At 05:20 PM 7/12/2004 +0000, Graham Davies wrote:
>--- In AVR-Chat@yahoogroups.com, David VanHorn <dvanhorn@c...> wrote:
>
>> [in reply to my statement] For best EMI
>> performance ... you need to form a grid
>
>> Well ... I have a lot of success with ...
>> providing one and ONLY one low impedance
>> path back to the source, and not allowing
>> RF current to spread out and take all
>> possible paths.
>
>The problem with this is as follows. If the return path is not close
>to the signal path, you are making a loop antenna. This both radiates
>EMI and picks up EMI from other places.
Read the other rule about minimizing loop areas.
>The way that a ground plane works is to provide a return path for
>every signal that is very close to the signal path.
A perfect ground plane would do that, but it also allows the current to take ALL possible paths in the ground plane. Given that any real board will have an imperfect plane, with non-zero impedance, this causes problems.
> The signal in the
>signal layer trace actually causes the high frequency components of
>the return current to flow right underneath the trace even though the
>ground plane covers the board.
Mostly, but not completely. In a perfect ground plane, it's not bad.
>But if you don't have a ground plane, what is best? First, make the
>antenna as small as possible. This means routing ground close to the
>signal.
In one printer, I have a printhead that uses 32V at 19A pulsed for 300uS every few mS. In this one, I routed printhead power and printhead power return as two 200 mil tracks, top and bottom of the PCB, power directly on top of the return. The only place the PH return contacts "system ground" is at the third of the three caps that provide the energy storage.
Picture, caps as values in uF:
/----+----+----+--------------------------------+---\
PS 330 330 330 5 inches or so 1 HEAD
\----+----+----+--------------------------------+---/
|
System ground
So, the energy mostly flows in a very tight loop between the three caps and the head. 1uF local bypass at the head turned out to be not all that useful.
Note, the caps weren't blobbed togther, they were routed exactly as drawn. This places more of the ripple current burden on the one closest to the printhead, and so that one is specified for high ripple. Even though they are as close as can be, there is a significant difference in ripple current. Verified with current probe.
If I had simply dumped the return current into the plane, this would have caused the return current to run through my logic section, and I guarantee you that board would have had glitchy problems.
> But, with signals running all over the place you can't do
>this unless you have ground all over too, which ends up being a grid.
Only if you let it. I use planed areas for each function, then connect the planed areas at the power supply, in a star topology.
I also handle power that way.
>
>Second, split the signal to go in opposite directions around two
>adjacent antennas. Both radiation and pickup then cancel. This also
>ends up being a grid.
I don't get what you mean here, exactly.