Vous êtes sur la page 1sur 114

The Method of Moments

V Jithesh, Scientist D
LRDE, Bangalore
Our Life Revolves Around This!!!
with boundary conditions
Evolution of EM - Experimental
Christian Oersted
Benjamin Franklin
Michael Faraday
Henry Cavendish
William Gilbert
Alessandro Volt
Evolution of EM - Theoretical
Carl Friedrich Gauss
Charles Agustin de
Coulomb
Georg Simon Ohm
Andre Marie
Ampere
Jean-Baptiste Biot
Felix Savart
Simeon Poisson
Evolution of EM - Theoretical
JC Maxwell
Oliver Heaviside Heinrich Hertz
Hendrik Lorentz
Hermann Helmholtz Ludwig Boltzmann
William Thomson
(Lord Kelvin)
Evolution of EM - Analytical
George Green
Joseph Fourier Pierre-Simon Laplace
Gustav Mie
John William Strutt
(Lord Rayleigh)
Adrien-Marie
Legendre
Friedrich Wilhem
Bessel
Richard Courand Boris Galerkin Walther Ritz
Arnold
Sommerfeld
Evolution of EM - Computational
L F Greengard
(FMM)
Olgierd Zienkiewicz
(FEM)
Kane S Yee
(FDTD)
JP Berenger
(PML)
Roger F
Harrington (MoM)
Peter P Silvester
(FEM for EM)
Kenneth K Mei
(MoM)
Allen Taflove
(FDTD)
Joseph B Keller
(GTD)
Prabhakar Pathak
(UTD)
RG Kouyoumjian
(UTD)
James Clerk Maxwell
1 Equation
1 Equation
3 Equations
3 Equations
6 Equations
3 Equations
3 Equations
20 Equations!!!
Historical Timeline
1785 Coulombs Law is published
1812 Poissons Law is published
1813 Gauss Divergence Theorem is discovered
1820 H.C. Orsted discovers that an electric current creates a magnetic field
1820 Andre-Marie Amperes work founds electrodynamics; Biot-Savart Law is discovered
1826 Ampres Law is published
1831 Faradays Law is published
1856 James Clerk Maxwell publishes "On Faradays lines of force"
1861 Maxwell publishes "On physical lines of force"
1865 Maxwell publishes "A dynamical theory of the electromagnetic field"
1873 Maxwell publishes Treatise on Electricity and Magnetism
1870 In this decade Lorentz and Helmholtz develop the electromagnetic theory of light
1880 In this decade Heaviside recasts equations into vector form
1888 Heinrich Hertz discovers radio waves
1940 Albert Einstein popularizes the name 'Maxwells Equations'
1966 Kane Yee introduces finite-difference time domain methods to solve Maxwells Equations
MaxwellHeaviside-Hertz Equations
The most elegant equations in the Universe: Maxwell - Heaviside - Hertz
In short, known as The Maxwells Equations
0
v
D
B
B
E M
t
D
H J
t
=
=

= +

*
,
,
source source
D E B H
J J E M M H


= =
= + = +
How to Solve ?
Back-of-the-envelope
Analytic
Semi-analytic
Computational
Discrete (Differential)
Integral
Asymptotic
Hybrid
Parallel
Actual solution for realistic problems is complex and requires simplifying assumptions
and/or numerical approximations
Solving EM Field Problems
Find electromagnetic field and/or source functions such
that they
obey Maxwells equations,
satisfy all boundary conditions,
satisfy all interface and material conditions,
satisfy all excitation conditions.
(In both time and space, or at one frequency in space)
Field solutions are unique
But numerical solution depends on
Physical Modeling Error
Discretization Error
Numerical Modeling Error
Measurement Error
Field Solving Methods
Analytical Methods
Exact explicit solutions (only a few ideal cases)
Semi-Analytical Methods
Explicit solutions requiring final numerical evaluation
Numerical solutions with analytical preprocessing
Approximate analytical models
Approximate analytical solutions for simplified
structures (provides physical insight)
Only practical way to handle very large electrical
structures
Numerical Methods
Differential or integral equations are transformed into
matrix equations by numerical approximations
(sampling) and solved iteratively or by matrix inversion
Closed Form Solutions
In the beginning of 1900s, many closed-form
solutions have been obtained:
Mie (1900): Mie series solution of scattering by a
sphere (Separation of variables)
Rayleigh (1897): Guided-wave solution in a hollow
waveguide (Separation of variables)
Rayleigh: Rayleigh scattering by small particles
Approximate Methods
High Frequency Methods
Geometrical Optics (GO)
Geometrical Theory of Diffraction (GTD)
Uniform Theory of Diffraction (UTD)
Physical Optics (PO)
Physical Theory of Diffraction (PTD)
Shooting and Bouncing Ray (SBR)
Hybrid Methods
Numerical Methods
Age of Numerical Methods: MoM, FDTD, FEM
Yee, 1966; Harrington, 1968; Silvester, 1972; Rao, Wilton
& Glisson, 1982; Raj Mittra, 1980+; Taflove 1980+
Differential Equation Solvers FDTD, FEM
The radiation condition at infinity is emulated by the use of
absorbing boundary conditions (ABC), such as Perfectly Matched
Layer (PML)
Integral Equation Solvers MoM, BEM
Many numerical methods started in the EM community and
later spread to other communities and become popular
(and vice versa)
Computational Electromagnetics
Solutions to Maxwells equations using numerical
approximations is known as the study of Computational
Electromagnetics (CEM)
20 Years back:
Computations: No one believes them, except the person who made them
Measurements: Everyone believes them, except the person who made them
- Prof Ben Munk, Ohio State University
Each method needed a Technological advance
MoM: MLFMM
FDTD: PML
FEM: Vector Elements
Now: Far greater faith in CEM results
Importance of CEM
Key to analysis, design & optimization of RF Systems.
Virtual prototyping.
Key to economical success of a product through
shortening of development time.
Only means for dealing with complex (non-canonical)
electromagnetic structures.
Theoretical models must be validated by experiments.
Theoretical and experimental work are of equal
importance.
Overview of CEM
Central to all CEM techniques is the idea of
discretizing some unknown EM property, for example:
In MoM the surface current is typically used
In FEM, the Electric Field
In FDTD, the Electric and Magnetic Field
Meshing is used to subdivide a large geometry into a
number of non-overlapping sub-regions or elements,
for example:
In two dimensional regions triangles maybe used
In three dimensional geometries a tetrahedral shape
may be used
Overview of CEM
Within each element, a simple functional dependence
(basis functions) is assumed for the spatial variation
of the unknown
CEM is a modeling process and therefore a study in
acceptable approximation and numerical solution
In other words, CEM replaces a real field problem with an
approximate one which causes physical (geometric) and
numerical limitations that one must keep in mind
Overview of CEM
During the creation of the approximate model,
assumptions and simplifications are generally introduced
so limitations on the solution accuracy, for example:
Assuming an infinite ground plane or substrate in an antenna
structure
Simplifying a thin wire by a current filament (dipole impedance)
When analyzing solutions generated from a CEM
techniques keep in mind limitations of the solution
introduced by manufacturing tolerances:
Small changes in dimensions may affect the performance
Frequency dependent (or unknown) material properties
Overview of CEM
Limitations are also introduced by the finite discretization
The mesh must be fine enough so that the basis functions can
adequately represent the electromagnetic fields
Fine mesh required for critical variations such as source
region
Numerical approximations and finite precision will limit
the analysis
Limited computational resources (i.e., mesh size)
Double precision accuracy will not help if the problem is ill
conditioned
Finer mesh is often the only choice
Classification of Methods
1-D Methods:
Fields and Voltage/current vary in one space dimension
(TL problems)
(Touchstone, SuperCompact, SPICE, Multisim, CRIPTE)
2-D Methods:
Fields and Voltage/current vary in two space dimensions
(Cross-section problems, TE
n0
waveguide problems)
(FEM-2D, MEFiSTTo-2D)
21/2-D Methods:
Fields vary in three space dimensions and Voltage /
current vary in two space dimensions
(Planar Multilayer Circuits)
(Sonnet, Momentum, Ensemble)
3-D Methods:
Fields and Voltage/current vary in three space dimensions
(General Propagation, Radiation & Scattering problems)
(FEKO, HFSS, CST, XFDTD, GEMS, GEMACS,
SEMCADX, WIPL-D, EMC Studio, PAM-CFEM FD/FE,
IE3D, Concerto, SuperNEC, Empire, MultiPhysics, efield)
What Have All Methods in Common?
In all methods, the unknown solution is
expressed as a sum of known functions
(expansion functions or basis functions).
The weight (coefficient) of each expansion
function is determined for best fit.
What distinguishes them?
the electromagnetic quantity approximated,
the expansion functions used,
the strategy employed for determining the
coefficients of the expansion functions
the numerical solution method.
Why Model in Frequency Domain?
Most microwave engineers are more familiar
with FD concepts than with TD concepts
Frequency domain simulations are steady-state
Complex notation is elegant and efficient
Specifications are traditionally formulated in the
FD (S-Parameters, loss tangent, dispersion)
Time domain information can be obtained by
inverse Fourier Transform
Dispersive materials and boundaries are easily
described by frequency-dependent parameters
Why Model in Time Domain?
Time domain simulations are life-like and
allow visualization of signal propagation
Virtual experiments are set up as in the lab
(Source, reference planes, output probes)
Cause and effect can be distinguished
One simulation can cover a wide bandwidth
Transient phenomena can be simulated
Dispersive materials and boundaries are
modeled in a more physical manner
Frequency domain information can be obtained
via Fourier transform
Solving a Complex Problem Needs
Electromagnetics / Physics:
A correct and efficient problem definition
A good physical insight within calculations
A good physical model can reduce complexity
Mathematics
A correct and efficient mathematical description
Mathematical analysis: convergence, stability, conditioning, error
analysis, error control
Computer Science
Efficient algorithms for the math problem
Efficient memory arrangement: shared memory and local memory
Parallelization of computers and inter-processor communications
Integral Equation
Methods
Differential Equation Methods
Finite-Difference
Time-Domain
(FDTD)
Finite-Difference
Frequency-Domain
(FDFD)
Method of Moments
(MoM)
Fast Multipole
Method (FMM)
Finite Element
Method (FEM)
Transmission Line
Matrix (TLM)
Finite Integration
Technique (FIT)
Computational Techniques
Frequency domain methods
Time-domain methods
Finite-Difference
Time-Domain
(FDTD)
Finite-Difference
Frequency-Domain
(FDFD)
Method of Moments
(MoM)
Fast Multipole
Method (FMM)
Finite Element
Method (FEM)
Transmission Line
Matrix (TLM)
Finite Integration
Technique (FIT)
Computational Techniques
Few Commercial CEM Packages
FDTD
MoM
FEM
FIT
HFSS
GEMS
FIDELITY
Applications of CEM
Over the past five decades CEM has been successfully
applied to several engineering areas, including:
Antennas
Biological electromagnetic (EM) effects
Medical diagnosis and treatment
Electronic packaging and high speed circuits
Superconductivity
Microwave devices and circuits
Law enforcement
Environmental issues
Avionics
Communications
Energy generation and conservation
Surveillance and intelligence gathering
Homeland Security
Signal Integrity
The Method of Moments (MoM)
Method of Moments usually a boundary element method - the
most popular method in antenna engineering.
For perfectly or highly conducting narrow-band structures, very
efficient.
Uses free-space (or geometry specific) Greens function,
incorporating Sommerfeld radiation condition.
Usually reduces problem dimensionality by at least one (surfaces),
sometimes two (wires).
MoM starts from an integral equation for either the electric
or magnetic field (MFIE or EFIE or CFIE).
e
m
: L
: L
EFIE J E
MFIE J H
=
=
MoM
MoM
MoM
MoM
MoM
MoM
MoM
Derive integral equation (IE)
Convert the IE into a matrix equation
using basis functions and weighting
functions
Evaluate matrix elements
Solve the matrix equation and obtain the
parameters of interest
MoM
Only the conducting surfaces of
the structure are descretized
Current in each cell is calculated
Best suited for metallic objects
Not very effective for dielectric
objects if proper basis functions are
not used
Frequency Domain Method
Every single frequency point needs
to be solved separately
Some Terminology
Integral Equation
Unknown quantity/function appears under the integral sign
Integral Equations of 1
st
2
nd
and 3
rd
type:
( ') ( , ') ' ( )
( ) ( ') ( , ') ' ( )
( ) ( ) ( ') ( , ') ' ( )
I z K z z dz F z
z I z K z z dz F z
a z z I z K z z dz F z

=
=
=

If the integration limits are constants, it is Fredholm IE


If the upper limits of integration are variables, it is Volterra IE
Here F(z) represents the source
(known), a(z) & are known
functions, K(z, z) is the Kernel and
is known and I(z) is the unknown
function (solution)
is a scalar/complex parameter
( ) z
Integral Equations in Electromagnetics represents a boundary condition
Some Terminology
Integral Equation
The kernel K(z, z) is the Greens Function
2 2 2
4
( ') ( ') ( ')
jkr
e
K
r
where r x x y y z z

=
= + +
A well-known Greens function is the free-space Greens Function defined
by
Example:
The Poisson Equation
2
( )
( )
r
r

=
has the following representation in integral form:
( ') 1
( ) '
4 '
v
V
r
r dv
r r

Some Terminology
g Lf =
We consider a general operator equation:
L is the operator.
may be an algebraic or
an integral or differential
operator (known)
f is the unknown to be determined
in EM problems typically the
electric / magnetic current
g is the excitation or source
in EM typically a plane wave
or source driving an antenna
(known)
An Example
( )
( ) ( ) 0 1 0
1 0
2
2
= =
=
f f
x x g
dx
f d
A differential operator
L= plus
the Boundary
Conditions (BC)
2
2
dx
d

unknown function f(x)


known function g(x)
Range and domain of the operator L
The range of L is the space of all functions g(x) in the interval
that we wish to consider as excitation functions.
The domain of L is the space of those functions in the interval
that satisfy the boundary conditions AND have second derivatives
in the range of L.
For differential operators the BC are required to define the
operator. Otherwise the solution is not unique.
The integral operator includes the boundary conditions (in
the limits).
An Example (contd)
An EM Operator Equation
By requiring that the total tangential electrical field vanish on a
wire conductor (except at the feed point) we can derive
Pocklingtons integral equation:
( ) ( ) ( )
( ) ( ) ( )
2
2 2
2
0
2
2
2
2
2
2
4 and
2
1
z z a a R d
R
e
z z G
a E j z d z z G k
z
z I
jkR
i
z z

+
|

\
|

= = =

= =

|
|

\
|
+

sin ,
,
'
/
/
l
l
Incident field on surface of wire
Unknown current
Inner Product
The inner product of two vectors is denoted symbolically as
v u,
The inner product must satisfy some typical relationships.
, ,
, , ,
, 0 , 0 0
u v v u
u v w u w v w
u u and u u if and only if u

=
+ = +
= =
( ) ( )

=
b
a
dx x v x u v u,
In MoM, we use the following inner product (of two functions)
defined as:
where [a, b] is the domain of the functions u & v
MOM
We start with
g Lf =
1
N
n n
n
f f
=
=

Now let the unknown f be expanded in a series of known functions f


n
Substitute into the operator equation.

=
=
n
n n
n
n n
g Lf
g f L

The
n
are unknown
constants (usually
complex).
The f
n
are called
basis functions or
expansion functions
and are known.
Linearity of the operator
The basic formulation of MoM is as follows:
MOM
We assume an inner product has been chosen
v u,
Now define a set of weighting functions or test functions w
m
and take the inner product of both sides of the previous equation.
This is sometimes called taking moments. Or identically, multiply
both sides by w
m
and take the inner product.
,..... , , , , 3 2 1 = =

m g w Lf w
n
m n m n

Note the indices


( ) ( )

=
b
a
dx x v x u v u,
Matrix equation
If we expand
(
(
(
(

=
(
(
(
(

(
(
(
(

.
.
,
,
.
.
.
.
, ,
. . , ,
g w
g w
Lf w Lf w
Lf w Lf w
2
1
2
1
2 2 1 2
2 1 1 1

,..... , , , , 3 2 1 = =

m g w Lf w
n
m n m n

We actually generate a matrix equation:


Za b =
In matrix shorthand
What happened?
What we have done is convert the operator equation
Lf = g into a matrix equation.
We can now use matrix algebra to solve for the unknown
vector
n
.
Knowledge of the coefficients
n
now allows f to be found
via f =
n
f
n
.
This is the essential idea of the method of moments.
NOTE: The MoM Matrix is dense
What do we need to do?
Choose some basis functions
Choose some weighting functions
Choose a suitable inner product
Load the matrix
Solve the equation system (invert).
How ?
How many basis functions are required ?
Or, in otherwors, what should be N ?
Test Functions
Sometimes easiest to set
w
m
=f
m
(i.e., test functions identical to basis functions)
This is known as Galerkins method.
A second approach is to choose the Dirac Delta function as
the test function. This approach is called Point matching.
In this case the computation of the inner products
But the accuracy is suffered in this case
And the matrix equation may not be satisfied at points
other than the match points
,
m n
w Lf
and
are easier since
( ) ( ) ( ) ( )
, ' ( ')
b b
m m
a a
w g w x g x dx x x g x dx g x = = =

,
m
w g
The Main Task is: Choice of f
n
f
n
should be linearly independent & chosen so that
superposition can approximate f. In other words we should
know a little about what we are expecting!
Factors are:
Accuracy of solution.
Ease of evaluation of matrix elements
Size of matrix
Well conditioned matrix
The requirement that the matrix elements should be finite
may rule out certain combinations of testing and basis
functions.
Also, smoothness of the basis/testing functions affects the
convergence and accuracy of the numerical solution.
Sub-sectional Basis Functions
x
1
x
2
.
0 1
Pulse (Window) Function
Can be differentiated only once
Spans one sub-domain/interval
OR
Sub-sectional Basis Functions
0
1
Triangular (Rooftop) Function
Type 1
Type 2
Can be differentiated twice
Spans two intervals/sub-domains
One interval/sub-domain is spanned by two basis functions
Sub-sectional Basis Functions
Piece-wise Sine (PWS) Function
Sub-sectional Basis Functions
Truncated Cosine Function
Sub-sectional Basis Functions
Spline Function
Order 1: Pulse Function
Order 2: Triangular Function
Order 3: Quadratic Spline
Spans three sub-intervals
A Quadratic Spline
Entire Domain Basis Functions
More suited to Regular Geometries (unloaded)
Very efficient if they are the eigen functions of the
problem
Bessel Function
Legendre Function
Chebyshev Polynomial
Polynomials (Power Series)
Two-Dimensional Basis Functions
Pulse-Pulse Function
Pulse-PWS Function
Pulse-Rooftop Function
Example 1
( ) ( ) 0 1 0 with 1 0 over 4 1
2
2
2
= = + = f f x x
dx
f d
The exact solution solution is: ( )
3 2 6
5
4 2
x x x
x f =
Try a polynomial basis function N n x x f
n
n
,... , , 3 2 1
1
= =
+
( ) ( )

+
=
N
n
n
x x x f
1
1

So,
Chosen to
satisfy B/C
Galerkin Solution
Galerkin solution implies
1 +
= =
n
n n
x x f w
Now the matrix entries:
( )
( )
( )
( )
2
1
2
1
1 1
0
, ,
1
1
m
n
mn m n
m n
d f
Z w Lf x x
dx
x x n n x dx
mn
m n
+
+
= =
= +
=
+ +

( )( )
( )
( )( )
1
1 2
0
,
1 4
3 8
2 4 2
m m m
m
b w g
x x x dx
m m
m m
+
=
= +
+
=
+ +

Matrix is symmetric!!
Convergence
N=1
10
11
30
11
3
1
1 1 11
= = = g Z ,
N=2
(
(
(

=
(

(
(
(

=
(

(
(
(

3
2
10
1
12
7
30
11
5
4
2
1
2
1
3
1
2
1
2
1

N=3
(
(
(
(
(

=
(
(
(

(
(
(
(
(
(

=
(
(
(

(
(
(
(
(
(

3
1
0
2
1
70
51
12
7
30
11
7
9
1
5
3
1
5
4
2
1
5
3
2
1
3
1
3
2
1
3
2
1

( )
2
10
11
x x f =
( ) ( )
3 2
3
2
10
1
x x x x f + =
( ) ( )
4 2
3
1
2
1
x x x x f + =
Results
0.00 0.20 0.40 0.60 0.80 1.00
0.00
0.10
0.20
0.30
N=1
N=2
N=3
f(x)
The N = 3
curve is exact
since
( ) ( )
3 2 6
5
3
1
2
1
4 2
4 2
x x x
x x x x
=
+
Point Matching
The integration in evaluating the matrix entries can be difficult to
perform in problems of practical interest. A simple way of
obtaining solutions is to require that the operator equation be
satisfied at discrete point. This is called point matching.
Mathematically, it is equivalent to using Dirac delta functions as
testing functions. E.g., the matrix entry would become
( ) ( )
( ) ( )
1
1
0
1
1
, 1
1 1
1
n
mn m n m
n
n
m
Z w Lf x x n n x dx
m
n n x n n
N

= = +
| |
= + = +
|
+
\

x
m
are evenly
spaced points
on [0,1].
Here matrix is not symmetric
For the case of Example 1,
Example 2
( ) ( )
2
2
6 2 over 0 1 with 0 1 0
d f
x x f f
dx
= = =
( )
2
(1 ) f x x x =
Try a basis function N n x x f
n
n
,... , , 3 2 1
1
= =
+
Galerkin approach implies
1 +
= =
n
n n
x x f w
The exact solution solution is:
Results - Galerkin
N = 1
N = 2
Results Point Matching
N = 1
N = 2
Example 3
( ) ( )
2
2
2
4 12 over 0 1 with 0 1 0
d f
x x f f
dx
= = =
( )
4 2
2 f x x x x = +
Try a basis function
2
(1 ) 1, 2, 3,...
n
n
f x x n N = =
Galerkin solution implies
2
(1 )
n
n n
w f x x = =
The exact solution solution is:
Results - Galerkin
N = 2
N = 5
N = 1
N = 10
Surface Charge - Square Plate
The Integral Equation is
Pulse Basis function & Point Matching
2 2
( ', ')
' '
4 ( ') ( ')
a a
a a
x y
V dx dy
x x y y


=
+

2
2 2
(2 )
,
4
2
log(1 2),
( ) ( )
mn
mn
mn m n m n
b
if m n
R
Z
b
if m n
R x x y y

= +
After simplification
b
m
= V = 1 volt
Matrix is symmetric!!
Surface Charge - Square Plate
Initialization
Surface Charge - Square Plate
Z Matrix Computation
B Matrix Computation
Solve Matrix Equation
Compute Capacitance
Surface Charge - Square Plate
Display of the Result
Surface Charge - Square Plate
Parallel Plate Capacitor
2
2 2 2
(2 )
,
4
2
log(1 2),
( ) ( ) ( )
mn
mn
mn m n m n m n
b
if m n
R
Z
b
if m n
R x x y y z z

= + +
Matrix is symmetric!!
Parallel Plate Capacitor
Thin-Wire Dipole Antenna
PWS Basis Function
The Integral Equation
2
2
2
1 ( , ')
( ') ( , ') ' ( ) 0
L
i
z
L
V z z
I z V z z dz E z
j z


(
+ + =
(

1
1
1
1
1
1
sin ( )
'
sin ( )
sin ( )
( ') '
sin ( )
0
n
n n
n n
n
n n n
n n
z z
for z z z
z z
z z
J z for z z z
z z
otherwise

+
+
+

Thin-Wire Dipole Antenna


N = 50
N = 80
Input Impedance:
Mag Frill: 82.6961 + j21.6556
Delta Gap: 85.9360 + j16.1577
Input Impedance:
Mag Frill: 76.4328 + j17.3315
Delta Gap: 82.2019 + j16.1063
Wire Length = 0.48
Radius: 0.005
RWG Basis Functions
$
( )
0
i s
n + = E E
on S
$ $
i s
n n = E E on S
s
j = E A
( )
tan
tan
i
j = + E A
( )
'
4
jkR
s
e
r ds
R

A J
( )
'
1
4
jkR
s
e
ds
R

r
s
j = J
But
and
i
L = J E
This leads to an equation of the form:
A PEC surface is considered here
RWG Basis Functions
( )
n
n
,
2
,
2
0,
n
n
n
n
l
if is in T
A
l
if is in T
A
otherwise
+
+

+
n
-
n n
r
f r r
The Basis Function (Known as RWG
basis function)
Triangles, T
n
+
and T
n
-
, correspond to the n
th
edge of triangulated surface modeling
of the scatterer. Point in T
n
+
may be designated either by the position vector r
defined with respect to O, or by the position vector defined with respect to the free
vertex of T
n
+
.
l
n
is the length of the edge and A
n

is the
area of the triangle T
n

.
RWG Basis Functions
n
n
,
,
0,
n
n
s n
n
l
if is in T
A
l
if is in T
A
otherwise
+
+

n
r
f r
MoM Formulation
Let
1
N
n
n
I
=
=
n
J f
Where f
n
are RWG basis functions and I
n
are unknowns
( )
m
Tm
ds A

c
m
r
( )
i i
m
Tm
ds A

c
m
E E r
The following integrals are approximated
where is a vector from origin to the centroid of T
n

i
j = + E A
, , ,
i
j = +
m m m
E f A f f
,
i i
s
ds =

m m
E f E f
( ) ( )
2
i c i c
m
l

(
+

c+ + c-
m m m m
E r E r
But
where are vectors between the free vertex and the centroid point of the
triangles T
m

MoM Formulation
= V
m
, m = 1, 2, , N
, . .
2 2
c c
i
m
l

(
= +
(

+
+ -
m m
m m m

E f E E
( )
i
=
c
m m
E E r
Now if we take
Similarly
,
s
S
ds =

m m
f f
( ) ( )
m
l
(
=

c+ c-
m m
r r
( ) ( )
,
2 2
c c
m
j jwl

(
+
(

+
c+ c-
m m
m m m

A f A r A r
so that
and
( ) ( )
,
2
c c
m
l

(
+

c+ + c-
m m m m m
A f A r A r
MoM Formulation
Now
'
1
4
jkR
N
n
n
s
e
ds and I
R

=
= =

n
A J J f
( ) ( )
'
4
m
jkR
m s
e
ds
R

+
=

c+ '
m
A r J r
m
R
+
=
c+ '
m
r r
1
N
n mn
n
I
+
=
=

A
( )
'
4
m
jkR
mn
m s
e
ds
R

+
+
=

'
n
A f r
( )
1
,
N
n mn
n
I

=
=

c-
m
A r A
( )
'
,
4
m
jkR
mn n m
m s
e
ds R
R

= =

' c- '
m
A f r r r
So that
and
MoM Formulation
Thus
1
,
2 2
c c
N
n m mn mn
n
jw I jwl

+
=
(
| |
= +
( |
\

+
m m
m

A f A A
( )
'
1
4
jkR
s
e
ds and s j
R

= =

r J
( ) ( )
'
1
1
4
m
jkR
N
n s
n
m s
e
I ds
j R


c '
m n
r f r
( )
'
1
4
m
jkR
s
m s
e
ds
j R

'
mn n
f r
1
,
N
m n
n
l I
+
=
( =

m mn mn
f
Hence
But
so that
1
N
n
n
I
=
=
n
J f
( ) ( )
,
m
l
(
=

c+ c-
m m m
f r r Now
and
MoM Formulation
Thus
1 1
2 2
c c
N N
m n m m n
n n
V I jwl l I

+
= =
(
| |
( = + +
( |

\


+
+
m m
mn mn mn mn

A A
mn
N
n
n
Z I

=
=
1
2 2
c c
mn m
Z l jw

+
| |
(
= + +
|
(

\
+
+
m m
mn mn mn mn

A A
where
Now
( ) ( ) ( )
( )
'
' '
'
,
4
jk
m m
s
e
g ds where g


+ +
= =

c+
m
r r
+ ' '
mn n
c+
m
A f r r r
r r
( ) ( ) ( ) ( )
' '
4 2 2
n n
c c
n n
n m n m
n n
T T
l l
g ds g ds
A A

+
+ +
+
(
= + (
(


+ ' ' ' '
r r r r
( ) ( ) ( ) ( )
' '
4 2 2
n n
c c
n n
n m n m
n n
T T
l l
g ds g ds
A A

+

+
(
= + (
(


+ ' ' ' '
mn
A r r r r
( )
( )
'
'
jk
m
e
g

c
m
r r
'
c
m
r
r r
and
MoM Formulation
Also
( ) ( )
'
1
4
s m
s
g ds
j
+ +
=

' '
mn n
f r r
( ) ( )
' '
n n
1
4
n n
m m
n n
T T
l l
g ds g ds
j A A
+
+ +
+
(
= (
(


' '
r r
( ) ( )
' '
n n
1
4
n n
m m
n n
T T
l l
g ds g ds
j A A
+

+
(
= (
(


' '
mn
r r
MoM Formulation
And thus
( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
'
'
' '
4 4
4
n
n
n n
c c
n
m n m m n m
n
T
m
mn
c c
n
m n m m n m
n
T
m n n
m m m m
n n
T T
l
g g ds
A
j l
Z
l
g g ds
A
l l l
g g ds g g ds
j A A

+
+
+
+

+ +
+
(
| |
( | + +
|
(
\
( = +
| | (
| + (
|
(
\

(
+ (
(


' + ' + ' + '
' ' + ' '
' ' ' '
r r r r
r r r r
r r r r
RWG for Dielectric Bodies
The basic RWG basis function can be used only for Conducting bodies.
However, modified RWG functions are available for handling such cases.
These basis functions are defined on tetrahedral pairs, instead of
Triangle pairs. This demands for a volumetric discretization of the body
and increases the unknowns considerably.

+ +
+
n
V
n
n
n
n
V
n
n
n
V
n
T in r
V
a
T in r
V
a
r f
r r
r r
r

3
3
) (
Computational Aspects
Stability:
Condition number of a matrix: Condition number of matrix A is
defined as
k(A) = ||A|| ||A
-1
||, where ||A|| denotes the norm of matrix A.
A good condition number is small, near to 1; and k = 1 for the identity
matrix. The condition number of a matrix increases with the size of the
matrix.
Meshing:
The mesh size should not exceed /10. This defines the number
of basis functions, N.
Computational Aspects
Inversing the Matrix:
Fast iterative algorithms like CGM and its variants may be
needed
Numeric Integration:
Romberg, Richardson, Simpson's
High Performance Computing and other
Strategies:
Parallel Computing, Out-of-Core Approach, Graphics
Processor Unit (GPU) based Computing, etc.,
Electrically Large Problems
MoM: very accurate and efficient for open boundary (radiating) problems, but does not
scale well for electrically large. (Suitable for mobile phone and base station antenna.)
MLFMM: same accuracy as the MoM, but requires up to several orders of magnitude less
memory. (Suitable for large antenna, e.g. array, reflector, antenna placement e.g. on a car.)
MoM/PO: very memory efficient, can be used on arbitrarily shaped structures but where
multiple reflections play a big role in desired parameters it is inefficient. (Suitable for
large reflector antennas.)
MoM/UTD: extremely memory and run-time efficient. Structures have to be modelled
with flat polygonal plates with minimum side length of about a wavelength. (Suitable
for antenna placement on huge structure.)
Method Formulation CPU-
time
Memory
MoM Current-based N
2..3
N
2
MLFMM Current-based N log
2
N N log N
PO Current-based N constant
UTD / GO Ray-based constant constant
Courtesy: M/s EMSS Pvt Ltd, SA
Resource Requirements
PEC Structure Simulation at 2 GHz instead of 1 GHz.
f 2f
N 4N
Complexity Factor
O(N
3
) 64
O(N
2
) 16
O(N) 4
O(N logN) 5
O(N log
2
N) 6
Courtesy: M/s EMSS Pvt Ltd, SA
MLFMM
With the MoM a large structure (e.g. ship) can only be
solved up to several MHz. (The actual frequency depends
on hardware, RAM, number of processors etc.)
The MLFMM extends the range over which accurate
simulations can be done without having to use
asymptotic techniques such as the PO, or the use of
partial models (i.e. the exclusion of certain geometries
known not to have a significant influence the results).
Courtesy: M/s EMSS Pvt Ltd, SA
MLFMM Application: Sphere
Bistatic RCS computation of a
PEC sphere:
Diameter 10.264
N = 100 005 unknowns
Memory requirement:
MLFMM 1.45 GByte
MoM 149.02 GByte
Run-time (XEON 3.4 GHz)
MLFMM 2.4 hours
MoM not solved
Courtesy: M/s EMSS Pvt Ltd, SA
Typical Memory Requirement
N MoM MLFMM Application
100 000 75 GByte 1 GByte Military aircraft at 690 MHz
Ship (115 m x 14 m) at 107 MHz
Reflector antenna with aperture size 19
200 000 300 GByte 2 GByte Military aircraft at 960 MHz
Ship (115 m x 14 m) at 150 MHz
Reflector antenna with aperture size 27
400 000 1.2 TByte 4.5 GByte Military aircraft at 1.37 GHz
Ship (115 m x 14 m) at 214 MHz
Reflector antenna with aperture size 38
1 000 000 7.5 TByte 12 GByte Military aircraft at 2.2 GHz
Ship (115 m x 14 m) at 340 MHz
Reflector antenna with aperture size 60
Memory usage for the MoM with and without MLFMM.
Courtesy: M/s EMSS Pvt Ltd, SA
Comparison
Comparison
How Reliable are CEM Tools?
Very much if properly applied (by
experienced users), on problems for which
codes were designed for
However remains easy to compute wrong
answers, especially for novices
Main Problems:
Insufficient understanding of the capabilities and
limitations of underlying algorithms
Inappropriate meshes: Inadequately refined, overly
simplified, etc.,
Inadequate appreciation of the effect of critical
tolerances (dimensions, material parameters) on real
engineered devices)
How Reliable are CEM Tools?
Using simulation tools is very useful in RF engineering
because efficient and accurate programs make the design
easier, less prototypes are needed and time and money are
saved.
Simulation tools are expensive but they will usually pay the
investment back
Use of simulator does NOT make thinking and understanding
of RF Engineering, EM & Circuit Theory unnecessary because
the user has to
Know what he/she puts into the simulator
Understand the simulation method used because the user
chooses the simulation tool and chooses the settings of
the simulation
Have suggestive knowledge about expected results
because the simulated results can be unreliable if the user
has made mistakes
The (Crude?) Fact
God Only Knows !!!
How my problem gets
solved in the software?
What Happens Inside the Black Box ?
The Myth of Procrustes
Procrustes, whose name means "he who stretches", was arguably the most
interesting of the hero Theseus's challenges. He kept a house by the side of
the road where he offered hospitality to passing strangers, who were invited in
for a pleasant meal and a night's rest in his very special bed. Procrustes
described it as having the unique property that its length exactly matched
whomsoever lay down upon it. What Procrustes didn't volunteer was the
method by which this "one-size-fits-all" was achieved, namely as soon as
the guest lay down Procrustes went to work upon him, stretching him on the
rack if he was too short for the bed and chopping off his legs if he was too
long. Theseus turned the tables on Procrustes, fatally adjusting him to fit his
own bed.
Is it Due to the Black Magic ?
Scattering from a small Dielectric Sphere (dia =13 / 200)
Courtesy: Prof Raj Mittra
Code 1
Code 2
Mie
Is it Due to the Black Magic ?
RCS of a PEC Sphere at Low Frequency
Courtesy: Prof Raj Mittra
Mie
Technique 1
Technique 2
Mie
Technique 1
Technique 2
The Big Questions ?
Is the software using the Right Technique?
The Big Questions ?
Are you using the software correctly ?
Actual Problem How the CAD
Modelling was done
How it was meshed How the Solver
Solved it
How the Post-processor
processed the results
How results were
displayed
How the Problem Was
Understood
How the CAD Import
was done
The Big Questions ?
How to ensure that the results are correct ?
IEEE Std 1597.1 2008: IEEE Standard for validation of Computational
Electromagnetics Computer Modeling and Simulations
Validation Approaches
Computational Technique Validation
Software Code Implementation Validation
Specific Model Validation
Validation using closed form equations
Validation using measurements
Validation using other modeling techniques
Validation using intermediate results
Validation using convergence
The Fundamental Rule
Use the Right Tool for the Right Job
References
1.Field Computation by Moment Methods R.F. Harrington
2.Understanding Electromagnetic Scattering Using Moment Method
Randy Bancroft
3.The Method of Moments in Electromagnetics Walton C Gibson
4.Antenna and EM Modelling Using MATLAB Sergey N Makarov
5.Electromagnetic Scattering by Surfaces of Arbitrary Shape
Rao, Wilton, Glisson (IEEE Trans. A&P, Vol. 30, No. 3, May 1982, pp 409 - 418)
6.A Tetrahedral Modeling Method for Electromagnetic Scattering by
Arbitrary Shaped Inhomogeneous Dielectric Bodies
Schaubert, Wilton, Glisson (IEEE Trans. A&P, Vol. 32, No. 1, Jam 1984, pp 77 - 85)
7.Simple Examples of the Method of Moments in Electromagnetics
Edward H Newman (IEEE Trans. Education, Vol. 31, No. 3, Aug 1988, pp 193 -
200)
8.Introducing Undergraduates to the Moment Method
Perry Wheless, Larry T Wurtz (IEEE Trans Education, Vol. 38, No. 4, Nov 1995, pp
385 - 390)

Vous aimerez peut-être aussi