You are on page 1of 10

IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS, VOL. 00, NO.

00, 2016 1

1 A Groundwave Propagation Model Using a Fast


2 Far-Field Approximation
3 Gokhan Apaydin, Cai-Cheng Lu, Levent Sevgi, and Weng Cho Chew

5
4 :Abstract—A fast far-field approximation (FAFFA), which is sim-
6 ple to use, is applied to groundwave propagation modeling from
7 a nonpenetrable surface with both soft and hard boundaries. The
8 results are validated against available reference models as well as
9 compared to other numerical methods such as split step parabolic

of
10 equation model and the method of moments.
11 Index Terms—Fast far-field approximation (FAFFA), ground-
12 wave propagation, irregular terrain, knife edge, method of mo-
13 ments (MoM), split step parabolic equation (SSPE), two-ray (2Ray)
14 model.

I. INTRODUCTION
15

16
17
18
19
20
21
22
G ROUNDWAVE propagation has long been an important
option for medium- and long-range communications
at medium frequency (MF) (0.3–3 MHz) and high frequency
(HF) (3–30 MHz) bands. Classical MF/HF broadcast and com-
munication systems, HF/VHF radars, intelligent transportation
systems, etc., require an understanding of propagation char-
acteristics over the Earth’s surface along realistic propagation
ro Fig. 1. Groundwave propagation scenario over irregular terrain using a
FAFFA. Horizontal field profile is above ground zero. Here, p and p  are the
centers of the pth and p  th groups, respectively. The FF and NF contribution of
the pth group are also shown.

decades [5]–[12]. There are also free propagation prediction 39


EP
tools (e.g., [8], [10], and [12]). 40
23 paths (the list will always be incomplete, but see, for example, Although there are few attempts in three dimensions (3-D), 41
24 [1]–[14]). the problem has long been modeled in 2-D and handled in 42
25 The International Telecommunication Union (ITU) regu- transverse-longitudinal (i.e., height-range) coordinates within a 43
26 lates electromagnetic (EM) groundwave propagation predic- homogenous medium illuminated by a line source located at 44
27 tion. ITU-R Recommendation P-368-9 [4] fully covers both the initial range (usually, z = 0) at a specified height (x = xs ). 45
28 homogeneous and mixed-path propagation problems and pro- Fig. 1 shows the simplest 2-D propagation environment with 46
29 vides a set of predicted field strength versus distance curves perfectly electrical conductor (PEC) ground. The environment 47
30 for vertically polarized EM waves in the MF and HF bands is open longitudinally (z → ±∞) and semiopen transversely 48
31 for a variety of ground conductivity (σ) and relative permit- (x → ∞). Although Fig. 1 shows that the horizontal field profile 49
32 tivity (εr ) values. This recommendation is based on early an- uses the height with respect to x = 0, the height above local 50
IEE

33 alytical approximate models such as the ray method [1], the terrain can also be used for this calculation. 51
34 mixed-path model [2], and the mode method [3] (note that The mathematical model of this scenario is given by the wave 52
35 a free package, GRWAVE, mentioned there, is available at equation [5] 53
36 http://www.itu.int/oth/R0A0400000F/en). Numerical methods
such as parabolic equation model (PEM) and method of mo-  
37
∂2 ∂2
38 ments (MoM) have also been used extensively for the last few + + k2 ψ(z, x) = 0 (1)
∂z 2 ∂x2
Manuscript received October 25, 2016; accepted November 29, 2016. Date
of publication; date of current version. where z and x are the longitudinal and vertical coordinates, 54
G. Apaydin is with Electromagnetic Consulting, Gaziantep 27560, Turkey respectively, k is the free-space wavenumber, and the wave 55
(e-mail: g.apaydin@gmail.com).
C.-C. Lu is with the Department of Electrical and Computer Engineering, Uni- function ψ(z, x) represents the electric and magnetic fields for 56
versity of Kentucky, Lexington, KY 40506 USA (e-mail: cclu@engr.uky.edu). the horizontal and vertical polarizations, respectively. Here, the 57
L. Sevgi is with the Department of Electrical and Electronics Engineering, electric/magnetic field has only one nonzero component Ey /Hy 58
Okan University, Istanbul 34959, Turkey (e-mail: levent.sevgi@okan.edu.tr).
W. C. Chew is with the Department of Electrical and Computer Engineering, for horizontal/vertical polarization, respectively. The boundary 59
University of Illinois at Urbana-Champaign, Urbana, IL 61801 USA (e-mail: condition (BC) along x → ∞ and z → ±∞ satisfies the radia- 60
w-chew@uiuc.edu). tion condition, and both the Dirichlet BC and Neumann BC are 61
Color versions of one or more of the figures in this letter are available online
at http://ieeexplore.ieee.org. taken into consideration at the PEC surface for horizontal and 62
Digital Object Identifier 10.1109/LAWP.2016.2636802 vertical polarizations, respectively. 63

1536-1225 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
2 IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS, VOL. 00, NO. 00, 2016

64 The simplest propagation scenario used in analytical model- The solution of MoM is time-consuming as the number of 104
65 ing is the flat earth with PEC surface under line source excitation. segments increases for long range and/or high-frequency prop- 105
66 The total field at the observer is simply obtained via vector ad- agation problems. For example, N = 100 000 for 100 km 2-D 106
67 dition of direct and ground-reflected rays. This model is called flat-earth scenario at 30 MHz. Direct solution of matrix inversion 107
68 two-ray (2Ray) model [15]. requires O(N 3 ) operations. On the other hand, unknowns can 108
69 On the single knife-edge problem, the ray summation ap- be iteratively solved for using conjugate gradient method with a 109
70 proach is based on the construction of four different rays, related computational complexity of O(N 2 ) per iteration. Fast iterative 110
71 reflection, and diffraction coefficients using Fresnel integrals solvers have been developed to speed up the matrix-vector prod- 111
72 [15]. uct operation. This letter applies the fast far-field approximation 112
(FAFFA) to accelerate the iterative solution [21], [22]. 113
73 II. MOM-BASED GROUNDWAVE PROPAGATION
III. FAFFA-BASED GROUNDWAVE PROPAGATION 114
74 An MoM technique can be used to find propagation of hori-

of
75 zontally and vertically polarized waves by using the electric field Consider EFIE given in (5), first, N unknowns are divided 115
76 integral equation (EFIE) and the magnetic field integral equation into groups with M unknowns 116
77 (MFIE), respectively [16]–[19]. Open region propagation over
 
N /M
78 irregular ground has been successfully modeled with MoM [8], Am n un = uinc (rm ) , m ∈ Gp , p = 1, 2, . . . , N/M.
79 [20]. In the classical MoM, the integral equation is converted p  =1 n ∈G p 
80 to the corresponding matrix equation via the discretization of (7)
81 the surface. Then, an N × N system of equations is constructed Here, Gp and Gp  denote groups p and p , respectively (see 117
82
83

84
85
and solved numerically [15].
The total field for EFIE is

u (r) = u (r) −

(1)
inc i


4 S ∂n
∂u (r  ) (1)

ro
H0 (k |r − r  |) dr (2)

where H0 is the first-kind Hankel function with order zero, r =


z ẑ + xx̂ is above the ground, r  = z  ẑ + x x̂ is on the surface S,
∂u (r  ) /∂n = n̂ · ∇u (r  ) are unknowns, and n̂ is the outward
Fig. 1). The idea of FAFFA is to use FF approximation for large
distances between groups [21]–[23]. The summation in (8) has
FF and near-field (NF) contribution

i  
4 
Δsn un
p ∈FF n ∈G p 

2 eik r n m

πi krn m
118
119
120
EP
86
normal unit vector. “ˆ” here implies unit vectors. The total field i   (1)
87 + Δsn un H0 (krn m ) = uinc (rm ) (8)
88 on the surface is zero for horizontal polarization; therefore, 4 
p ∈NF n ∈G p 
89 placing r on the surface leads to an integral equation of the first
90 order for horizontal polarization as follows: where Δsn is the segment length, r n m = r n p  + r p  p + r pm . 121
 Under rp  p  rn p  and rp  p  rpm , the FF contribution is ap- 122
i (1) ∂u (r  ) 
H0 (k |r − r  |) dr = uinc (r) . (3) proximated by [22] 123
4 S ∂n ⎛ ⎞
Q1 91 The discretized form of (4) is ⎜ ⎛ ⎞⎟
⎜ ⎟

N ⎜ ⎜ ⎟⎟
⎜ ⎜ ⎟ ⎟
Am n un = uinc (rm ) ⎜  ⎟
(4) ⎜eik r̂ p  p ·rp m ⎜

i ik r p  p
Δs ik r̂ p  p ·r n p ⎟⎟

IEE

⎜ e u e .
⎟⎟
n n
n =1 ⎜ 8πkrp  p
p  ∈FF⎜ ⎝ n ∈G p  ⎠ ⎟
⎜   ⎟
92 where A is the N × N impedance matrix of the ground. Solution ⎜ ⎟
⎝ Aggregation ⎠
93 of this system yields the unknown segment currents. Superposi-  
94 tion of the contributions of the segment currents via the Green’s 
Translation

95 function of the problem yields the ground-scattered field. Fi- Disaggregation
96 nally, the total field is obtained by adding the incident field. (9)
97 The total field for MFIE is described as Equation (9) has three steps. The first step, aggregation, com- 124
 putes the total field at a group center from the subscatterers of the 125
i ∂ (1)
u (r) = u (r) +
inc
u (r  ) H (k |r − r  |) dr (5)
4 S ∂n 0 group. The second step translates the field from one group center 126
to another. Then, the third step, disaggregation, distributes the 127
98 and placing r on the surface, the integral equation of the second field at another group center to each subscatterers of the group 128
99 order for vertical polarization is obtained as [23]. Fig. 2 shows the CPU time growth of classical MoM and 129

u (r) i ∂ (1) FAFFA per iteration with the number of unknowns (N ). The data 130
− p.v. u (r  ) H0 (k |r − r  |) dr = uinc (r) .
2 4 S ∂n for this figure are simulated on an Intel Core i5 CPU at 2.5 GHz. 131
(6) The classical MoM may be more efficient for small problems 132
100 Here, the unknown u (r  ) is available instead of ∂u (r  ) /∂n, (e.g., when N is less than a thousand). In order to further reduce 133
101 and p.v. means the Cauchy principal value of the integral. the computational cost, interpolation and smoothing techniques 134
102 Note that 25λ to 30λ additional surface length is enough for are used for aggregation and disaggregation steps, respectively. 135
103 infinite surface assumption in the MoM-based application. Hence, the computational complexity is of O(N 4/3 ) and the 136
APAYDIN et al.: GROUNDWAVE PROPAGATION MODEL USING A FAFFA 3

of
Fig. 2. CPU time growth of classical MoM and FAFFA per iteration as the
number of unknowns is increased. Fig. 5. Field versus height at two specified ranges using FAFFA, SSPE, and
4Ray models (horizontal polarization, f = 30 MHz, nonpenetrable flat surface
with 100-m-high knife edge at 800-m range illuminated by a line source at
200 m height, Δs = λ/10, N = 2300, M = 14 for FAFFA, Δz = 10 m, Δx
= 0.5 m, X max = 2000 m for SSPE).

ro
EP
Fig. 3. Field versus range at 8-m height using FAFFA, SSPE, and 2Ray models
(vertical polarization, f = 10 MHz, nonpenetrable flat surface illuminated by
a line source at 100 m height, Δs = λ/10, N = 20 600, M = 28 for FAFFA,
Δz = 40 m, Δx = 0.5 m, X max = 5000 m for SSPE).
IEE

Fig. 6. Field versus range at 8 m height above local terrain using FAFFA
and SSPE models (vertical polarization, f = 10 MHz, nonpenetrable Gaussian
surface with 100- and 500-m heights illuminated by a line source at 100 m
height, Δs = λ/10, N = 13 600, M = 24 for FAFFA, Δz = 20 m, Δx = 0.5
m, X max = 5000 m for SSPE).

Fig. 4. Field versus range at 30-m height using FAFFA, two-way SSPE, and agation over PEC flat earth is taken into account where the field 142
4Ray models (horizontal polarization, f = 10 MHz, nonpenetrable flat surface at a chosen observer for a given source location is simply formed 143
with 90-m-high knife edge at 12 km range illuminated by a line source at
100 m height, Δs = λ/10, N = 6600, M = 19 for FAFFA, Δz = 40 m, Δx
by the addition of direct and ground-reflected rays. This is called 144
= 0.5 m, X max = 3000 m for SSPE). 2Ray model. Fig. 3 shows field versus range variation. Here, the 145
results of the well-known split step parabolic equation (SSPE) 146
model are also included. The computations were performed us- 147
137 total field memory requirement is of O(N ) for M = N 1/3 [22],
ing MATLAB-based free codes given in [15]. As observed, very 148
138 [23].
good agreement is obtained among the three models. 149
The second example used in tests and comparisons is the 150
139 IV. NUMERICAL EXAMPLES AND COMPARISONS
knife-edge problem and the 4Ray model (see [15, Sec. 12.6] 151
140 The new FAFFA-based groundwave propagation model has for this model and free MATLAB codes). Figs. 4 and 5 be- 152
141 been tested against various scenarios and calibrated. First, prop- long to this scenario. A 90-m-high knife edge is located at a 153
4 IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS, VOL. 00, NO. 00, 2016

15◦ ; therefore, the FAFFA-based model can be used to achieve 187


more accurate results while considering higher slope changes 188
of terrain in 2-D and 3-D propagation problems. 189

REFERENCES 190

[1] K. A. Norton, “The propagation of radio waves over the surface of the 191
earth and in the upper atmosphere-Part I,” Proc. Inst. Radio Eng., vol. 24, 192
no. 10, pp. 1367–1387, Oct. 1936. 193
[2] G. Millington, “Ground-wave propagation over an inhomogeneous smooth 194
earth,” Proc. IEE (London), vol. 96, no. 39, pp. 53–64, Jan. 1949. 195
[3] J. R. Wait, Electromagnetic Waves in Stratified Media. New York, NY, 196
USA: Pergamon, 1962. 197
[4] ITU-R, Recommendations P-368–9, “Groundwave Propagation Curves 198

of
for Frequencies between 10 kHz and 30 MHz,” International Telecommu- 199
nications Union, Geneva, Switzerland, Feb. 2007. 200 Q2
[5] M. F. Levy, Parabolic Equation Methods for Electromagnetic Wave Prop- 201
agation. London, U.K.: IEE, 2000. 202
[6] L. Sevgi and L. B. Felsen, “A new algorithm for ground wave propagation 203
based on a hybrid ray-mode approach,” Int. J. Numer. Model., vol. 11, 204
no. 2, pp. 87–103, Mar. 1998. 205
[7] L. Sevgi, F. Akleman, and L. B. Felsen, “Groundwave propagation model- 206
Fig. 7. Field versus range at 8-m height above local terrain using FAFFA ing: Problem-matched analytical formulations and direct numerical tech- 207
and SSPE models (vertical polarization, f = 10 MHz, nonpenetrable irregular niques,” IEEE Antennas Propag. Mag., vol. 44, no. 1, pp. 55–75, Feb. 208

154
155
156
157
158
ro
surface illuminated by a line source at 100 m height, Δs = λ/10, N = 15 600,
M = 25 for FAFFA, Δz = 20 m, Δx = 0.5 m, X max = 5000 m for SSPE).

distance of 12 km from the line source over a PEC flat earth.


The height of the source is 100 m. The frequency is chosen as
10 MHz. Again, SSPE model is also used in the comparisons.
Field versus range at 30 m fixed height is plotted in Fig. 4. Simi-
lar field versus height plots at 30 MHz are given in Fig. 5 at short
2002.
[8] F. Akleman and L. Sevgi, “A novel MoM- and SSPE-based groundwave-
propagation field-strength prediction simulator,” IEEE Antennas Propag.
Mag., vol. 49, no. 5, pp. 69–82, Oct. 2007.
[9] L. Sevgi, “Groundwave modeling and simulation strategies and path loss
prediction virtual tools,” IEEE Trans. Antennas Propag., vol. 55, no. 6,
pp. 1591–1598, Jun. 2007.
[10] G. Apaydin and L. Sevgi, “FEM-based surface wave multi-mixed-path
propagator and path loss predictions,” IEEE Antennas Wireless Propag.
Lett., vol. 8, pp. 1010–1013, 2009.
209
210
211
212
213
214
215
216
217
218
EP
[11] G. Apaydin and L. Sevgi, “Numerical investigations of and path loss 219
159 ranges beyond the 100-m-high knife edge located at 800 m. As predictions for surface wave propagation over sea paths including hilly 220
160 observed, the agreement among the models is impressive. Note island transitions,” IEEE Trans. Antennas Propag., vol. 58, no. 4, pp. 1302– 221
161 that the interference of the direct and ground-reflected waves is 1314, Apr. 2010. 222
[12] O. Ozgun, G. Apaydin, M. Kuzuoglu, and L. Sevgi, “PETOOL: MATLAB- 223
162 significant before the knife edge at 90 m, as seen in Fig. 4. based one-way and two-way split-step parabolic equation tool for ra- 224
163 The other two examples belong to terrain modeling along diowave propagation over variable terrain,” Comput. Phys. Commun., 225
164 PEC flat earth without analytical models where comparisons vol. 182, no. 12, pp. 2638–2654, Dec. 2011. 226
[13] G. Apaydin, O. Ozgun, M. Kuzuoglu, and L. Sevgi, “A novel two-way 227
165 only against numerical models are possible. First, a Gaussian- finite-element parabolic equation groundwave propagation tool: Tests with 228
166 shaped mountain problem is used. Fig. 6 shows field versus canonical structures and calibration,” IEEE Trans. Geosci. Remote Sens., 229
167 range at 8 m fixed height above local terrain using FAFFA and vol. 49, no. 8, pp. 2887–2899, Aug. 2011. 230
[14] L. Zhou, X. Xi, J. Liu, and N. Yu, “LF ground-wave propagation over 231
168 SSPE models. Here, Gaussian-shaped PEC hills with different irregular terrain,” IEEE Trans. Antennas Propag., vol. 59, no. 4, pp. 1254– 232
IEE

169 heights (100 and 500 m) are located at a distance of 20 km. The 1260, Apr. 2011. 233
170 line source’s height is also 100 m. As observed, FAFFA results [15] L. Sevgi, Electromagnetic Modeling and Simulation. Piscataway, NJ, 234
USA: IEEE Press, 2014. 235
171 are in very good agreement with the SSPE results. [16] R. F. Harrington, Field Computation by Moment Method. New York, NY, 236
172 The last example presented in Fig. 7 belongs to an arbitrary- USA: IEEE Press, 1993 (1st ed. 1968). 237
173 shaped irregular terrain profile. As observed, the agreement be- [17] J. T. Johnson, “On the canonical grid method for two-dimensional scatter- 238
ing problems,” IEEE Trans. Antennas Propag., vol. 46, no. 3, pp. 297–302, 239
174 tween FAFFA and SSPE is very good. Mar. 1998. 240
[18] E. Arvas and L. Sevgi, “A tutorial on the method of moments,” IEEE 241
Antennas Propag. Mag., vol. 54, no. 3, pp. 260–275, Jun. 2012. 242
175 V. CONCLUSION [19] G. Apaydin, F. Hacivelioglu, L. Sevgi, and P. Y. Ufimtsev, “Wedge 243
diffracted waves excited by a line source: Method of moments (MoM) 244
176 A FAFFA-based groundwave propagation model, which is modeling of fringe waves,” IEEE Trans. Antennas Propag., vol. 62, no. 8, 245
177 easy to apply, is introduced and tested against various propa- pp. 4368–4371, Aug. 2014. 246
178 gation models with and without analytical reference solutions. [20] C. Brennan and J. Cullen, “Application of the fast far-field approximation 247
to the computation of UHF path loss over irregular terrain,” IEEE Trans. 248
179 The FAFFA-based model is very promising in terms of accuracy, Antennas Propag., vol. 46, no. 6, pp. 881–890, Jun. 1998. 249
180 memory requirements, and computation times. It is also much [21] W. C. Chew, T. J. Cui, and J. M. Song, “A FAFFA-MLFMA algorithm 250
181 simpler than the multilevel fast multipole algorithm [21]–[23]. for electromagnetic scattering,” IEEE Trans. Antennas Propag., vol. 50, 251
no. 11, pp. 1641–1649, Nov. 2002. 252
182 Although the SSPE model is useful and fast for long-range [22] C. C. Lu and W. C. Chew, “Fast far field approximation for calculating 253
183 groundwave propagation analysis, there are some drawbacks the RCS of large objects,” Microw. Opt. Technol. Lett., vol. 8, no. 5, 254
184 such as the introduction of nonphysical fluctuations caused by pp. 238–241, Apr. 1995. 255
[23] W. C. Chew, J. M. Jin, E. Michielssen, and J. M. Song, Fast and Effi- 256
185 the incorporation of an upper boundary and the inability of the cient Algorithms in Computational Electromagnetics. Boston, MA, USA: 257
186 model to calculate waves having propagation angles greater than Artech House, 2001. 258
QUERIES 259

Q1. Author: Please confirm if “(4)” is the correct equation number in line 90 since the equation directly below it is numbered as 260
“(4)”. 261
Q2. Author: Please check whether Ref. [4] is ok as set. 262

of
ro
EP
IEE
IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS, VOL. 00, NO. 00, 2016 1

1 A Groundwave Propagation Model Using a Fast


2 Far-Field Approximation
3 Gokhan Apaydin, Cai-Cheng Lu, Levent Sevgi, and Weng Cho Chew

5
4 :Abstract—A fast far-field approximation (FAFFA), which is sim-
6 ple to use, is applied to groundwave propagation modeling from
7 a nonpenetrable surface with both soft and hard boundaries. The
8 results are validated against available reference models as well as
9 compared to other numerical methods such as split step parabolic

of
10 equation model and the method of moments.
11 Index Terms—Fast far-field approximation (FAFFA), ground-
12 wave propagation, irregular terrain, knife edge, method of mo-
13 ments (MoM), split step parabolic equation (SSPE), two-ray (2Ray)
14 model.

I. INTRODUCTION
15

16
17
18
19
20
21
22
G ROUNDWAVE propagation has long been an important
option for medium- and long-range communications
at medium frequency (MF) (0.3–3 MHz) and high frequency
(HF) (3–30 MHz) bands. Classical MF/HF broadcast and com-
munication systems, HF/VHF radars, intelligent transportation
systems, etc., require an understanding of propagation char-
acteristics over the Earth’s surface along realistic propagation
ro Fig. 1. Groundwave propagation scenario over irregular terrain using a
FAFFA. Horizontal field profile is above ground zero. Here, p and p  are the
centers of the pth and p  th groups, respectively. The FF and NF contribution of
the pth group are also shown.

decades [5]–[12]. There are also free propagation prediction 39


EP
tools (e.g., [8], [10], and [12]). 40
23 paths (the list will always be incomplete, but see, for example, Although there are few attempts in three dimensions (3-D), 41
24 [1]–[14]). the problem has long been modeled in 2-D and handled in 42
25 The International Telecommunication Union (ITU) regu- transverse-longitudinal (i.e., height-range) coordinates within a 43
26 lates electromagnetic (EM) groundwave propagation predic- homogenous medium illuminated by a line source located at 44
27 tion. ITU-R Recommendation P-368-9 [4] fully covers both the initial range (usually, z = 0) at a specified height (x = xs ). 45
28 homogeneous and mixed-path propagation problems and pro- Fig. 1 shows the simplest 2-D propagation environment with 46
29 vides a set of predicted field strength versus distance curves perfectly electrical conductor (PEC) ground. The environment 47
30 for vertically polarized EM waves in the MF and HF bands is open longitudinally (z → ±∞) and semiopen transversely 48
31 for a variety of ground conductivity (σ) and relative permit- (x → ∞). Although Fig. 1 shows that the horizontal field profile 49
32 tivity (εr ) values. This recommendation is based on early an- uses the height with respect to x = 0, the height above local 50
IEE

33 alytical approximate models such as the ray method [1], the terrain can also be used for this calculation. 51
34 mixed-path model [2], and the mode method [3] (note that The mathematical model of this scenario is given by the wave 52
35 a free package, GRWAVE, mentioned there, is available at equation [5] 53
36 http://www.itu.int/oth/R0A0400000F/en). Numerical methods
such as parabolic equation model (PEM) and method of mo-  
37
∂2 ∂2
38 ments (MoM) have also been used extensively for the last few + + k2 ψ(z, x) = 0 (1)
∂z 2 ∂x2
Manuscript received October 25, 2016; accepted November 29, 2016. Date
of publication; date of current version. where z and x are the longitudinal and vertical coordinates, 54
G. Apaydin is with Electromagnetic Consulting, Gaziantep 27560, Turkey respectively, k is the free-space wavenumber, and the wave 55
(e-mail: g.apaydin@gmail.com).
C.-C. Lu is with the Department of Electrical and Computer Engineering, Uni- function ψ(z, x) represents the electric and magnetic fields for 56
versity of Kentucky, Lexington, KY 40506 USA (e-mail: cclu@engr.uky.edu). the horizontal and vertical polarizations, respectively. Here, the 57
L. Sevgi is with the Department of Electrical and Electronics Engineering, electric/magnetic field has only one nonzero component Ey /Hy 58
Okan University, Istanbul 34959, Turkey (e-mail: levent.sevgi@okan.edu.tr).
W. C. Chew is with the Department of Electrical and Computer Engineering, for horizontal/vertical polarization, respectively. The boundary 59
University of Illinois at Urbana-Champaign, Urbana, IL 61801 USA (e-mail: condition (BC) along x → ∞ and z → ±∞ satisfies the radia- 60
w-chew@uiuc.edu). tion condition, and both the Dirichlet BC and Neumann BC are 61
Color versions of one or more of the figures in this letter are available online
at http://ieeexplore.ieee.org. taken into consideration at the PEC surface for horizontal and 62
Digital Object Identifier 10.1109/LAWP.2016.2636802 vertical polarizations, respectively. 63

1536-1225 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
2 IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS, VOL. 00, NO. 00, 2016

64 The simplest propagation scenario used in analytical model- The solution of MoM is time-consuming as the number of 104
65 ing is the flat earth with PEC surface under line source excitation. segments increases for long range and/or high-frequency prop- 105
66 The total field at the observer is simply obtained via vector ad- agation problems. For example, N = 100 000 for 100 km 2-D 106
67 dition of direct and ground-reflected rays. This model is called flat-earth scenario at 30 MHz. Direct solution of matrix inversion 107
68 two-ray (2Ray) model [15]. requires O(N 3 ) operations. On the other hand, unknowns can 108
69 On the single knife-edge problem, the ray summation ap- be iteratively solved for using conjugate gradient method with a 109
70 proach is based on the construction of four different rays, related computational complexity of O(N 2 ) per iteration. Fast iterative 110
71 reflection, and diffraction coefficients using Fresnel integrals solvers have been developed to speed up the matrix-vector prod- 111
72 [15]. uct operation. This letter applies the fast far-field approximation 112
(FAFFA) to accelerate the iterative solution [21], [22]. 113
73 II. MOM-BASED GROUNDWAVE PROPAGATION
III. FAFFA-BASED GROUNDWAVE PROPAGATION 114
74 An MoM technique can be used to find propagation of hori-

of
75 zontally and vertically polarized waves by using the electric field Consider EFIE given in (5), first, N unknowns are divided 115
76 integral equation (EFIE) and the magnetic field integral equation into groups with M unknowns 116
77 (MFIE), respectively [16]–[19]. Open region propagation over
 
N /M
78 irregular ground has been successfully modeled with MoM [8], Am n un = uinc (rm ) , m ∈ Gp , p = 1, 2, . . . , N/M.
79 [20]. In the classical MoM, the integral equation is converted p  =1 n ∈G p 
80 to the corresponding matrix equation via the discretization of (7)
81 the surface. Then, an N × N system of equations is constructed Here, Gp and Gp  denote groups p and p , respectively (see 117
82
83

84
85
and solved numerically [15].
The total field for EFIE is

u (r) = u (r) −

(1)
inc i


4 S ∂n
∂u (r  ) (1)

ro
H0 (k |r − r  |) dr (2)

where H0 is the first-kind Hankel function with order zero, r =


z ẑ + xx̂ is above the ground, r  = z  ẑ + x x̂ is on the surface S,
∂u (r  ) /∂n = n̂ · ∇u (r  ) are unknowns, and n̂ is the outward
Fig. 1). The idea of FAFFA is to use FF approximation for large
distances between groups [21]–[23]. The summation in (8) has
FF and near-field (NF) contribution

i  
4 
Δsn un
p ∈FF n ∈G p 

2 eik r n m

πi krn m
118
119
120
EP
86
normal unit vector. “ˆ” here implies unit vectors. The total field i   (1)
87 + Δsn un H0 (krn m ) = uinc (rm ) (8)
88 on the surface is zero for horizontal polarization; therefore, 4 
p ∈NF n ∈G p 
89 placing r on the surface leads to an integral equation of the first
90 order for horizontal polarization as follows: where Δsn is the segment length, r n m = r n p  + r p  p + r pm . 121
 Under rp  p  rn p  and rp  p  rpm , the FF contribution is ap- 122
i (1) ∂u (r  ) 
H0 (k |r − r  |) dr = uinc (r) . (3) proximated by [22] 123
4 S ∂n ⎛ ⎞
Q1 91 The discretized form of (4) is ⎜ ⎛ ⎞⎟
⎜ ⎟

N ⎜ ⎜ ⎟⎟
⎜ ⎜ ⎟ ⎟
Am n un = uinc (rm ) ⎜  ⎟
(4) ⎜eik r̂ p  p ·rp m ⎜

i ik r p  p
Δs ik r̂ p  p ·r n p ⎟⎟

IEE

⎜ e u e .
⎟⎟
n n
n =1 ⎜ 8πkrp  p
p  ∈FF⎜ ⎝ n ∈G p  ⎠ ⎟
⎜   ⎟
92 where A is the N × N impedance matrix of the ground. Solution ⎜ ⎟
⎝ Aggregation ⎠
93 of this system yields the unknown segment currents. Superposi-  
94 tion of the contributions of the segment currents via the Green’s 
Translation

95 function of the problem yields the ground-scattered field. Fi- Disaggregation
96 nally, the total field is obtained by adding the incident field. (9)
97 The total field for MFIE is described as Equation (9) has three steps. The first step, aggregation, com- 124
 putes the total field at a group center from the subscatterers of the 125
i ∂ (1)
u (r) = u (r) +
inc
u (r  ) H (k |r − r  |) dr (5)
4 S ∂n 0 group. The second step translates the field from one group center 126
to another. Then, the third step, disaggregation, distributes the 127
98 and placing r on the surface, the integral equation of the second field at another group center to each subscatterers of the group 128
99 order for vertical polarization is obtained as [23]. Fig. 2 shows the CPU time growth of classical MoM and 129

u (r) i ∂ (1) FAFFA per iteration with the number of unknowns (N ). The data 130
− p.v. u (r  ) H0 (k |r − r  |) dr = uinc (r) .
2 4 S ∂n for this figure are simulated on an Intel Core i5 CPU at 2.5 GHz. 131
(6) The classical MoM may be more efficient for small problems 132
100 Here, the unknown u (r  ) is available instead of ∂u (r  ) /∂n, (e.g., when N is less than a thousand). In order to further reduce 133
101 and p.v. means the Cauchy principal value of the integral. the computational cost, interpolation and smoothing techniques 134
102 Note that 25λ to 30λ additional surface length is enough for are used for aggregation and disaggregation steps, respectively. 135
103 infinite surface assumption in the MoM-based application. Hence, the computational complexity is of O(N 4/3 ) and the 136
APAYDIN et al.: GROUNDWAVE PROPAGATION MODEL USING A FAFFA 3

of
Fig. 2. CPU time growth of classical MoM and FAFFA per iteration as the
number of unknowns is increased. Fig. 5. Field versus height at two specified ranges using FAFFA, SSPE, and
4Ray models (horizontal polarization, f = 30 MHz, nonpenetrable flat surface
with 100-m-high knife edge at 800-m range illuminated by a line source at
200 m height, Δs = λ/10, N = 2300, M = 14 for FAFFA, Δz = 10 m, Δx
= 0.5 m, X max = 2000 m for SSPE).

ro
EP
Fig. 3. Field versus range at 8-m height using FAFFA, SSPE, and 2Ray models
(vertical polarization, f = 10 MHz, nonpenetrable flat surface illuminated by
a line source at 100 m height, Δs = λ/10, N = 20 600, M = 28 for FAFFA,
Δz = 40 m, Δx = 0.5 m, X max = 5000 m for SSPE).
IEE

Fig. 6. Field versus range at 8 m height above local terrain using FAFFA
and SSPE models (vertical polarization, f = 10 MHz, nonpenetrable Gaussian
surface with 100- and 500-m heights illuminated by a line source at 100 m
height, Δs = λ/10, N = 13 600, M = 24 for FAFFA, Δz = 20 m, Δx = 0.5
m, X max = 5000 m for SSPE).

Fig. 4. Field versus range at 30-m height using FAFFA, two-way SSPE, and agation over PEC flat earth is taken into account where the field 142
4Ray models (horizontal polarization, f = 10 MHz, nonpenetrable flat surface at a chosen observer for a given source location is simply formed 143
with 90-m-high knife edge at 12 km range illuminated by a line source at
100 m height, Δs = λ/10, N = 6600, M = 19 for FAFFA, Δz = 40 m, Δx
by the addition of direct and ground-reflected rays. This is called 144
= 0.5 m, X max = 3000 m for SSPE). 2Ray model. Fig. 3 shows field versus range variation. Here, the 145
results of the well-known split step parabolic equation (SSPE) 146
model are also included. The computations were performed us- 147
137 total field memory requirement is of O(N ) for M = N 1/3 [22],
ing MATLAB-based free codes given in [15]. As observed, very 148
138 [23].
good agreement is obtained among the three models. 149
The second example used in tests and comparisons is the 150
139 IV. NUMERICAL EXAMPLES AND COMPARISONS
knife-edge problem and the 4Ray model (see [15, Sec. 12.6] 151
140 The new FAFFA-based groundwave propagation model has for this model and free MATLAB codes). Figs. 4 and 5 be- 152
141 been tested against various scenarios and calibrated. First, prop- long to this scenario. A 90-m-high knife edge is located at a 153
4 IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS, VOL. 00, NO. 00, 2016

15◦ ; therefore, the FAFFA-based model can be used to achieve 187


more accurate results while considering higher slope changes 188
of terrain in 2-D and 3-D propagation problems. 189

REFERENCES 190

[1] K. A. Norton, “The propagation of radio waves over the surface of the 191
earth and in the upper atmosphere-Part I,” Proc. Inst. Radio Eng., vol. 24, 192
no. 10, pp. 1367–1387, Oct. 1936. 193
[2] G. Millington, “Ground-wave propagation over an inhomogeneous smooth 194
earth,” Proc. IEE (London), vol. 96, no. 39, pp. 53–64, Jan. 1949. 195
[3] J. R. Wait, Electromagnetic Waves in Stratified Media. New York, NY, 196
USA: Pergamon, 1962. 197
[4] ITU-R, Recommendations P-368–9, “Groundwave Propagation Curves 198

of
for Frequencies between 10 kHz and 30 MHz,” International Telecommu- 199
nications Union, Geneva, Switzerland, Feb. 2007. 200 Q2
[5] M. F. Levy, Parabolic Equation Methods for Electromagnetic Wave Prop- 201
agation. London, U.K.: IEE, 2000. 202
[6] L. Sevgi and L. B. Felsen, “A new algorithm for ground wave propagation 203
based on a hybrid ray-mode approach,” Int. J. Numer. Model., vol. 11, 204
no. 2, pp. 87–103, Mar. 1998. 205
[7] L. Sevgi, F. Akleman, and L. B. Felsen, “Groundwave propagation model- 206
Fig. 7. Field versus range at 8-m height above local terrain using FAFFA ing: Problem-matched analytical formulations and direct numerical tech- 207
and SSPE models (vertical polarization, f = 10 MHz, nonpenetrable irregular niques,” IEEE Antennas Propag. Mag., vol. 44, no. 1, pp. 55–75, Feb. 208

154
155
156
157
158
ro
surface illuminated by a line source at 100 m height, Δs = λ/10, N = 15 600,
M = 25 for FAFFA, Δz = 20 m, Δx = 0.5 m, X max = 5000 m for SSPE).

distance of 12 km from the line source over a PEC flat earth.


The height of the source is 100 m. The frequency is chosen as
10 MHz. Again, SSPE model is also used in the comparisons.
Field versus range at 30 m fixed height is plotted in Fig. 4. Simi-
lar field versus height plots at 30 MHz are given in Fig. 5 at short
2002.
[8] F. Akleman and L. Sevgi, “A novel MoM- and SSPE-based groundwave-
propagation field-strength prediction simulator,” IEEE Antennas Propag.
Mag., vol. 49, no. 5, pp. 69–82, Oct. 2007.
[9] L. Sevgi, “Groundwave modeling and simulation strategies and path loss
prediction virtual tools,” IEEE Trans. Antennas Propag., vol. 55, no. 6,
pp. 1591–1598, Jun. 2007.
[10] G. Apaydin and L. Sevgi, “FEM-based surface wave multi-mixed-path
propagator and path loss predictions,” IEEE Antennas Wireless Propag.
Lett., vol. 8, pp. 1010–1013, 2009.
209
210
211
212
213
214
215
216
217
218
EP
[11] G. Apaydin and L. Sevgi, “Numerical investigations of and path loss 219
159 ranges beyond the 100-m-high knife edge located at 800 m. As predictions for surface wave propagation over sea paths including hilly 220
160 observed, the agreement among the models is impressive. Note island transitions,” IEEE Trans. Antennas Propag., vol. 58, no. 4, pp. 1302– 221
161 that the interference of the direct and ground-reflected waves is 1314, Apr. 2010. 222
[12] O. Ozgun, G. Apaydin, M. Kuzuoglu, and L. Sevgi, “PETOOL: MATLAB- 223
162 significant before the knife edge at 90 m, as seen in Fig. 4. based one-way and two-way split-step parabolic equation tool for ra- 224
163 The other two examples belong to terrain modeling along diowave propagation over variable terrain,” Comput. Phys. Commun., 225
164 PEC flat earth without analytical models where comparisons vol. 182, no. 12, pp. 2638–2654, Dec. 2011. 226
[13] G. Apaydin, O. Ozgun, M. Kuzuoglu, and L. Sevgi, “A novel two-way 227
165 only against numerical models are possible. First, a Gaussian- finite-element parabolic equation groundwave propagation tool: Tests with 228
166 shaped mountain problem is used. Fig. 6 shows field versus canonical structures and calibration,” IEEE Trans. Geosci. Remote Sens., 229
167 range at 8 m fixed height above local terrain using FAFFA and vol. 49, no. 8, pp. 2887–2899, Aug. 2011. 230
[14] L. Zhou, X. Xi, J. Liu, and N. Yu, “LF ground-wave propagation over 231
168 SSPE models. Here, Gaussian-shaped PEC hills with different irregular terrain,” IEEE Trans. Antennas Propag., vol. 59, no. 4, pp. 1254– 232
IEE

169 heights (100 and 500 m) are located at a distance of 20 km. The 1260, Apr. 2011. 233
170 line source’s height is also 100 m. As observed, FAFFA results [15] L. Sevgi, Electromagnetic Modeling and Simulation. Piscataway, NJ, 234
USA: IEEE Press, 2014. 235
171 are in very good agreement with the SSPE results. [16] R. F. Harrington, Field Computation by Moment Method. New York, NY, 236
172 The last example presented in Fig. 7 belongs to an arbitrary- USA: IEEE Press, 1993 (1st ed. 1968). 237
173 shaped irregular terrain profile. As observed, the agreement be- [17] J. T. Johnson, “On the canonical grid method for two-dimensional scatter- 238
ing problems,” IEEE Trans. Antennas Propag., vol. 46, no. 3, pp. 297–302, 239
174 tween FAFFA and SSPE is very good. Mar. 1998. 240
[18] E. Arvas and L. Sevgi, “A tutorial on the method of moments,” IEEE 241
Antennas Propag. Mag., vol. 54, no. 3, pp. 260–275, Jun. 2012. 242
175 V. CONCLUSION [19] G. Apaydin, F. Hacivelioglu, L. Sevgi, and P. Y. Ufimtsev, “Wedge 243
diffracted waves excited by a line source: Method of moments (MoM) 244
176 A FAFFA-based groundwave propagation model, which is modeling of fringe waves,” IEEE Trans. Antennas Propag., vol. 62, no. 8, 245
177 easy to apply, is introduced and tested against various propa- pp. 4368–4371, Aug. 2014. 246
178 gation models with and without analytical reference solutions. [20] C. Brennan and J. Cullen, “Application of the fast far-field approximation 247
to the computation of UHF path loss over irregular terrain,” IEEE Trans. 248
179 The FAFFA-based model is very promising in terms of accuracy, Antennas Propag., vol. 46, no. 6, pp. 881–890, Jun. 1998. 249
180 memory requirements, and computation times. It is also much [21] W. C. Chew, T. J. Cui, and J. M. Song, “A FAFFA-MLFMA algorithm 250
181 simpler than the multilevel fast multipole algorithm [21]–[23]. for electromagnetic scattering,” IEEE Trans. Antennas Propag., vol. 50, 251
no. 11, pp. 1641–1649, Nov. 2002. 252
182 Although the SSPE model is useful and fast for long-range [22] C. C. Lu and W. C. Chew, “Fast far field approximation for calculating 253
183 groundwave propagation analysis, there are some drawbacks the RCS of large objects,” Microw. Opt. Technol. Lett., vol. 8, no. 5, 254
184 such as the introduction of nonphysical fluctuations caused by pp. 238–241, Apr. 1995. 255
[23] W. C. Chew, J. M. Jin, E. Michielssen, and J. M. Song, Fast and Effi- 256
185 the incorporation of an upper boundary and the inability of the cient Algorithms in Computational Electromagnetics. Boston, MA, USA: 257
186 model to calculate waves having propagation angles greater than Artech House, 2001. 258
QUERIES 259

Q1. Author: Please confirm if “(4)” is the correct equation number in line 90 since the equation directly below it is numbered as 260
“(4)”. 261
Q2. Author: Please check whether Ref. [4] is ok as set. 262

of
ro
EP
IEE

You might also like