Re: gEDA-user: 3 questions on a 4 layer board

[Dan McMahill <dan@xxxxxxxxxxxx>]

Rick Collins wrote:
> At 04:33 PM 4/29/2008, you wrote:
>> Dan McMahill wrote:
>>> If my traces were carrying ground referenced analog signals I wouldn't
>>> want the power plane being my "ground" plane.
>>>
>>> And yes, I've seen a real life case with some rather frustrating
>>> behavior that the correct stackup totally fixed.  There was a moment of
>>> clarity when a bypass cap way on the other end of the board made a much
>>> larger than expected difference.
>> Would you mind elaborating on this?  I could imagine a scenario where
>> the power plane could be a source of AC noise - especially if components
>> aren't bypassed adequately.  The AC noise could then couple to the
>> signal traces in its quest to complete a low impedance circuit.  Is this
>> what you observed, or am I way off?
>>
>> -Ethan
> 
> 
> I don't want to come off as argumentative, but if you both think 
> about what has been said here, you can see the fallacy.  If the 
> frequency of the noise is such that it can easily couple into a 
> signal trace, then that same noise would be well decoupled between 
> the power and ground planes and so would not be present to any 
> significant degree.  The decoupling would not be through caps as it 
> would be from the very high frequency capacitance of the planes themselves.

The difference is I think you're talking more about signal integrity 
issues around high speed digital and I'm talking about sensitive analog 
stuff.  The key phrase in what you wrote is "any significant degree". 
The particular board I had in mind had a coupling of perhaps -60 to -70 
dB which was too much for my application but would have been totally 
insignificant for a digital circuit.  And in this case, the coupling 
path was capacitance from a signal trace to the power plane and then the 
power plane back to a different signal trace.  This I was able to 
demonstrate conclusively in that case.

> Ground or power noise in circuits is nearly always due to 
> ground/power bounce and only affects the signals in or out of the 
> chip that has the "bounce" problem.  Even if a board has inadequate 
> decoupling of the power plane, that noise is not going to 
> significantly couple into signals.  Of course, because of the 
> ground/power bounce on the chip making the noise, its signals will 
> look like they are coupled to the noise because they are... but not 
> directly from the power plane to the signal traces, through the power 
> pins on the chip.

Again, you have to careful of what type of design.  If you have single 
ended analog circuits, you have to define what the reference is 
("ground") and you'll almost always find that the supply rejection is 
better or worse depending on the exact circuit.  Yes fully differential 
circuits can help, but lots of practical RF circuits tend to be single 
ended.

You just can't count on ground/power noise from one chip not causing 
major headaches for other unrelated circuits.  I helped someone fix a 
problem in an L-band receiver once where it was the 13th harmonic of a 
clock oscillator coupling through the power supply over into his RF 
signal chain.  The solution was easy, local series impedance in the 
supply and shunt bypassing with a well defined current path to the load. 
  Its been enough years now that I don't recall what the coupling 
mechanism was on the receiver end.  I also don't recall the details of 
his power routing (plane vs traces, how many supplies, etc).

About 15 years ago or so, EDN magazine had a special issue all about EMI 
problems.  It was a pretty good read and one of the simple facts that 
they point out has proven very useful in practice.  To have an EMI 
problem, you need a source, a path, and a receiver.  You may not always 
be able to identify all 3, but if you can break at least one, you win. 
In cases where you can identify more than one and improve them, you win 
more.  It sounds simple, but it's the first thing I remind myself of 
when I have an EMI problem.

-Dan







_______________________________________________
geda-user mailing list
geda-user@xxxxxxxxxxxxxx
http://www.seul.org/cgi-bin/mailman/listinfo/geda-user