Re: limit of method of moments

cashioli@ansoft.com
Wed, 29 May 1996 08:52:52 -0500

I agree with you that your friend is wrong in stating the entire structure
has to
be less 1/10 wavelength to use MoM. Instead, the structure has to be
discretized
in such a way that each element is very small compared to wavelength.
Typically,
1/10 wavelength is a good rule of thumb. One may go as high as 1/20
wavelength
to achieve higher accuracy. This is because many MoM formulation uses so
called
locally supported basis function to represent electric currents. Those
basis
are typically low order polynomials. Hence, it only can represent small
change
in currents on a local area such as a triangle or rectangle.

The issue of limit of MoM is a hard one in general. Since you are
referring to
HP's momentum, I can list a few limiting factors:

(1). Accuracy in computing Green's function. Momentum uses dyatic Green's
function
for multilayer dielectrics. To speed up the computation, HP uses
precomputed
Green's function and does interpolation when the Green's function is
needed.

(2). Mesh densities. In order to achieve high accuracy, local refined mesh
is
necessary. I am not sure Momentum has this capability.

(3). Deembeding techniques. In order to extract circuite parameters such
as
scattering parameters, it is necessary to extend the ports to certain
length
and perform certain operations. This has a strong influence on the
accuracy
of circuit parameters.

(4). Low frequency limit. Momentum uses mixed potential integral equation
technique. It generally breaks down at very low frequency unless special
care is
taken.

If you would like a practical example on the limit of accuracy for code
such as HP
Momentum, I would refer you to an artical published in European Microwave
Journal earlier this year which compared the measured results with the
numercial
soltions obtained by various simulators. For matched microstrip line, |S11|
= 0.0
analytically. Numerically, it is in the range of 0.01 to 0.001.

-Xingchao Yuan
Cadence Design Systems