Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
8, AUGUST 2003
Abstract—Starting from a parabolic approximation to the recently—rather than working with the Helmholtz equation
Helmholtz equation, a three–dimensional (3-D) vector parabolic (elliptic type) that leads to bidirectional waves and global cou-
equation technique for calculating path loss in an urban environ- pling of fields, some formulations have considered the parabolic
ment is presented. The buildings are assumed to be polygonal in
cross section with vertical sides and flat rooftops and the terrain approximation to the Helmholtz equation, which assumes for-
is assumed to be flat. Both buildings and ground are allowed to ward scattering at the outset [16], [10]. Under the assumption of
be lossy and present impedance-type boundary condition to the forward scattering, the fields are only coupled along the principal
electromagnetic field. Vector fields are represented in terms of the direction, but not in the opposite direction. The problem of single
two components of Hertzian potentials and depolarization of the
fields is automatically included in the formulation. A split-step or multiple knife-edge diffraction under the Kirchhoff’s ap-
algorithm is presented for marching the aperture fields along the proximation is one such example. Despite its seemingly flagrant
range. Boundary conditions on the building surfaces are treated assumption, the parabolic equation yields very useful results
by using a local Fourier representation of the aperture fields. for long-distance propagation problems, as observed by several
Several test cases are considered to check the boundary treatment investigators (see [16] for a comprehensive reference list). Most
used in the technique as well as to validate the overall approach.
Comparison is shown with uniform theory of diffraction (UTD), of the formulations of parabolic equation are two-dimensional
exact solutions, as well as with measurements. (2-D) in nature, in that they assume propagation to take place in
Index Terms—Outdoor models, split-step algorithm, three– a 2-D horizontal or vertical plane. The 2-D PE model presented
dimensional (3-D) propagation, vector parabolic equation. in [10] had an average error of 10 dB and a standard deviation of
8–10 dB compared with field measurements. It is generally be-
lieved that a full 3-D formulation that includes both vertically and
I. INTRODUCTION laterally propagating waves would not only reduce the mean error
gate uncoupled on flat ground, but become coupled in the pres- On horizontal edges formed by the intersection of the rooftops
ence of lossy and finite-height buildings. In this work, we use and vertical sides of buildings, we combine the boundary
impedance boundary conditions to represent the lossy nature of conditions on the two faces. Of course, such a combination of
building surfaces and ground. If is the unit outward normal boundary conditions will be approximate and ignores the edge
on a boundary with normalized surface impedance , then the condition that the true near fields have to satisfy. It may be
impedance boundary condition states that [19] noted that the satisfaction of edge condition by the near fields is
not necessary to effectively predict the far-zone diffracted field.
(9)
Diffraction by knife-edge under Kirchhoff’s approximation is a
For the most general case, the surface impedance will depend simple example. Hence, we regard the above approximation as
upon the material constitutive parameters , where no more severe than the impedance boundary condition itself,
is the dielectric constant and is the conductivity of the which will also become invalid near edges.
medium, and the incidence angle of the incoming wave. How- If the buildings and ground are treated as perfect conductors
ever, if the material is dense enough , the (PEC), then , , and the boundary conditions
transmitted wave will be approximately normal in the medium become
and it is reasonable to take the value of for normally propa-
gating waves. The surface impedance for normally propagating ground and building rooftops (16)
plane waves is [2]
(10)
building sidewalls (17)
We will assume this approximation to hold on both the ground Combining these two we get and on the hor-
and building surfaces, but with different values of material pa- izontal edges for a PEC building. It is seen from (16) and (17)
rameters. (If propagation takes place only over a flat ground, it that the TE and TM modes propagate uncoupled when the build-
may be more appropriate to take the impedance to be that of a ings and ground are treated as perfect conductors. Even for per-
grazing wave [11].) On the flat ground with , , fect conductors, polarization coupling will arise if the buildings
and , (9) gives have sloping rooftops or nonvertical sides. But such geometries
are not considered in this paper. For the lossy case, mode cou-
(11) pling will occur even for flat rooftops and vertical sides.
Expressing the various fields in terms of the potentials using A. Parabolic Approximation
(3)–(6), the boundary conditions on the ground at reduce
Equations (7) and (8) together with the boundary conditions
to
(12)–(15) and a radiation boundary condition at infinity con-
(12) stitute the exact boundary value problem. However, their nu-
merical solution for long-range propagation problems requires
and its dual excessive computational resources, which current technology
cannot provide. This is due to the occurrence of full matrices that
(13) are the result of global coupling of fields. Reasonable answers
for path loss can still be obtained by assuming forward propa-
The boundary conditions on the rooftop of a building will take gation at the outset [10]. To do this, a principal direction is first
the same forms as (12) and (13) with replaced by chosen along which only forwardly propagating waves are per-
mitted. Let this axis be the axis and be known as the range axis.
In the transverse directions (i.e., along and ) waves are still
where and are the material constants for the rooftop. allowed to propagate in both directions. In a source-free region,
On the vertical sides of a building, assuming , the Helmholtz equation , with a second-order
, , , and derivative along the axis, is replaced with the wide-angle par-
expressing the various fields in terms of the potentials, we have abolic equation (PE)
from (9) that
(18)
Fig. 2. Plan view of aperture plane marching over buildings. The dashed lines show the outline of the buildings, whereas the solid vertical lines show the
intersections with the aperture plane.
this equation outside the source region. In the numerical solution is the wavenumber along the range axis for a plane wave trav-
of the parabolic equation by the split-step Fourier technique, the eling in the direction, and is the 2-D Fourier trans-
aperture field at is determined in terms of the aperture form of the aperture field that satifies the impedance boundary
field at by decomposing the field in a spectrum of plane waves condition over ground
[20]. In the spectral representation, both incoming and outgoing
plane waves along the -direction are chosen to automatically
satisfy the impedance boundary condition on the ground.
An important point to recognize is that when a wave field
modeled by a uniaxial parabolic equation encounters a vertical (21)
obstacle in range, it only affects a local change to the aperture
field. This is due to the causal nature of the forward-propagating These equations follow from extending the results for the ax-
fields along the range that prevents information about a future isymmetric case presented in [11] to a more general case with
(in range) obstacle to flow backward. Boundary conditions on transverse (i.e., -) variation. The propagator is
the building surface are then imposed on the field as the aper- due to the accumulation of phase by a plane wave traveling in the
ture plane crosses it. Due to the finite range-step size, they are, positive -direction over a distance . The quantity is
however, only applied at discrete values of the range as shown simply the plane-wave reflection coefficient over an impedance
in Fig. 2. Between two adjacent aperture planes, we assume that ground and is
the field marches as if it were propagating over flat ground. The
situation is not unlike finite-differencing of differential opera- TM mode
(22)
tors on a discrete grid. The next subsection discusses the treat- TE mode.
ment of object boundary conditions in more detail.
We assume that the frequency is high enough so that the sur- Equations (20) and (21) are used for both and with the
face wave on the ground could be neglected. In the far zone, the appropriate value of chosen from (22). As with all para-
potential at over flat ground could be written in terms bolic equation formulations, they will continue to be used near
of the potential at as [11] the source even though they have been derived for the far zone.
The error due to this assumption will be most severe for plane
waves propagating at very large angles near the source, but will
(20) decrease as the waves become shallow with respect to the axis.
It can also be shown that, under this approximation, the initial
field is [11]
where
(23)
1720 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 51, NO. 8, AUGUST 2003
(30)
D. Propagation Factor
For reference, we will compare the field in an urban environ-
A similar equation is obtained for . Inserting (28) and ment with the field in free space. If the antenna source function
its dual into the boundary conditions (14) and (15) for at height is expressed in terms of a baseband func-
and into (12)–(15) for and evaluating tion by means of , then the
the various derivatives, we arrive at two coupled sets of linear far-zone potential in free space is
equations in the unknown vectors
(33)
where
(34)
(35)
1722 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 51, NO. 8, AUGUST 2003
and
(36)
The -component of the free-space electric field in the far zone
is
(37)
III. NUMERICAL RESULTS Fig. 4. Normalized field over flat ground versus lateral displacement showing
the angular performance of wide-angle PE.
Using the theory presented in Section II, we have computed
the normalized field for several test geometries and compared
the results with those obtained by using series solution, or , we get , , and
uniform theory of diffraction (UTD) or with measurements. . At a frequency of 1 GHz, the maximum range
The test geometry for series solution is a lossy or perfectly step over flat ground is about 362 m. In urban areas, will
conducting circular cylinder. The geometries considered for be governed more by the building density and its representation
UTD comparisons were an infinitely tall, perfectly conducting, rather than (40). Nevertheless, (40) gives the maximum range
offset circular cylinder, an infinitely tall, inclined strip, a step one can ever use using the split-step algorithm.
finite-height strip, a series of finite-height strips, a finite-height We would first like to illustrate the dependence of the wide
circular cylinder, and a finite-height square cylinder. The angle capability of the algorithm on the mesh size ( , ).
geometries involved in the comparisons with measurements Fig. 4 shows the normalized field over flat ground at a distance
include a centered, lossy rectangular block and an offset, of 1 km from a 15 source located at . The numerical
lossy rectangular building. Some of the geometries chosen are parameters chosen in the PE computations are shown in the
general enough without having symmetries with respect to the inset. The PE solution was computed out to a distance of 1 km
source and axis planes that they provide the same challenging and the results are shown for a horizontal cut. Comparison
conditions that one encounters in practice. is shown with a two-ray model that includes a direct ray and
The important parameters which will affect the numerical a ground-reflected ray. The variable on the bottom abscissa
accuracy the PE solution are the grid sizes , , and is the lateral displacement, whereas the variable on the top
along , , and directions, respectively, and the 2-D FFT size abscissa is the equivalent propagation angle as computed from
. The increments and required on the aper- (34). As stated previously, the maximum propagation angle
ture plane are dictated by the maximum propagation angle being treated without aliasing is 20.7 for . This is
modeled. If is the maximum propagation angle with respect clearly seen from the close agreement of the PE result with the
to the positive axis and if , then, we have two-ray model until about . Beyond about 21 , aliasing
completely destroys the PE solution and only smaller mesh
(39)
sizewill remedy the solution. A similar observation was also
Equation (39) can be used to predict the maximum propagation seen in the vertical cut.
angle for a given mesh size. For example, if , We now consider 2–D obstacles placed asymmetrically with
then and the maximum propagation angle modeled respect to the source to provide rich conditions for lateral reflec-
without aliasing is 20.7 . tion and diffraction. The purpose of the following two test cases
Using and , it can be is to bring out the effectiveness (or otherwise) of the boundary
shown that the range step size over flat ground satisfies treatment discussed in Section II-B. Both geometries will have
vertical surfaces that are not parallel to the axis planes. The first
(40) test case we consider is that of propagation around a perfectly
conducting circular cylinder with infinite dimension along the
The sizes , are chosen large enough to prevent spurious axis. No ground plane is assumed in this case. A line source is
reflections off of the aperture boundaries from reaching the located at and radiates a vertically polarized field toward
useful computational domain. As an example, with , a circular cylinder of radius and center at , as shown
JANASWAMY: PATH LOSS PREDICTIONS IN THE PRESENCE OF BUILDINGS ON FLAT TERRAIN 1723
Fig. 5. Direct and reflected rays from a line source exciting a PEC circular
cylinder.
Fig. 7. A line source exciting an inclined conducting strip. The various shadow
boundaries are shown as dashed lines.
Fig. 8. Reflected and diffracted fields of a line source exciting an offset and Fig. 10. Propagation factor versus lateral displacement around a finite screen.
inclined conducting strip.
the result due to a 2-D parabolic equation, run for different ra-
dials about the transmitter. The 2-D model assumes propagation be an aberration. (We have utilized numerical electromagnetics
to take place in the meridian planes. If a radial line intersects code-basic scattering code (NEC-BSC) to compute the UTD re-
the screen, then an infinite-width screen perpendicular to the sults.) Once again the 3-D PE results are in good agreement with
radial line is erected at the intersecting point. In the shadow re- the UTD results both in the lit and shadow regions. We have
gion, waves will reach the receiver only via diffraction from the also considered multiple screens of varying widths and a per-
top edge. In the 3-D model, there will be diffraction from the fectlyconducting square cylinder and compared the PE results
side and top edges of the screen. At 500 m, the incident with UTD results. The agreement obtained there was similar to
shadow boundary exists at 100 m. It is seen that the 3-D the ones shown in Figs. 10 and 11.
PE results agrees very well with the UTD results both in the lit The next two examples we consider bring out the effective-
and shadow regions. In contrast, the 2-D model only predicts the ness of the boundary treatment in calculating the cross-polar
average field variation and misses all of the interference lobes. component. The first example is that of a circular conducting
Fig. 11 shows the field variation in the vertical cut as a func- building of radius and height placed over perfect ground
tion of the receiver height at . Fine fluctuations, due to at a distance of along the axis from the transmitter. The
the constructive and destructive interference of laterally prop- and components of the electric field are calculated at
agating waves, are seen in both the UTD and PE results. The 500 m and for a –directed Gaussian source placed at
large spike seen in the UTD result at is believed to . Fig. 12 shows the comparison of the transverse vari-
JANASWAMY: PATH LOSS PREDICTIONS IN THE PRESENCE OF BUILDINGS ON FLAT TERRAIN 1725
Fig. 12. Copolar and cross-polar components of the electric field behind a Fig. 13. Depolarization due to oblique incidence on an impedance cylinder of
perfectly conducting circular building. finite height.
ation of the fields with UTD. All fields have been normalized to
in free space. In this example, the cross-polar component re-
sults from itself and no is excited. The circular cylinder
was intersected by the marching plane times. Other pa-
rameters are shown in the figure inset. It is seen that the agree-
ment with UTD is within a few decibels both for and .
It is well known that an infinite dielectric cylinder produces
depolarization of waves when the incident wave is oblique to
the cylinder axis [22]. The next example we consider is that
of oblique scattering by a cylindrical building having a surface
impedance . A Gaussian source producing a main beam at el- Fig. 14. Field strength measurement behind a lossy block-shaped obstacle,
evation angle 30 and located at is used for this = =
w 30 cm, d 7.7 cm, H =50 cm, x =
68.9 cm, x =
25.2 cm, H =
purpose. The cylindrical building has its axis at , a radius 46 cm, H = 40 cm, and =114.8 cm.
, and a height . The copolar field and the depolarized
field (resulting from the excitation of ) are calculated at performed scaled model measurements and measured the field
. The elevation beam width of the Gaussian source strength along a circular arc behind a lossy, block-shaped ob-
and the building height are chosen such that there is essentially stacle at a frequency of 50 GHz. The geometry of the measure-
no field incident on the building rooftop. Consequently, prop- ment setup is shown in Fig. 14. A vertically polarized signal was
agation is almost entirely off of the building side walls. This generated using open-ended WR-90 waveguide. A comparison
is done so that the exact series solution of a infinite cylinder of propagation factor is shown in Fig. 15 as a function of the az-
could be used for comparison. The Green’s function for a point imuth angle . Measured data have been read off the plots given
source and an infinite impedance cylinder was derived in terms in [21]. Both copolar, , and cross-polar, , components were
of cylindrical harmonics and integrated over the Gaussian aper- computed using the 3-D PE algorithm described in the previous
ture to produce the exact field. The received fields are calculated sections. The constitutive parameters used for the building ma-
in the direction of the main beam at terial and ground are indicated in the figure text. In the PE com-
and normalized to produced in free space. Fig. 13 shows putations, we had a variable such that there were
the comparison of the results for 50.85 m and sections before the block, sections inside the block, and
200 m. Other parameters are shown in the figure inset. In the sections beyond the block. For the boundary treatment
PE computations, the ground impedance is taken to be the same we chose . The results were rather insensitive to the
as . Considering the fact that the depolarized field is about choice of over the range . It is seen that the
40 dB below the copolar field, the agreement seen in the figure PE results match very closely with measurements having a mean
is excellent over the wide range of azimuth angles considered error of only 0.5 dB and a standard deviation of error of 4.6 dB.
(0–26.6 ). The example further testifies the effectiveness of the Most of the contribution to the standard deviation comes from
boundary treatment of Section II-B. the deep shadow region where the fields are already attenuated
We will next a show comparison with measurement for a 30 dB. Furthermore, it is seen for this case that the cross-polar
lossy building-shaped obstacle. Van Dooren and Herben [21] component is about 30 dB below the copolar component.
1726 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 51, NO. 8, AUGUST 2003
APPENDIX
(61)
With
(62)
(63)
(41)
(42)
(64)
1728 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 51, NO. 8, AUGUST 2003
(65) [17] A. Papoulis, The Fourier Integral and its Applications. New York: Mc-
Graw-Hill, 1962.
[18] K. Rizk, R. Valenzuela, S. Fortune, D. Chizhik, and F. Gardiol, “Lateral,
(66) Full 3D and Vertical Plane Propagation in Microcells and Small Cells,”
European Cooperation in the Field of Scientific and Technical Research,
(67) Switzerland, COST 259 Tech. Doc., TD(98)-47, 1998.
[19] T. B. A. Senior and J. L. Volakis, Approximate Boundary Conditions in
(68) Electromagnetics. Piscataway, NJ: IEEE Press, 1995.
[20] F. Tappert, “The parabolic equation method,” in Wave Propagation in
Underwater Acoustics, J. B. Keller and J. S. Papadakis, Eds. New
(69) York: Springer-Verlag, 1977, pp. 224–287.
[21] G. A. J. van Dooren and M. H. A. J. Herben, “Field strength prediction
(70) behind lossy dielectric objects by using UTD,” Electron. Lett., vol. 29,
no. 11, pp. 1016–1018, May 1993.
where denotes element-by-element array multiplication. [22] J. R. Wait, “Scattering of a plane wave from a circular dielectric cylinder
at oblique incidence,” Canad. J. Phys., vol. 33, no. 1, pp. 189–195, Jan.
1955.
REFERENCES [23] A. A. Zaporozhets and M. F. Levy, “Modeling of radiowave propagation
in urban environment with parabolic equation method,” Electron. Lett.,
[1] H. L. Bertoni, Radio Propagation for Modern Wireless Systems. Upper
vol. 32, no. 17, pp. 1615–1616, Aug. 1996.
Saddle River, NJ: Prentice Hall PTR, 2000.
[24] A. A. Zaporozhets, “Application of the vector parabolic equation
[2] L. M. Brekhovskikh, Waves in Layered Media. San Diego, CA: Aca-
method to urban propagation problems,” Proc. Inst. Elec. Eng., Pt. H,
demic, 1980.
vol. 146, pp. 253–256, 1999.
[3] C. Brennan and P. J. Cullen, “Tabulated interaction method for UHF
[25] C. A. Zelley and C. C. Constantinou, “A three-dimensional parabolic
terrain propagation problems,” IEEE Trans. Antennas Propagat., vol.
equation applied to VHF/UHF propagation over irregular terrain,” IEEE
46, pp. 738–739, May 1998.
Trans. Antennas Propagat., vol. 47, pp. 1586–1596, Oct. 1999.
[4] W. C. Chew, Waves and Fields in Inhomogeneous Media. Piscataway,
NJ: IEEE Press, 1995.
[5] F. Collino and P. Joly, “Splitting of operators, alternate directions, and
paraxial approximations for the three-dimensional wave equation,”
SIAM J. Sci. Comp., vol. 16, no. 5, pp. 1019–1048, Sept. 1995.
[6] C. J. Haslett, “Modeling and measurements of the diffraction of mi- Ramakrishna Janaswamy (S’82–M’83–SM’93–
crowaves by buildings,” IEE Proc. Microwawe Antennas Propag., vol. F’03) received the Bachelor’s degree in 1981 from
141, no. 5, pp. 397–401, Oct. 1994. REC Warangal, India, the Master’s degree in 1983
[7] J. T. Hviid, J. B. Andersen, J. Toftgard, and J. Bojer, “Terrain based from IIT Kharagpur, India, both in electronics and
propagation model for rural area–An integral equation approach,” IEEE communications engineering, and the Ph.D. degree
Trans. Antennas Propagat., vol. 43, pp. 41–46, Jan. 1995. in electrical engineering in 1986 from the University
[8] R. Janaswamy, Radiowave Propagation & Smart Antennas for Wireless of Massachusetts, Amherst.
Communications. Boston, MA: Kluwer, 2000. From August 1986 to May 1987, he was an
[9] , “A fast finite difference method for propagation predictions over Assistant Professor of Electrical Engineering at
irregular, inhomogeneous terrain,” IEEE Trans. Antennas Propagat., Wilkes University, Wilkes Barre, PA. From August
vol. 42, pp. 1257–1267, Sept. 1994. 1987 to August 2001, he was on the faculty of the
[10] R. Janaswamy and J. B. Andersen, “Path loss predictions in urban areas Department of Electrical and Computer Engineering, Naval Postgraduate
with irregular terrain topography,” Wireless Personal Commun., vol. 12, School, Monterey, CA. In September 2001, he joined the Department of
no. 3, pp. 255–268, Mar. 2000. Also see the addendum in Wireless Per- Electrical and Computer Engineering, University of Massachusetts, Amherst,
sonal Commun., vol. 14, no. 3, pp. 303-304, Sept.2000.. where is a Full Professor. He was a Visiting Researcher at the Center for
[11] R. Janaswamy, “Radio wave propagation over a nonconstant immittance PersonKommunikation, Aalborg, Denmark from September 1997 to June 1998.
plane,” Radio Sci., vol. 36, no. 3, pp. 387–405, May-June 2001. His research interests include deterministic and stochastic radio wave propa-
[12] F. B. Jensen, W. A. Kuperman, M. B. Porter, and H. Schmidt, Compu- gation modeling, antenna analysis and design, and applied electromagnetics.
tational Ocean Acoustics. New York: Amer. Institute Phys., 1994. He is the author of the book Radiowave Propagation and Smart Antennas for
[13] J. T. Johnson et al., “A method of moments model for VHF propaga- Wireless Communications (Boston, MA: Kluwer, 2000) and a contributing
tion,” IEEE Trans Antennas Propagat., vol. 45, pp. 115–125, Jan. 1997. author in Handbook of Antennas in Wireless Communications, L. Godara (Ed.)
[14] S.-C. Kim et al., “Radio propagation measurement and prediction using (Boca Raton, FL: CRC, 2001).
three-dimensional ray tracing in urban environments at 908 MHz and Dr. Janaswamy was the recipient of the R. W. P. King Prize Paper Award of the
1.9 GHz,” IEEE Trans. Veh. Technol., vol. 48, pp. 931–946, May 1999. IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION in 1995 and a recipient
[15] T. Kurner, D. J. Cichon, and W. Wiesbek, “Concepts and results for 3D of the IEEE 3rd Millennium Medal from the Santa Clara Valley Section in 2000.
digital terrain based wave propagation models–An overview,” IEEE J. He also received Certificates of Recognition for Research Contributions from
Select. Areas Commun., vol. 11, pp. 1002–1012, July 1993. ONR/ASEE in 1995 and from the Naval Postgraduate School, Monterey, CA
[16] M. F. Levy, Parabolic Equation Methods for Electromagnetic Wave in 1991. He is an elected member of U.S. National Committee of International
Propagation. London, U.K.: IEE Press, 2000. Union of Radio Science, Commissions B and F.