Introduction to Computational

Electromagnetics for RF and Microwave

Applications

Prof David B Davidson

SKA Research Chair
Dept. Electrical and Electronic Engineering
Univ. Stellenbosch, South Africa

Faculty of Engineering
Outline of talk

• Brief outline of electromagnetics & computational

electromagnetics (CEM) techniques –
• Full wave methods – MoM, FEM, FDTD
• Asymptotic methods
 Maxwell’s equations


• Controlling equations in  B
classical EM are Maxwell’s E  
t
eqns. 
   D
• Two curl eqns (Faraday and H  J 
Ampere’s laws). t
• Two divergence eqns (Gauss’s 
law). D  

• Constitutive (material) B  0
parameters ε and μ.
 
D E
 
B  H
 Using Maxwell’s equations

• From late 19th century, these

have formed basis for
understanding of EM wave
phenomena.
• Classical methods of
mathematical physics yielded
solutions for canonical problems
– sphere, cylinders, etc (Mie
series opposite).
• Astute use of these, physical
insight and measurements
produced great advances in
understanding of antennas, EM
radiation etc.
 Computational Electromagnetics (CEM)

• In common with Comp

Sci & Engr, CEM has its
genesis in 1960s as a new
paradigm.
• First methods were MoM
(circa 1965), FDTD
(1966), FEM (1969).
CEM formulations

• Solutions to Maxwell’s eqns have been sought in time and

frequency domains (d/dt → j ω, aka phasor domain).
• Full-wave formulations have included:

• Finite difference (usually in time domain - FDTD)

• Finite element method (FEM - traditionally frequency, now
increasingly time domain)
• Green’s function based (boundary element, volume element;
known as Method of Moments in CEM). (Usually frequency
domain).
• FEM and MoM both require solution of matrix equation;
FDTD is matrix-free.
• Asymptotic methods have also been used (typically ray-optic
based methods, eg geometrical theory of diffraction). Very
powerful for a limited class of problems (reflectors!)
 MoM, FDTD, FEM – basics

• Left: MoM (usually) meshes surfaces; uses a Green’s function which

couples all elements (global).
• Centre: FDTD meshes volumes with cuboidal elements; directly
approximates Maxwell’s curl equation using finite differences (local).
• Right: FEM (usually) meshes volumes with tetrahedral elements; uses
Galerkin or variational functional formulation (local).
CEM as a viable design tool

• Elevation of CEM to equal • RF & microwave industry:

partner of analysis & • General telecoms
measurement only since • Cell phone designers &
1990s. operators
• Widespread adoption of • Radio networks
CEM for general industrial • Terrestrial & satellite
RF & microwave use broadcasting;
delayed by computational • Radar and aerospace
cost of 3D simulations. applications (esp. defence –
which is where much of SA’s
• 1990s saw first
current expertise
commercial products originated)
emerge (eg FEKO, HFSS, • Radio astronomy.
MWS), and 2000s has seen
these products become • CEM is now an industry in its
industry standards. own right.
 Key advances needed for 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.
(Attributed to the late Prof
Ben Munk, OSU).
• Each method needed a key
advance, circa late 80s early • Moore’s law of CEM –
90s: processor capacity doubling every
• MoM: MLFMM two years or faster; algorithm
• FEM: vector elements
speed doubling at least as fast.
• FDTD: PML
• Now: far greater (but still • Over a decade, this predicts a
sometimes misplaced) faith in speed-up of around 1000!
CEM results! • Also, memory became very
cheap.
Method of Moments (MoM)

• Method of Moments – usually a boundary element

method - still most popular method in antenna
engineering.
• For perfectly or highly conducting narrow-band
structures, very efficient.
• Uses free-space (or geometry specific, eg stratified
media) Green’s function, incorporating Sommerfeld
radiation condition.
• Usually reduces problem dimensionality by at least
one (surfaces), sometimes two (wires).
 MoM formulation – (very) basics
• Thin-wires one of earliest
applications.
• Based on integral eq:

1  2 ( z, z ')
E ( z)     ( z, z ')]I z ( z ')dz
inc 2
[ k
j 0 L z
z 2

e  jkR
 ( z, z ') 
4 R
• Generates full, complex-valued
interaction matrix: O(f 6) asymptotic
computational cost for surfaces;
O(f 4) memory.
• Key advance: Fast methods, esp
MLFMM.
 Multilevel Fast Multipole Method (MLFMM)
one level

N
N

two levels
K1
K2 Multiple levels in the limit:

Memory requirement: O (N log N)

N Run-time: O (N log2 N)
(per iteration)
MLFMM application example: Sphere (FEKO)

Bistatic RCS computation

of a PEC sphere:
diameter 10.264 
N=100005 unknowns

Memory requirement:
MLFMM 406 MByte
MoM (est) 149 GByte

Run-time (Intel Core 2

E8400):
MLFMM 5 mins
MoM not solved
FEM in CEM

• FEM in CEM shares much • Based on “minimizing”

with computational variational functional:
mechanics.
 1 1    
• Along with FDTD, FEM F ( E )    (  E )  (  E )  ki  r E  E  dS
2
2 S  r 
shares simple handling of
different materials.
• Key advance: “edge based”
• FEM offers systematic unknowns, via vector basis
approach to higher-order functions:
elements and different

elements shapes. wij   i j   j i
• Less computationally
efficient than FDTD, but
uses degrees of freedom
more efficiently.
 FEM – Vector elements

Whitney basis function N2={−y,x}

1.2

• Vector elements provide 0.8

constant tangential/linear
0.6

y
0.4

normal approximation along 0.2

edges.
0

−0.2

−0.2 0 0.2 0.4 0.6 0.8 1 1.2

• Recovering full vector field

x

Whitney basis function N3={−y;−1+x}

1.2

requires summing all elements 1

– not interpolatory.
0.8

0.6

y
0.4

0.2

−0.2

−0.2 0 0.2 0.4 0.6 0.8 1 1.2

x

Whitney basis function N1={1−y,x}

1.2

0.8

0.6

y
0.4

0.2

−0.2

−0.2 0 0.2 0.4 0.6 0.8 1 1.2

x
FEM – “Spurious modes”

• Vector elements
kc =137.377
1.2

revolutionized FEM – largely

1

0.8

0.6 solved “spurious mode”

0.4
problem.
0.2

0 • Top - first TE mode in rect. X-

−0.2
−0.5 0 0.5 1 1.5 2 2.5
band guide (top). kc=137.4
1.2
kc =2.6982e−006
• Bottom - first ``spurious’’
1

modes. Approximates zero

eigen-frequency very well:
0.8

0.6

0.4 kc=2.7x10-6.
0.2

−0.2
−0.5 0 0.5 1 1.5 2 2.5
 Higher order vector elements

• Higher-order elements have CT/LN 1 per 

wij   i j   j i
seen many approaches. edge
• Hierarchal esp. attractive for p-
refinement. LT/LN 1 per   i j 





• Webb’s 1999 scheme shown edge

opp. Include complete elements. LT/QN 2 per   
 j wij i w jk
• More rapid convergence – face
O(h2p) with p highest
(incomplete) polynomial order. QT/QN 1 per 
  i j [ i   j ] 
edge;
  i j k 
 

1 per  

face
FDTD method (1)

• Finite Difference Time Domain

(FDTD) currently most popular
full-wave method overall.
• Usually refers to a specific
formulation – [Yee 66], right.
• Uses central-difference spatial and
temporal approximation of
Maxwell curl equations on “Yee
cell”. (2D eg below)
• Basic Yee leap-frog
implementation simple & 2nd order
accurate with explicit time
integration.

t
E x (i, j, n  1)  E x (i , j , n )  [ H z (i, j, n )  H z (i, j  1, n )]
s
FDTD Radiation boundary conditions & PML

• Biggest single drawback of FDTD was handling open regions.

• FEM has same problem, but FEM/MoM solves this elegantly.
• What is needed is a radiation boundary conditions - widely
used early ones based on properties of wave operators
(Higdon, Bayliss-Turkel).
• Berenger (1994) proposed what is in essence an artificial
uniaxial absorber to create the Perfectly Matched Layer (PML).
• The PML revolutionized applications of FDTD, as very high
performance (>70dB dynamic range of good anechoic
chambers) could now be added to codes.
 FDTD method-MWS example

• Rat-race coupler in
microstrip, 1.8 GHz
center frequency.
• “Open boundaries” –
Perfectly Matched Layer –
used to terminate upper
space.
 Asymptotic (aka optical methods)

• As opposed to full-wave • PO uses approximation

methods which are for surface current as
“numerically exact”, Js ≈ 2 n x Hi
asymptotic methods • PTD improves this with
approximate the basic edge current (Fock
physics. currents, etc)
• Widely used methods are
• Physical Optics and
Physical Theory of • UTD/GTD is ray-based,
Diffraction (PO/PTD) and adds diffracted rays
• Uniform Geometrical (in the Keller cone) to
Theory of Diffraction the direct and reflected
rays.
 Asymptotic methods contd

• PO/PTD is available in • However - because the

FEKO, and of course in asymptotic methods
GRASP. (typically) decouple the
• UTD codes tend to be in- feedhorn and reflector,
house and special purpose; small but significant effects
some elements of UTD such as standing waves
have been implemented in between these in prime-
FEKO. fed systems (eg KAT-7)
• Aysmpotic methods are cannot be predicted with
usually fast, and UTD does these.
not inherently scale with f.
2nd edition of text:
How reliable are CEM tools?

• Properly applied, by experienced users, on

problems for which codes were designed – very!
• 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, over-
simplified, even over-complex!
• Inadequate appreciation of effect of critical
tolerances (dimensions, material parameters) on
real engineered devices.
In summary

• Full-wave CEM has matured and moved into

routine use in RF& microwave industries since
early 90s.
• Previous generation of many devices (eg radio
telescopes) were designed in essentially pre-CEM
era (nonetheless many superb designs!)
• New tools permit investigation of effects that
previously could only be measured on final
prototype.
• Eg. is MeerKAT - one of the first radio telescopes
designs to benefit from this.