|
Practical Papers, Articles and Application Notes
|
|
Comments on EMI Measurements and Modeling I read the article by Colin and Bronwyn Brench in the Practical Papers, Articles and Application Notes in the Summer 2001 issue of the EMC Newsletter. It rekindled some thoughts I have had about the issue of modeling versus experimentation. I realize that Colin and Bronwyn were addressing modeling in the context of modern Computational Electromagnetics Modeling (CEM) by numerical solution of Maxwell’s equations as represented by Method of Moments (MOM), Finite Element Method (FEM), Finite-Difference Time-Domain (FDTD), and the like but I think the definition and concept of modeling should be broadened. The other part of their article that struck a chord with me is that they bring up the issue of what seems to be a longstanding divide between “modelers” and “experimentalists”. This creates an artificial division between practitioners of EMC that is unfortunate and results in a form of “distrust” between the two camps. I would like to address both issues in order to
bring us closer together.
The thing that distinguishes the EMC discipline
from other traditional Electrical Engineering disciplines such as
antenna designers, microwave circuit designers, etc. is that the
nonideal aspects of a system cannot be avoided or designed out of
the system. For example, parasitic effects are one of the dominant
effects that determine how a practical system behaves from an EMC
standpoint. Granted microwave circuit designers have to consider
these parasitic effects but, for the most part, these are known
beforehand and can be incorporated into their simulations which
give highly accurate correlation with measured results. But for
the EMC practitioner, the nonideal aspects that must be confronted
are not something we can design around but are a “given”
that we have little control over. Assymetries in the physical system
tend to cause common-mode currents. In EMC these common-mode currents
are often the critical aspect in determining the radiated emissions
from a product. In the operation of a microwave circuit, these common-mode
currents are present but their effect is dominated by the desired
or differential-mode currents in the functional performance of the
device. Common-mode currents are extremely difficult to predict,
whereas differential-mode currents are predictable with considerably
more accuracy and repeatability. These nonideal aspects make modeling
of EMC systems much more difficult and imprecise, and it’s
a fact that we must expect and live with. It’s also an aspect
of EMC that initially intrigued me and continues to stimulate my
interest in the discipline. I think the concept of “modeling” and
“experimentation” are a matter of degree. I firmly believe
that all of us in EMC should consider ourselves as being both modelers
and experimentalists. For example, a board designer whose responsibility
is the +5V power distribution does a lot of Ohm’s law “modeling”.
Another example is sizing a bypass capacitor to divert a 100MHz
noise signal from a cable where it would have radiated and caused
radiated emissions problems. In determining the proper value of
that bypass capacitor we shouldn’t use a value of capacitor
that “seems right” but should calculate a value of capacitance
that will give, say, a 1 ohm impedance at 100MHz. This is modeling
although not as detailed as what Colin and Bronwyn were discussing.
I always tell my students that you must “do the numbers”.
On the other hand, we all at one time or another perform experiments.
These also range in degree. For example, final testing of a product
for compliance requires a great deal of skill if we expect to get
valid data. My good friend, the late Don Bush, was the most gifted
experimentalist I know. He could make measurement gear “talk”.
Being a ham radio enthusiast, he also was very skilled in constructing
test jigs and making measurements that gave reliable results, i.e.,
what he was measuring was indeed the desired item rather than some
other variable that was corrupted by test probes, etc. In order
to understand the basic mechanisms responsible for radiated and
conducted emissions, he and I constructed many simple experiments.
These experiments were not made on complete systems such as a laser
printer but were simple items such as a pair of lands on a printed
circuit board driven by a canned DIP oscillator. A great deal of
what I understand about these basic phenomena came from what we
learned from those simple experiments. But he and I did not stop
there, we always constructed mathematical models (generally only
simple models were needed because of the simple nature of the experiment)
so that we could ascertain whether the “predictions” of
the model matched the experimental results. We certainly didn’t
expect 100% accuracy, but 3dB error was sufficient to indicate that
our understanding was in the “ballpark”. Don used to say
that “Anyone can construct a mathematical model and generate
data. But if the predictions of your mathematical model do not match
experimental data either your model is worthless or your measurements
are not done properly.” Once you can get both to agree within
a reasonable degree you can study the mathematical model you developed
and be assured that it will tell you what is really going on. I
grant that this kind of experimentation/modeling is not of the same
degree as comparing measured data from an entire system to a computer
simulation of the entire system (which is not feasible today and
probably won’t be in the future). But it is a combination of
experimentation and modeling to understand what factors influence
the basic phenomenon. I strongly believe that all of us involved
in EMC should combine modeling and experimentation to some degree,
and so we should not divide into two camps. We should respect the
skills and abilities of modelers and experimentalists even though
we are more proficient in one aspect and not so proficient in the
other. It’s a matter of degree.
And finally, there are simple problems and there
are difficult problems. We should all be aware of this difference.
For example, in today’s digital systems having GHz clock rates
and picosecond rise/fall times, factors such as interconnect lands
that were inconsequential in systems twenty years ago are critical
today. The modeling of these factors becomes more complicated and
there is nothing we can do about it. Kirchhoff’s laws become
invalid for structures that are electrically large (i.e., a significant
portion of a wavelength), and numerical solution of Maxwell’s
equations (or an appropriate approximation of them such as the solution
of the transmission line equations) is the only way to generate
data that matches the physical world. We have no choice. For example,
a pair of lands on a FR4 printed circuit board (PCB) that are 10cm
in length (about 4 inches) will have a propagation delay of about
0.6ns or 600ps. Twenty years ago with clock speeds in the tens of
MHz these interconnect delays were inconsequential; the propagation
delay through the gates was the only concern. In today’s high-speed
designs this delay caused by the interconnect lands is becoming
extremely important. For simple problems where the structure is
electrically small, we do not need to utilize the sophisticated
full wave models; simple models will do. The key thing here is to
be aware of when a model of simple sophistication and detail is
adequate and to use it. But when a simple model is invalid, recognize
that a sophisticated model must be used and use it (or get a person
skilled in its use such as Colin to use it properly for you).
This cooperation and trust between modelers
and experimentalists is increasingly needed in today’s digital
system designs and more so in the future. We should respect each
other’s skills. In our earlier association, Don was the more
skilled at experimentation and I tended to do the modeling. But
as we worked together, I learned from him and he learned from me.
We increased our proficiency in each other’s primary area of
expertise. We should respect each other’s primary skills and
make use of them. We’re all in this together and will increasingly
need each other’s skills in the future. We cannot continue
to progress in successful EMC design without this cooperation and
respect for each other’s primary skills. Clayton R. Paul The summer issue of the EMC newsletter started,
as we may hope, a series on practical papers. The first discussion
had an interesting subject, the comparison of measurements results
and calculations; the latter often called a numerical model (NM).
To the points brought forward by the authors, one could only add
a few.
All numerical modeling is based on a theory. Maxwell’s
equations have been formulated to describe experiments in electricity
and magnetism carried out earlier in the 18th and 19th century.
The validity of the theory was shortly after its appearance beautifully
demonstrated by Herz’s radiation experiments. Since that time,
the electromagnetic theory is generally accepted as a highly accurate
description of the physical EM phenomena in the macroscopic world.
The equations are simple to write down; their full solution is a
formidable task except for a free space or other simple boundary
conditions.
EM models build on an approximation of Maxwell’s
theory. The computing power increases steadily, as does the complexity
of systems one wishes to analyze: chips, PCB’s or large installations.
Ultimately, the model that fits within limits of time and budget
and is accurate enough for the goals set, is the one that will be
preferred mostly. But the limits of validity for the approximations
should be carefully and constantly regarded. For instance, at power
frequencies a grounding electrode in the soil can be described by
a current distribution determined by the soil resistivity. As the
system becomes larger or the frequency increases, inductance cannot
be neglected anymore. A similar situation occurs in the steadily
faster micro-processors. Also, induction phenomena have to be taken
into account added to the usual RC transmissions lines.
A good model is able to predict, perhaps a posteriori,
the outcome of an experiment; practical circumstances should reflect
the model boundary conditions. But for many Ôreal world’
problems, the circumstances of the experiment may be less known.
For instance, in safety grounding or lightning protection, the amount
of known buried conductors depends on the good bookkeeping of an
installation throughout its lifetime. If other parties bury their
metal contraptions nearby, this will influence the envisaged safety,
either in a positive or in a negative way. Modifications of an existing
installation, ageing and corrosion affect even the known conductors.
In such a case, NM may provide us with the sensitivity of the safety
with respect to uncertainty in or degradation of the installation.
But in the case of lightning protection of a critical installation,
e.g. a nuclear power plant, a real measurement on the actual installation
would be useful to convince ourselves and others, until sufficient
experience has been gained to trust the next installation without
further tests. Such a sensitivity analysis can be quite sobering.
The EMI coupling of PCBs often shows sharply peaked resonances.
Minor changes in the NM parameters shift the calculated resonance
frequencies. If the PCB circuits operate near a resonance, large
differences may occur between NM and the measured EMI, and perhaps
also between various measured samples.
Many EMC solutions for acute industrial problems
have to be given without any time for modeling, not to mention a
full analysis. This Ôfire brigade’ type of activity is
quite common for EMC consultants. Apparently, a fast mental picture
is possible for the behavior of currents that guides one effectively
to EMC solutions. Modeling, even afterwards, is very useful to extend
this mental database. It is sometimes advocated to compare the results
of various NM codes, to find out whichever suits better. Such enterprise
is certainly useful to enhance the quality of the codes. However,
a carefully designed experiment requires as much diligence as modeling,
but a different experience. So the effort of the authors to combine
NM and measurements should really be appreciated. After all, the
modeling is carried out to make a real piece of electronics work.
Lex van Deursen
|
|
If you would
like to contact the IEEE Webmaster |
 |