Electromagnetic Problems and Numerical Simulation Techniques

Levent SEVGI1,2
1
Raytheon Systems Canad aLtd, Integrated Systems Division, Waterloo, On, N2J 4K CANADA
6
2
MAM-ITRI Information Technologies Research Institute Gebze-Istanbul
, TU RKEY
[email protected] m http://www.ehb.itu.edu.tr/~sevgi

Abstract: - In this paper, examples of current electromagnetic (EM) problems and available numerical
modeling techniques are reviewed. These techniques include the Finite-Difference Time-Domain (FDTD),
Transmission Line Matrix (TLM) and the Method of Moments (MoM).

Key-Words: - Computational Electromagnetics, Numerical Simulations, FDTD, TLM, MoM

1 Introduction examples presented are listed below together with

Computer techniques have revolutionized the way in the most suitable numerical techniques:
which EM problems are analyzed. EM engineers • Waveguide analysis (FDTD and TLM)
rely heavily on computer methods to analyze, for • EMC/EMI Modeling (FDTD and TLM)
example, complex antenna systems, planar • Cellular phone–human head interaction in
microwave devices, EMC/EMI (compatibility and terms of biomedical effects and antenna
interference) problems, etc. A number of different performance (FDTD)
numerical techniques for solving these EM problems • RCS Modeling (FDTD and MoM)
are available. They are mostly based on full-wave
analysis, either in the time or frequency domains, 2 Numerical Techniques in EM
where one or two differential equations plus FDTD and TLM are time domain techniques, which
problem-specific boundary conditions are of have been used for more than a decade. MoM is a
interest. Each numerical technique is well suited for frequency domain technique and has been used for
the analysis of a particular class of EM problem. nearly thirty years. All three techniques are widely
EM engineers must be very careful in applying used and well understood. Therefore, only some of
these pure numerical techniques. Although they are their practical aspects are mentioned in this section.
powerful and can be applied to variety of EM
problems, their output may frequently mislead or be 2.1 FDTD Method
misinterpreted. As long as there are no numerical This method is based on the discretization of
errors such a s overflow or underflow, computers Maxwell’ s two curl equations directly in time and
always give numbers as solutions. The problem is spatial domains and dividing the volume of interest
whether these numbers correspond t real o physics of into unit (Yee) cells [1]) as shown in Fig. 1.
the problem at hand. EM engineers must always be In applying the FDTD method the following
aware of the assumptions made in the numerical observations should be taken into account:
technique that is being used. Under what conditions • The physical volume is subdivided into small
is this technique derived? What kind of a problem or cells.
problems can be handled via this technique? Are
• There are three electric and three magnetic field
there any parametric limitations? What are the
components in each Yee cell distinguished by
accuracy and numerical error limits? Without
(i,j,k) label. The time and spatial discretization
knowing the answers of these questions it is very
steps ar e∆t an d ∆x, ∆y, ∆z, respectively.
dangerous to use these techniques.
Here, three well-known techniques, FDTD, TLM • Although field components in each cell are
and MoM, are reviewed and applied to variety of labeled with the same (i,j,k) numbers (such as
important EM problems. Each EM problem is Ex(i,j,k) or Hz(i,j,k)), their locations are
handled via one or two of these techniques so that different (see Fig.1).
the results can be compared with each other. The • There is a ∆t/2 time difference between E and H
field components in the cell.
 • Any object may be simulated by the medium • Between the nodes, time domain propagation
parameters ε, µ, σ. is simulated via the connection of incident and
• Although electric and magnetic field reflected pulses by the scattering matrix. The
scattering matrix [S], relating the reflected
components are updated during the time voltages, Vr to the incident voltages, Vi is a
simulation, voltages and currents in (i,j,k) cell 12×12 matrix for the node structure given in
are obtained directly from Gauss and Faraday Fig.1, but becomes 18×18 to model any
laws. inhomogeneity in the medium of interest[2].
• Applying effective absorbing boundary • Lossy structures can be modeled by 21×18
simulations allows open region analysis (e.g., scattering matrix.
antenna and RCS analysis). • The medium parameters, ε and µ are modeled
• Narrow and broad band responses may be by open- and short- circuited stubs with the
readily obtained via FDTD simulations. lengths of ∆l/2. But, σ can be modeled by
infinitely long (matched terminated) stubs[2].
One extra stub per electric field component is
necessary for lossy dielectric. The total number
of ports is therefore 21.
• Twelve pulses incident upon the node via the
link transmission lines, produce 12 reflected
pulses. The incident and reflected pulses
appear on the terminals of the transmission
lines at ports which are numbered and directed
according to the voltages shown in Fig. 2.
• 12 voltage pulses given in Fig. 2 describe the
Electric and magnetic field components. For
example, the pulses contributing to the Ex field
are at ports 1, 2, 9 and 12.

Fig.1: FDTD unit cell in rectangular coordinates

2.2 TLM Method

The TLM [2] is another time domain technique. The
Symmetrical Condensed Node (SCN) is the most
common version, where the unit cell is reconstructed
to overcome asymmetry and asynchronous
problems. The SCN-TLM cell for a homogeneous
medium is given in Fig.2.
For simulations using the SCN-TLM, these points
must be taken into account:
• The TLM method is based on the network
theory where voltages and currents are the
independent observation parameters.
• The physical environment is sub-divided into
Fig.2: SCN-TLM cell in rectangular coordinates
small cells as shown in Fig.2. Each cell is
distinguished with (i,j,k) label. The label (n) is
used to represent the current simulation time. 2.3 MOM Technique
As a result, V(x,y,z;t) is replaced with One of the most powerful techniques in the
V(i,j,k;n), where x= i×∆x, y= j×∆y, z= k×∆z and frequency domain is the MoM [3]. The primary
t=n×∆t. formulation of MoM is an integral equation obtained
• The total simulation volume extends from through the use of Green’s functions. The technique
(0,0,0) to (X,Y,Z) which is subdivided into is based on solving complex integral equations by
Nx×Ny×Nz nodes where ∆x=X/Nx, ∆y=Y/Ny and reducing them to a system of linear equations and on
∆z=Z/Nz. applying method of moments (weighted residuals).
• In any direction, the propagation is simulated All method of moments techniques begin by
via two pairs of transmission lines, which do establishing a set of trial solutions with one or more
not couple with each other (these variable parameters. The residuals are a measure of
transmission lines are totally separated in
the difference between the trial and the true
space).
solutions. The variable parameters are determined in
a manner that guarantees a best fit of the trial variation of a Gaussian pulse in an empty waveguide
functions based on a minimization of the residuals. (having 3dB bandwidth of 2GHz around f0=10GHz
Depending on the form of the integral equation center frequency) is plotted.
used, MoM can be applied to
Irises Open Circuit
• conductors only,
• homogeneous dielectrics only, or
• very-specific conductor-dielectric
configurations. MoM techniques do an excellent
job of analyzing a variety of important three-
dimensional radiation and scattering problems. Source Observation
General purpose MoM codes are particularly Fig.4: Rectangular waveguide model used in FDTD
efficient of modeling wire antennas or wires and TLM calculations
attached to large conductive surfaces. MoM 1
techniques applied to integral equations are not
very effective when applied to arbitrary
configurations with complex geometries or
inhomogeneous dielectrics. Two typical discrete

Ey(t)
0
MoM models are pictured in Fig.3.

-1
Time [ns]
Fig.5: Time variation of a narrow duration pulse
around waveguide cutoff

The frequency content of this pulse extends from

Fig.3: Typical wire-grid models for (a) a vehicle DC up to 20GHZ (for %10-amplitude degradation).
antenna (b) a quadlet array over ground screen Therefore, components below cut-off (approx.
7.5GHz) do not propagate and diminish within a few
In Fig.3, a vertical wire antenna over the body of a wavelengths. On the other hand, multiple reflections
car and a monopole quadlet (four vertical wires) occur among the walls of the waveguide. These
with a ground screen are presented. These two effects are presented in Fig.5. Trials showed that at
examples are too complex to handle via analytical least λmin/30 spatial discretisation is required to
solutions. Structures such as shown in this figure obtain good agreement between the two techniques.
can easily be modeled via MoM. Major limitations The effects of the two irises placed inside the
in MoM are the number of segments and diffraction waveguide (see Fig.4) are given in Fig.6.
effects.

3 EM Applications
In this section, characteristic examples modeled via
these three techniques are presented.

3.1 Waveguide Analysis

Time and frequency domain modeling of wave
propagation through rectangular waveguides is a
complex problem. The waveguide acts as a high-
pass filter, where the cut-off frequency depends on
the cross-section. Its frequency characteristics may
be controlled via thin capacitive or inductive irises
as shown in Fig.4. Here, the time domain pulse
propagation through an X-band rectangular
waveguide (near cut-off) and the tuning effects of Fig.6: S-parameters calculated via FDTD and TLM
different inside discontinuities are simulated via the
FDTD and TLM techniques [4]. In Fig.5, time
Here, the frequency variations of the S-parameters inside the box, a very long simulation period is
are calculated via time domain simulation [4]. As required in the time domain. That is the main reason
shown in Fig.6, the irises control transmission of poor agreement in the first plot. Using long
characteristics of the waveguide. simulation periods in FDTD and TLM is very time
consuming. Therefore, instead of this, windowing is
3.2 EMI Modeling applied before Fourier transform to suppress aliasing
Parallel to the increase in the number of electronic effects. This example indicates that pre and/or post
devices that we use, EM Interference (EMI) has processing of simulation data may be required in
become a major EM engineering discipline. Almost order to obtain reliable results.
all of the EMI problems are extremely complex in Once the algorithm is built and tested, any kind of
nature, and can only be handled via pure numerical geometry can be handled via these techniques. This
techniques. Here, a typical EMI problem is modeled is shown in Fig.8, where the results of SE analysis
via both FDTD and TLM techniques. for multi-aperture cases are presented.
A rectangular box having apertures on its different
faces is modeled [4]. Its shielding effectiveness (SE)
for a desired frequency band is calculated. SE is the
measure of isolation, which can be calculated in two
steps:
• First, the field, E(t) of a source inside the box,
is calculated at a point across the aperture
(outside).
• Then, the box is removed and the same
procedure is repeated. The free-space field,
E0(t) is obtained.
The ratio (i.e., 20log10[(E0 / E], dB) in the frequency
domain gives the SE(f) of the rectangular box. If the
source is a pulse, then E0(t) and E(t) are pulse Frequency [GHz]
responses, where broad band SE behaviors can be
obtained via discrete Fourier transforms. A typical Fig.8: SE versus frequency of a multi-aperture box
result is plotted in Fig.7. (solid: multi-aperture, dashed: single-aperture)

3.3 Human Head –Mobile Phone Interaction

Mobile phone-human interaction is another current
EM research topic[5-6]. It is important from both
human health and antenna performance points of
view. A typical FDTD simulation of human head-
mobile phone interaction is pictured in Fig.9.

Frequency [GHz]

Fig.7: SE versus frequency of a box with aperture

Simulations with pure numerical techniques require

great care. This is clearly observed in Fig.7. Here,
first, the results of two techniques are directly Fig.9: FDTD computation volume and far field
plotted (top figure). Then, the simulation data is extrapolation from the simulated near fields
filtered with a Hamming window and plotted again
(bottom figure). Because of the ringing effects [4] Total field FDTD code is used in these simulations.
The simulation procedure is as follows:
 • A discrete model of human head and mobile obtained via a single time domain simulation.
phone is located inside the FDTD volume and
near fields are simulated. PIFA-phone Eθ
• Radiated power is calculated by applying the
surface integration of Poynting vector over a
virtual surface.
• Absorbed power is calculated by applying a z
Eφ
volume integration of lossy cells (i.e., σ|E|2)
• Far fields are extrapolated in time-domain via y
x
near-to-far field transformation routines [5]. Phone-Head
• Frequency behaviors and radiation patterns are
obtained via off-line DFT analysis plus
cartesian-polar transformation. Solid : Eθ
symbols: Eφ
A typical example related to power simulations is Horizontal Plane Vertical Plane
plotted in Fig.10. The transient effects and the
steady-state regime (which is reached after 4-8 Fig.12: Vertical and horizontal radiation patterns of
periods) in FDTD volume is clearly observed in the Planar Inverted F-antenna (PIFA)
figure.
It should be noted that, biomedical modeling via
these techniques is very difficult. Only Specific
Absorption Rate (SAR) of human tissues can be
calculated [5]. SAR is a measure of EM energy
converted into heat in tissues. Discrete tissues are
modeled with their electrical parameters (σ and εr),
which are supplied by EM measurements. Different
EM groups use quite different values. There is also
discrepancies among the limits of SAR values
declared to be safe by the international health
Fig.10: Radiated and absorbed powers versus time organizations. Simulation results must therefore be
carefully analyzed when human health is the
Fig.11 and 12 illustrates performances of two concern.
different mobile phone antennas [5].
3.3 RCS Modeling
Electromagnetic reflectivity of objects is another
Dipole-phone
major subject area that uses pure numerical
techniques. Here, examples are given related to RCS
modeling of different targets. The FDTD and MoM
z techniques are used and their results are compared.
In both FDTD and MoM, scattered field based
y
representations are used. Targets are modeled as
x discrete blocks and as wire meshes in FDTD and
Phone-Head
MoM, respectively. They are located inside the
computation volumes, far fields are obtained all
around and mono and bi-static RCS patterns are
Solid : Eθ
obtained.
symbols: Eφ The first example is given in Fig.13. Here, bi-
Horizontal Plane
static RCS patterns of a metallic rectangular prism
are plotted at different frequencies. Its longest
Fig.11: Vertical and horizontal radiation patterns of
dimension is l=30m (10m×30m×10m) and is
a quarter-wavelength monopole
illuminated by a vertically polarized EM wave. The
illumination is chosen in such a way that that E field
As shown in these figures, broad-band antenna
is parallel to one of the edges.
characteristics (e.g., voltage, current, power,
impedance, gain, radiation patterns) can easily be
It is easy to calculate RCS of complex geometries
via these techniques at low and medium frequencies
but extremely difficult at high frequencies. As the
frequency increases, the number of cells in FDTD
and number of wire meshes in MoM for the discrete
models increases drastically. Nevertheless, these
techniques are especially good for RCS modeling in
the resonance regime, where the wavelength and FDTD

target size are in the same order. MoM

Edge Illumination l=λ/2

27
27 26 Fig.14: Backscatter RCS versus frequency of a the
test target.

Although limited with only idealized geometries,

analytical solutions are very important to understand
the physics behind the problem at hand. It is only
then possible to use pure numerical techniques in
l=λ analyzing complex EM problems.
32 35 The state-of-the-art in numerical modeling is
progressing rapidly. On the other hand, practical EM
problems are also becoming more and more
complicated. It is, therefore, essential that EM
engineers should
• have strong analytical background
• use numerical as well as analytical techniques
l=2λ at the same time.
38 39 Finally, it may be concluded that the trend in
numerical simulation techniques is towards using
some hybrid forms of analytical approximate and
numerical methods.

References:
FDTD NEC [1] K. S. Yee, “Numerical solution of initial
boundary value problems involving Maxwell’s
equations,” IEEE Trans. AP, V-14, no. 3, pp.
Fig.13: Bi-static RCS behaviors at different 302-307, May 1966
frequencies [2] P. B. Johns and R. L. Beurle, “Numerical
solution of two-dimensional scattering problems
using TLM,” Proc. IEE, vol. 118, pp. 1203-
Frequency variation of the backscatter RCS of this 1208, 1971
test object is plotted in Fig.14 for the illumination [3] R. F. Harrington, “Field Computation by
Moment Methods”, The Macmillan Co., New
mentioned in the figure. Better agreement between York 1968
MoM and FDTD results can be obtained if [4] M. Orhan Ozyalcin & L. Sevgi, "Comparisons of
parameters are optimized. Once the parameters are FDTD and TLM methods in Shielding effect
optimized and the limitations and accuracies are analysis", Proc. of IEEE CEFC'98 the Eight
Biennial IEEE Conference on Electromagnetic
tested, any object or group of objects can be used as field Computation, June 1-3, Tuscon, Arizona,
a target. Then, not only isolated RCS behaviors, but USA
also mutual RCS interaction can be calculated. [5] F. Akleman & L. Sevgi, "FDTD Analysis of
Human Head–Mobile Phone Interaction in
Terms of SSAR Calculations and Antenna
4 Conclusion Design", Proc. of IEEE-APS Conference on
Examples of current electromagnetic research topics Antennas & Propagation for Wireless Comm.,
pp. 85-88, November 2-4, Waltham, MA, USA
analyzed via powerful numerical techniques, are [6] L. Sevgi & S. Paker, "FDTD Based RCS
discussed in this study. The attention that must be Calculations and Antenna Simulations", AEU,
paid during the numerical simulations is discussed. International J. of Electronics and Commun.,
Vol.52, No.2, pp.65-75, March 1998