Вы находитесь на странице: 1из 20

220 IEEE Antennas and Propagation Magazine, Vol. 56, No.

2, April 2014
Testing Ourselves
Levent Sevgi
Dou University
Electronics and Communications Eng. Dept.
Zeamet Sokak, No 21, Acibadem Kadiky
Istanbul - Turkey
E-mail: lsevgi@dogus.edu.tr, levent.sevgi@ieee.org
http://www3.dogus.edu.tr/lsevgi
W
e have introduced many computer codes and virtual
tools for electromagnetic modeling and simulation in
the Magazine for nearly a decade. Im very glad to see that
those tools are used in several universities and institutions,
even in national research centers, from USA to Japan, Europe
to Australia and Africa. Most of those tools can be downloaded
from http://modsim.dogus.edu.tr (and may also be requested
from the authors), and be used in teaching/training in virtual
undergraduate labs as well in graduate-level research. Now,
were happy to announce they can also be downloaded from my
new Web site (http://leventsevgi.net). The list of the tutorials
we have introduced since February 2007 may also be found
there.
We have received several requests and questions on some
of our Finite-Difference Time-Domain (FDTD)-based virtual
tools from our readers, who have experienced MATLAB-based
coding/compiling/version problems. I have assigned Miss
Gizem Toroglu, the youngest research and teaching assistant
in our department at Dou University, to re-shape (as well as
redevelop) a collection of MATLAB-based core FDTD codes
in two dimensions (2D), without the need for any toolbox
and/or special command/macro, and to present them in her
interdepartmental seminar this semester. I liked the way she
tailored and presented these codes. Although there were a
number of FDTD codes and packages, I therefore decided to
share them with our readers, through the tutorial we prepared
for this purpose in this issue (by the way, the tutorial on novel
RCS measurement approaches by B. Fisher is on the way).
Those codes mentioned in this issues tutorial are already there,
under EM Virtual Tools, at leventsevgi.net. I hope the readers
will enjoy having them and fnd them useful.

We have already discussed Statistical Decision Making
[1] and Biostatistics with hypothetical tests on cell-phone
users using statistical decision making [2]. What about
Strategic Decision Making? The study of strategic decision
making is called Game Theory. We, engineers, have mostly
been familiar with game theory after a wonderful movie, A
Beautiful Mind, a 2001 American biographical drama flm
based on the life of John Nash, a Nobel Laureate in Econom-
ics. John Nash, who introduced the Nash equilibrium concept,
was played by Russell Crowe. Im glad to announce that I have
fnally convinced Prof. Benan Zeki Orbay, former Dean of the
Faculty of Economics and Administrative Sciences, and cur rent
Chair of the Department of Economics and Finance at Dou
University, to prepare a tutorial on game theory. She is going
to give a presentation entitled Game Theory and Engi neering
Applications in one of our inter-departmental semi nars in
April 2014. Hopefully, well extend it to an interesting tutorial.
References
1. L. Sevgi, Hypothesis Testing and Decision Making: Con-
stant-False-Alarm Rate, IEEE Antennas and Propagation
Magazine, 51, 3, June 2009, pp. 218-224.
2. L. Sevgi, Biostatistics and Epidemiology: Hypothetical
Tests on Cell Phone Users, IEEE Antennas and Propagation
Magazine, 52, 1, February 2010, pp. 267-273.

Finite-Difference Time-Domain (FDTD)
MATLAB Codes for First- and Second-Order
EM Differential Equations
Gizem Torolu, Levent Sevgi
Electronics and Communications Engineering Department
Dou University
Zeamet Sokak 21, Acbadem Kadky, 34722 Istanbul Turkey
E-mail: lsevgi@dogus.edu.tr
Abstract
A set of two-dimensional (2D) electromagnetic (EM) MATLAB codes, using both rst-order coupled differential (Maxwell)
equations and second-order decoupled (wave) equations, are developed for both transverse-magnetic (TM) and
transverse-electric (TE) polarizations. Second-order MUR type absorbing boundary conditions are used to simulate
free space. Metamaterial (MTM) modeling is also included. Performance tests in terms of computational times, memory
requirements, and accuracies were done for simple EM scenarios with magnetic eld, current, and voltage comparisons.
The codes may be used for teaching and research purposes.
Keywords: Maxwell equations; nite-difference time-domain; FDTD; wave equation; absorbing boundary conditions;
MUR conditions; transverse electric; TE; transverse magnetic; TM; metamaterials; MTM; MATLAB
1. Introduction
T
he Finite-Difference Time-Domain (FDTD) method is one
of the most powerful numerical approaches widely used in
solving a broad range of electromagnetic (EM) prob lems since
its frst introduction [1] (a quick Internet search will list tens of
thousands of FDTD studies). A few of the many useful books
written on the FDTD are [2-8]. Information related to the FDTD
may also be found in Wikipedia [9]. The books on the parallel
FDTD [10] and FDTD-based metamate rial (MTM) modeling
[11] are also worth mentioning. We have also presented many
useful tutorials, and have shared our codes and virtual tools for
a long time [12-19]. Table 1 lists these free FDTD-based virtual
tools, with short explanations. These and many more can be
found in the IEEE Press/John Wiley book recently published
within the Press series on EM Wave Theory [20].
The MATLAB-based codes and virtual tools in [12] use
the one-dimensional FDTD for the plane-wave propagation
modeling and simulation through inhomogeneous media, and
in [13] for voltage/current wave transmission and refection
along a transmission line (TL) under different termination and
impedance-mismatch conditions. The TDRMeter virtual tool
in [13] can be used for the visualization of both transmis sion/
refections and fault identifcation.
A general-purpose two-dimensional FDTD virtual tool,
MGL-2D [14], and its modifed version, MTM-FDTD [15],
can be used in the modeling and simulation of EM waves in
two dimensions. A variety of electromagnetic problems, from
indoor/outdoor radiowave urban/rural propagation to electro-
magnetic compatibility (EMC), from resonators to closed/open
periodic structures, linear and planar arrays of radiators can be
simulated easily with MGL-2D. The beauty of MGL-2D comes
from its visualization power, as well as its easy-to-use design
steps. Similarly, MTM-FDTD may be used for the visualiza-
tion of EM waves interacting with different metamaterials.
Snapshots during these interactions may be taken. Scenarios
with normal and oblique incidences, demonstrating focusing
beams in planar metamaterials and the existence of a negative
refractive angle, respectively, may be observed in the time
domain. In addition, video clips of wave- metamaterial inter-
actions may easily be recorded.
The MATLAB-based virtual tool WedgeFDTD was devel-
oped to investigate EM scattering on the canonical non-pene-
AP_Mag_Apr_2014_Final.indd 220 5/31/2014 3:51:23 PM
IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014 221
Testing Ourselves
Levent Sevgi
Dou University
Electronics and Communications Eng. Dept.
Zeamet Sokak, No 21, Acibadem Kadiky
Istanbul - Turkey
E-mail: lsevgi@dogus.edu.tr, levent.sevgi@ieee.org
http://www3.dogus.edu.tr/lsevgi
W
e have introduced many computer codes and virtual
tools for electromagnetic modeling and simulation in
the Magazine for nearly a decade. Im very glad to see that
those tools are used in several universities and institutions,
even in national research centers, from USA to Japan, Europe
to Australia and Africa. Most of those tools can be downloaded
from http://modsim.dogus.edu.tr (and may also be requested
from the authors), and be used in teaching/training in virtual
undergraduate labs as well in graduate-level research. Now,
were happy to announce they can also be downloaded from my
new Web site (http://leventsevgi.net). The list of the tutorials
we have introduced since February 2007 may also be found
there.
We have received several requests and questions on some
of our Finite-Difference Time-Domain (FDTD)-based virtual
tools from our readers, who have experienced MATLAB-based
coding/compiling/version problems. I have assigned Miss
Gizem Toroglu, the youngest research and teaching assistant
in our department at Dou University, to re-shape (as well as
redevelop) a collection of MATLAB-based core FDTD codes
in two dimensions (2D), without the need for any toolbox
and/or special command/macro, and to present them in her
interdepartmental seminar this semester. I liked the way she
tailored and presented these codes. Although there were a
number of FDTD codes and packages, I therefore decided to
share them with our readers, through the tutorial we prepared
for this purpose in this issue (by the way, the tutorial on novel
RCS measurement approaches by B. Fisher is on the way).
Those codes mentioned in this issues tutorial are already there,
under EM Virtual Tools, at leventsevgi.net. I hope the readers
will enjoy having them and fnd them useful.

We have already discussed Statistical Decision Making
[1] and Biostatistics with hypothetical tests on cell-phone
users using statistical decision making [2]. What about
Strategic Decision Making? The study of strategic decision
making is called Game Theory. We, engineers, have mostly
been familiar with game theory after a wonderful movie, A
Beautiful Mind, a 2001 American biographical drama flm
based on the life of John Nash, a Nobel Laureate in Econom-
ics. John Nash, who introduced the Nash equilibrium concept,
was played by Russell Crowe. Im glad to announce that I have
fnally convinced Prof. Benan Zeki Orbay, former Dean of the
Faculty of Economics and Administrative Sciences, and cur rent
Chair of the Department of Economics and Finance at Dou
University, to prepare a tutorial on game theory. She is going
to give a presentation entitled Game Theory and Engi neering
Applications in one of our inter-departmental semi nars in
April 2014. Hopefully, well extend it to an interesting tutorial.
References
1. L. Sevgi, Hypothesis Testing and Decision Making: Con-
stant-False-Alarm Rate, IEEE Antennas and Propagation
Magazine, 51, 3, June 2009, pp. 218-224.
2. L. Sevgi, Biostatistics and Epidemiology: Hypothetical
Tests on Cell Phone Users, IEEE Antennas and Propagation
Magazine, 52, 1, February 2010, pp. 267-273.

Finite-Difference Time-Domain (FDTD)
MATLAB Codes for First- and Second-Order
EM Differential Equations
Gizem Torolu, Levent Sevgi
Electronics and Communications Engineering Department
Dou University
Zeamet Sokak 21, Acbadem Kadky, 34722 Istanbul Turkey
E-mail: lsevgi@dogus.edu.tr
Abstract
A set of two-dimensional (2D) electromagnetic (EM) MATLAB codes, using both rst-order coupled differential (Maxwell)
equations and second-order decoupled (wave) equations, are developed for both transverse-magnetic (TM) and
transverse-electric (TE) polarizations. Second-order MUR type absorbing boundary conditions are used to simulate
free space. Metamaterial (MTM) modeling is also included. Performance tests in terms of computational times, memory
requirements, and accuracies were done for simple EM scenarios with magnetic eld, current, and voltage comparisons.
The codes may be used for teaching and research purposes.
Keywords: Maxwell equations; nite-difference time-domain; FDTD; wave equation; absorbing boundary conditions;
MUR conditions; transverse electric; TE; transverse magnetic; TM; metamaterials; MTM; MATLAB
1. Introduction
T
he Finite-Difference Time-Domain (FDTD) method is one
of the most powerful numerical approaches widely used in
solving a broad range of electromagnetic (EM) prob lems since
its frst introduction [1] (a quick Internet search will list tens of
thousands of FDTD studies). A few of the many useful books
written on the FDTD are [2-8]. Information related to the FDTD
may also be found in Wikipedia [9]. The books on the parallel
FDTD [10] and FDTD-based metamate rial (MTM) modeling
[11] are also worth mentioning. We have also presented many
useful tutorials, and have shared our codes and virtual tools for
a long time [12-19]. Table 1 lists these free FDTD-based virtual
tools, with short explanations. These and many more can be
found in the IEEE Press/John Wiley book recently published
within the Press series on EM Wave Theory [20].
The MATLAB-based codes and virtual tools in [12] use
the one-dimensional FDTD for the plane-wave propagation
modeling and simulation through inhomogeneous media, and
in [13] for voltage/current wave transmission and refection
along a transmission line (TL) under different termination and
impedance-mismatch conditions. The TDRMeter virtual tool
in [13] can be used for the visualization of both transmis sion/
refections and fault identifcation.
A general-purpose two-dimensional FDTD virtual tool,
MGL-2D [14], and its modifed version, MTM-FDTD [15],
can be used in the modeling and simulation of EM waves in
two dimensions. A variety of electromagnetic problems, from
indoor/outdoor radiowave urban/rural propagation to electro-
magnetic compatibility (EMC), from resonators to closed/open
periodic structures, linear and planar arrays of radiators can be
simulated easily with MGL-2D. The beauty of MGL-2D comes
from its visualization power, as well as its easy-to-use design
steps. Similarly, MTM-FDTD may be used for the visualiza-
tion of EM waves interacting with different metamaterials.
Snapshots during these interactions may be taken. Scenarios
with normal and oblique incidences, demonstrating focusing
beams in planar metamaterials and the existence of a negative
refractive angle, respectively, may be observed in the time
domain. In addition, video clips of wave- metamaterial inter-
actions may easily be recorded.
The MATLAB-based virtual tool WedgeFDTD was devel-
oped to investigate EM scattering on the canonical non-pene-
AP_Mag_Apr_2014_Final.indd 221 5/31/2014 3:51:23 PM
222 IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014
trable wedge problem with the FDTD method [16]. Diffracted
felds may easily be extracted and compared with the results
of high-frequency asymptotic (HFA) models. Some interesting
applications of the two-dimensional FDTD method were also
discussed in one of our tutorials [17]. There, FDTD-based path
planning and segmentation were modeled and implemented.
Finally, full-wave, 3D-FDTD EM virtual tools have been
prepared and reviewed in tutorials [18] and [19] for realistic
problem modeling and simulations. In [18], MSTRIP was
introduced for the investigation of a variety of microstrip cir-
cuits. MSTRIP is a 3D-FDTD EM simulator that uses the pow-
erful perfectly matched layer terminations (PML) [21]. The
user needs only to render the microstrip circuit via a computer
mouse on a rectangular grid, and to specify basic dimensions
and supply operational requirements, such as the frequency
band and simulation length. The rest is handled by MSTRIP. It
is easy-to-use, strengthened with visualization and video-clip
capabilities, and can handle very complex single- and double-
layer microstrip structures. Time-domain visualization is pos-
sible during the simulations and video clips may be recorded.
The S parameters are automatically calculated, and may be
displayed online.
In [19], a three-dimensional FDTD-based RCS prediction
virtual analysis tool (MGL-RCS) was introduced. It can be used
to design any kind of a PEC target using basic blocks, such as
a rectangular prism, cone, cylinder, sphere, etc. A col lection of
pre-designed surface and air targets stored in 3DS format fles,
are also supplied. Time-domain near scattered felds can be
simulated around the object under investigation, and transients
can be recorded as video clips. Far felds are then extrapolated,
and RCS as a function of frequency and RCS as a function of
angle plots can be produced (FORTRAN source codes of this
package may also be found in [4]).
2. The Two-Dimensional FDTD Models
The FDTD method [1] discretizes Maxwell equations by
replacing derivatives with their fnite-difference approxima-
tions, directly in the time domain. It is simple, easy to code,
but has the open-form (iterative) solution. It is therefore con-
ditionally stable: one needs to satisfy a stability condition. The
FDTD volume is fnite, and therefore may model only closed
regions. Free-space simulation is an important task in FDTD,
and various effective boundary terminations have been devel-
oped for the last two decades (see [22] for the second-order
MUR-type terminations used here). Broadband (pulse) excita-
tion is possible in the FDTD, but inherits the numerical-disper-
sion problem. Finally, only near felds can be simulated around
the object under investigation; far felds can be extrapolated
using the Equivalence Principle (e.g., the Stratton-Chu equa-
tions) [4].
2.1 First-Order Coupled Equations
The assumption of a continuous translational symmetry
along z lets us reduce the three-dimensional problem into two
dimensions on the xy plane. Maxwell equations in such an
environment are characterized with three parameters (the per-
mittivity, , permeability, , and conductivity, ):

H
E
t

, (1)

E
H E
t

= +


. (2)
Table 1. Free FDTD-based EM Virtual Tools presented in the IEEE Antennas and Propagation Magazine.
Virtual Tool Explanation
1DFDTD
A MATLAB-based 1D FDTD simulation of plane wave propagation in time domain through single, double
or three-layer media. EM parameters are supplied by the user [12].
TDRMeter
A virtual time-domain refectometer virtual tool. It is used to locate and identify faults in all types of
metallic paired cable. Fourier and Laplace analyzes are also possible [13].
MGL2D
A general purpose 2D FDTD package for both TE and TM type problems. Any 2D scenario may be created
by the user by just using the mouse [14].
MTM-FDTD
Modifed version of MGL-2D to simulate cylindrical wave propagation through MeTaMaterials (MTM)
[15].
WedgeFDTD
*
A 2D MATLAB-based simulator for the modeling of EM diffraction from a semi-infnite non-penetrable
wedge using high frequency asymptotics and FDTD [16] (
*
published in ACES).
MSTRIP
A 3D FDTD-based EM simulator for the broadband investigation of microstrip circuits. The user only needs
to picture the microstrip circuit via computer mouse on a rectangular grid, to specify basic dimensions and
operational needs such as the frequency band, simulation length [18].
MGL-RCS
A 3D FDTD-based EM simulator for RCS prediction. The user only needs to locate a 3D image fle of
the target in 3DS graphics format, specify dimensions and supply other user parameters. The simulator
predicts RCS vs. angle and/or RCS vs. frequency [19].
These reduce to two sets of scalar equations (i.e., TM
z
and
TE
z
) in two dimensions under the assumption 0 z , and
can be given as [23]
SET #1: TM
z
( 0
z
H )

x z
H E
t y


=

, (3a)

y
z
H
E
t x



=

, (3b)

y
x z
z
H
H E
E
t x y


=

, (3c)
SET #2: TE
z
( 0
z
E )

x z
x
E H
E
t y


=

, (4a)

y
z
y
E
H
E
t x


=

, (4b)

y
x z
E
E H
t x y



=

. (4c)
As observed, knowing the
z
E (
z
H ) component is enough to
derive all the other feld components for the TM
z
( TE
z
) prob-
lem. The discretized FDTD iteration equations then reduce to

SET #1: TM
z
( 0
z
H )
( ) ( ) ( ) ( )
1
, , , , 1
n n n n
x x z z
t
H i j H i j E i j E i j
y


(
=

,
(5a)
( ) ( ) ( ) ( )
1
, , , 1,
n n n n
y y z z
t
H i j H i j E i j E i j
x


(
= +

,
(5b)
( ) ( )
1
2
, ,
2
n n
z z
t
E i j E i j
t


+
| |
=
|
+
\ .

( ) ( ) , 1,
2
2
n n
y y
H i j H i j
t
t x
(

( +
+ (



( ) ( ) , , 1
2
2
n n
x x
H i j H i j
t
t y
(

(
+
(

,
(5c)
SET #2: TE
z
( 0
z
E )
( ) ( )
1
2
, ,
2
n n
x x
t
E i j E i j
t




=
+

( )
( ) ( )
2
, , 1
2
n n
z z
t
H i j H i j
t y

(


+
,
(6a)
( ) ( )
1
2
, ,
2
n n
z z
t
E i j E i j
t




=
+

( )
( ) ( )
2
, 1,
2
n n
z z
t
H i j H i j
t x

(
+

+

(6b)
( ) ( )
1
, ,
n n
z z
H i j H i j

=

( ) ( )
0
, 1,
n n
y y
E i j E i j
t
x
(

( +
(


( ) ( )
0
, , 1
n n
x x
E i j E i j
t
y
(

(

.
(6c)
2.2 Second-Order Decoupled Equations
Two of the three feld components in Equations (3) and (4)
can be eliminated, and a second-order differential (wave)
equation with a single feld component can be obtained. For
example, the following wave equation for the TM
z
problem
can be directly obtained from Equation (3c) using Equa-
tions (3a) and (3b):

2 2 2
2 2 2
1
0
z
E
t
x y t

(

+ =
(

(

. (7)

This equation, defned for

0; 0 , 0
max max
t x X y Y , (8)
together with the boundary conditions
( ) ( )
1
0, , ,
z
E y t g y t = for 0, 0
max
x y Y = , (9a)
( ) ( )
2
, 0, ,
z
E x t g x t = for 0, 0
max
y x X = , (9c)
( ) ( )
3
, , ,
z max
E X y t g y t = for , 0
max max
x X y Y =
(9b)
( ) ( )
4
, , ,
z max
E x Y t g x t = for , 0
max max
y Y x X = ,
(9d)
and, the initial conditions
AP_Mag_Apr_2014_Final.indd 222 5/31/2014 3:51:24 PM
IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014 223
trable wedge problem with the FDTD method [16]. Diffracted
felds may easily be extracted and compared with the results
of high-frequency asymptotic (HFA) models. Some interesting
applications of the two-dimensional FDTD method were also
discussed in one of our tutorials [17]. There, FDTD-based path
planning and segmentation were modeled and implemented.
Finally, full-wave, 3D-FDTD EM virtual tools have been
prepared and reviewed in tutorials [18] and [19] for realistic
problem modeling and simulations. In [18], MSTRIP was
introduced for the investigation of a variety of microstrip cir-
cuits. MSTRIP is a 3D-FDTD EM simulator that uses the pow-
erful perfectly matched layer terminations (PML) [21]. The
user needs only to render the microstrip circuit via a computer
mouse on a rectangular grid, and to specify basic dimensions
and supply operational requirements, such as the frequency
band and simulation length. The rest is handled by MSTRIP. It
is easy-to-use, strengthened with visualization and video-clip
capabilities, and can handle very complex single- and double-
layer microstrip structures. Time-domain visualization is pos-
sible during the simulations and video clips may be recorded.
The S parameters are automatically calculated, and may be
displayed online.
In [19], a three-dimensional FDTD-based RCS prediction
virtual analysis tool (MGL-RCS) was introduced. It can be used
to design any kind of a PEC target using basic blocks, such as
a rectangular prism, cone, cylinder, sphere, etc. A col lection of
pre-designed surface and air targets stored in 3DS format fles,
are also supplied. Time-domain near scattered felds can be
simulated around the object under investigation, and transients
can be recorded as video clips. Far felds are then extrapolated,
and RCS as a function of frequency and RCS as a function of
angle plots can be produced (FORTRAN source codes of this
package may also be found in [4]).
2. The Two-Dimensional FDTD Models
The FDTD method [1] discretizes Maxwell equations by
replacing derivatives with their fnite-difference approxima-
tions, directly in the time domain. It is simple, easy to code,
but has the open-form (iterative) solution. It is therefore con-
ditionally stable: one needs to satisfy a stability condition. The
FDTD volume is fnite, and therefore may model only closed
regions. Free-space simulation is an important task in FDTD,
and various effective boundary terminations have been devel-
oped for the last two decades (see [22] for the second-order
MUR-type terminations used here). Broadband (pulse) excita-
tion is possible in the FDTD, but inherits the numerical-disper-
sion problem. Finally, only near felds can be simulated around
the object under investigation; far felds can be extrapolated
using the Equivalence Principle (e.g., the Stratton-Chu equa-
tions) [4].
2.1 First-Order Coupled Equations
The assumption of a continuous translational symmetry
along z lets us reduce the three-dimensional problem into two
dimensions on the xy plane. Maxwell equations in such an
environment are characterized with three parameters (the per-
mittivity, , permeability, , and conductivity, ):

H
E
t

, (1)

E
H E
t

= +


. (2)
Table 1. Free FDTD-based EM Virtual Tools presented in the IEEE Antennas and Propagation Magazine.
Virtual Tool Explanation
1DFDTD
A MATLAB-based 1D FDTD simulation of plane wave propagation in time domain through single, double
or three-layer media. EM parameters are supplied by the user [12].
TDRMeter
A virtual time-domain refectometer virtual tool. It is used to locate and identify faults in all types of
metallic paired cable. Fourier and Laplace analyzes are also possible [13].
MGL2D
A general purpose 2D FDTD package for both TE and TM type problems. Any 2D scenario may be created
by the user by just using the mouse [14].
MTM-FDTD
Modifed version of MGL-2D to simulate cylindrical wave propagation through MeTaMaterials (MTM)
[15].
WedgeFDTD
*
A 2D MATLAB-based simulator for the modeling of EM diffraction from a semi-infnite non-penetrable
wedge using high frequency asymptotics and FDTD [16] (
*
published in ACES).
MSTRIP
A 3D FDTD-based EM simulator for the broadband investigation of microstrip circuits. The user only needs
to picture the microstrip circuit via computer mouse on a rectangular grid, to specify basic dimensions and
operational needs such as the frequency band, simulation length [18].
MGL-RCS
A 3D FDTD-based EM simulator for RCS prediction. The user only needs to locate a 3D image fle of
the target in 3DS graphics format, specify dimensions and supply other user parameters. The simulator
predicts RCS vs. angle and/or RCS vs. frequency [19].
These reduce to two sets of scalar equations (i.e., TM
z
and
TE
z
) in two dimensions under the assumption 0 z , and
can be given as [23]
SET #1: TM
z
( 0
z
H )

x z
H E
t y


=

, (3a)

y
z
H
E
t x



=

, (3b)

y
x z
z
H
H E
E
t x y


=

, (3c)
SET #2: TE
z
( 0
z
E )

x z
x
E H
E
t y


=

, (4a)

y
z
y
E
H
E
t x


=

, (4b)

y
x z
E
E H
t x y



=

. (4c)
As observed, knowing the
z
E (
z
H ) component is enough to
derive all the other feld components for the TM
z
( TE
z
) prob-
lem. The discretized FDTD iteration equations then reduce to

SET #1: TM
z
( 0
z
H )
( ) ( ) ( ) ( )
1
, , , , 1
n n n n
x x z z
t
H i j H i j E i j E i j
y


(
=

,
(5a)
( ) ( ) ( ) ( )
1
, , , 1,
n n n n
y y z z
t
H i j H i j E i j E i j
x


(
= +

,
(5b)
( ) ( )
1
2
, ,
2
n n
z z
t
E i j E i j
t


+
| |
=
|
+
\ .

( ) ( ) , 1,
2
2
n n
y y
H i j H i j
t
t x
(

( +
+ (



( ) ( ) , , 1
2
2
n n
x x
H i j H i j
t
t y
(

(
+
(

,
(5c)
SET #2: TE
z
( 0
z
E )
( ) ( )
1
2
, ,
2
n n
x x
t
E i j E i j
t




=
+

( )
( ) ( )
2
, , 1
2
n n
z z
t
H i j H i j
t y

(


+
,
(6a)
( ) ( )
1
2
, ,
2
n n
z z
t
E i j E i j
t




=
+

( )
( ) ( )
2
, 1,
2
n n
z z
t
H i j H i j
t x

(
+

+

(6b)
( ) ( )
1
, ,
n n
z z
H i j H i j

=

( ) ( )
0
, 1,
n n
y y
E i j E i j
t
x
(

( +
(


( ) ( )
0
, , 1
n n
x x
E i j E i j
t
y
(

(

.
(6c)
2.2 Second-Order Decoupled Equations
Two of the three feld components in Equations (3) and (4)
can be eliminated, and a second-order differential (wave)
equation with a single feld component can be obtained. For
example, the following wave equation for the TM
z
problem
can be directly obtained from Equation (3c) using Equa-
tions (3a) and (3b):

2 2 2
2 2 2
1
0
z
E
t
x y t

(

+ =
(

(

. (7)

This equation, defned for

0; 0 , 0
max max
t x X y Y , (8)
together with the boundary conditions
( ) ( )
1
0, , ,
z
E y t g y t = for 0, 0
max
x y Y = , (9a)
( ) ( )
2
, 0, ,
z
E x t g x t = for 0, 0
max
y x X = , (9c)
( ) ( )
3
, , ,
z max
E X y t g y t = for , 0
max max
x X y Y =
(9b)
( ) ( )
4
, , ,
z max
E x Y t g x t = for , 0
max max
y Y x X = ,
(9d)
and, the initial conditions
AP_Mag_Apr_2014_Final.indd 223 5/31/2014 3:51:24 PM
224 IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014
( ) ( )
1
, , 0 ,
z
E x y f x y = , (10a)

( )
( )
2
, , 0
,
z
E x y
f x y
t

, (10b)
are enough to solve for
z
E and the other feld components.
Equation (7) can therefore also be used in the FDTD modeling
and simulations. The FDTD discretized form of Equation (7) is
( )
( )
( ) ( )
1 1
4 1
, , ,
n n n
z z z
p q
t
E i j E i j E i j
g g
+

=
( ) ( )
2
1, 1,
n n
z z
p
E i j E i j
g
(
+ + +


( ) ( )
2
, 1 , 1
n n
z z
q
E i j E i j
g
(
+ + +


(11)
where

2
v t
p
x
| |
|

\ .
,
(12a)

2
v t
q
y
| |
|

\ .


2
2 g v t + ,
(12b)

2
2 t v t +

1
v

= . (12c)
Note that the dispersion and stability conditions, as well as the
source injection in time, are handled just like the frst-order
coupled FDTD equations. On the other hand, the values at the
frst two time instants of
z
E (i.e., ( )
0
,
z
E i j and ( )
1
,
z
E i j ) must
be supplied for the spatial source injection.
2.3 Basic Features of the FDTD Equations
The observations listed below are important for the
numerical implementation of the frst-order coupled (FOC)
FDTD model:
There are three feld components (
x
H ,
y
H ,
z
E for
TM
z
and
x
E ,
y
E , and
z
H for TE
z
) in each cell,
and they are distinguished by the ( ) , i j label for the
frst-order coupled FDTD model.
The discretization steps are , x y , and t , and the
physical quantities are calculated from x i x = ,
y j y = , and t n t = .
Since the FDTD equations are iterative (i.e., open-
form solutions), they are conditionally stable. The
Courant stability condition, which states that the
time step cannot be arbitrarily specifed once the
spatial discretization is done, must be satisfed.
Although the same notations, ( ) ,
n
x
E i j and
( ) ,
n
z
H i j , are used, their locations are different in
the classical Yee cell [1] (see Figure 1), and there is
a half-time-step difference between the E and H
feld computation times. That is, the magnetic-feld
components are calculated at time steps 2 t t = ,
3 2 t , 5 2 t , ..., but the electric felds are calcu-
lated at time steps , 2 , 3 ,... t t t t = .
Figure 1. The Yee cells for the (a) TM
z
and (b) TE
z
prob-
lems.
Only neighboring magnetic-feld values and
( )
1
,
n
z
E i j

are required to update ( ) ,


n
z
E i j . Simi-
larly, neighboring electric-feld values and
( )
1
,
n
x
H i j

are required to update ( ) ,


n
x
H i j .
Both magnetic- and electric-feld components in any
cell may be moved to the origin by just cell
averaging. This is accomplished via
( ) ( ) ( ) , 0.5 , 1,
x x x
H i j H i j H i j = + + (

for mag-
netic felds, but four electric-feld components are
required for this purpose:
( ) ( ) ( ) , 0.25 , 1,
z z z
E i j E i j E i j = + +

( ) ( ) , 1 1, 1
z z
E i j E i j + + + + + (

.
Any object may be modeled by giving , , and
. Two of these, and , appear in the electric-
feld components, and the third, , appears in the
magnetic-feld components.
Three different and values may be assigned for
three electric-feld components, so that different
objects may be located within the Yee cell. Simi-
larly, different values may be given for H-feld
components

for the same purpose.
The important aspects of the second-order decoupled
(SOD) FDTD model are as follows:
There is only one feld component, and its location
may be anywhere in the unit cell.
The models and discrete equations are identical for
the TM
z
and TE
z
problems.
The past two values are needed in every cell.
FDTD iterations yield only
z
E ( TM
z
) or
z
H
( TE
z
). One therefore needs to write down another
discrete (Maxwell) equation for the other two com-
ponents, i.e.,
x
H ,
y
H

( TM
z
) or
x
E ,
y
E ( TE
z
).
2.4 Absorbing Boundary Conditions
To make it simple in this tutorial, the second-order MUR
terminations [22] are used. Table 2 lists equations that must
be satisfed along the boundaries (see Figure 2). The discrete
iteration equations will then be
At 0 x = (
x
N N = )
( ) ( )
1 1
1, 2,
n n
z z
E j E j
+
=
( ) ( )
1 1
2, 1,
n n
z z
c t x
E j E j
c t x
+
(
(
+ +
(

+

( ) ( )
2
2, 1,
n n
z z
x
E j E j
c t x
(
(
+ +
(

+

( )
( )
( ) ( ) ( )
2
2
2, 1 2 2, 2, 1
2
n n n
z z z
c t x
E j E j E j
y c t x
(

(
( + + +

( +

( )
( )
( ) ( ) ( )
2
2
1, 1 2 1, 1, 1
2
n n n
z z z
c t x
E j E j E j
y c t x
(

(
( + + +

( +

(13)
At
max
x X = (
x
N N = )
( ) ( )
1 1
, 1,
n n
z z
E N j E N j
+
=
( ) ( )
1 1
1, ,
n n
z z
c t x
E N j E N j
c t x
+
(
(
+ +
(

+

( ) ( )
2
1, ,
n n
z z
x
E N j E N j
c t x
(
(
+ +
(

+


( )
( )
( )
2
2
1, 1
2
n
z
c t x
E N j
y c t x
(


( + +

( +

( ) ( ) 2 1, 1, 1
n n
z z
E N j E N j
(
+

( )
( )
( ) ( ) ( )
2
2
, 1 2 , , 1
2
n n n
z z z
c t x
E N j E N j E N j
y c t x
(

(
( + + +

( +


(14)
Table 2. Differential equations for the second-order MUR
terminations.
( )
2 2 2
2 2
0
1
0, , 0
0 2
z
max
x
c
E y t
y Y x t c
t y
( =

+ =
(

(

( )
2 2 2
2 2
1
, , 0
0 2
max
z max
max
x X
c
E X y t
y Y x t c
t y
( =

+ =
(

(

( )
2 2 2
2 2
0
1
, 0, 0
0 2
z
max
y
c
E x t
x X y t c
t x
( =

+ =
(

(

( )
2 2 2
2 2
1
, , 0
0 2
max
z max
max
y Y
c
E x Y t
x X y t c
t x
( =

+ =
(

(

AP_Mag_Apr_2014_Final.indd 224 5/31/2014 3:51:25 PM
IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014 225
( ) ( )
1
, , 0 ,
z
E x y f x y = , (10a)

( )
( )
2
, , 0
,
z
E x y
f x y
t

, (10b)
are enough to solve for
z
E and the other feld components.
Equation (7) can therefore also be used in the FDTD modeling
and simulations. The FDTD discretized form of Equation (7) is
( )
( )
( ) ( )
1 1
4 1
, , ,
n n n
z z z
p q
t
E i j E i j E i j
g g
+

=
( ) ( )
2
1, 1,
n n
z z
p
E i j E i j
g
(
+ + +


( ) ( )
2
, 1 , 1
n n
z z
q
E i j E i j
g
(
+ + +


(11)
where

2
v t
p
x
| |
|

\ .
,
(12a)

2
v t
q
y
| |
|

\ .


2
2 g v t + ,
(12b)

2
2 t v t +

1
v

= . (12c)
Note that the dispersion and stability conditions, as well as the
source injection in time, are handled just like the frst-order
coupled FDTD equations. On the other hand, the values at the
frst two time instants of
z
E (i.e., ( )
0
,
z
E i j and ( )
1
,
z
E i j ) must
be supplied for the spatial source injection.
2.3 Basic Features of the FDTD Equations
The observations listed below are important for the
numerical implementation of the frst-order coupled (FOC)
FDTD model:
There are three feld components (
x
H ,
y
H ,
z
E for
TM
z
and
x
E ,
y
E , and
z
H for TE
z
) in each cell,
and they are distinguished by the ( ) , i j label for the
frst-order coupled FDTD model.
The discretization steps are , x y , and t , and the
physical quantities are calculated from x i x = ,
y j y = , and t n t = .
Since the FDTD equations are iterative (i.e., open-
form solutions), they are conditionally stable. The
Courant stability condition, which states that the
time step cannot be arbitrarily specifed once the
spatial discretization is done, must be satisfed.
Although the same notations, ( ) ,
n
x
E i j and
( ) ,
n
z
H i j , are used, their locations are different in
the classical Yee cell [1] (see Figure 1), and there is
a half-time-step difference between the E and H
feld computation times. That is, the magnetic-feld
components are calculated at time steps 2 t t = ,
3 2 t , 5 2 t , ..., but the electric felds are calcu-
lated at time steps , 2 , 3 ,... t t t t = .
Figure 1. The Yee cells for the (a) TM
z
and (b) TE
z
prob-
lems.
Only neighboring magnetic-feld values and
( )
1
,
n
z
E i j

are required to update ( ) ,


n
z
E i j . Simi-
larly, neighboring electric-feld values and
( )
1
,
n
x
H i j

are required to update ( ) ,


n
x
H i j .
Both magnetic- and electric-feld components in any
cell may be moved to the origin by just cell
averaging. This is accomplished via
( ) ( ) ( ) , 0.5 , 1,
x x x
H i j H i j H i j = + + (

for mag-
netic felds, but four electric-feld components are
required for this purpose:
( ) ( ) ( ) , 0.25 , 1,
z z z
E i j E i j E i j = + +

( ) ( ) , 1 1, 1
z z
E i j E i j + + + + + (

.
Any object may be modeled by giving , , and
. Two of these, and , appear in the electric-
feld components, and the third, , appears in the
magnetic-feld components.
Three different and values may be assigned for
three electric-feld components, so that different
objects may be located within the Yee cell. Simi-
larly, different values may be given for H-feld
components

for the same purpose.
The important aspects of the second-order decoupled
(SOD) FDTD model are as follows:
There is only one feld component, and its location
may be anywhere in the unit cell.
The models and discrete equations are identical for
the TM
z
and TE
z
problems.
The past two values are needed in every cell.
FDTD iterations yield only
z
E ( TM
z
) or
z
H
( TE
z
). One therefore needs to write down another
discrete (Maxwell) equation for the other two com-
ponents, i.e.,
x
H ,
y
H

( TM
z
) or
x
E ,
y
E ( TE
z
).
2.4 Absorbing Boundary Conditions
To make it simple in this tutorial, the second-order MUR
terminations [22] are used. Table 2 lists equations that must
be satisfed along the boundaries (see Figure 2). The discrete
iteration equations will then be
At 0 x = (
x
N N = )
( ) ( )
1 1
1, 2,
n n
z z
E j E j
+
=
( ) ( )
1 1
2, 1,
n n
z z
c t x
E j E j
c t x
+
(
(
+ +
(

+

( ) ( )
2
2, 1,
n n
z z
x
E j E j
c t x
(
(
+ +
(

+

( )
( )
( ) ( ) ( )
2
2
2, 1 2 2, 2, 1
2
n n n
z z z
c t x
E j E j E j
y c t x
(

(
( + + +

( +

( )
( )
( ) ( ) ( )
2
2
1, 1 2 1, 1, 1
2
n n n
z z z
c t x
E j E j E j
y c t x
(

(
( + + +

( +

(13)
At
max
x X = (
x
N N = )
( ) ( )
1 1
, 1,
n n
z z
E N j E N j
+
=
( ) ( )
1 1
1, ,
n n
z z
c t x
E N j E N j
c t x
+
(
(
+ +
(

+

( ) ( )
2
1, ,
n n
z z
x
E N j E N j
c t x
(
(
+ +
(

+


( )
( )
( )
2
2
1, 1
2
n
z
c t x
E N j
y c t x
(


( + +

( +

( ) ( ) 2 1, 1, 1
n n
z z
E N j E N j
(
+

( )
( )
( ) ( ) ( )
2
2
, 1 2 , , 1
2
n n n
z z z
c t x
E N j E N j E N j
y c t x
(

(
( + + +

( +


(14)
Table 2. Differential equations for the second-order MUR
terminations.
( )
2 2 2
2 2
0
1
0, , 0
0 2
z
max
x
c
E y t
y Y x t c
t y
( =

+ =
(

(

( )
2 2 2
2 2
1
, , 0
0 2
max
z max
max
x X
c
E X y t
y Y x t c
t y
( =

+ =
(

(

( )
2 2 2
2 2
0
1
, 0, 0
0 2
z
max
y
c
E x t
x X y t c
t x
( =

+ =
(

(

( )
2 2 2
2 2
1
, , 0
0 2
max
z max
max
y Y
c
E x Y t
x X y t c
t x
( =

+ =
(

(

AP_Mag_Apr_2014_Final.indd 225 5/31/2014 3:51:25 PM
226 IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014
At 0 y = (
y
N N = )
( ) ( )
1 1
,1 , 2
n n
z z
E i E i
+
=
( ) ( )
1 1
, 2 ,1
n n
z z
c t y
E i E i
c t y
+
(
(
+ +
(

+

( ) ( )
2
, 2 ,1
n n
z z
y
E i E i
c t y
(
(
+ +
(

+

( )
( )
( ) ( ) ( )
2
2
1, 2 2 , 2 1, 2
2
n n n
z z z
c t y
E i E i E i
x c t y
(

(
( + + +

( +

( )
( )
( ) ( ) ( )
2
2
1,1 2 ,1 1,1
2
n n n
z z z
c t y
E i E i E i
x c t y
(

(
( + + +

( +

(15)
At
max
y Y = (
y
N N = )
( ) ( )
1 1
, , 1
n n
z z
E i N E i N
+
=
( ) ( )
1 1
, 1 ,
n n
z z
c t y
E i N E i N
c t y
+
(
(
+ +
(

+

( ) ( )
2
, 1 ,
n n
z z
y
E i N E i N
c t y
(
(
+ +
(

+


( )
( )
( )
2
2
1, 1
2
n
z
c t y
E i N
x c t y
(


( + +

( +

( ) ( ) 2 , 1 1, 1
n n
z z
E i N E i N
(
+

( )
( )
( ) ( ) ( )
2
2
1, 2 , 1,
2
n n n
z z z
c t y
E i N E i N E i N
x c t y
(

(
( + + +

( +


(16)
2.5 Parameter Selection in
FDTD Simulations
FDTD modeling and simulations are usually preferred
because of the ability for handling complex EM environments
and broadband behavior. Running simulations requires
parameter optimization. The spatial mesh sizes, x and y ,
the time step, t , the total simulation period (
max
T n t = ), the
source bandwidth, B, and the pulse duration are character istic
parameters that should be optimally selected prior to the
simulation [4].
FDTD simulations are generally performed in obtaining
the frequency characteristics of a given EM structure, for
example, the radiation characteristics or input impedance of an
antenna structure, the RCS behavior of a chosen target, the
transmission and/or refection characteristics of a microstrip
network, the propagation characteristics of a waveguide, the
resonance frequencies of a closed enclosure, the shielding
effectiveness of an aperture, etc. One therefore needs to start
with the frequency requirements (the minimum/maximum fre-
quency of interest,
min
f /
max
f , and the frequency resolution,
f ). The time-domain discrete simulation parameters ( x ,
y , t ,
max
T n t = , source bandwidth B, etc.) are then
accordingly specifed.
Suppose the problem was to fnd the frequency characteris-
tics of refections from a free-space/dielectric inter face, from
dc to 1 GHz with 50 MHz frequency steps. Starting from the
frequency-analysis requirements and sampling crite ria, the
parameter-optimization steps can be listed as follows:
Choose the source waveform with a duration that
contains the maximum frequency of interest.
According to the properties of the fast Fourier trans-
form (FFT), the maximum frequency deter mines the
minimum time step, i.e., ( ) 1 2
FFT max
t f = . This
is the hard limit for the frequency analysis. A 1 GHz
maximum frequency corresponds to a 0.5 ns
FFT
t

.
There are two important points in choosing the
maximum simulation (observation) time. First, the
frequency sensitivity, f , which determines the
observation time should be 1
max
f T = . Second,
the simulation should continue until all the tran-
sients are over. Therefore,
max
T is chosen to satisfy
both requirements. Since f was given as 10 MHz,
max
T was determined to be 100 ns. The number of
time steps, n, will then be 200. If all transients decay
after 200 time steps, then this will be enough for the
simulation time. If the structure under investigation
is some kind of resonant struc ture, which
Figure 2. The boundary cells used in MUR terminations.
corresponds to ringing effects in the time domain,
then a much longer observation period will be
required.
Two important issues in the time-domain simula-
tions are the Courant stability criteria and numeri cal
dispersion.
The spatial mesh sizes, x and y , are chosen
according to numerical dispersion requirements.
This is nothing but satisfying the Nyquist sampling
criteria in the spatial domain. The minimum wave-
length,
min
, must be sampled with at least two
sam ples, i.e., { } max , 2
min
x y . In practice, at
least 10
min
is required for acceptable results.
Depending on the problem at hand, as much as
100
min
to 120
min
may be required in order to
get rid of numerical-dispersion effects. Since
min

was 30 cm, 1 x y = = cm may be chosen if
30
min
is good enough for eliminating numerical-
dispersion effects.
The time step,
FDTD
t , may be directly chosen
from the Courant stability criteria. Since x y =
and this is equal to 1 cm, t may be chosen to be
( )
2 x c , where c is the speed of light. This gives
24 t ps. In general,
FDTD
t

is much less than
FFT
t , and

therefore
FDTD
t

is taken into account.
With this time step, the simulation time was
5000 n = .
3. Tests and Comparisons
Simple MATLAB codes were developed for the frst-order
coupled FDTD (FOC-FDTD) and the second-order decoupled
FDTD (SOD-FDTD) models, for both the TE and TM prob-
lems. Table 3 lists these codes and their explanations (visit
http://leventsevgi.net for these codes).

Tests with frst-order coupled FDTD and second-order
decoupled FDTD were done in terms of memory requirements
and computational times. Table 4 shows some numerical results
for these comparisons. As observed, the computational times
were of the same order, but the second-order decoupled FDTD
was slightly faster. Note that the classical loop philoso phy used
in MATLAB coding drastically slowed down the computation.
This means that the use of For/End loops had a signifcant
impact on the computation time (two loops almost doubled the
computation time of one loop). The frst-order coupled FDTD
had three loops, one inside the other, whereas one loop was used
for the second-order decoupled FDTD. The frst-order coupled
FDTD lasted roughly three times longer than the second-order
decoupled FDTD with the classical coding approach. Their
computational times were almost the same when the For/
End loops were removed. (For example, observe in the table
that simulations in a 400 400 FDTD area lasted 8.85 s and
8.56 s with the frst-order coupled FDTD and second-order
decoupled FDTD models, respectively. On the other hand,
these values were 523 s and 202 s, respectively, if the classical
For/End loops were used in the MATLAB codes). However,
the memory allocation of the second-order decoupled FDTD
was considerably higher than for the frst-order coupled FDTD,
because of the requirements of the two past time values of the
felds at every cell.
Table 3. The rst-order coupled FDTD and SOD_FDTD MATLAB codes.
FrstOrder_TM_FDTD_MUR.m
2D-FDTD MATLAB codes for TM problem under MUR terminations (
x
H ,
y
H ,
z
E )
FrstOrder_TM_FDTD_MUR_INH.m 2D-FDTD MATLAB codes for TM problem under MUR terminations (
x
H ,
y
H ,
z
E )
having a rectangular lossy layer
FrstOrder_TM_FDTD_MUR_MTM.m 2D-FDTD MATLAB codes for TM problem under MUR terminations (
x
H ,
y
H ,
z
E )
having a rectangular MTM layer
FrstOrder_TE_FDTD_PEC.m
2D-FDTD MATLAB codes for TE problem under PEC terminations (
x
E ,
y
E ,
z
H )
FrstOrder_TE_FDTD_MUR_MTM.m 2D-FDTD MATLAB codes for TE problem under MUR terminations (
x
E ,
y
E ,
z
H )
having a rectangular MTM layer
ScndOrder_TM_FDTD_MUR.m
2D Second order FDTD MATLAB codes for TM problem under MUR terminations
(
z
E )
ScndOrder_TM_FDTD_MUR_INH.m
2D Second order FDTD MATLAB codes for TM problem under MUR terminations
(
z
E ) having a rectangular lossy layer
ScndOrder_TE_FDTD_MUR.m
2D Second order FDTD MATLAB codes for TE problem under MUR terminations
(
z
H )
AP_Mag_Apr_2014_Final.indd 226 5/31/2014 3:51:26 PM
IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014 227
At 0 y = (
y
N N = )
( ) ( )
1 1
,1 , 2
n n
z z
E i E i
+
=
( ) ( )
1 1
, 2 ,1
n n
z z
c t y
E i E i
c t y
+
(
(
+ +
(

+

( ) ( )
2
, 2 ,1
n n
z z
y
E i E i
c t y
(
(
+ +
(

+

( )
( )
( ) ( ) ( )
2
2
1, 2 2 , 2 1, 2
2
n n n
z z z
c t y
E i E i E i
x c t y
(

(
( + + +

( +

( )
( )
( ) ( ) ( )
2
2
1,1 2 ,1 1,1
2
n n n
z z z
c t y
E i E i E i
x c t y
(

(
( + + +

( +

(15)
At
max
y Y = (
y
N N = )
( ) ( )
1 1
, , 1
n n
z z
E i N E i N
+
=
( ) ( )
1 1
, 1 ,
n n
z z
c t y
E i N E i N
c t y
+
(
(
+ +
(

+

( ) ( )
2
, 1 ,
n n
z z
y
E i N E i N
c t y
(
(
+ +
(

+


( )
( )
( )
2
2
1, 1
2
n
z
c t y
E i N
x c t y
(


( + +

( +

( ) ( ) 2 , 1 1, 1
n n
z z
E i N E i N
(
+

( )
( )
( ) ( ) ( )
2
2
1, 2 , 1,
2
n n n
z z z
c t y
E i N E i N E i N
x c t y
(

(
( + + +

( +


(16)
2.5 Parameter Selection in
FDTD Simulations
FDTD modeling and simulations are usually preferred
because of the ability for handling complex EM environments
and broadband behavior. Running simulations requires
parameter optimization. The spatial mesh sizes, x and y ,
the time step, t , the total simulation period (
max
T n t = ), the
source bandwidth, B, and the pulse duration are character istic
parameters that should be optimally selected prior to the
simulation [4].
FDTD simulations are generally performed in obtaining
the frequency characteristics of a given EM structure, for
example, the radiation characteristics or input impedance of an
antenna structure, the RCS behavior of a chosen target, the
transmission and/or refection characteristics of a microstrip
network, the propagation characteristics of a waveguide, the
resonance frequencies of a closed enclosure, the shielding
effectiveness of an aperture, etc. One therefore needs to start
with the frequency requirements (the minimum/maximum fre-
quency of interest,
min
f /
max
f , and the frequency resolution,
f ). The time-domain discrete simulation parameters ( x ,
y , t ,
max
T n t = , source bandwidth B, etc.) are then
accordingly specifed.
Suppose the problem was to fnd the frequency characteris-
tics of refections from a free-space/dielectric inter face, from
dc to 1 GHz with 50 MHz frequency steps. Starting from the
frequency-analysis requirements and sampling crite ria, the
parameter-optimization steps can be listed as follows:
Choose the source waveform with a duration that
contains the maximum frequency of interest.
According to the properties of the fast Fourier trans-
form (FFT), the maximum frequency deter mines the
minimum time step, i.e., ( ) 1 2
FFT max
t f = . This
is the hard limit for the frequency analysis. A 1 GHz
maximum frequency corresponds to a 0.5 ns
FFT
t

.
There are two important points in choosing the
maximum simulation (observation) time. First, the
frequency sensitivity, f , which determines the
observation time should be 1
max
f T = . Second,
the simulation should continue until all the tran-
sients are over. Therefore,
max
T is chosen to satisfy
both requirements. Since f was given as 10 MHz,
max
T was determined to be 100 ns. The number of
time steps, n, will then be 200. If all transients decay
after 200 time steps, then this will be enough for the
simulation time. If the structure under investigation
is some kind of resonant struc ture, which
Figure 2. The boundary cells used in MUR terminations.
corresponds to ringing effects in the time domain,
then a much longer observation period will be
required.
Two important issues in the time-domain simula-
tions are the Courant stability criteria and numeri cal
dispersion.
The spatial mesh sizes, x and y , are chosen
according to numerical dispersion requirements.
This is nothing but satisfying the Nyquist sampling
criteria in the spatial domain. The minimum wave-
length,
min
, must be sampled with at least two
sam ples, i.e., { } max , 2
min
x y . In practice, at
least 10
min
is required for acceptable results.
Depending on the problem at hand, as much as
100
min
to 120
min
may be required in order to
get rid of numerical-dispersion effects. Since
min

was 30 cm, 1 x y = = cm may be chosen if
30
min
is good enough for eliminating numerical-
dispersion effects.
The time step,
FDTD
t , may be directly chosen
from the Courant stability criteria. Since x y =
and this is equal to 1 cm, t may be chosen to be
( )
2 x c , where c is the speed of light. This gives
24 t ps. In general,
FDTD
t

is much less than
FFT
t , and

therefore
FDTD
t

is taken into account.
With this time step, the simulation time was
5000 n = .
3. Tests and Comparisons
Simple MATLAB codes were developed for the frst-order
coupled FDTD (FOC-FDTD) and the second-order decoupled
FDTD (SOD-FDTD) models, for both the TE and TM prob-
lems. Table 3 lists these codes and their explanations (visit
http://leventsevgi.net for these codes).

Tests with frst-order coupled FDTD and second-order
decoupled FDTD were done in terms of memory requirements
and computational times. Table 4 shows some numerical results
for these comparisons. As observed, the computational times
were of the same order, but the second-order decoupled FDTD
was slightly faster. Note that the classical loop philoso phy used
in MATLAB coding drastically slowed down the computation.
This means that the use of For/End loops had a signifcant
impact on the computation time (two loops almost doubled the
computation time of one loop). The frst-order coupled FDTD
had three loops, one inside the other, whereas one loop was used
for the second-order decoupled FDTD. The frst-order coupled
FDTD lasted roughly three times longer than the second-order
decoupled FDTD with the classical coding approach. Their
computational times were almost the same when the For/
End loops were removed. (For example, observe in the table
that simulations in a 400 400 FDTD area lasted 8.85 s and
8.56 s with the frst-order coupled FDTD and second-order
decoupled FDTD models, respectively. On the other hand,
these values were 523 s and 202 s, respectively, if the classical
For/End loops were used in the MATLAB codes). However,
the memory allocation of the second-order decoupled FDTD
was considerably higher than for the frst-order coupled FDTD,
because of the requirements of the two past time values of the
felds at every cell.
Table 3. The rst-order coupled FDTD and SOD_FDTD MATLAB codes.
FrstOrder_TM_FDTD_MUR.m
2D-FDTD MATLAB codes for TM problem under MUR terminations (
x
H ,
y
H ,
z
E )
FrstOrder_TM_FDTD_MUR_INH.m 2D-FDTD MATLAB codes for TM problem under MUR terminations (
x
H ,
y
H ,
z
E )
having a rectangular lossy layer
FrstOrder_TM_FDTD_MUR_MTM.m 2D-FDTD MATLAB codes for TM problem under MUR terminations (
x
H ,
y
H ,
z
E )
having a rectangular MTM layer
FrstOrder_TE_FDTD_PEC.m
2D-FDTD MATLAB codes for TE problem under PEC terminations (
x
E ,
y
E ,
z
H )
FrstOrder_TE_FDTD_MUR_MTM.m 2D-FDTD MATLAB codes for TE problem under MUR terminations (
x
E ,
y
E ,
z
H )
having a rectangular MTM layer
ScndOrder_TM_FDTD_MUR.m
2D Second order FDTD MATLAB codes for TM problem under MUR terminations
(
z
E )
ScndOrder_TM_FDTD_MUR_INH.m
2D Second order FDTD MATLAB codes for TM problem under MUR terminations
(
z
E ) having a rectangular lossy layer
ScndOrder_TE_FDTD_MUR.m
2D Second order FDTD MATLAB codes for TE problem under MUR terminations
(
z
H )
AP_Mag_Apr_2014_Final.indd 227 5/31/2014 3:51:26 PM
228 IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014
3.1 Current Computations
Alternating currents produce surrounding magnetic felds
and can be calculated using Amperes Law. This means cur-
rents can be extracted from known magnetic felds. Assume an
infnite thin wire is located in the simulation area (see Fig ure 3).
The current passing through this wire can be calculated using
Amperes Law:
I Hdl Hdl = =

( ) ( )
2 2
1 1
1 2
, ,
x y
x y
i x j y
H i y dx H x j dx
= =
= +

( ) ( )
2 2
1 1
2 1
, ,
x y
x y
i x j y
H i y dx H x j dx
= =



(17)
Short scripts were added to the frst-order coupled and
second-order decoupled FDTD codes for the computation of
a current fowing on a thin wire. The sample scenario prepared
for this purpose is pictured in Figure 4. At 300 MHz, a 25 m
25 m simulation area was assumed. A PEC rectangular object
(5 m 2.5 m) was placed at Node (200,225). A hard pulse line
source was injected from Node (350,200). An infnitely long,
thin wire was located in the simulation area. The number of
time steps was 500.
Figure 5 shows a current as a function of time compari son
of the frst-order coupled FDTD and second-order decoupled
FDTD models. The computational times are also given on the
graph. The wire was enclosed by a rectangle sized four cells
along the x direction and fve cells along the y direc tion. The
magnetic felds were calculated along this arbitrary loop. The
y
H components were considered for the right and left edges of
the rectangle, whereas the
x
H components were used for the
top and bottom edges. The sum of the magnetic felds was
multiplied by the cell size. Finally, the fowing cur rent was
obtained.
Table 4. The Time and memory requirements for the rst-
order coupled FDTD and second-order decoupled FDTD
models ( 500 n = , FOC:
x
H ,
y
H ,
z
E ; SOD:
z
E ).
Simulation
Area
Time (s) Memory (MB)
FOC SOD FOC SOD
300300 7.19 5.27 2.11 345
400400 8.85 8.56 3.71 920
600600 17.3 13.4 8.3 1384
10001000 47.48 37.48 22.9 3845
Figure 3. An application of Amperes Law.
Figure 4. A sample scenario for current simulations.
3.2 Voltage Computations
Potential differences cause electric felds. This means that
the voltage induced by EM felds between any terminals may
also be computed during the FDTD simulations. Faradays Law
may be used for this purpose. The necessary equation for the
voltage computations in the scenario shown in Figure 6 is

( )
2
1
,
y
z a
j y
V Edl E x j dy
=
= =

. (18)
Here, two parallel PEC plates were inserted horizontally, and
the voltage between them was simulated (see the scenario
in Figure 7a). Two thin parallel plates, of a size of 1 100
seg ments, were inserted into the simulation area. They were
sepa rated from each other by 50 cells. At a selected node (the
150th cell), the voltage value was computed. A hard pulse
source was then applied at node (250,250). Figure 7b shows
an instant (a screen capture) during the FDTD simulations. The
wave components marked 1, 2, 3, and 4 corresponded to the
cylindrical incident feld, refections from the top plate, the top-
edge diffracted felds, and the bottom-edge diffracted felds,
respectively. All the E felds from the 150th to the 200th cells
Figure 5. The current as a function of time obtained with
both models.
Figure 6. An application of Faradays Law.
along the y direction were added, and the sum was multiplied
by the cell size. A comparison of the voltage as a function of
time is given in Figure 8.
Note that the codes listed in the Appendix were used for
the scenario given in Figure 7, and produced the results in Fig-
ure 8. The codes in Appendices 6.1 and 6.2 generated the same
Figure 8 using the frst-order coupled FDTD and second-order
decoupled FDTD models, respectively. The code in Appendix
6.3 could be used after the other two. It loaded the recorded
FDTD data, applied the FFT, and compared the two models in
both the time and frequency domains. (All these codes and the
others may also be downloaded from leventsevgi.net.)
4. Metamaterial (MTM) Modeling
The EM response of a material is determined to a large
extent by its electrical properties. A material/medium with both
permittivity and permeability greater than zero ( 0 > , 0 > )
is called double positive (DPS). A medium with permit tivity
less than zero and permeability greater than zero ( 0 < ,
0 > ) is designated as -negative (ENG). Materials with
permittivity and permeability both less than zero ( 0 < ,
0 < ) are called double-negative (DNG) media.
Recent research on FDTD modeling has covered a hot
topic of artifcial materials. Russian scientists [24] came up
with the idea of negative-index materials in 1967 [25]. They put
negative-index materials into Maxwells equations and
investigated them. The electric feld ( E

), magnetic feld ( H

)
and the wave vector ( k

) originally obeyed the right-handed


rule ( E H k

). However, when metamaterials are consid-
ered, they realized that these materials satisfy the left-handed
rule. The opposite directions of the wave vector and the pointing
vector results in backwards wave propagation and also focusing.
When both the real parts of the permittivity and the permeability
are negative, the material is then called a double-negative
(DNG) media [11].
One of the many applications of metamaterials is in medi-
cal/optical imaging with high focusing capabilities. Another
application is antenna design. Antenna sizes may be reduced,
and higher-directivity gains can be achieved by using meta-
materials [11, 25].
Metamaterials can be characterized in terms of their disper-
sive properties [11, 15]. The virtual tools listed in Table 1 and
presented in [15] use the auxiliary differential-equation (ADE)
based FDTD approach [26]. In this approach, two of Maxwells
equations for the dispersive materials in the frequency domain
are given as
( ) ( ) ( ) ( ) ( )
0 e
D E E

= = + (

, (19a)
( ) ( ) ( ) ( ) ( )
0 m
B H H

= = + (

, (19b)
AP_Mag_Apr_2014_Final.indd 228 5/31/2014 3:51:27 PM
IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014 229
3.1 Current Computations
Alternating currents produce surrounding magnetic felds
and can be calculated using Amperes Law. This means cur-
rents can be extracted from known magnetic felds. Assume an
infnite thin wire is located in the simulation area (see Fig ure 3).
The current passing through this wire can be calculated using
Amperes Law:
I Hdl Hdl = =

( ) ( )
2 2
1 1
1 2
, ,
x y
x y
i x j y
H i y dx H x j dx
= =
= +

( ) ( )
2 2
1 1
2 1
, ,
x y
x y
i x j y
H i y dx H x j dx
= =



(17)
Short scripts were added to the frst-order coupled and
second-order decoupled FDTD codes for the computation of
a current fowing on a thin wire. The sample scenario prepared
for this purpose is pictured in Figure 4. At 300 MHz, a 25 m
25 m simulation area was assumed. A PEC rectangular object
(5 m 2.5 m) was placed at Node (200,225). A hard pulse line
source was injected from Node (350,200). An infnitely long,
thin wire was located in the simulation area. The number of
time steps was 500.
Figure 5 shows a current as a function of time compari son
of the frst-order coupled FDTD and second-order decoupled
FDTD models. The computational times are also given on the
graph. The wire was enclosed by a rectangle sized four cells
along the x direction and fve cells along the y direc tion. The
magnetic felds were calculated along this arbitrary loop. The
y
H components were considered for the right and left edges of
the rectangle, whereas the
x
H components were used for the
top and bottom edges. The sum of the magnetic felds was
multiplied by the cell size. Finally, the fowing cur rent was
obtained.
Table 4. The Time and memory requirements for the rst-
order coupled FDTD and second-order decoupled FDTD
models ( 500 n = , FOC:
x
H ,
y
H ,
z
E ; SOD:
z
E ).
Simulation
Area
Time (s) Memory (MB)
FOC SOD FOC SOD
300300 7.19 5.27 2.11 345
400400 8.85 8.56 3.71 920
600600 17.3 13.4 8.3 1384
10001000 47.48 37.48 22.9 3845
Figure 3. An application of Amperes Law.
Figure 4. A sample scenario for current simulations.
3.2 Voltage Computations
Potential differences cause electric felds. This means that
the voltage induced by EM felds between any terminals may
also be computed during the FDTD simulations. Faradays Law
may be used for this purpose. The necessary equation for the
voltage computations in the scenario shown in Figure 6 is

( )
2
1
,
y
z a
j y
V Edl E x j dy
=
= =

. (18)
Here, two parallel PEC plates were inserted horizontally, and
the voltage between them was simulated (see the scenario
in Figure 7a). Two thin parallel plates, of a size of 1 100
seg ments, were inserted into the simulation area. They were
sepa rated from each other by 50 cells. At a selected node (the
150th cell), the voltage value was computed. A hard pulse
source was then applied at node (250,250). Figure 7b shows
an instant (a screen capture) during the FDTD simulations. The
wave components marked 1, 2, 3, and 4 corresponded to the
cylindrical incident feld, refections from the top plate, the top-
edge diffracted felds, and the bottom-edge diffracted felds,
respectively. All the E felds from the 150th to the 200th cells
Figure 5. The current as a function of time obtained with
both models.
Figure 6. An application of Faradays Law.
along the y direction were added, and the sum was multiplied
by the cell size. A comparison of the voltage as a function of
time is given in Figure 8.
Note that the codes listed in the Appendix were used for
the scenario given in Figure 7, and produced the results in Fig-
ure 8. The codes in Appendices 6.1 and 6.2 generated the same
Figure 8 using the frst-order coupled FDTD and second-order
decoupled FDTD models, respectively. The code in Appendix
6.3 could be used after the other two. It loaded the recorded
FDTD data, applied the FFT, and compared the two models in
both the time and frequency domains. (All these codes and the
others may also be downloaded from leventsevgi.net.)
4. Metamaterial (MTM) Modeling
The EM response of a material is determined to a large
extent by its electrical properties. A material/medium with both
permittivity and permeability greater than zero ( 0 > , 0 > )
is called double positive (DPS). A medium with permit tivity
less than zero and permeability greater than zero ( 0 < ,
0 > ) is designated as -negative (ENG). Materials with
permittivity and permeability both less than zero ( 0 < ,
0 < ) are called double-negative (DNG) media.
Recent research on FDTD modeling has covered a hot
topic of artifcial materials. Russian scientists [24] came up
with the idea of negative-index materials in 1967 [25]. They put
negative-index materials into Maxwells equations and
investigated them. The electric feld ( E

), magnetic feld ( H

)
and the wave vector ( k

) originally obeyed the right-handed


rule ( E H k

). However, when metamaterials are consid-
ered, they realized that these materials satisfy the left-handed
rule. The opposite directions of the wave vector and the pointing
vector results in backwards wave propagation and also focusing.
When both the real parts of the permittivity and the permeability
are negative, the material is then called a double-negative
(DNG) media [11].
One of the many applications of metamaterials is in medi-
cal/optical imaging with high focusing capabilities. Another
application is antenna design. Antenna sizes may be reduced,
and higher-directivity gains can be achieved by using meta-
materials [11, 25].
Metamaterials can be characterized in terms of their disper-
sive properties [11, 15]. The virtual tools listed in Table 1 and
presented in [15] use the auxiliary differential-equation (ADE)
based FDTD approach [26]. In this approach, two of Maxwells
equations for the dispersive materials in the frequency domain
are given as
( ) ( ) ( ) ( ) ( )
0 e
D E E

= = + (

, (19a)
( ) ( ) ( ) ( ) ( )
0 m
B H H

= = + (

, (19b)
AP_Mag_Apr_2014_Final.indd 229 5/31/2014 3:51:27 PM
230 IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014
Figure 7b. A screen capture during the FDTD simulations.
Figure 7a. A voltage simulation scenario.
Figure 8. The voltage as a function of time between the PEC
plates.

where ( )
e
, ( )
m
are the electric and magnetic suscepti-
bilities, and

are the complex relative permittivities.


According to the Lorentz model, these susceptibility are given
as
( )
2
2
k
e
k
k k
k k
i





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

, (20a)
( )
2
2
k
m
k
k k
k k
i





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

. (20b)
The combination of Equations (16a) and (17a) yields
( ) ( )
0
2
2
k
k
k k
k k
D E
i

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

,
(21)
which may also be written as
( ) ( ) ( )
k
k
D E S

= +

, (22)
where

2
( ) ( )
2
k
k
k k
k k
S E
i





=
| | | |
+
| |
\ . \ .
, (23)
or
( ) ( ) ( ) ( )
2 2 2
2
k k k k k k k k k
S i S S E + + = .
(24)
The inverse Fourier transform of Equation (21) is


( ) ( )
( ) ( )
2
2 2
2
2
k k
k k k k k k k
S t S t
S t E t
t
t


+ + =

,
(25)
which is suitable for the FDTD discretization:

( )
( )
2 2
1
2
1
k k
n n
k k
k k
t
S S
t


+

=
+

( )
( )
( )
( )
2 2
1
1
1 1
k k
k k n n
k k
k k k k
t
t
S E
t t



+
+ +
+ +
, (26)


, ,
1
n n n
x y x y k
k
E D S

(
=
(
(

, (27)
( ) ( )
1
, ,
n n
x x
D i j D i j
+
=
( ) ( ) , , 1
n n
z z
t
H i j H i j
y

(
+

, (28a)
( ) ( )
1
, ,
n n
y y
D i j D i j
+
=
( ) ( ) , 1,
n n
z z
t
H i j H i j
x

. (28b)
Similarly; starting with Equations (16b) and (17b), one ends up
with the other set of equations:

( )
( )
2 2
1
2
1
k k
n n
k k
k k
t
S S
t


+

=
+


( )
( )
( )
( )
2 2
1
1
1 1
k k
k k n n
k k
k k k k
t
t
S H
t t



+
+ +
+ +
, (29)

1
n n n
z z k
k
H B S

(
=
(
(

, (30)
( ) ( )
1
, ,
n n
z z
B i j B i j
+
=
( ) ( ) , 1 ,
n n
x x
t
E i j E i j
y

(
+ +

( ) ( ) 1, ,
n n
y y
t
E i j E i j
x

(
+

.
(31)
The classical FDTD procedure is based on three-step
iterations: update E felds, update source and boundary condi-
tions, update H felds. Figure 9 pictures the classical time loop
of the FDTD procedure. Here, the source was injected to the
magnetic-feld component, but any other feld component may
also be used, depending on the type of the source (such as a line
source, a horizontal dipole, a vertical dipole, an array, etc.).
The Lorentzian metamaterial-FDTD time loop, which
is a little bit complicated, is pictured in Figure 10 and can be
described step by step as follows:
Use Equation (26) and update
kx
S and
ky
S using
the
x
E and
y
E components, respectively, inside the
metamaterial region. Remember, the two past time
Figure 9. The classical FDTD time loop.
Figure 10. The Lorentzian metamaterial (MTM) FDTD
time loop.
AP_Mag_Apr_2014_Final.indd 230 5/31/2014 3:51:27 PM
IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014 231
Figure 7b. A screen capture during the FDTD simulations.
Figure 7a. A voltage simulation scenario.
Figure 8. The voltage as a function of time between the PEC
plates.

where ( )
e
, ( )
m
are the electric and magnetic suscepti-
bilities, and

are the complex relative permittivities.


According to the Lorentz model, these susceptibility are given
as
( )
2
2
k
e
k
k k
k k
i





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

, (20a)
( )
2
2
k
m
k
k k
k k
i





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

. (20b)
The combination of Equations (16a) and (17a) yields
( ) ( )
0
2
2
k
k
k k
k k
D E
i

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

,
(21)
which may also be written as
( ) ( ) ( )
k
k
D E S

= +

, (22)
where

2
( ) ( )
2
k
k
k k
k k
S E
i





=
| | | |
+
| |
\ . \ .
, (23)
or
( ) ( ) ( ) ( )
2 2 2
2
k k k k k k k k k
S i S S E + + = .
(24)
The inverse Fourier transform of Equation (21) is


( ) ( )
( ) ( )
2
2 2
2
2
k k
k k k k k k k
S t S t
S t E t
t
t


+ + =

,
(25)
which is suitable for the FDTD discretization:

( )
( )
2 2
1
2
1
k k
n n
k k
k k
t
S S
t


+

=
+

( )
( )
( )
( )
2 2
1
1
1 1
k k
k k n n
k k
k k k k
t
t
S E
t t



+
+ +
+ +
, (26)


, ,
1
n n n
x y x y k
k
E D S

(
=
(
(

, (27)
( ) ( )
1
, ,
n n
x x
D i j D i j
+
=
( ) ( ) , , 1
n n
z z
t
H i j H i j
y

(
+

, (28a)
( ) ( )
1
, ,
n n
y y
D i j D i j
+
=
( ) ( ) , 1,
n n
z z
t
H i j H i j
x

. (28b)
Similarly; starting with Equations (16b) and (17b), one ends up
with the other set of equations:

( )
( )
2 2
1
2
1
k k
n n
k k
k k
t
S S
t


+

=
+


( )
( )
( )
( )
2 2
1
1
1 1
k k
k k n n
k k
k k k k
t
t
S H
t t



+
+ +
+ +
, (29)

1
n n n
z z k
k
H B S

(
=
(
(

, (30)
( ) ( )
1
, ,
n n
z z
B i j B i j
+
=
( ) ( ) , 1 ,
n n
x x
t
E i j E i j
y

(
+ +

( ) ( ) 1, ,
n n
y y
t
E i j E i j
x

(
+

.
(31)
The classical FDTD procedure is based on three-step
iterations: update E felds, update source and boundary condi-
tions, update H felds. Figure 9 pictures the classical time loop
of the FDTD procedure. Here, the source was injected to the
magnetic-feld component, but any other feld component may
also be used, depending on the type of the source (such as a line
source, a horizontal dipole, a vertical dipole, an array, etc.).
The Lorentzian metamaterial-FDTD time loop, which
is a little bit complicated, is pictured in Figure 10 and can be
described step by step as follows:
Use Equation (26) and update
kx
S and
ky
S using
the
x
E and
y
E components, respectively, inside the
metamaterial region. Remember, the two past time
Figure 9. The classical FDTD time loop.
Figure 10. The Lorentzian metamaterial (MTM) FDTD
time loop.
AP_Mag_Apr_2014_Final.indd 231 5/31/2014 3:51:28 PM
232 IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014
values and current
x
E (
y
E ) value are neces sary.
The
x
E and
y
E felds are then updated every where
as in the classical FDTD procedure.
Derive the electric fux densities,
x
D and
y
D ,
inside the metamaterial region using Equa tions (28a)
and (28b).
Use Equation (27) and update the
x
E and
y
E

com-
ponents inside the metamaterial region.
Repeat the same procedure for the magnetic-feld
and fux components using Equations (29) to (31).
Inject the source to the related feld component. The
source may be a pulsed line source, a short dipole,
an array, or a beam.
A directive beam was simulated here. A narrow band
(sinusoidal) beam was generated as explained in
Figure 11. First, determine the number of nodes that
produce line sources individually (it is better if this
is an odd number). The distance, , between these
nodes is stated. The selected cells are then excited
by the sinusoidal source ( )
0
sin 2 f t . The
amplitudes are changed using a suitable weighting
function
( )
2 2
exp x nT , where nT controls the
beamwidth of the source. The beam angle is also
controlled by shifting the nodes along the x and y
directions. The selected seven red nodes are shown
in Figure 11a, whereas the blue nodes show the
shifted beam. To achieve this, the x component (y
component) was multiplied with
2
cos a ( sin cos a a )
(see Figure 11b).
Finally, satisfy the MUR absorbing boundary condi-
tions for the
z
H component.
The two MATLAB codes (FrstOrder_TM_MUR_MTM.m
and FrstOrder_TE_MUR_MTM.m), available at http://
leventsevgi.net, were prepared for the investigation of wave
propagation inside a metamaterial layer. The example given in
Figure 12 was for the TE
z
problem. The metamate rial values
were taken from [15], and are given in Table 5. Here, a Gaussian
beam-type sinusoidal source was used. The angle of incidence
was 60. Figure 9a visualizes a sinusoidal array source
propagation in free space. A dielectric object was placed in
Figure 12b. A wave was propagated conforming to Snells law
of refraction. Figures 12c and 12d belonged to metamaterials
the refractive indices of which were negative. Backward wave
propagation and wave focusing were observed.
The next example given in Figure 13 belonged to wave
propagation through a metamaterial layer with 2 n = for the
TE
z
problem. Here, a Gaussian-beam-type sinusoidal source
was used. The angle of incidence was 90. Note that the beams
Figure 11. Beamforming using multiple line sources: (a) the
source nodes and beam directions; (b) beam angle con trol
with shifts along the x and y components.
in Figures 12 and 13 were formed with 25 line sources with
inter-element distances of 5 .
The last example belonged to wave propagation through a
metamaterial layer with 2 n = for the TM
z
problem (see
Figure 14). A single-node line source was used as an incident
feld. The E felds at the selected three nodes were recorded
during the FDTD simulations. Point 1 at node (200,150) and
Point 3 at node (200,350) were in the free-space regions before
and after the metamaterial layer, respectively. Point 2 at node
(200,250) was inside the metamaterial region. Figure 15 shows
the E feld as a function of time at these three nodes. As
observed, the once-differentiated Gaussian pulse reached
Point 1 frst, and then refections from the free-space/
metamaterial interface were observed. After these two, there
appeared a strong sinusoidal ringing. On the other hand, sine-
modulated signals were recorded at Points 2 and 3. Fre quency
spectra of the recorded signals at these three points are plotted
in Figure 16. The frequency spectra of the source is also given
in this fgure. Note that the FDTD parameters of the examples
presented here are listed in Table 6.
5. Conclusions
Two-dimensional FDTD modeling and simulation was
reviewed. MATLAB codes for the frst-order coupled and sec-
ond-order decoupled EM equations for both TM- and TE-type
problems were prepared. The current fowing on a thin wire
and the voltage between parallel plates were also calculated
during FDTD simulations. MATLAB codes for Lorentz-type
metamaterial models were developed. Lecturers who teach
courses such as Computational Electromagnetics, Finite-
Figure 12d Wave propagation in a metamaterial ( TE
z

problem, material parameters are given in Table 5).
Figure 12a. Wave propagation in free space ( TE
z
prob lem,
material parameters are given in Table 5).
Figure 12b. Wave propagation in a dielectric ( TE
z
prob-
lem, material parameters are given in Table 5).
Figure 12c Wave propagation in a metamaterial ( TE
z

problem, material parameters are given in Table 5).
Table 5. The metamaterial Lorentz parameters used in Figure 12.
Figure
k

k
( )

( )
k k

12a ( 1 n = )
1 0 0 1 0
12b ( 2 n = )
1 0 0 4 0
12c ( 1 n = )
1 0 7.535 MHz 1 48
12d ( 2 n = )
1 0 9.418 MHz 1 45
AP_Mag_Apr_2014_Final.indd 232 5/31/2014 3:51:28 PM
IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014 233
values and current
x
E (
y
E ) value are neces sary.
The
x
E and
y
E felds are then updated every where
as in the classical FDTD procedure.
Derive the electric fux densities,
x
D and
y
D ,
inside the metamaterial region using Equa tions (28a)
and (28b).
Use Equation (27) and update the
x
E and
y
E

com-
ponents inside the metamaterial region.
Repeat the same procedure for the magnetic-feld
and fux components using Equations (29) to (31).
Inject the source to the related feld component. The
source may be a pulsed line source, a short dipole,
an array, or a beam.
A directive beam was simulated here. A narrow band
(sinusoidal) beam was generated as explained in
Figure 11. First, determine the number of nodes that
produce line sources individually (it is better if this
is an odd number). The distance, , between these
nodes is stated. The selected cells are then excited
by the sinusoidal source ( )
0
sin 2 f t . The
amplitudes are changed using a suitable weighting
function
( )
2 2
exp x nT , where nT controls the
beamwidth of the source. The beam angle is also
controlled by shifting the nodes along the x and y
directions. The selected seven red nodes are shown
in Figure 11a, whereas the blue nodes show the
shifted beam. To achieve this, the x component (y
component) was multiplied with
2
cos a ( sin cos a a )
(see Figure 11b).
Finally, satisfy the MUR absorbing boundary condi-
tions for the
z
H component.
The two MATLAB codes (FrstOrder_TM_MUR_MTM.m
and FrstOrder_TE_MUR_MTM.m), available at http://
leventsevgi.net, were prepared for the investigation of wave
propagation inside a metamaterial layer. The example given in
Figure 12 was for the TE
z
problem. The metamate rial values
were taken from [15], and are given in Table 5. Here, a Gaussian
beam-type sinusoidal source was used. The angle of incidence
was 60. Figure 9a visualizes a sinusoidal array source
propagation in free space. A dielectric object was placed in
Figure 12b. A wave was propagated conforming to Snells law
of refraction. Figures 12c and 12d belonged to metamaterials
the refractive indices of which were negative. Backward wave
propagation and wave focusing were observed.
The next example given in Figure 13 belonged to wave
propagation through a metamaterial layer with 2 n = for the
TE
z
problem. Here, a Gaussian-beam-type sinusoidal source
was used. The angle of incidence was 90. Note that the beams
Figure 11. Beamforming using multiple line sources: (a) the
source nodes and beam directions; (b) beam angle con trol
with shifts along the x and y components.
in Figures 12 and 13 were formed with 25 line sources with
inter-element distances of 5 .
The last example belonged to wave propagation through a
metamaterial layer with 2 n = for the TM
z
problem (see
Figure 14). A single-node line source was used as an incident
feld. The E felds at the selected three nodes were recorded
during the FDTD simulations. Point 1 at node (200,150) and
Point 3 at node (200,350) were in the free-space regions before
and after the metamaterial layer, respectively. Point 2 at node
(200,250) was inside the metamaterial region. Figure 15 shows
the E feld as a function of time at these three nodes. As
observed, the once-differentiated Gaussian pulse reached
Point 1 frst, and then refections from the free-space/
metamaterial interface were observed. After these two, there
appeared a strong sinusoidal ringing. On the other hand, sine-
modulated signals were recorded at Points 2 and 3. Fre quency
spectra of the recorded signals at these three points are plotted
in Figure 16. The frequency spectra of the source is also given
in this fgure. Note that the FDTD parameters of the examples
presented here are listed in Table 6.
5. Conclusions
Two-dimensional FDTD modeling and simulation was
reviewed. MATLAB codes for the frst-order coupled and sec-
ond-order decoupled EM equations for both TM- and TE-type
problems were prepared. The current fowing on a thin wire
and the voltage between parallel plates were also calculated
during FDTD simulations. MATLAB codes for Lorentz-type
metamaterial models were developed. Lecturers who teach
courses such as Computational Electromagnetics, Finite-
Figure 12d Wave propagation in a metamaterial ( TE
z

problem, material parameters are given in Table 5).
Figure 12a. Wave propagation in free space ( TE
z
prob lem,
material parameters are given in Table 5).
Figure 12b. Wave propagation in a dielectric ( TE
z
prob-
lem, material parameters are given in Table 5).
Figure 12c Wave propagation in a metamaterial ( TE
z

problem, material parameters are given in Table 5).
Table 5. The metamaterial Lorentz parameters used in Figure 12.
Figure
k

k
( )

( )
k k

12a ( 1 n = )
1 0 0 1 0
12b ( 2 n = )
1 0 0 4 0
12c ( 1 n = )
1 0 7.535 MHz 1 48
12d ( 2 n = )
1 0 9.418 MHz 1 45
AP_Mag_Apr_2014_Final.indd 233 5/31/2014 3:51:28 PM
234 IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014
Figure 13. Wave propagation through a metamaterial layer
( TE
z
problem, metamaterial parameters were 1
k
= ,
0
k
= , 9.42
k
= MHz, 1

= , 45
k
= ).
Figure 16. The frequency spectra of the E-eld data that are
highlighted in Figure 14.
Figure 14. Wave propagation through a metamaterial layer
( TM
z
problem, metamaterial parameters were 1
k
= ,
0
k
= , 9.42
k
= MHz, 1

= , 45
k
= ).
Figure 15. The E eld as a function of time in three regions.
Difference Time-Domain Method, Numerical Modeling and
Simulation, etc., may download and use all the MATLAB codes
listed at http://leventsevgi.net.
Those of you who use the FDTD method in your research
are strongly advised to develop your own codes. This is cer tainly
required at some stage of graduate-level numerical mod eling
and simulation studies. Commercial packages are won derful;
use them as much as possible, but preferably for vali dation. You
can start with the simple codes discussed in this tutorial and
extend them into three dimensions. You may also need to add
important modules such as a perfectly matched layer (PML)
[21] and/or near-to-far-feld (NTFF) transforma tion [4], etc., in
order to deal with real-life, complex engi neering problems.
Table 6. The FDTD discretization parameters used in the examples.
FDTD Parameters
Figure 4
Current
Comparison
Figure 7
Voltage
Comparison
Figures 12
and 13
MTM-TE
Figure 14
MTM-TM
Cell Size
dx
0.049 m 0.049 m 2.49 m 2.49 m
dy
0.049 m 0.049 m 2.49m 2.49 m
Time step
dt
11.2 ns 11.2 ns 5.6 ns 5.6 ns
# of cells
Nx
500 500 400 400
Ny
500 500 400 400
# of time steps n 700 1000 1000 1000
6. Appendix
6.1 Code for Figures 7 and 8 Using
First-Order Coupled FDTD
The MATLAB code for Figures 7 and 8 using frst-order
coupled FDTD is given in Figure 17.
AP_Mag_Apr_2014_Final.indd 234 5/31/2014 3:51:29 PM
IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014 235
Figure 13. Wave propagation through a metamaterial layer
( TE
z
problem, metamaterial parameters were 1
k
= ,
0
k
= , 9.42
k
= MHz, 1

= , 45
k
= ).
Figure 16. The frequency spectra of the E-eld data that are
highlighted in Figure 14.
Figure 14. Wave propagation through a metamaterial layer
( TM
z
problem, metamaterial parameters were 1
k
= ,
0
k
= , 9.42
k
= MHz, 1

= , 45
k
= ).
Figure 15. The E eld as a function of time in three regions.
Difference Time-Domain Method, Numerical Modeling and
Simulation, etc., may download and use all the MATLAB codes
listed at http://leventsevgi.net.
Those of you who use the FDTD method in your research
are strongly advised to develop your own codes. This is cer tainly
required at some stage of graduate-level numerical mod eling
and simulation studies. Commercial packages are won derful;
use them as much as possible, but preferably for vali dation. You
can start with the simple codes discussed in this tutorial and
extend them into three dimensions. You may also need to add
important modules such as a perfectly matched layer (PML)
[21] and/or near-to-far-feld (NTFF) transforma tion [4], etc., in
order to deal with real-life, complex engi neering problems.
Table 6. The FDTD discretization parameters used in the examples.
FDTD Parameters
Figure 4
Current
Comparison
Figure 7
Voltage
Comparison
Figures 12
and 13
MTM-TE
Figure 14
MTM-TM
Cell Size
dx
0.049 m 0.049 m 2.49 m 2.49 m
dy
0.049 m 0.049 m 2.49m 2.49 m
Time step
dt
11.2 ns 11.2 ns 5.6 ns 5.6 ns
# of cells
Nx
500 500 400 400
Ny
500 500 400 400
# of time steps n 700 1000 1000 1000
6. Appendix
6.1 Code for Figures 7 and 8 Using
First-Order Coupled FDTD
The MATLAB code for Figures 7 and 8 using frst-order
coupled FDTD is given in Figure 17.
AP_Mag_Apr_2014_Final.indd 235 5/31/2014 3:51:29 PM
236 IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014
6.2 Code for Figures 7 and 8 Using
Second-Order Decoupled FDTD
The MATLAB code for Figures 7 and 8 using second-order decoupled FDTD is given in Figure 18.
AP_Mag_Apr_2014_Final.indd 236 5/31/2014 3:51:31 PM
IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014 237
6.2 Code for Figures 7 and 8 Using
Second-Order Decoupled FDTD
The MATLAB code for Figures 7 and 8 using second-order decoupled FDTD is given in Figure 18.
AP_Mag_Apr_2014_Final.indd 237 5/31/2014 3:51:32 PM
238 IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014
6.3 Code for Voltage as a Function of
Time and Frequency Calculations
The MATLAB code for calculations of voltage as a func-
tion of time and frequency is given in Figure 19.
7. References
1. K. S. Yee, Numerical Solution of Initial Boundary Value
Problems Involving Maxwells Equations in Isotropic Media,
IEEE Transactions on Antennas and Propagation, AP-14, 3,
1966, pp. 302-307.
2. K. A. Kunz and Raymond J. Luebbers, The Finite Differ ence
Time Domain Method for Electromagnetics, Boca Raton, FL,
CRC Press, 1993.
3. A. Tafove, Computational Electrodynamics: The nite-
Difference Time-Domain Method, Norwood, MA, Artech
House, 1995.
4. L. Sevgi, Complex Electromagnetic Problems and Numeri-
cal Simulation Approaches, New York, IEEE Press/John Wiley,
2003.
5. A. Tafove and S. C. Hagness, Computational Electrody-
namics: The Finite-Difference Time-Domain Method, Nor-
wood, MA, Artech House, 2005.
6. D. Sullivan, Electromagnetic Simulation Using the FDTD
Method, Second Edition, New York, IEEE Press/John Wiley,
2013.
7. J. B. Schneider, Understanding the FDTD Method, 2010,
(http://www.eecs.wsu.edu/~schneidj/ufdtd/).
8. A. Z. Elsherbeni and V. Demir, The Finite Difference Time
Domain Method for Electromagnetics: With MATLAB Simu-
lations, Rayleigh, NC, SciTech, 2009.
9. Finite-Difference Time-Domain Method, http://
en.wikipedia.org/wiki/Finite-difference_time-domain_method.
10. W. Yu, R. Mittra, T. Su, Y. Liu, and X. Yang, Parallel Finite-
Difference Time-Domain Method, Norwood, MA, Artech
House, 2006.
11. Y. Hao and R. Mittra, FDTD Modeling of Metamaterials:
Theory and Applications, Norwood, MA, Artech House, 2008.
12. L. Sevgi, Review of Discrete Solutions of Poisson,
Laplace, and Wave Equations, IEEE Antennas and Propaga-
tion Magazine, 50, 1, February 2008, pp. 246-254.
13. L. Sevgi and . Uluk, A MATLAB-Based Transmis-
sion-Line Virtual Tool: Finite-Difference Time-Domain
Refectometer, IEEE Antennas and Propagation Magazine,
48, 1, February 2006, pp. 141-145.
14. G. akr, M. akr, and L. Sevgi, A Multipurpose FDTD-
Based Two-Dimensional Electromagnetic Virtual Tool, IEEE
Antennas and Propagation Magazine, 48, 4, August 2006, pp.
142-151.
15. M. akr, G. akr, and L. Sevgi, A Two-dimensional
FDTD-Based Virtual Metamaterial Wave Interaction Visu-
alization Tool, IEEE Antennas and Propagation Magazine, 50,
3, June 2008, pp. 166-175.
16. M. A. Uslu and L. Sevgi, MATLAB-Based Virtual
Wedge Scattering Tool for the Comparison of High Frequency
Asymptotics and FDTD Method, ACES Journal on Applied
Computational Electromagnetics, 27, 9, September 2012, pp.
697-705.
17. M. akr and L. Sevgi, Path Planning and Image Seg-
mentation Using the FDTD Method, IEEE Antennas and
Propagation Magazine, 53, 2, April 2011, pp. 230-245.
18. G. akr, M. akr, and L. Sevgi, A Novel Virtual FDTD-
Based Microstrip Circuit Design and Analysis Tool, IEEE
Antennas and Propagation Magazine, 48, 6, December 2006,
pp. 161-173.
19. G. akr, M. akr, and L. Sevgi, Radar Cross Section
(RCS) Modeling and Simulation: Part II A Novel FDTD-
Based RCS Prediction Virtual Tool, IEEE Antennas and
Propagation Magazine, 50, 2, April 2008, pp. 81-94.
20. L. Sevgi, Electromagnetic Modeling and Simulation, New
York, IEEE Press/John Wiley, 2014.
21. J. P. Berenger, A Perfectly Matched Layer for the
Absorption of Electromagnetic Waves, J. Comput. Phys., 114,
1994, pp. 185-200.
22. G. Mur, Absorbing Boundary Conditions for the Finite
Difference Approximation of the Time-Domain Electromag-
netic-Field Equations, IEEE Transactions on Electromag netic
Compatibility, EMC-23, 4, 1981, pp. 377-382.
23. L. Sevgi, Guided Waves and Transverse Fields: Trans verse
to What? IEEE Antennas and Propagation Magazine, 50, 6,
December 2008, pp. 221-225.
24. V. G. Veselago and P. N. Lebedev The Electrodynamics
of Substances with Simultaneously Negative Values of Per-
mittivity and Permeability, Soviet Physics, 10, 4, January
1968, pp. 509-514.
25. T. J. Cui, D. R. Smith, and R. Liu (eds.), Metarmaterials:
Theory, Design and Applications, New York, Springer, USA,
2009.
26. T. Kashiwa and I. Fukai, A Treatment by FDTD Method
of Dispersive Characteristics Associated with Electronic
Polarization, Microwave and Optical Technology Letters, 3,
1990, pp. 203-205.
AP_Mag_Apr_2014_Final.indd 238 5/31/2014 3:51:33 PM
IEEE Antennas and Propagation Magazine, Vol. 56, No. 2, April 2014 239
6.3 Code for Voltage as a Function of
Time and Frequency Calculations
The MATLAB code for calculations of voltage as a func-
tion of time and frequency is given in Figure 19.
7. References
1. K. S. Yee, Numerical Solution of Initial Boundary Value
Problems Involving Maxwells Equations in Isotropic Media,
IEEE Transactions on Antennas and Propagation, AP-14, 3,
1966, pp. 302-307.
2. K. A. Kunz and Raymond J. Luebbers, The Finite Differ ence
Time Domain Method for Electromagnetics, Boca Raton, FL,
CRC Press, 1993.
3. A. Tafove, Computational Electrodynamics: The nite-
Difference Time-Domain Method, Norwood, MA, Artech
House, 1995.
4. L. Sevgi, Complex Electromagnetic Problems and Numeri-
cal Simulation Approaches, New York, IEEE Press/John Wiley,
2003.
5. A. Tafove and S. C. Hagness, Computational Electrody-
namics: The Finite-Difference Time-Domain Method, Nor-
wood, MA, Artech House, 2005.
6. D. Sullivan, Electromagnetic Simulation Using the FDTD
Method, Second Edition, New York, IEEE Press/John Wiley,
2013.
7. J. B. Schneider, Understanding the FDTD Method, 2010,
(http://www.eecs.wsu.edu/~schneidj/ufdtd/).
8. A. Z. Elsherbeni and V. Demir, The Finite Difference Time
Domain Method for Electromagnetics: With MATLAB Simu-
lations, Rayleigh, NC, SciTech, 2009.
9. Finite-Difference Time-Domain Method, http://
en.wikipedia.org/wiki/Finite-difference_time-domain_method.
10. W. Yu, R. Mittra, T. Su, Y. Liu, and X. Yang, Parallel Finite-
Difference Time-Domain Method, Norwood, MA, Artech
House, 2006.
11. Y. Hao and R. Mittra, FDTD Modeling of Metamaterials:
Theory and Applications, Norwood, MA, Artech House, 2008.
12. L. Sevgi, Review of Discrete Solutions of Poisson,
Laplace, and Wave Equations, IEEE Antennas and Propaga-
tion Magazine, 50, 1, February 2008, pp. 246-254.
13. L. Sevgi and . Uluk, A MATLAB-Based Transmis-
sion-Line Virtual Tool: Finite-Difference Time-Domain
Refectometer, IEEE Antennas and Propagation Magazine,
48, 1, February 2006, pp. 141-145.
14. G. akr, M. akr, and L. Sevgi, A Multipurpose FDTD-
Based Two-Dimensional Electromagnetic Virtual Tool, IEEE
Antennas and Propagation Magazine, 48, 4, August 2006, pp.
142-151.
15. M. akr, G. akr, and L. Sevgi, A Two-dimensional
FDTD-Based Virtual Metamaterial Wave Interaction Visu-
alization Tool, IEEE Antennas and Propagation Magazine, 50,
3, June 2008, pp. 166-175.
16. M. A. Uslu and L. Sevgi, MATLAB-Based Virtual
Wedge Scattering Tool for the Comparison of High Frequency
Asymptotics and FDTD Method, ACES Journal on Applied
Computational Electromagnetics, 27, 9, September 2012, pp.
697-705.
17. M. akr and L. Sevgi, Path Planning and Image Seg-
mentation Using the FDTD Method, IEEE Antennas and
Propagation Magazine, 53, 2, April 2011, pp. 230-245.
18. G. akr, M. akr, and L. Sevgi, A Novel Virtual FDTD-
Based Microstrip Circuit Design and Analysis Tool, IEEE
Antennas and Propagation Magazine, 48, 6, December 2006,
pp. 161-173.
19. G. akr, M. akr, and L. Sevgi, Radar Cross Section
(RCS) Modeling and Simulation: Part II A Novel FDTD-
Based RCS Prediction Virtual Tool, IEEE Antennas and
Propagation Magazine, 50, 2, April 2008, pp. 81-94.
20. L. Sevgi, Electromagnetic Modeling and Simulation, New
York, IEEE Press/John Wiley, 2014.
21. J. P. Berenger, A Perfectly Matched Layer for the
Absorption of Electromagnetic Waves, J. Comput. Phys., 114,
1994, pp. 185-200.
22. G. Mur, Absorbing Boundary Conditions for the Finite
Difference Approximation of the Time-Domain Electromag-
netic-Field Equations, IEEE Transactions on Electromag netic
Compatibility, EMC-23, 4, 1981, pp. 377-382.
23. L. Sevgi, Guided Waves and Transverse Fields: Trans verse
to What? IEEE Antennas and Propagation Magazine, 50, 6,
December 2008, pp. 221-225.
24. V. G. Veselago and P. N. Lebedev The Electrodynamics
of Substances with Simultaneously Negative Values of Per-
mittivity and Permeability, Soviet Physics, 10, 4, January
1968, pp. 509-514.
25. T. J. Cui, D. R. Smith, and R. Liu (eds.), Metarmaterials:
Theory, Design and Applications, New York, Springer, USA,
2009.
26. T. Kashiwa and I. Fukai, A Treatment by FDTD Method
of Dispersive Characteristics Associated with Electronic
Polarization, Microwave and Optical Technology Letters, 3,
1990, pp. 203-205.
AP_Mag_Apr_2014_Final.indd 239 5/31/2014 3:51:33 PM

Вам также может понравиться