You are on page 1of 13

RADIO SCIENCE, VOL. 38, NO. 1, 1005, doi:10.

1029/2001RS002595, 2003

FDTD-SUPML-ADE simulation for


Ground-Penetrating Radar modeling
Fayçal Rejiba, Christian Camerlynck, and Pierre Mechler
Department of Applied Geophysics, Université Pierre et Marie Curie, UMR 7619, Sisyphe, CNRS, Paris, France

Received 15 December 2001; revised 18 July 2002; accepted 11 September 2002; published 16 January 2003.
[1] GPR data simulations require above all an efficient forward modeling algorithm. A
Finite Difference Time Domain code is presented with Simplified Unsplit Perfect
Matched Layer that leads to insignificant reflections on border with few modifications of
the core algorithm. In addition, the Auxiliary Differential Equation method allows GPR
simulations including dispersive phenomena that could not be neglected at radar
frequencies. The numerical validation is done in a classic way for amplitude/time wave
propagation and SUPML, and using the nonconventional Dispersion Analysis technique
for physical dispersion behavior. Full 3-D and 2-D forward modeling results are
finally presented for lossy, dispersive and random media in GPR conditions. INDEX
TERMS: 0644 Electromagnetics: Numerical methods; 0689 Electromagnetics: Wave propagation (4275); 3210
Mathematical Geophysics: Modeling; 0684 Electromagnetics: Transient and time domain; KEYWORDS:
Ground-Penetrating Radar, Finite Difference-Time Domain, Perfect Matched Layer, Auxiliary Differential
Equation, physical dispersion
Citation: Rejiba, F., C. Camerlynck, and P. Mechler, FDTD-SUPML-ADE simulation for Ground-Penetrating Radar
modeling, Radio Sci., 38(1), 1005, doi:10.1029/2001RS002595, 2003.

1. Introduction an open media [Mur, 1981; Mei and Fang, 1992]. At


present time, Bérenger’s PML [Bérenger, 1994; Katz et
[2] Ground Penetrating Radar (GPR) is widely used in al., 1994] is by far the most interesting ABC. In this
shallow geophysical exploration, in civil engineering and paper, a simplified version of Unsplit PML is used
water research. It is also efficient for detecting buried [Sacks et al., 1995; Gedney, 1996; Sullivan, 1997;
objects such as land mines [Montoya and Smith, 1999], Teixeira et al., 1998].
or archeological structures and artifacts [Conyers and [5] In addition, frequency dependence of material
Goodman, 1997]. In favorable conditions, the transmis- properties cannot be dismissed, and we count several
sion of short electromagnetic pulses (few nanoseconds) realistic relaxation models for geological structures as the
allows the investigation up to several tens of meters for Cole Cole relation [Xu and McMechan, 1997], the
low loss geological materials. Jonscher relation [Hollender and Tillard, 1998] or a
[3] The simulation of GPR is generally achieved by visco-elastic analogy formulation [Carcione, 1996].
several techniques like Ray Tracing [Cai and McMe- Throughout this paper, the dispersion will refers to a
chan, 1995], the Moment Method [Tabbagh, 1995; simple Debye model, in addition to a static electrical
Vitebsky et al., 1997], the Spectral and Pseudo-Spectral conductivity term, knowing that at GPR frequencies the
Methods [Bitri and Grandjean, 1998; Carcione et al., electrical conductivity is frequency independent and real
1999; Qing Huo and Guo-Xin, 1999] and FDTD method [Lysne, 1983].
[Yee, 1966; Bourgeois and Smith, 1996; Wang and Tripp, [6] Four main methods are usually used in FDTD
1996; Roberts and Daniels, 1997; Teixeira et al., 1998; schemes to implement frequencies dependences: the
Bergmann et al., 1998; Chen and Huang, 1998; Gürel Z-transform method [Sullivan, 1992], the Auxiliary
and Ugur, 2000]. Differential Equation (ADE) method [Gandhi et al.,
[4] The FDTD method requires what is generally 1993], the Recursive Convolution method [Luebbers et
called Absorbing Boundary Condition (ABC) to simulate al., 1990; Luebbers and Hunsberger, 1992] and the
Piecewise Linear Recursive Convolution technique [Kel-
ley and Luebbers, 1996; Teixeira et al., 1998].
Copyright 2003 by the American Geophysical Union. [7] In this paper we showed the validity of a 3-D
0048-6604/03/2001RS002595 FDTD-SUPML-ADE implementation to simulate a
5-1
5-2 REJIBA ET AL.: FDTD-SUPML-ADE SIMULATION

Figure 1. Comparison between the 3-D free space analytical solution and the FDTD-Unsplit PML
formulation. The point source is a 2 GHz Ey derived gaussian pulse (DFG) measured after 150 time
steps in the vacuum. Model discretization dx = dy = dz = 1 cm, dt = 16 ps.

rffiffiffiffiffi
GPR acquisition in realistic media. In the followed ~¼ e0
algorithm, the classic PML exponential functions are E E ð1:1Þ
m0
replaced by simplified functions over the entire domain,
that are constant in the computation domain, and hyper- 1
~ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi
D D ð1:2Þ
bolic in the SUPML lossy domain: such an implementa- e0  m0
tion leads to an equivalent efficiency and avoids the
heavy implementation due to the PML/computation Then Maxwell’s equations are
domain interface. ~ ¼ c0 r  H
[8] The propagation scheme deals with H (magnetic jwD ð1:3Þ
field) and D (electric flux density) instead of the classic ~ ðwÞ ¼ e*r ðwÞE
D ~ ðwÞ ð1:4Þ
H and E (electric field) update technique, and allows an
easy implementation of the constitutive relation between ~
jwH ¼ c0 r  E ð1:5Þ
D and E by the Auxiliary Differential Equation method.
In addition to the numerical validations needed, we use The equivalent dielectric permittivity includes a real
in this paper, an original approach to validate the electric conductivity term as follows:
physical dispersion that consist in a Dispersion Analysis es  e1 s
generally used in acoustic wave propagation [He, 2000]. e*r ¼ e1 þ þ ð1:6Þ
1 þ j  w  t0 j  w  e0
where es, e1, c0, s, and w are respectively the static
2. Simplified-UPML Formulation permittivity, the high frequency permittivity, the vacuum
[9] We start with the following transformation [Sulli- velocity (m/s), the conductivity (S/m), and the pulsation
van, 1996]: (rad/s).
REJIBA ET AL.: FDTD-SUPML-ADE SIMULATION 5-3

Figure 2. Snapshots of a Ey front wave propagation during 150 time steps on a 60 cm segment
lying the 40 cm thickness PML border. A 1 GHz centered DFG Ey source is emitted. Model
discretization dx = dy = dz = 2 cm, dt = 33 ps.

[10] The FDTD formulation consists in a generaliza- [12] Concerning D ~ x (and Hx), some algebra due to the
tion of the Sullivan [1997] approach for the electro- equivalent integral operator jw1 term gives when applied a
magnetic field components using dt and dx, dy, dx for second order centered FDTD scheme:
time and spatial discretization. The time step is chosen 2 n1=2 n1=2
3
H ðiþ1=2; jþ1=2; kÞHz ðiþ1=2; j1=2; kÞ
according the Courant stability usually used for plane @Hz @Hy 4 z 
 , dy 5
wave propagation: @y @z Hy
n1=2
ðiþ1=2; j; kþ1=2ÞHy
n1=2
ðiþ1=2; j; k1=2Þ
1 dz
dt ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð1:7Þ
c 1=dx þ 1=dy2 þ 1=dz2
2 rot h
where c is the maximum velocity encountered in the IDn~ xði þ 1=2; j; kÞ ¼ IDn1
~ ði þ 1=2; j; kÞ
media. x

[11] The SUPML are characterized by an anisotropic þ dtc0 gi1 ði þ 1=2Þrot h


implementation for each direction. In what follows, D ~z
~
and Dx components are expressed for the x direction ~ nx ði þ 1=2; j; kÞ ¼ D
~ n1
D x ði þ 1=2; j; kÞ þ dtc0 rot h
from equations (1.3) to (1.5):
    þ dtc0 IDn~ ði þ 1=2; j; kÞ ð1:11Þ
~ sD ðxÞ 1 @Hz @Hy x

jwDx 1 þ ¼ c0  ð1:8Þ The specificity of the SUPML consists in simplified


jwe0 @y @z
 1   attenuation functions g(i,j,k)(1,2,3) and f(i,j,k)(1,2,3). Those
~ z 1 þ sD ðxÞ @Hy @Hx functions that shrink the field’s amplitude in the PML are
D ¼ c0  ð1:9Þ defined to restore the classical propagation in the
jwe0 @x @y
computation domain, and range between 0 and 1:
We use a staggered grid using a Yee [1966] scheme to
express the derivation operators. That explains the 1/2  
increment in the spatial coordinate and time index as 1 npml  i a
fncðiÞ ¼ ; i ¼ 1; . . . ; npml
~ z (and Hz):
follow for D b npml
~ nz ði; j; k þ 1=2Þ ¼ gi3 ðiÞD ~ n1 f i1 ðiÞ ¼ gi1 ðiÞ ¼ fncðiÞ
D z ði; j; k þ 1=2Þ þ gi2 ðiÞdtc0
0 n1=2 n1=2
1 1
Hy ðiþ1=2; j; kþ1=2ÞHy ði1=2; j; kþ1=2Þ gi2 ðiÞ ¼ f i2 ðiÞ ¼
 1 þ fncðiÞ
@ n1=2 dx
n1=2
A ð1:10Þ
Hx ði; jþ1=2; kþ1=2ÞHx ði; j1=2; kþ1=2Þ
dy 1  fncðiÞ
gi3 ðiÞ ¼ f i3 ðiÞ ¼ ð1:12Þ
~ y and Hy are developed in a similar manner.
D 1 þ fncðiÞ
5-4 REJIBA ET AL.: FDTD-SUPML-ADE SIMULATION

Figure 3. (a) r values in the calculation domain including PML response in the inlet. (b) r values in
the reference domain without PML influence. (c) Evaluation of percentage of reflection following
the (1.18) expression with a maximum average reflection of 0.8%, corresponding to the (Figure 2).

where npml is the number of layers in the SUPML region eikR


and (a, b) are real parameters. For other directions, G3D ¼ ð1:14Þ
4pR
equivalent formulation concerning the attenuation func-
tions and field equations update are easily expressed. The The shape of the current density is a y-polarized derived
numerical stability for the FDTD-SUPML scheme is gaussian pulse (DFG):
  "   ! #
guaranteed for b less than 3 [Sullivan, 1997], and its t t 2
validity is demonstrated by a simple comparison with the Jy ðtÞ ¼  exp  1 þ =2 ð1:15Þ
analytical S free space solution (Figure 1): dttp dttp
 
SðtÞ ¼ TF1 G3D *Jy ðwÞ ð1:13Þ where tP is the number of time step from the beginning of
the pulse to the first maximum.
TF1 is the inverse Fourier transform, and notation G3D [13] The geometrical model consist in dx = dy = dz = 1
is the Green function defined at the distance R from a point cm, dt = 16 ps, and the source is 2 GHz DFG pulse. The
source: source Jy is added in the FDTD scheme by the following
way:
REJIBA ET AL.: FDTD-SUPML-ADE SIMULATION 5-5

Figure 4. (a) Dispersion model (complex dielectric permittivity) used for the dispersion analysis.
It consist in a Debye
pffiffiffiffiffiffiffipole
ffi with es = 1, e1 = 3 and t0 = 0.1 ns. (b) Corresponding phase velocity
curve vphase ¼ 1= ereal .

~ ny ði; j þ 1=2; kÞ ¼ D
D ~ ny ði; j þ 1=2; kÞ þ Jy ðtÞ ð1:16Þ [15] To quantify the efficiency of the SUPML with
such a point source, we calculate, at each time step, the
mean of the average E ~ y (Figure 2) across the upper
 y of E
PML (60 cm  60 cm ! 30dx  30dy).
3. Evaluation of SUPML Xns Xns 
 y ðnpml; tÞ ¼ 1
E E
~ ðx; y; npml; tÞj
[14] PML is now well mastered, and even if SUPML 30 2 x¼1 y¼1 y
differs from the original formulation, it does not really ð1:17Þ
need any validation for plane wave propagation. Never- y
theless, for a point source emitted inside the computation A percentage of reflection is evaluated by comparing E
domain, difficulties arise due to the variation of the angle with what is obtained in a wider enough reference
of incidence inherent to the spherical-shape of the wave. domain in order to avoid PML interference:
We thus consider a 1 GHz Ey DFG point source emitted Ref ¼ 100
in vacuum (equivalent results are obtained with others
polarization). The simulation is made for a 2 cm isotropic  y ðnpml; tÞ
E E  ðnpml; tÞ
discretization with 40 cm of SUPML which is an optimal   domain y
reference calculation
 domain
Max E  y ðnpml; tÞ
reference domain
thickness regarding the pulse width, with the following
parameters (dx = dy = dz = 2 cm, dt = 33 ps). ð1:18Þb ¼ 5
5-6 REJIBA ET AL.: FDTD-SUPML-ADE SIMULATION

Figure 5. Both traces recorded at the probes during the simulation (the geometrical model is
presented at left). The source is a Ey 10 GHz DFG emitted in the air on the Debye-like half space
(Figure 4). Model discretization dx = dy = dz = 1 mm, dt = 1,6 ps.

For our example we observe less than 0.8% of reflection same way [Powers and Olhoeft, 1994; Xu and McMe-
(Figure 3). chan, 1997]).
[17] Some algebra between (1.4) and (1.6) using (1.19)
and (1.20) lead to the following expression for the E ~x
4. Dispersion and Attenuation With ADE component:
Method
[16] The FDTD scheme is used for D ~ and H update, by     n   
~ nþ1
a1 D ~ n1
~ x þ a3 D
þ a2 D ~ nx  b3E
 b2E ~ n1
this way, the ADE method looks to be particularly ~ nþ1
x x x
E ¼
adapted to take into account for physical dispersion in x
b1
general and so on for an equivalent Debye model. This
method simply consists in expressing the relation (1.21)
~ and the electric field
between the electric flux density D with
~ thanks to following equivalences:
E
e0 t0 e0
@A Anþ1  An a1 ¼ þ
jwA , ¼ ð1:19Þ 2dt dt2
@t dt
2  t0 e0
@ 2 A Anþ1  2An þ An1 a2 ¼ 
w2 A , 2 ¼ ð1:20Þ dt2
@t dt2 e0 t0 e0
a3 ¼  þ 2
~ or D
A is E ~ 2:dt dt
Let us consider for convenience the dielectric dispersive es e0 þ st0 t0 e1 e0 s
b1 ¼ þ þ
model (1.6) (refined model could be implemented in the 2dt dt2 2
REJIBA ET AL.: FDTD-SUPML-ADE SIMULATION 5-7

Figure 6. Magnitude for both traces (Figure 5): optimal accuracy on phase velocity is obtained
between 500 MHz (before 500 MHz the fourier transform is badly evaluated) and 1.5 GHz (after
1.5 GHz the spatial discretization, 2.5 cm, becomes too coarse).

2t0 e1 e0 [19] Rdt is calculated by cross correlation and led to


b2 ¼ 
dt2 superimposed traces (Figure 5), while the spectrum of
respective traces (Figure 6) is used to define the interval
es e0 þ st0 t0 e1 e0 s of frequencies where the interpretation can be done: a
b3 ¼  þ þ
2dt dt2 2 minimum frequency under which there is not enough
accuracy on phase values, and a maximum frequency
5. Validation by Dispersion Analysis above which the spatial discretization is no more
[18] The 3-D analytical validation for numerical and adapted. Rdp is evaluated thanks to the angle difference
physical dispersion is a tough problem for a point source f(w) between the traces, after having shifted the later
emitted inside the computation domain. For this reason, a trace by Rdt (Figure 7).
Dispersion Analysis approach is used. A 10 GHz DFG fðwÞ
pulse in the air and two probes are placed in a Debye Rdp ðf Þ ¼ ð1:22Þ
2pw
dispersive half space (Figure 4) with the following
The calculated phase velocity (Figure 8) is then
discretization properties: dx = dy = dz = 1 mm, dt =
1.6 ps. The probes are near enough to produce less than a D
difference angle of p, in an interval of frequencies VðwÞ ¼ ð1:23Þ
Rdt þ Rdp ðwÞ
corresponding to the representative spectral power of
the source. We know that the delay between the traces is
composed of a delay Rdt inherent to the distance D 6. Real Case Applications
between both probes, and a delay Rdp(w) that depends of
the interspectra of both traces measured at their [20] To illustrate the following implementation in a
respective probe [Mari et al., 1997]. nondispersive media, let us consider first a constant
5-8 REJIBA ET AL.: FDTD-SUPML-ADE SIMULATION

Figure 7. Angle difference f(w) calculated before Rdp evaluation. The range of frequency where
the difference is calculated is determined thanks to the magnitude curve (Figure 6).

offset acquisition with a 225 MHz Pulse Ekko radar on a 1. The reflections are less marked in the material-3
20/40 gravel wall buried in clay (Figure 9), with follow- because the peak frequency is situated at a maximum of
ing equivalent electromagnetic parameters (the dielectric the imaginary part of the dielectric permittivity (Figure
permittivity are obtained by pointing the respective 11), and arrive earlier because the real permittivity at that
arrivals on the real radargram): top soil: e = 10, m = m0, frequency is lower (Figure 11) than in material-4.
s = 0.02 S/m; clay: e = 16, m = m0, s = 0.1 S/m; gravel 2. Top and bottom reflections for the water filled
wall: e = 2.25, m = m0, s = 0.005 S/m. pipes are well distinguished thanks to a pronounced
[21] A 3-D FDTD simulation (dx = dy = dz = 2.5 cm, contrast of permittivity (the water acts like a low filter
dt = 40 ps) is done on two perpendicular profiles, and because the spatial discretization is no longer adapted for
shows the validity of the approach. It requires heavy such a permittivity).
computational resources for this simple case: the profiles 3. The reflections on both interfaces between materi-
(Figure 10) calculated in a 200  200  140 domain al-1/material-2 and material-2/material-3 or 4 are well
during 550 time steps require about 20 hours of con- defined. Furthermore, we can see the slight separation
nection on a CRAY SV1 working with four 300 MHz between material-3 and 4, due to their respective
CPUs. characteristic time.
[22] Here a 2-D example on a realistic problem, con- [23] That simulation had required 6h30 on CRAY SV1
cerning pipes detection in distinct dispersive Debye supercomputer with four 300MHz CPUs.
media, for a constant offset acquisition (offset = 1 cm,
spacing = 20 cm) with 100 MHz bistatic antennas
(Figure 11). The TM mode simulation (point source
7. Conclusion
Ez) (Figure 12) is done on a 200  200 points domain [24] The FDTD-SUPML-ADE has shown to be well
(dx = dy = 10 cm) during 1000  dt = 160 ns (dt = 0.16 adapted to simulate GPR acquisition, even on highly
ns). A random distribution in material-2 is defined to add heterogeneous media. The possibilities for 3-D GPR
a realistic noise in the synthetic radargram. simulations are obvious, even if large computational
REJIBA ET AL.: FDTD-SUPML-ADE SIMULATION 5-9

Figure 8. Comparison between the analytical expression of phase velocity from the dispersive
model (Figure 4) and the dispersion analysis results, on a wide enough frequency interval (several
simulations are done). The difference between the analytical (continuous line) and the calculated
phase velocity is due to numerical dispersion (about 5% for that simulation).

resources are needed for a somewhat restricted volume of Bitri, A., and G. Grandjean, Frequency-wavenumber modeling
interest. Furthermore, the validation by a Dispersion and migration of 2D GPR data in moderately heterogeneous
Analysis has showed to be an original alternative to the dispersive media, Geophysics, 46, 287 – 301, 1998.
analytical procedure commonly used for plane wave Bourgeois, J. M., and G. S. Smith, A fully three-dimensional
source. simulation of a ground-penetrating radar FDTD theory com-
pared with experiment, IEEE Trans. Geosci. Remote Sens.,
[25] Acknowledgments We wish to thank Albane Sainte- 34, 36 – 44, 1996.
noy (Département des Sciences de la Terre, Université PARIS Cai, J., and G. A. McMechan, Ray-based synthesis of bistatic
XI) for useful suggestions regarding this paper. All computer ground-penetrating radar profiles, Geophysics, 60, 87 – 96,
time was provided by Centre de Calcul Recherche et Réseau 1995.
Jussieu, Université PARIS VI. Carcione, J. M., Ground-penetrating radar, wave theory and
numerical simulation in lossy anisotropic media, Geophy-
sics, 61, 1664 – 1677, 1996.
References
Carcione, J. M., G. Lenzi, and S. Valle, GPR modelling by the
Bérenger, J. P., A perfect matched layer for the absorption of Fourier method: Improvement of the algorithm, Geophys.
electromagnetic waves, J. Comput. Phys., 114, 185 – 200, Prospect., 47, 1015 – 1029, 1999.
1994. Chen, H. W., and T. M. Huang, Finite-difference time-domain
Bergmann, T., J. O. A. Robertsson, and K. Holliger, Finite- simulation of GPR data, J. Appl. Geophys., 40, 139 – 163,
difference modeling of electromagnetic wave propagation 1998.
in dispersive and attenuating media, Geophysics, 63, 856 – Conyers, L. B., and D. Goodman, Ground Penetrating Radar:
867, 1998. An Introduction for Archaeologists, AltaMira, London, 1997.
5 - 10 REJIBA ET AL.: FDTD-SUPML-ADE SIMULATION

Grandjean, G., et al., Evaluation of GPR techniques for civil-


engineering applications: Study on a test site, J. Appl. Geo-
phys., 45, 141 – 156, 2000.
Gürel, L., and O. Ugur, Three-dimensional FDTD modeling of
a ground-penetrating radar, IEEE Trans. Geosci. Remote
Sens., 38, 1513 – 1521, 2000.
He, P., Measurement of acoustic dispersion using both trans-
mitted and reflected pulse, J. Acoust. Soc. Am., 107, 801 –
807, 2000.

Figure 9. 225 MHz pulse Ekko radar 3-D acquisition


on a gravel wall buried in clay with 50 cm between the
antennas and a spacing of 5 cm along each profile. Some
sections are presented in order to situate the wall and was
used to build an equivalent model (Figure 10) from trial
and errors forward modeling.

Gandhi, O. P., B. Q. Gao, and J. Y. Chen, A frequency depen- Figure 10. On top, the equivalent geometrical model
dent finite difference time domain formulation for general (Top soil: e = 10, m = m0, s = 0.02 S/m, Clay: e = 16, m =
dispersive media, IEEE Trans. Microwave Theory Tech., 41, m0, s = 0.1 S/m, Gravel wall: e = 2.25, m = m0, s = 0.005
792 – 797, 1993. S/m). On bottom, the 3-D FDTD Unsplit PML simula-
Gedney, S. D., An anisotropic PML absorbing media for the tion on two profiles, with a 225 MHz DFG Ey emitter-
FDTD simulation of fields in lossy and dispersive media, receiver. Model discretization dx = dy = dz = 2.5 cm, dt =
Electromagnetic, 16, 399 – 415, 1996. 40 ps.
REJIBA ET AL.: FDTD-SUPML-ADE SIMULATION 5 - 11

Figure 11. Debye model for material 3 and 4. For a 100 MHz peak frequency the equivalent soil
properties are: ereal = 2.5, eimaginary = 13 for material 3 (red line) and ereal = 0, eimaginary = 15 for
material 4 (blue line).

Hollender, F., and S. Tillard, Modeling ground-penetrating ra- Mari, J. L., F. Glangeaud, and F. Coppens, Traitement du signal
dar wave propagation and reflection with the Jonsher para- pour géologues et géophysiciens, editions technip, 460 pp.,
meterization, Geophysics, 62, 1933 – 1942, 1998. Inst. Fr. du Petrole, Rueil-Malmaison, France, 1997.
Katz, D. S., E. T. Thiele, and A. Taflove, Validation and exten- Mei, K. K., and J. Fang, Superabsorption—A method to im-
sion to three dimensions of the Berenger PML absorbing prove absorbing boundary conditions, IEEE Trans. Antennas
boundary condition for FD-TD meshes, IEEE Microwave Propag., 40, 1001 – 1010, 1992.
Guided Wave Lett., 4, 268 – 270, 1994. Montoya, T. P., and G. S. Smith, Land mine detection using a
Kelley, D. F., and R. J. Luebbers, Piecewise linear recursive ground penetrating radar based on resistively loaded Vee di-
convolution for dispersive media using FDTD, IEEE Trans. poles, IEEE Trans. Antennas Propag., 47, 1795 – 1806, 1999.
Antennas Propag., 44, 792 – 797, 1996. Mur, G., Absorbing boundary conditions for wave-like equa-
Kunz, K. S., and R. J. Luebbers, The Finite Difference Time tions, IEEE Trans. Electromagn. Compat., 23, 377 – 382,
Domain for Electromagnetic, CRC Press, Boca Raton, Fla., 1981.
1993. Powers, M. H., and G. R. Olhoeft, Modeling dispersive ground
Luebbers, R. J., and F. Hunsberger, FDTD for N-th order dis- penetrating radar data, paper presented at 5th International
persive media, IEEE Trans. Antennas Propag., 40, 1297 – Conference on Ground Penetrating Radar, Kitchener, Ont.,
1301, 1992. Canada, 12 – 16 June 1994.
Luebbers, R. J., F. Hunsberger, K. S. Katz, R. B. Standler, and Qing Huo, L., and F. Guo-Xin, Simulations of GPR in disper-
M. Schneider, A frequency-dependent finite-difference time- sive media using a frequency dependant PSTD algorithm,
domain formulation for dispersive materials, IEEE Trans. IEEE Trans. Geosci. Remote Sens., 37, 2317 – 2324, 1999.
Electromagn. Compat., 32, 222 – 227, 1990. Roberts, R. L., and J. J. Daniels, Modeling near-field GPR in
Lysne, P. C., A model for the high-frequency electrical response three dimensions using the FDTD method, Geophysics, 62,
of wet rocks, Geophysics, 48, 775 – 786, 1983. 1114 – 1126, 1997.
5 - 12 REJIBA ET AL.: FDTD-SUPML-ADE SIMULATION

Figure 12. Model including several types of pipes used for 2-D, TM mode simulation. The source
is a Ez 100 MHz DFG, (offset = 1 m, spacing = 5 cm). Model discretization: dx = dy = 2 cm, dt =
33 ps.

Sacks, Z. S., D. M. Kingsland, R. Lee, and J. F. Le, A perfectly Sullivan, D. M., An unsplit step 3D PML for use with the
matched anisotropic absorber for use as an absorbing bound- FDTD method, IEEE Microwave Guided Wave Lett., 7,
ary condition, IEEE Trans. Antennas Propag., 43, 1460 – 184 – 186, 1997.
1463, 1995. Tabbagh, A., The response of a three dimensionnal magnetic
Sullivan, D. M., Frequency-dependent FDTD metods using Z and conductive body in shallow depth electromagnetic pro-
transforms, IEEE Trans. Antennas Propag., 40, 1223 – 1230, specting, Geophys. J. R. Astron. Soc., 81, 215 – 230, 1995.
1992. Taflove, A., Computational Electromagnetic: The Finite Differ-
Sullivan, D. M., A simplified PML for use with the FDTD meth- ence Time Domain Method, Artech House, Norwood, Mass.,
od, IEEE Microwave Guided Wave Lett., l6, 97 – 99, 1996. 1995.
REJIBA ET AL.: FDTD-SUPML-ADE SIMULATION 5 - 13

Teixeira, F. L., W. C. Chew, M. Straka, M. L. Oristaglio, and Xu, T., and G. A. McMechan, GPR attenuation and its numer-
T. Wang, Finite difference time domain simulation of ground ical simulation in 2.5D dimensions, Geophysics, 62, 403 –
penetrating radar on dispersive, inhomogeneous and conduc- 414, 1997.
tive soils, IEEE Trans. Geosci. Remote Sens., 36, 1928 – Yee, K. S., Numerical solution of initial boundary value pro-
1937, 1998. blems involving Maxwell’s equation in isotropic media,
Vitebsky, S., L. Carin, M. A. Ressler, and F. H. Le, Ultra-wide- IEEE Trans. Antennas Propag., 14, 302 – 307, 1966.
band, short-pulse ground penetrating radar: Simulation and
measurement, IEEE Trans. Geosci. Remote Sens., 35, 762 – 
772, 1997. C. Camerlynck, P. Mechler, and F. Rejiba, Department of
Wang, T., and A. C. Tripp, FDTD simulation of EM wave Applied Geophysics, Université Pierre et Marie Curie, UMR
propagation in 3-D media, Geophysics, 61, 1097 – 1106, 7619, Sisyphe, CNRS, Case 105, 4 Pl. Jussieu, 75252 Paris
1996. Cedex 05, France. (rejiba@ccr.jussieu.fr)

You might also like