PCB gaps lead to unwanted emissions
At the 2010 IEEE EMC Symposium held last week in Ft. Lauderdale, consultant Kenneth Wyatt used a piece of PCB material to show how gaps in a signal’s return plane can emit unwanted interference. The figure (click to enlarge) shows how he attached two BNC connectors to the board and soldered their shields to the board’s copper plane surface. Using a Dremmel tool, he etched a 5-cm. gap in the plane starting from one edge of the board. A wire connected the center of each BNC to a 50-Ω load at the other end of the board. One wire passed over the center of the gap while the other didn’t.Wyatt then drove the wire passing over the gap with a 10-MHz comb generator whose signal source is capacitively coupled to its output port. The coupling produces narrow pulses. With a magnetic-field (H-field) probe, he measured the emissions along the wire with an oscilloscope’s channel 2 triggered by the comb-generator’s output (channel 1). At the gap, the phase of the field changed relative to the pulses. The experiment proved that gap in ground planes produce unwanted emissions because they alter the path of current returning to its source.
See “Oscillators and emissions” for an experiment on how I/O cables can produce emissions.
KenW commented:
I'd like to address some of the excellent comments and suggestions posted.
Jiri: It surprises me as well that spilt planes or gaps are not generally understood. Present company excepted, of course! I suspect there are many new EMC (not RF) engineers that are new to this. In fact, split planes are one of the top issues I run into as a consultant.
Duncan: You present several good questions. (1) yes, there is a second plane (unconnected) under the plane in question. (2) While probing with the h-field probe, the amplitude comes close to doubling over the gap. Since I have simply taped a wire down to the return plane, there is greater leakage flux around the wire than if I had used a printed trace. There would be a much bigger difference in that case. (3) the lower plane is not connected in any way. That would have possibly affected the demo. (4) Not sure I understand what you're asking - sorry (5) Absolutely!
Wei: Good question. If I understand your question, we have the trace traveling between two planes, with a gap in one of them. In this case, it depends on which plane is connected to the source return. If the return connects to the ungapped plane, I wouldn't think there would be that much of a change. TDRing would probably show a small "impedance bump", though. In the case where the return was connected to the gapped plane, some of the return current would return through the capacitively-coupled floating plane. In the case where both planes were connected to the source return, I suspect the outcome would be very similar to the first case. Remember, the key to PC board EMC and associated common-mode current generation is that return currents will tend to flow in the path of least IMPEDANCE. Just trace out where the return current wants to go and make sure there are no gaps in the way.
Marian: I suspect things start to get interesting when the two planes are connected together at the source and load ends. Without a gap, the second plane has no effect, connected or not. Because of skin effect, all the source and return currents stay on the one side of the board. NOW, if we introduce a gap, there will be some return current flowing on the top-side of the lower plane, due to capacitive coupling between the planes. The amount of current flowing through the lower plane would be dependent on the plane-plane capacitance and the clock frequency, so, I'm not sure you could state that 50% would flow. Assuming a "small" loop (
Marian Stofka commented:
The previous poster poses an interesting theoretical question. Let's assume, that both proposed copper planes serve as current return paths and they are conductively coupled by two conductive through-holes right at ends of the resistor-terminated wires; also at position of BNC connectors.
Again a similar effect can be expected as originally described; but of no more than 50% of magnitude - in the case that the slotted Cu plane is closer to wires and the non-slotted Cu plane is invisible from side of the wires.
Even if the positions of the Cu planes were interchanged; some percentage of the described effect could be expected; although much smaller than 50% of the original amplitude. Wires and plane (planes) form transmission lines (almost perfectly homogeneous for wire 2 and containing a local inhomogeneity in case of wire 1).
It might be a challenge for the author of the paper to prove the above.
Pengyu Wei commented:
I want to know what about the situation that signal line within 2 cooper planes, and make the gap on 1 of the planes.
Duncan Gray commented:
1. Does this have a plane under the split plane?
2. I don't see any measure of the absolute H amplitude. Though I hope it's a dramatic difference too.
3. If this does have a plane under it, then have you tagged the two planes with strategically (not periodically spaced or valued) caps.
3. Could you do some bridge caps across the gap to see how much a cap might reduce the emissions.
4. Could you try the difference between a 2-mil dielectric parallel plane vs a 3 mil dielectric spaced plane.
5.. Could you put a TDR on the BNC to show the Z-bump at the gap and the ring after that gap as the return finds its way home.
This is very old news. I've been dealing with it for a decade. I don't have any simple experiments like this, though, to give me a quantitative basis for sensibly dealing with the necessary split planes in any design.
Thanks, great presentation.
Jiri Polivka commented:
This "new" experiment has been known for a long time among RF engineers. What surprised me is that so many engineers see this as something new and unknown.
There are also filters and couplers utilizing gaps in ground plane, and antennas...
KenW commented:
I wanted to add a couple clarifying points. First, the board design has been duplicated by several others in the past, but this particular design was originally conceptualized by my friend, and fellow consultant, Doug Smith. There were a couple points I was trying to demonstrate; (1) that running the h-field probe along the signal trace with the cut showed a high peak at the return plane gap, and (2) that offsetting the probe so that it passed from one side of the gap to the other shows a phase change in the scope indicating that the return current flows away from the trace on the near side and then towards the trace on the far side from the source. The un-gapped transmission line, of course, shows an even signal level and no phase reversals. This is a great demo to show how structural unbalances in trace layout can cause common-mode currents to form. Any nearby I/O cable to the gap can radiate due to these "antenna" currents coupling to the cable. Thanks for posting this little experiment!
