[SI-LIST] : EMI idea or heresy

Mellitz, Richard (mellitz@xgate.columbiasc.ncr.com)
Fri, 21 Nov 1997 14:40:27 -0500

I have an idea. Others may have thought about this before, I'm sure,
and can tell me why it can't work.

... Behavioral EM Models.

Or better yet, what can work and what working means.

The stage for EMI is somewhat different than for signal integrity. I'd
like my crosstalk predictions to be of no more that a few percent off.
However, it I would be useful to be able to predict EMI to 3 or 6 dB.
i.e. a factor or 2 or 4. So the criteria for "working" is somewhat
different. I feel first attempts at behavioral EM models will be
inaccurate because of errors introduced by juxtaposition but still may
be useful. The key word here is "useful" not perfection!

There is a common problem the system designers are often faced with.
Many times the box or enclosure is designed before boards are. I would
like take a top down approach. Lets say each of the components in the
box have behavioral EM models. Lets say we could figure out hierarchical
algorithms associated with placement. Then put all this is put together
with the mechanical design of our enclosure. We could have behavioral
models of cables and seals. In cases where were simulated models don't
exist maybe we could use measured models. If all this existed we could
affect board design and enclosure design when the decisions were needed,
at the concept stage. It could also offer fodder for the "what-if"
trade offs that occur during a development process.

I know there are a lot of tools that help once you've got your boards
all designed. There are tools that model connectors. There are tools
that model ICs. There are tools that model IC packages. There are tools
that allow you to determined fields from enclosures. It would be nice
if we could use these tools to produce standard behavioral models. This
all worked for IBIS which was seeded with a basic concept that V/I can
be used to predict circuit behavior in digital systems. What simple
notion exists for EM?

My first naive thoughts are focused around:
1) X-Y near field a vector map for relevant frequencies
2) Ports that have a certain transfer impedance from the near field
data.

Do any of you EM experts out there have any ideas? Perhaps this could
be the start of an unified EM standard.

Rich Mellitz, NCR
803-939-6240