Professional Documents
Culture Documents
AbstractA novel two-way finite-element parabolic equation a good understanding of electromagnetic (EM) wave behav-
(PE) (2W-FEMPE) propagation model which handles both for- ior in the presence of irregular terrain and inhomogeneous
ward and backward scattering effects of the groundwave propaga- atmosphere. Therefore, a generally applicable all-purpose prop-
tion above the Earths surface over irregular terrain paths through
inhomogeneous atmosphere is introduced. A Matlab-based propa- agation prediction method/tool must be able to predict the
gation tool for 2W-FEMPE is developed and tested against math- effect of such complex environmental factors/uncertainties on
ematical exact and asymptotic solutions as well as the recently the performance of radio communication and radar systems
introduced two-way split-step PE model through a canonical val- [1]. However, since the complexity of the realistic propagation
idation, verification, and calibration process for the first time in scenarios may place limitations in the range of validity of
literature.
the prediction models/tools, the fidelity of the models to the
Index TermsAtmospheric refractivity, Claerbout equation, underlying physics must be assessed by means of, for example,
ducting, electromagnetic (EM) propagation, fast Fourier trans- validation, verification, and calibration (VV&C) tests. This
form (FFT), finite-element method (FEM), Matlab, narrow angle,
split-step parabolic equation (PE) (SSPE), terrain effect, wave process of determining whether the right model is built (or
equation, wide angle. solving the right equations) is called validation, whereas
the verification assessment examines if the model is built
I. I NTRODUCTION right. The final step is the calibration, which is the process
of quantitatively defining the performance of the model with
differential equations, both of which are in PWE form, belong- ing the direction of propagation in the standard PE method
ing to forward and backward propagating waves. In the standard in a forwardbackward manner. Recently, a two-way SSPE
PE method, only the PWE corresponding to the forward part (2W-SSPE) has been proposed to account for multiple-
is solved, which makes the method a one-way forward-scatter reflection effects over arbitrary staircase-approximated irregu-
model. The standard PE method has been first introduced by lar terrain profiles [36] and then calibrated by means of various
Leontovich and Fock [2]. However, the development of the systematic tests and comparisons [37], [38]. In this paper, the
Fourier split-step algorithm by Hardin and Tappert [3] has two-way finite-element PE (2W-FEMPE) propagation tool is
initiated the wide-spread usage of the method in various prop- presented and calibrated in certain canonical and/or complex
agation scenarios. Generally speaking, the Fourier split-step scenarios for the first time in literature. The VV&C of the
PE (SSPE) algorithm is a marching-type initial-value problem, 2W-FEMPE model is performed against the analytical data, the
which computes an initial field from a known antenna pattern Image Method and the Geometric Optic (GO) + Asymptotic
located at a reference range and determines the field along Theory of Diffraction [39][41], as well as the recently pro-
the vertical direction at each range step by using the field posed 2W-SSPE model [36][38]. Narrow- and wide-angle PE
at the preceding range step through the utilization of Fourier models are also discussed here, for the sake of completeness.
transformations between transverse spatial and wavenumber It is generally accepted that the simplest conceptual radiowave
domains. propagation model is GO. The EM field is locally a plane wave
The step-by-step solution of the standard PWE has also in a medium that changes slowly, and wave energy propagates
been achieved by employing finite-difference methods [8][11] along rays that are perpendicular to the wavefronts. Snells
and finite-element methods (FEMs) [18][22] along each verti- law can be applied repeatedly to find out ray trajectories in
cal direction. Although the SSPE algorithm is a more robust this medium. Unfortunately, GO fails near caustics and it also
and faster algorithm and can handle larger range increments ignores the effects of diffraction. The PE model is based on
(in terms of wavelength), the major advantage of the finite parabolic approximation of the wave equation which admits
methods, over the SSPE, is the easiness in handling arbitrary a very feasible numerical marching solution. The PE model
boundary conditions (BCs). In particular, the FEM has been contains information missing in the GO model such as the
widely used in the EM modeling due to its adaptability to behavior near caustics and shadow boundaries. Also, the PE
arbitrary geometries and material inhomogeneities. Therefore, model is frequency dependent, but GO is frequency indepen-
to make a compromise, the finite methods can be preferred in dent. The GO needs precise calculation of optical ray paths and
complex propagation environments involving different surface coherent integration of the family of rays passing through the
impedances [20][22]. same observation point.
The standard PWE is a narrow-angle approximation and This paper is organized as follows: Section II summarizes
restricts the accuracy to propagation angles up to 10 15 the standard one-way PE method and its implementation via
from the paraxial direction [9]. Usually, this is not a serious the Fourier SSPE algorithm. Section III introduces the deriva-
restriction since most of the long-range propagation scenarios tions of the FEMPE model together with the two-way algo-
encounter propagation angles that are usually less than a few rithm. Canonical tests, comparisons, and calibration results are
degrees. On the other hand, short-range propagation problems demonstrated in Section IV. Finally, Section V draws some
as well as the problems involving multiple reflections and conclusions.
diffractions because of hills and valleys with steep slopes ne- Throughout this paper, the suppressed time dependence of
cessitate models that are effective for larger propagation angles. the form exp(it) is assumed.
Wide-angle PE propagators have been introduced to handle
propagation angles up to 40 45 [11], [23][26].
All of the aforementioned approaches offer one-way prop- II. PE M ODEL
agation modeling and neglect the backward propagation and The scalar wave function (x, z) can be obtained as the
multipath effects. Ignoring backward waves may not be serious solution of the following 2-D Helmholtz equation:
for long-range propagation problems, where the energy carried 2
by the forward propagating waves dominates. However, obsta- 2 2 2
+ 2 + k0 n (x, z) (x, z) = 0 (1)
cles and/or irregular terrain along the propagation path neces- x2 z
sitate the well prediction of the backward reflected, refracted,
and diffracted waves to achieve reliable results. The two-way subject to given BCs. Here, n(x, z) is the refractive index,
versions of the SSPE algorithm have also been introduced for k0 = 2/ is the free space wavenumber ( is the wave-
different applications and with implementation schemes [9], length), and z and x stand for the transverse (height above
[27][38]; pioneer works have been studied in [27] and [28] ground) and longitudinal (range) coordinates, respectively. Fur-
on the construction of analytical solutions to the integral and thermore, (x, z) corresponds to either electric or magnetic
functional equations based on two-way PE (2W-PE) providing field components for horizontal and vertical polarizations. Note
the exact solution for the current on a perfectly conducting strip that some researchers use the terms perpendicular and parallel
(refer to [29][31] for details), in acoustic wave propagation in polarizations, whereas others prefer horizontal or vertical polar-
elastic media [32], then in multiple scattering modeling [33], izations. Here, we assume an xz plane as the 2-D environment.
in multiple diffraction modeling [34], and in multiple knife- The TM (vertical polarization) and TE (horizontal polarization)
edge modeling [35]. The underlying idea, in all these, is switch- equations use (Ex , Hy , Ez ) and (Hx , Ey , Hz ), respectively
APAYDIN et al.: TWO-WAY FINITE-ELEMENT PARABOLIC EQUATION GROUNDWAVE PROPAGATION TOOL 2889
where Q = (1 + q)1/2 and q = k02 2 /z 2 + (n2 1) if the In general, a Cauchy-type impedance BC (i.e., lossy Earths
refractive index is independent of range.1 The first and sec- surface),
which is introduced via = ik 0 1 and =
ond parts of (2) correspond to the forward and backward ik0 1/ for horizontal and vertical polarizations, respec-
propagating waves, respectively (see Fig. 1). If the backward tively. Here, = r + i60 is computed in terms of the con-
propagation is ignored, (2) reduces to ductivity () and relative permittivity ( r ) of the ground [43].
For the perfect electric conductor (PEC) surface, Dirichlet and
3 2 Neumann BCs (DBC and NBC) have been taken into considera-
A0 + A1 2 + A2 + A3 2 u(x, z) = 0 (3)
x z x z tion for horizontal and vertical polarizations, respectively. In the
FFT-based SSPE algorithm, to satisfy the BCs over the Earths
with the help of (1 + q)1/2 (a0 + a1 q)/(b0 + b1 q) approx- surface, the surface is removed by taking a mirror copy of the
imation; therefore, A0 = b0 + b1 (n2 1), A1 = b1 k02 , A2 = initial vertical field profile with respect to the surface (odd/even
ik0 ((b0 a0 ) + (b1 a1 )(n2 1)), and A3 = ik01 (b1 a1 ). symmetric for DBC/NBC, respectively). Another choice is
If the angle of propagation measured from paraxial direction to use one-sided sine/cosine transforms (sinFFT/cosFFT) for
is less than 15 , the standard PE is obtained with the help DBC/NBC, respectively. Since the propagation problem in-
of square root approximation ( 1+q 1+q/2); therefore, volves a physical domain extending vertically to infinity, an
A0 = 1, A1 = 0, A2 = 0.5ik0 (n2 1), and A3 = 0.5ik01 . abrupt truncation is required at certain height, and therefore,
If the propagation angle is more than 15 , the Claerbout the upper BC must be satisfied to avoid nonphysical reflections.
equationis obtained by using the first-order Pad approxi- Such artificial reflections can be obviated by introducing ab-
mation ( 1+q (1+0.75q)/(1+0.25q)); therefore, A0 = 1+ sorbing layers above the height of interest [9].
0.25(n2 1), A1 = 0.25k02 , A2 = 0.5ik0 (n2 1), and A3 =
0.5ik01 to satisfy the propagation angles up to 40 [37].
The standard fast Fourier transform (FFT)-based SSPE so- III. FEMPE P ROPAGATION T OOL
lution for the narrow and wide angles, respectively, are shown
A. 1W-FEMPE Model
in [9], [23][26]
The initial phase of the FEM-based PWE procedure is to
u(x+x, z) divide the transverse domain between the ground and the user-
= exp 0.5ik0 (n2 1)x defined maximum height (Zmax ) into a number of elements.
F 1 exp ikz2 x /(2k0 ) F {u(x, z)} (4) Next, starting from the initial field at x = 0, the approximated
u(x+x, z) field values at the selected discrete nodes in the vertical domain
0 (n1)x]
= exp [ik are propagated longitudinally by applying the CrankNicholson
1
approach, which is based on the improved Euler method [44]
F exp ikz2 x / k0 1+ 1kz2 /k02 [46]. Although the CrankNicholson technique is inherently
F {u(x, z)}} (5) fast, it requires that both height and range step sizes, namely,
z and x, respectively, should be chosen as small as nec-
1 This assumption does not violate the applicability of PWE to range-
essary to overcome numerical oscillation problems, which ob-
dependent refractivity profiles, because it is valid for each range step during viously decelerate the speed of the method (refer to [22] for
the split-step solution of PWE, which will be clear in the sequel. detailed discussions).
2890 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST 2011
To formulate the one-way FEMPE (1W-FEMPE), we start Using CrankNicholson approximation based on the im-
with (3) whose coefficients are chosen by considering narrow- proved Euler method for k = 2, . . . , Nx where Nx is the num-
and wide-angle approximations. Note that some coefficients are ber of nodes in the horizontal domain, the new coefficients for
functions of z because of the refractivity index n. However, the next step are obtained as
the coefficients are assumed to be constant for each element by
finding the value of the refractive index at the midpoint of each
x x
element. By multiplying (3) by a smooth test function v(z) with A+B ck = AB ck1 (12)
2 2
the given BC in (6) and by using the integration-by-parts rule,
one obtains
which yields an unconditionally stable system and accu-
Zmax
rate method with the discretization error O(x2 ). The co-
u(x, z) v(z)
A0 u(x, z)v(z) A1 efficients of the initial field c1 at x = 0 can be generated
x z z from the Gaussian antenna pattern specified by its height
0
(hv ), 3-dB beamwidth angle (bw ), and tilt angle (tilt ).
u(x, z) v(z) Although CrankNicholson provides a fast solution, it has
+ A2 u(x, z)v(z) A3 dz
z z some disadvantages since oscillation occurs for large x
[22], [45].
u(Zmax ) u(0)
+ A1 v(Zmax ) v(0)
x z z
B. 2W-FEMPE Model
u(Zmax ) u(0)
+ A3 v(Zmax ) v(0) = 0. (7) The 2W-FEMPE approach is the iterative implementation
z z of the 1W-FEMPE algorithm by simply rotating the direction
of propagation in a forwardbackward manner to estimate
The last two terms of (7) should be taken into consideration
the multiple-reflection effects. The principle idea is based on
according to BCs. Replacing u(x, z) with the approximated
the 2W-SSPE proposed in [32][34]. The algorithm can be
solution
applied to a variable terrain by using staircase approximations,
ne
2 as shown in Fig. 1. If the vertical field meets a vertical
uap (x, z) = cej (x)Bje (z) (8) terrain facet, it is split into two components propagating
e=1 j=1 in forward and backward directions. The forward field
continues in the usual way after setting it to zero on the
where ne is the number of elements and cej (x) indicates the vertical terrain facet. In other words, the fields (at x + x)
coefficients of the unknown function. Moreover, the basis obtained from the previous vertical field (at x) are set to zero
functions are chosen as the linear piecewise Lagrange poly- inside the terrain (between z = 0 and z = zter(x0 + x)).
nomials as follows: B1e (z) = (z2e z)/(z2e z1e ) and B2e (z) = However, it is evident that the field must be partially
(z z1e )/(z2e z1e ), where e stands for the element between reflected from the terrain facet. Therefore, first the initial
nodes z1e and z2e . By choosing the test function v(z) as the same field of the backward field is obtained by imposing the
as the basis function Bm (z) for m = 1, 2, the matrix form of BCs at the facet (i.e., the tangential field must be zero on
(7) is the PEC facet), and then, this initial field is marched back
c in the x direction by reversing the signs of k0 and x.
A + Bc = 0 (9) Namely, the backward propagating waves are initiated from
x
the waves between z = zter(x) and z = zter(x0 + x)
with A = A0 M A1 K and B = A2 M A3 K + A3 BC and then propagated in the reverse direction. Note that
or (A0 Mmje
A1 Kmj
e
){cej /x} + (A2 Mmje
A3 Kmj
e
+ the same form of the PWE is derived, as expected,
A3 BCmj ){cj } = {0} for e = 1, . . . , ne , m = 1, 2, and
e e for the reduced function in the backward propagation,
j = 1, 2 with but the original field is expressed as b (x, z) =
ub (x, z) exp(ik0 x).
z2 z2
e e
e B e Both forward and backward fields continue to march out in
e e Bm j
Mmj = Bm Bje dz e
Kmj = dz (10) their own propagation directions. At each time the wave hits a
z z terrain facet, it is again split into forward and backward com-
z1e z1e
ponents. The total field inside the domain is then determined
and BCmje
is taken into consideration with BCs as BC11 1
= by the superposition of the backward and forward fields at
for each step in x, x + x, . . .. The ground properties are each range step. It is useful to note that the convergence of
incorporated into the FEM with the help of the matrix BC. the algorithm is achieved because, as the iterations are carried
The elemental matrices between nodes z1e and z2e for the linear out, the field contributions
of the multiple reflections decrease
piecewise Lagrange polynomials are obtained as with regard to the 1/ r term in the 2-D Greens function.
The convergence of the algorithm is checked against a certain
e z 2 1 e 1 1 1 threshold criterion, which compares the total fields at each
[M ] = [K ] = . (11)
6 1 2 z 1 1 iteration.
APAYDIN et al.: TWO-WAY FINITE-ELEMENT PARABOLIC EQUATION GROUNDWAVE PROPAGATION TOOL 2891
2
N q
u(x, z) = cq sin z exp(iq x) (13)
d q=1 d
where N is the total number of modes, q = k02 (q/d)2
is the longitudinal propagation constant for the mode q, and cq
is the modal excitation coefficient numerically derived from the
vertical orthonormality condition as
d q
cq = g(z) sin z dz. (14)
d
0
2 1/2
Here, g(z) = (2 ) exp[(z hv )2 /(2 2 )] and is the
spatial width of the Gaussian source. The Gaussian source
pattern is often used in applications since it represents various
antenna types. The Gaussian antenna pattern can also be defined
in the vertical wavenumber domain as
kz2 ln2
g(kz ) = exp . (15)
2k02 sin2 (bw /2)
Fig. 2. Parallel-plate PEC waveguide modeled with narrow-angle propagator.
The tilt (or elevation) angle tilt is introduced by shifting the (Top) FEMPE (z = 5 mm and x = 17 mm). (Bottom) SSPE (z =
antenna pattern, i.e., g(kz ) g(kz k0 sin tilt ). The vertical 20 mm and x = 67 mm) with 30 tilted waves at 1 GHz (hv = 4 m,
field distribution in the spatial domain is then obtained by bw = 7 , and d = 8 m).
taking the inverse Fourier transform of (15).
The limiting case for the source function u(x, z) is the downward, inside an 8-m-wide parallel-plate PEC waveguide
line-source representation which requires an infinite number under DBC, the performances of narrow- and wide-angle PE
of modes (N ) in the modal summation. The number tools with respect to the analytical result are shown in Figs. 2
of modes would be finite for any other source function. It is and 3 by means of 3-D field maps. The two field maps in
common to choose a vertically extending Gaussian function Fig. 2 alone (generated with the narrow-angle 1W-FEMPE and
with arbitrary location having vertical tilt angle in the range of 1W-SSPE tools) seem to be logical and physical. A down-
90 . Note that the modal excitation coefficient cq is real for tilted Gaussian beam is bouncing up and down while lon-
a real source function without any tilt and becomes complex if gitudinally propagating with the interference patterns exactly
there is a downward/upward tilt no matter if the source function like that shown in the figure. On the other hand, the true 3-D
is real or not. The number of modes, which are necessary for wave patterns are shown in Fig. 3. The analytical exact so-
the modal summation representation for a vertical Gaussian lution is exactly the same as the FEMPE and SSPE results
source function located at midheight of an 8-m-wide parallel- of Fig. 3 and is not shown here. Hence, these results validate
plate waveguide, is listed in Table I. As observed, the number the accuracy of the wide-angle FEMPE and SSPE and show
of modes is less than a few hundreds even for 45 vertical that the narrow-angle FEMPE and SSPE cannot handle large
tilts. The higher the tilt angles (or frequency), the higher the tilt angles. To validate the narrow-angle model, the angle must
number of modes required in the modal summation. Moreover, be constrained to be less than 10 15 while considering both
the number of modes increases as the vertical source beamwidth beamwidth and tilt angles. Figs. 4 and 5 show horizontal field
gets narrower. profiles computed via analytical, FEMPE, and SSPE models
Assuming a horizontally polarized Gaussian source at 1 GHz at two different heights. In Fig. 4, the results computed via
with 3-dB beamwidth of 7 , located at midheight and tilted 30 narrow-angle FEMPE and SSPE tools are compared against
2892 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST 2011
Fig. 9. GO+UTD modeling and some possible direct, reflected, and diffracted
field components.
Fig. 11. The PF versus height at two specified ranges with the same pa-
rameters as in Fig. 10. SSPE (z = 0.33 m and x = 100 m) and FEMPE
(z = 0.33 m and x = 12.5 m).
Fig. 10. Two-way propagation over PEC ground with one 150-m-height wall
at 40-km range. The PF versus range/height for a given source (at 250-m height
and 0-km range). (Top) GO+UTD. (Bottom) FEMPE method, f = 3 GHz.
Fig. 13. PF versus height at two specified ranges with the same parameters as
in Fig. 12. SSPE (z = 0.33 m and x = 100 m). FEMPE (z = 0.33 m
and x = 12.5 m).
Fig. 17. PF versus height at two specified ranges with the same parameters
as in Fig. 16. SSPE (z = 0.16 m and x = 100 m) and FEMPE (z =
0.33 m and x = 12.5 m).
Fig. 16. Two-way propagation for vertical polarization over PEC ground with
two infinite-height walls at 0 and 60 km. The PF versus range/height for a
given source (at 30-km range and 100-m height). (Top) Image method and
(bottom) FEMPE method, where f = 3 GHz, z = 0.33 m, x = 25 m.
# of images = 5, and # of reflection = 1.
V. C ONCLUSION
A 2W-FEMPE algorithm has been developed, tested, and
calibrated against scenarios with analytical exact solutions,
such as the modal summation, image method, and asymptotic
models, such as GO+UTD models. Tests and comparisons
are also performed against the 2W-SSPE algorithm. Both
narrow- and wide-angle PE models are used during these tests
and comparisons. Excellent agreement obtained from these
tests/comparisons among all the models used show that both of
the 2W-FEMPE and 2W-SSPE models can be used effectively
in complex propagation scenarios above the Earths surface
through inhomogeneous atmosphere.
The FEMPE model needs much smaller range steps than
the SSPE model and therefore necessitates longer computation
times when long-range propagation is of interest. According to
the last example (shown in Fig. 19), the computation time of
the FEMPE is at least 14 times longer as compared with that of
SSPE. As a result, the number of nodes used in the transverse
Fig. 19. Two-way propagation over arbitrary terrain through inhomogeneous and range coordinates should be optimized with respect to the
atmosphere. The PF versus range/height for a given source (hv = 400 m, following parameters: the operating frequency, the irregular
bw = 0.5 , tilt = 0.5 ). (Top) FEMPE and (bottom) SSPE, with f = terrain structure, inhomogeneous atmosphere, and the selection
300 MHz, z = 1 m, and x = 200 m, with refractivity slope of 200
Munit/km between 300 and 400 m (ducting case). tfempe = 7 min and of the initial antenna pattern specified by its height, beamwidth,
tsspe = 30 s. and tilt. Moreover, the line-source excitation is a serious prob-
lem in the FEMPE model, but this might be overcome by
constructing the initial field profile at a few wavelengths away
from the line source by using some other models such as the
SSPE or GO, etc., or using smooth, such as Gaussian-type,
source patterns. On the other hand, FEMPE model handles all
kinds of BC at the surface much more easily when compared
with the SSPE model.
R EFERENCES
[1] S. Grosdidier, A. Baussard, and A. Khenchaf, HFSW radar model: Sim-
ulation and measurement, IEEE Trans. Geosci. Remote Sens., vol. 48,
no. 9, pp. 35393549, Sep. 2010.
[2] M. A. Leontovich and V. A. Fock, Solution of propagation of electro-
magnetic waves along the Earths surface by the method of parabolic
equation, J. Phys. USSR, vol. 10, pp. 1323, 1946.
[3] R. H. Hardin and F. D. Tappert, Applications of the split-step Fourier
method to the numerical solution of nonlinear and variable coefficient
wave equations, SIAM Rev., vol. 15, p. 423, 1973.
Fig. 20. Two-way propagation over arbitrary terrain through inhomogeneous [4] J. R. Kuttler and G. D. Dockery, Theoretical description of the parabolic
atmosphere. The PF versus range at z = 250 m for the given source (hv = approximation/Fourier split-step method of representing electromagnetic
400 m, bw = 0.5 , and tilt = 0.5 ) using 2W-SSPE and 2W-FEMPE propagation in the troposphere, Radio Sci., vol. 26, no. 2, pp. 381393,
(f = 300 MHz, z = 1 m, and x = 200 m, with refractivity slope of 1991.
200 Munit/km between 300 and 400 m (ducting case), tfempe = 7 min, and [5] A. E. Barrios, Parabolic equation modeling in horizontally inhomoge-
tsspe = 30 s). neous environments, IEEE Trans. Antennas Propag., vol. 40, no. 7,
pp. 791797, Jul. 1992.
[6] A. E. Barrios, A terrain parabolic equation model for propagation in the
paraxial approximation). Yet, analytical exact solutions have troposphere, IEEE Trans. Antennas Propag., vol. 42, no. 1, pp. 9098,
Jan. 1994.
not appeared for the irregular terrain models. The asymptotic [7] R. Janaswamy, A curvilinear coordinate-based split-step parabolic equa-
GO+UTD model is not applicable (in its current form) to tion method for propagation predictions over terrain, IEEE Trans. Anten-
problems containing irregular terrains (which is beyond the nas Propag., vol. 46, no. 7, pp. 10891097, Jul. 1998.
[8] C. A. Zelley and C. C. Constantinou, A three-dimensional parabolic
scope of this study). To support this speculation, the PF versus equation applied to VHF/UHF propagation over irregular terrain, IEEE
range extracted from the 3-D field maps shown in Fig. 19 Trans. Antennas Propag., vol. 47, no. 10, pp. 15861596, Oct. 1999.
is plotted in Fig. 20. Here, the horizontal PF is plotted from [9] M. F. Levy, Parabolic Equation Methods for Electromagnetic Wave Prop-
agation. London, U.K.: IEE, 2000.
20 to 60 km, which is the range of interference of forward- [10] C. Mias, Fast computation of the nonlocal boundary condition in finite
and backward-propagated waves. As observed, there is a good difference parabolic equation radiowave propagation simulations, IEEE
agreement between the results of the 2W-FEMPE and 2W- Trans. Antennas Propag., vol. 56, no. 6, pp. 16991705, Jun. 2008.
[11] P. D. Holm, Wide-angle shift-map PE for a piecewise linear terrain
SSPE models, but this is not as good as the agreement obtained A finite-difference approach, IEEE Trans. Antennas Propag., vol. 55,
in other examples. no. 10, pp. 27732789, Oct. 2007.
2898 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST 2011
[12] D. J. Donohue and J. R. Kuttler, Propagation modeling over terrain using Propagation USNC/URSI Nat. Radio Sci. Meet., Toronto, ON, Canada,
the parabolic wave equation, IEEE Trans. Antennas Propag., vol. 48, July 1117, 2010, pp. 14.
no. 2, pp. 260277, Feb. 2000. [38] O. Ozgun, G. Apaydin, M. Kuzuoglu, and L. Sevgi, Two-way Fourier
[13] L. Sevgi, F. Akleman, and L. B. Felsen, Groundwave propagation split step algorithm over variable terrain with narrow and wide angle prop-
modeling: Problem-matched analytical formulations and direct numerical agators, in Proc. IEEE Int. Symp. Antennas Propagation USNC/URSI
techniques, IEEE Antennas Propag. Mag., vol. 44, no. 1, pp. 5575, Nat. Radio Sci. Meet., Toronto, ON, Canada, July 1117, 2010, pp. 14.
Feb. 2002. [39] J. B. Keller, Geometrical theory of diffraction, J. Opt. Soc. Amer.,
[14] L. Sevgi, Complex Electromagnetic Problems and Numerical Simulation vol. 52, no. 2, pp. 116130, 1962.
Approaches. Piscataway, NJ: IEEE Press, Jun. 2003. [40] R. G. Kouyoumjian and P. H. Pathak, A uniform geometrical theory of
[15] L. Sevgi, C. Uluisik, and F. Akleman, A Matlab-based two-dimensional diffraction for an edge in a perfectly conducting surface, Proc. IEEE,
parabolic equation radiowave propagation package, IEEE Antennas vol. 62, no. 11, pp. 14481461, Nov. 1974.
Propag. Mag., vol. 47, no. 4, pp. 164175, Aug. 2005. [41] P. Y. Ufimtsev, Fundamentals of the Physical Theory of Diffraction.
[16] L. Sevgi, Groundwave modeling and simulation strategies and path loss Hoboken, NJ: Wiley, 2007.
prediction virtual tools, IEEE Trans. Antennas Propag., vol. 55, no. 6, [42] L. Sevgi, Guided waves and transverse fields: Transverse to what?,
pp. 15911598, Jun. 2007, (Special issue on Electromagnetic Wave Prop- IEEE Antennas Propag. Mag., vol. 50, no. 6, pp. 221225, Dec. 2008.
agation in Complex Environments: A Tribute to L. B. Felsen). [43] J. R. Wait, The scope of impedance boundary conditions in radio prop-
[17] D. A. Hill and J. R. Wait, HF ground wave propagation over mixed land, agation, IEEE Trans. Geosci. Remote Sens., vol. 28, no. 4, pp. 721723,
sea, and sea-ice paths, IEEE Trans. Geosci. Remote Sens., vol. GRS-19, Jul. 1990.
no. 4, pp. 210216, Oct. 1981. [44] J. Volakis, A. Chatterjee, and L. Kempel, Finite Element Method for
[18] D. Huang, Finite element solution to the parabolic wave equation, Electromagnetics: Antennas, Microwave Circuits, and Scattering Appli-
J. Acoust. Soc. Am., vol. 84, no. 4, pp. 14051413, Oct. 1988. cations. Piscataway, NJ: IEEE Press, 1998.
[19] K. Arshad, F. A. Katsriku, and A. Lasebae, An investigation of tropo- [45] J.-M. Jin, The Finite Element Method in Electromagnetics. New York:
spheric radio wave propagation using finite elements, WSEAS Trans. Wiley, 2002.
Communications, vol. 4, no. 11, pp. 11861192, Nov. 2005. [46] W. Y. Yang, W. Cao, T. Chung, and J. Morris, Applied Numerical Methods
[20] G. Apaydin and L. Sevgi, FEM-based surface wave multi-mixed-path Using Matlab. Hoboken, NJ: Wiley, 2005.
propagator and path loss predictions, IEEE Antennas Wireless Propag. [47] G. Apaydin and L. Sevgi, Validation, verification and calibration in ap-
Lett., vol. 8, pp. 10101013, 2009. plied computational electromagnetics, in Proc. 26th Int. Rev. Progr. Appl.
[21] G. Apaydin and L. Sevgi, Numerical investigations of and path loss pre- Comput. Electromagn., Tampere, Finland, Apr. 2529, 2010, pp. 679684.
dictions for surface wave propagation over sea paths including hilly island [48] C. A. Tunc, A. Altintas, and V. B. Erturk, Examination of existent
transitions, IEEE Trans. Antennas Propag., vol. 58, no. 4, pp. 1302 propagation models over large inhomogeneous terrain profiles using fast
1314, Apr. 2010. integral equation solution, IEEE Trans. Antennas Propag., vol. 53, no. 9,
[22] G. Apaydin and L. Sevgi, The split step Fourier and finite element pp. 30803083, Sep. 2005.
based parabolic equation propagation prediction tools: Canonical tests, [49] C. Tunc, F. Akleman, V. Erturk, A. Altintas, and L. Sevgi, Fast integral
systematic comparisons, and calibration, IEEE Antennas Propag. Mag., equation solutions: Application to mixed path terrain profiles and compar-
vol. 52, no. 3, pp. 6679, Jun. 2010. isons with parabolic equation method, Complex Comput. Netw., vol. 104,
[23] D. J. Thomson and N. R. Chapman, A wide-angle split-step algorithm Springer Proc. Phys., pp. 5563, Jan. 2006.
for the parabolic equation, J. Acoust. Soc. Am., vol. 74, no. 6, pp. 1848 [50] A. Yagbasan, C. A. Tunc, V. B. Erturk, A. Altintas, and R. Mittra, Char-
1854, Dec. 1983. acteristic basis function method for solving electromagnetic scattering
[24] M. D. Collins, A split-step Pade solution for the parabolic wave equa- problems over rough terrain profiles, IEEE Trans. Antennas Propag.,
tion, J. Acoust. Soc. Am., vol. 93, no. 4, pp. 17361742, Apr. 1993. vol. 58, no. 5, pp. 15791589, May 2010.
[25] J. R. Kuttler, Differences between the narrow-angle and wide-angle [51] F. Akleman and L. Sevgi, A novel finite-difference time-domain wave
propagators in the split-step Fourier solution of the parabolic wave equa- propagator, IEEE Trans. Antennas Propag., vol. 48, no. 5, pp. 839841,
tion, IEEE Trans. Antennas Propag., vol. 47, no. 7, pp. 11311140, May 2000.
Jul. 1999. [52] F. Akleman and L. Sevgi, Time and frequency domain wave propa-
[26] M. F. Levy, Diffraction studies in urban environment with wide-angle gators, ACES J. Special Issue Comput. Electromagn. Techn. Wireless
parabolic equation method, Electron. Lett., vol. 28, no. 16, pp. 1491 Commun., vol. 15, no. 3, pp. 186201, Nov. 2000.
1492, Jul. 1992. [53] M. O. Ozyalcin, F. Akleman, and L. Sevgi, A novel TLM-based time-
[27] L. A. Weinstein, Open Resonators and Open Waveguides. Boulder, CO: domain wave propagator, IEEE Trans. Antennas Propag., vol. 51, no. 7,
Golem Press, 1969. pp. 16791680, Jul. 2003.
[28] P. Y. Ufimtsev, Current waves in a thin wire and in a ribbon, U.S.S.R., [54] F. Akleman and L. Sevgi, Realistic surface modeling for a finite-
Comput. Math. Math. Phys., vol. 8, no. 6, pp. 250261, 1968. difference time-domain wave propagator, IEEE Trans. Antennas
[29] P. Y. Ufimtsev, Theory of Edge Diffraction in Electromagnetics. Encino, Propag., vol. 51, no. 7, pp. 16751679, Jul. 2003.
CA: Tech Science Press, 2003.
[30] P. Y. Ufimtsev, Theory of Edge Diffraction in Electromagnetics: Origina-
tion and Validation of the Physical Theory of Diffraction. Raleigh, NC:
SciTech Publishing, Inc., 2009.
[31] P. Y. Ufimtsev and S. A. P. Krasnozhen, Current waves in a straight thin
wire resonators with finite conductivity, Electromagnetics, vol. 12, no. 2,
pp. 121132, 1992.
[32] M. D. Collins, A two-way parabolic equation for elastic media,
J. Acoust. Soc. Amer., vol. 93, no. 4, pp. 18151825, Apr. 1993.
[33] J. F. Lingevitch, M. D. Collins, M. J. Mills, and R. B. Evans, A two-way
parabolic equation that accounts for multiple scattering, J. Acoust. Soc. Gkhan Apaydin (M08SM11) received the B.S.,
Am., vol. 112, no. 2, pp. 476480, Aug. 2002. M.S., and Ph.D. degrees in electrical and electron-
[34] M. J. Mills, M. D. Collins, and J. F. Lingevitch, Two-way parabolic equa- ics engineering from Bogazici University, Istanbul,
tion techniques for diffraction and scattering problems, Wave Motion, Turkey, in 2001, 2003, and 2007, respectively.
vol. 31, no. 2, pp. 173180, Feb. 2000. From 2001 to 2005, he was a Teaching and
[35] H. Orazi and S. Hosseinzadeh, Radio-wave-propagation modeling in the Research Assistant with Bogazici University. From
presence of multiple knife edges by the bidirectional parabolic-equation 2005 to 2010, he was a Project and Research
method, IEEE Trans. Veh. Technol., vol. 56, no. 3, pp. 10331040, Engineer with the Applied Research and Devel-
May 2007. opment, University of Technology Zurich, Zurich,
[36] O. Ozgun, Recursive two-way parabolic equation approach for model- Switzerland. Since 2010, he has been with Zirve
ing terrain effects in tropospheric propagation, IEEE Trans. Antennas University, Gaziantep, Turkey. He has been working
Propag., vol. 57, no. 9, pp. 27062714, Sept. 2009. on several research projects on electromagnetic (EM) propagation, the develop-
[37] G. Apaydin, O. Ozgun, M. Kuzuoglu, and L. Sevgi, Two-way split- ment of a finite-element method for EM computation, positioning, filter design,
step Fourier and finite element based parabolic equation propagation and related areas. He has (co)authored 14 journals and 27 conference papers
tools: Comparisons and calibration, in Proc. IEEE Int. Symp. Antennas and technical reports at the Hochschule fr Technik Zrich, Zrich.
APAYDIN et al.: TWO-WAY FINITE-ELEMENT PARABOLIC EQUATION GROUNDWAVE PROPAGATION TOOL 2899
Ozlem Ozgun (M05) received the B.Sc. and M.Sc. Levent Sevgi (M99SM02F09) received the
degrees in electrical engineering from Bilkent Uni- Ph.D. degree from Istanbul Technical University,
versity, Ankara, Turkey, in 1998 and 2001, respec- Istanbul, Turkey, and Polytechnic Institute of New
tively, and the Ph.D. degree in electrical engineering York University, Brooklyn, in 1990. Prof. Leo Felsen
from the Middle East Technical University (METU), was his advisor.
Ankara, in 2007. He was with Istanbul Technical University from
She is currently an Assistant Professor with 1991 to 1998; TUBITAK-MRC, Information Tech-
METUNorthern Cyprus Campus, Mersin, Turkey. nologies Research Institute, Gebze/Kocaeli, Turkey,
From 2007 to 2008, she was with the Electro- from 1999 to 2000; Weber Research Institute/
magnetic Communication Laboratory, Pennsylvania Polytechnic University of New York, from 1988 to
State University, University Park; ASELSAN Inc., 1990; Scientific Research Group of Raytheon Sys-
Ankara, from 2004 to 2005; TUBITAK-UEKAE, National Research Institute of tems, Canada, from 1998 to 1999; and the Center for Defense Studies, ITUV-
Electronics and Cryptology, Ankara, from 2000 to 2004; and Bilkent University SAM, from 1993 to 1998 and from 2000 to 2002. Since 2001, he has been with
from 1998 to 2000. Her main research interests include computational elec- Dogus University, Istanbul. He has been involved with complex electromag-
tromagnetics, finite-element method, domain decomposition, electromagnetic netic (EM) problems and systems for more than 20 years. His research study
(EM) propagation and scattering, metamaterials, and stochastic EM problems. has focused on propagation in complex environments, analytical and numerical
methods in electromagnetics, EMC/EMI modeling and measurement, radar
and integrated surveillance systems, surface-wave HF radars, finite-difference
Mustafa Kuzuoglu (M92) received the B.Sc., time domain, TLM, finite-element method, split-step parabolic equation, and
M.Sc., and Ph.D. degrees in electrical engineer- Method of Moments techniques and their applications, radar cross-section
ing from the Middle East Technical University modeling, and bioelectromagnetics. He is also interested in novel approaches in
(METU), Ankara, Turkey, in 1979, 1981, and 1986, engineering education and teaching electromagnetics via virtual tools. He also
respectively. teaches popular science lectures such as science, technology, and society.
He is currently a Professor with METU. His re- Prof. Sevgi is the Writer/Editor of the Testing ourselves Column in the
search interests include computational electromag- IEEE Antennas and Propagation Magazine, a member of the IEEE Antennas
netics, inverse problems, and radars. and Propagation Society Education Committee, and the Scientific Literacy
Column Writer of the IEEE Region 8 Newsletter.