layer prototype.
Also, EMC is sensitive to little things, so for serious comments we would need to see layout and ideally schematic.
-
There are a lot of _single_ layer switch mode supplies out there.
I suggest you find an EMC lab and ask them to do a ‘debug’ radiated emissions test on your 2 layer prototype.
Also, EMC is sensitive to little things, so for serious comments we would need to see layout and ideally schematic.
Thank you. Yes, I know, I've worked on more single sided designs than anything else. If you think about "normal" power densities for "typical" switchmode power supplies in 2018, this is about double or three times that. Like escaping a BGA, when you want to put a lot of parts very close to each other, you need multiple layers. (Nothing you don't already know.) But it's not just a matter of making the parts fit. As a side effect, high EMI parts/sections might find themselves closer to the input filter, board edges, etc. than you'd ideally want.
We (the company I work for) takes engineering samples to EMC labs all the time, but I don't think a debug run is the typical protocol and I don't think I have the authority to order that myself. We just get the data back, and try to find the EMI aggressors with an H-field probe, spectrum analyzer, etc.
Sorry for being vague. I'd like to disclose more details, as I'm generally quite open, but don't want to get into hot water by publicly posting design files.
I suppose I'm just asking: Could there be any logic in using the ground plane less as a deliberate return path, but more as a shield? Similar to putting a metal plate under a PCB, or enclosing a project in a metal box...
(And I suppose I already know the answer: "Maybe. It depends.")
Thanks.
-
I mean, that's exactly what it is, you're talking equivalent properties there. 8)
Tim
-
I mean, that's exactly what it is, you're talking equivalent properties there. 8)
Tim
Assuming I'm reading your response right, then yeah I guess so, that's what the ground plane is supposed to do.
In which case, I'll probably keep high current returns out of the inner ground planes...not that it's impossible to do correctly, but I'm not too sure about my own ability to figure that out.
Thanks.
-
In RF work the ground plane is used as a shield all the time, and particularly noisy or sensitive traces can be sandwiched between two ground planes and some via stitching for what amounts to effectively in-board coax. If you have a specific traces producing most of the noise, it could be a viable option.
That said, using inner traces for power also means they need to be a lot wider, since the usually use thinner copper (half ounce is common) and since they're embedded in the board, they don't dissipate as much heat as normal outside layers, so it could be tricky to encase primary traces in your design. The plane itself would act as a shield (albeit not an incredible one if you have nothing else enclosing the board), and it would add a bit of capacitance to everything else (thinner layers also means higher capacitance between layers than a standard 2 layer) which could take the edge off of some of your high frequency stuff.
What gets radiated is also dependent on trace geometry, since those are the parts can acting like antennae (though coils and stuff can do that as well). Avoiding long straight traces is a good general idea, especially when you know they're a hard-edge switching or high current trace. If the issue is nearfield stuff, maybe look into using shielded magnetics?
And if it's tough to tell where it's coming from, knowing what frequencies are the issues and what parts operate on that frequency or lower harmonics of it could help point to the part of the design causing the issues.
-
In RF work the ground plane is used as a shield all the time, and particularly noisy or sensitive traces can be sandwiched between two ground planes and some via stitching for what amounts to effectively in-board coax. If you have a specific traces producing most of the noise, it could be a viable option.
That said, using inner traces for power also means they need to be a lot wider, since the usually use thinner copper (half ounce is common) and since they're embedded in the board, they don't dissipate as much heat as normal outside layers, so it could be tricky to encase primary traces in your design. The plane itself would act as a shield (albeit not an incredible one if you have nothing else enclosing the board), and it would add a bit of capacitance to everything else (thinner layers also means higher capacitance between layers than a standard 2 layer) which could take the edge off of some of your high frequency stuff.
What gets radiated is also dependent on trace geometry, since those are the parts can acting like antennae (though coils and stuff can do that as well). Avoiding long straight traces is a good general idea, especially when you know they're a hard-edge switching or high current trace. If the issue is nearfield stuff, maybe look into using shielded magnetics?
And if it's tough to tell where it's coming from, knowing what frequencies are the issues and what parts operate on that frequency or lower harmonics of it could help point to the part of the design causing the issues.
Thank you. All good tips to remember. The trace geometry thing is something I'm not so knowledgeable in. Resonant stubs and that kind of stuff? I suppose all boards have lots of little antennae just by nature of existing. The magnetics are shielded pretty well, about as good as they can get without crazy self-heating due to eddy currents.
Yeah, pinpointing the main offender(s) has been challenging. I built a magnetic field probe from this app note: www.analog.com/media/en/technical-documentation/application-notes/AN118fb.pdf (http://www.analog.com/media/en/technical-documentation/application-notes/AN118fb.pdf)
It has been useful, for sure, but still a bit finicky.
On that note, are there systems which do automated EMI characterization? That would be an interesting product. Basically a CNC H-field/E-field probe which records the FFT with respect to X,Y position and then plots the intensity in 3D. Would be a good visual. Maybe signal/power integrity simulation largely takes care of this these days.
-
If this is your first board, I would highly suggest to look how other people/companies are doing it.
Try to figure out and understand why they do it and were your problems are.
For cheap power supplies they mostly just put a faraday cage around it.
What a good example is all depends on your constraints.
4-layer board is nicer (in fact easier to route imo), but also more expensive.