Received 28 March 2023; revised 21 May 2023; accepted 3 June 2023.

Date of publication 12 June 2023; date of current version 29 June 2023.

Digital Object Identifier 10.1109/OJAP.2023.3285233

Physics-Based Coherent Modeling of Long-Range

Millimeter-Wave Propagation and Scattering in Rain
BEHZAD YEKTAKHAH (Member, IEEE), AND KAMAL SARABANDI (Life Fellow, IEEE)
Department of Electrical Engineering and Computer Science, University of Michigan at Ann Arbor, Ann Arbor, MI 48103, USA
CORRESPONDING AUTHOR: B. YEKTAKHAH (e-mail: [email protected])

ABSTRACT Precipitation and in particular rain present a random medium with constituent particles
having dimensions comparable to the wavelength at millimeter-wave frequencies. Due to considerable
scattering of waves from such particles, significant signal attenuation and phase front aberration can take
place which are important factors that must be considered in modern communication system or radar
design. This paper presents a fast and accurate numerical method for phase coherent modeling of wave
propagation and scattering in random media, like rain. In this approach, the random medium is divided
into sufficiently large finite slabs and each slab is modeled as a network with multiple input/output ports
where ports represent rays with different polarizations propagating in different directions entering and
leaving the slabs. The cascaded networks of slabs can be used to find the response of the random medium
to any arbitrary source. The proposed method considers multiple scattering among all scatterers and thus
the entire physics of wave-particle interactions is accounted for. In addition, the method can model both
the mean and the fluctuating parts of waves in the random medium. To demonstrate the applicability of
the proposed method, it is used for the calculation of the specific attenuation in rain media as a function
of rain rates and frequency within the millimeter-wave band. The results show a good match with the
available rain attenuation data in the literature.

INDEX TERMS Communication system, millimeter-wave, radar, rain, random media, specific attenuation.

I. INTRODUCTION With short wavelengths, ranging from 1 mm to 10 mm,

T HE WIDE spectrum available at the millimeter-wave

(mm-wave) band is the main reason for the develop-
ment of the future high data rate wireless communication
waves at mm-wave frequencies exhibit higher interactions
with the objects in the environment and thus the propagation
characteristics at mm-wave frequencies are significantly dif-
systems (such as 5G) [1] and Internet of Things (IoT) [2]. ferent from those at lower frequencies (the microwave band
The small wavelength at mm-wave frequencies and low and lower) in terms of path loss, foliage loss, phase aberra-
atmospheric absorption at frequencies around 77 GHz enable tion, and diffraction [6], [7], [8], [9]. This can pose many
design of high resolution short and long-range radars with challenges in design of mm-wave communication and radar
compact form factor for industrial applications (such as systems. While the free space path loss can be compensated
storage tank level measurement) and future advanced driver- by utilizing higher gain antennas, the propagation loss due to
assistance systems (ADAS). In automotive industry, radars rain, foliage, and the resulting aberration in phase front must
are key sensors [3] that provide target position and velocity be carefully considered in design of mm-wave systems. This
with high accuracy in addition to being able to operate in is more critical in case of non-line-of-sight links in cluttered
inclement weather conditions with a much lower cost com- environments such as in urban areas [10]. Various empiri-
pared to LIDAR. Efforts are being made to introduce higher cal models have been proposed for estimating attenuation of
mm-wave frequencies for automotive radars to reduce size mm-waves in line-of-sight and non-line-of-sight propagation
and/or improve resolution to be competitive with LIDAR in urban areas [11], [12]. In case of attenuation through rain,
resolution [4], [5]. several measurement results have been reported for various

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

588 VOLUME 4, 2023

rain rates at different frequencies [13], [14]. These results a certain range, the fluctuating part of the field becomes
are used for design of mm-wave communication systems. dominant that invalidates the specific attenuation predicted
In mm-wave radar design, specifically in coherent systems by single scattering approximation. This shows the need for
such as synthetic aperture and MIMO radars, in addition to a better technique to accurately predict the mean and fluc-
the attenuation, the phase fluctuation of the received signal tuating parts of fields propagating in rain and experiencing
in rain across the antenna array plays an important role for multiple scattering.
accurate modeling of the radar performance in the presence In [23], propagation and scattering in a 2-D random
of rain. medium is modeled by a scattering matrix (S-matrix) to
Rain can be viewed as a discrete random medium con- find the specific attenuation over long propagation distance
sisting of particles of different sizes with a distribution in spectral domain. In this method, first, the random medium
function that depends on the rain rate. The size distribution is divided into slabs and then each slab, similar to a network
of rain droplets is described by well-known models such as with multiple ports, is described by an S-matrix. Then,
Marshall-Palmer [15] and gamma distribution [16]. Due to the S-matrices of different slabs are cascaded to obtain an
high dielectric constant of water and the size of the droplets S-matrix representing the whole random medium.
that can be a considerable fraction of wavelength, a strong This paper extends the idea of S-matrix method to 3-D
scattering can occur in the medium. The strong scattering random media to analyze coherent wave propagation and
from particles results in significant attenuation, absorption, scattering at mm-wave band in rainy atmosphere. The
and phase aberration due to multiple scattering. Thus, the method considers multiple scattering and provides amplitude
received signal propagated through rain has random phase and phase data for both forward propagating and backscat-
and amplitude fluctuation which can be decomposed into tered waves for arbitrary sources and thus eliminating the
two components: a) the mean field which accounts for the need for full wave simulations which is not possible for very
average attenuation through the rain and b) fluctuating or large propagation and scattering problems. In the method, the
incoherent part that accounts for the phase and amplitude random medium is divided into slabs with a certain thickness
fluctuations in the received signal. Several models have been and for each slab, the input and output wave fronts at two
developed to model the mean and the fluctuating parts of the interfaces of the slab are represented by the superposition
fields at different frequency ranges and for different types of plane waves with different polarizations and propagating
of random media [17], [18], [19], [20]. For modeling of the in different directions. Each plane wave specifies a port for
mean field, dielectric mixing formula can be used at low the slab equivalent network. Then, the S-matrix is defined to
frequencies to model the random medium as an effective relate the amplitude/phase of each output port to the ampli-
medium. In case of week permittivity fluctuations, pertur- tude/phase of all input ports at both interfaces. Then the
bation techniques such as Born approximation can be used. S-matrices of different slabs are cascaded to calculate scat-
In case of sparse discrete random media, methods based on tering from and propagation through the random medium
single scattering approximation such as Foldy’s method [17] (including the phase information) for any source. This way
have been developed. multiple scattering among scatterers in different slabs are
For modeling the fluctuating part of the field, radiative taken into consideration. This model is used to calculate
transfer (RT) theory and T-matrix method can be applied. the specific attenuation of mm-wave signals as a function
RT method is applicable to sparse discrete random media of rain rate and frequency over long distances. The simu-
and is based on the law of conservation of energy and sin- lations are performed for different rain rates, ranging from
gle scattering approximation [18]. In [19], a method based few mm/hr to tens of mm/hr, within the frequency range
on ABCD-matrix is developed for analysis of propagation from 28 GHz (5G) to 230 GHz. Comparison of the simu-
(including phase information) in 2D periodic random media. lation results with the data available in literature shows the
Full-wave T-matrix method can provide phase information of accuracy of the presented method.
the waves; however, it is only applicable to small 3D prob- S-matrix method is a general method that can be applied to
lems involving small number of particles. T-matrix method any dense/sparse random media. In this paper, rain media is
applicability also depends on the volume fraction of the ran- selected to demonstrate the capability of S-matrix method,
dom medium. A discussion of T-matrix method limitation in since rain media have been thoroughly studied and vari-
3D problems can be found in [20]. In [21], a Monte Carlo ous empirical models and measurement data are available
technique was presented for analyses of multi-scattering for propagation and scattering in rain at different frequency
in rain considering spherical shape for rain particles and bands. Other examples of the applicability of the method are
excluding the phase and polarization information of waves. analysis of propagation and scattering in foliage, sandstorm
For long propagation through rain, multiple scattering (useful in applications such as radar imaging, target acqui-
effects become significant and single scattering approxima- sition, and helicopter landing [9]), as well as cloud/fog or
tion such as Foldy’s method can only provide an upper limit murky water at optical frequencies.
for the specific attenuation. In addition, the single scattering The advantages of the S-matrix method are: 1) it is based
model is valid only in short range propagation where the on full-wave analysis; 2) it is applicable to dense or sparse
mean field is dominant. It is shown in [22], [23] that after random media; 3) it tracks the polarization of the wave

VOLUME 4, 2023 589

YEKTAKHAH AND SARABANDI: PHYSICS-BASED COHERENT MODELING

Az component can be obtained from the other two

components of A:
  kx Ax + ky Ay kx Ax + ky Ay
A±
z kx , ky = − =−  . (4)
kz± ± k2 − k2 − k2 x y

In (4), A+
z is for the wave that is travelling in +z direction
and A− z is for the wave that is travelling in − z direction.
Assuming a large but finite lateral dimensions for the slabs
FIGURE 1. Random medium layered in z-direction.
and periodic arrangement within xy-plane, discrete values for
along the propagation; 4) it tracks phase aberration to the kx and ky must be used for plane waves in S-matrix method.
phase front needed for assessing the performance of coherent In this case, discrete IFT can be applied to find A(kx , ky ) for
imaging systems (such as SAR or MIMO radars) operating discrete values of kx and ky . In the S-matrix method only
in random media; 5) it can be used for random media with propagating waves with real kz are included in the analyses
scatterers having arbitrary shapes. and thus kx and ky must be sampled only in the circular
This paper is organized as follows. In Section II, the region kx2 + ky2 ≤ k2 (according to (2) for obtaining real
method based on the S-matrix extended to 3-D is intro- values for kz ). To include all the propagating waves in the
duced. Calculation of the elements of S-matrix is detailed in expansion, the spatial sampling interval (d) must be less
Section III. S-matrix simulation results for rain are given in than or equal to λ/2 (λ is the wavelength in the medium and
Section IV and compared to the available data in literature. is equal to 2π/k). Since, in general, the spectrum of fields is
Concluding remarks are given in Section V. not zero outside the region kx2 + ky2 ≤ k2 , aliasing in spectral
domain may occur. To reduce aliasing, d must be chosen
II. 3-D S-MATRIX METHOD a value less than λ/2. However, smaller d requires larger
The first step in S-matrix method is to divide the random computational resources and processing time. In this paper,
medium to multiple slabs with a certain thickness as shown d is selected to be λ/4. This value results in significant
in Fig. 1. In this paper, the direction of propagation is suppression of aliasing effect in the region kx2 + ky2 ≤ k2
assumed to be in z-direction and the random medium is while does not require large computational resources. To
divided along the direction of propagation into Ns slabs and further reduce the effect of aliasing, a window function can
slabs’ interfaces are parallel to xy-plane. The interface with be applied to the fields before expansion to reduce high
larger z-position value is denoted by z+ and the other one frequency spectral components.
is denoted by z− (Fig. 1). The time dependency of waves According to the discrete Fourier transform properties,
is ejωt . spacing of k between adjacent samples of kx and ky results
At the two interfaces of each slab, the incoming and out- in a periodic reconstructed field in spatial domain with period
going electric field waves are first expanded by plane waves 2π/k along both x and y directions. In this paper, k is
traveling in different directions. Electric field vector can be equal to k/35. This results in lateral dimensions of 35λ×
represented by infinite number of plane waves as: 35λ for uniquely calculating the fields from their spectral
 +∞  +∞ representation. With this spacing, total of 71×71 (=5041)
1  
E(x, y, z) = A kx , ky e−j(kx x+ky y+kz z) dkx dky , (1) samples are generated for kx and ky in the grid −k ≤ kx ,
4π 2 −∞ −∞
ky ≤ k. Among them, 3841 samples are inside the visible
where A is a vector in the spectral domain. The integrand region (kx2 + ky2 ≤ k2 ) representing propagating waves. In
in (1) represents a plane wave traveling in the direction the rest of the paper, N is the number of propagating plane
kx
x + kyy + kz
z and since ∇ · E = 0, it is noted that A. waves in the wave spectral expansion.
(kx
x+ky y+kzz) = 0. Substituting (1) into the wave equation, A wave propagating in + z-direction, which is called
kz is obtained as: Forward Propagating Wave, is expanded by N plane waves
 + +
kz = kz± = ± k2 − kx2 − ky2 (2) with wave vector denoted by kn (=kx,n  x + ky,n
y + kz,n z) for
the nth (n = 1, 2, . . . , N) plane wave. The superscript “+”
In (2), k is the wave number of the background medium indicates the quantity is related to the forward propagating
in which the discrete scatterers of the random medium is wave. The forward propagating wave at z− interface repre-
included. Also, kz = kz+ is for the waves that are traveling sents the incident wave on a slab with the corresponding
in +z direction, and kz = kz− is for the waves that are +
electric field vector denoted by An . The forward propagat-
traveling in −z direction. ing wave at z+ interface represents the wave exiting the slab
The tangential components of A (Ax and Ay ) can be +
with corresponding electric field vector denoted by Bn .
found by inverse Fourier transform (IFT) of the tangential The field propagating in − z-direction, which is referred
components of E (Ex and Ey ) at a constant plane z = z0 : to as Backward Propagating Wave, is expanded by N plane
 +∞  +∞ − −
  waves with wave vector denoted by kn (=kx,n  x+ky,n y+kz,n z)
Ax,y kx , ky = ejkz z0 Ex,y (x, y, z0 )ej(kx x+ky y) dxdy. (3)
−∞ −∞ for the nth (n = 1, 2, . . . , N) plane wave. The superscript

590 VOLUME 4, 2023

 vertical and horizontal unit vectors for nth plane wave is
constant and as a result only the complex coefficients A± n,v ,
± ± ±
An,h , Bn,v , and Bn,h in (6) are modified by propagation of
waves and their interactions with scatterers.
In the S-matrix method, each slab is modeled as a network
with 4N inputs and 4N outputs as illustrated in Fig. 2(a).
± ±
In this case, the coefficients A± ±
n,v , An,h , Bn,v , and Bn,h are
related by a 4N× 4N matrix, S, as follows:
⎛ − ⎞ ⎛ + ⎞
B1,v A1,v
⎜ B− ⎟ ⎜ A+ ⎟
⎜ 2,v ⎟ ⎜ 2,v ⎟
⎜ . ⎟ ⎜ . ⎟
⎜ .. ⎟ ⎜ .. ⎟
⎜ ⎟ ⎜ ⎟
⎜B− ⎟ ⎜A+ ⎟
⎜ N,v ⎟ ⎜ N,v ⎟
⎜ B− ⎟ ⎜ A+ ⎟
⎜ 1,h ⎟ ⎜ 1,h ⎟
⎜ − ⎟ ⎜ + ⎟
⎜ B2,h ⎟ ⎜ A2,h ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. ⎟ ⎜ .. ⎟
⎜ . ⎟ ⎜ . ⎟
⎜ − ⎟ ⎜ ⎟
⎜BN,h ⎟  ⎜A+ ⎟
⎜ ⎟ ⎜ N,h ⎟
⎜ ⎟ = [S11 ] [S12 ] ⎜ ⎟. (7)
⎜ + ⎟ ⎜ ⎟
⎜ B ⎟  [S21 ][S22 ] ⎜ A− ⎟
FIGURE 2. (a) Waves entering and exiting the slab. (b) Cascading the slabs. ⎜ 1,v ⎟ ⎜ 1,v ⎟
(c) direct and scattered waves in a slab. ⎜ B+ ⎟ ⎜ A− ⎟
⎜ 2,v ⎟ S
⎜ 2,v ⎟
⎜ . ⎟ ⎜ . ⎟
⎜ . ⎟ ⎜ . ⎟
“−” indicates the quantity is related to the backward ⎜ . ⎟ ⎜ . ⎟
⎜ + ⎟ ⎜ − ⎟
propagating wave. The backward propagating wave at z+ ⎜BN,v ⎟ ⎜AN,v ⎟
⎜ + ⎟ ⎜ − ⎟
interface indicates an incident wave on slab and the cor- ⎜ B1,h ⎟ ⎜ A1,h ⎟
− ⎜ + ⎟ ⎜ − ⎟
responding electric field vector is denoted by An . The ⎜B ⎟ ⎜A ⎟
− ⎜ 2,h ⎟ ⎜ 2,h ⎟
backward propagating wave at z interface indicates a wave ⎜ . ⎟ ⎜ . ⎟
⎝ .. ⎠ ⎝ .. ⎠
exiting from the slab and the corresponding electric field
−
vector is denoted by Bn . B+
N,h A−
N,h
To simplify the calculation of interaction of the plane Similar to the network theory, if the z− interface is con-
waves with scatterers within a slab, it is convenient to sidered as port 1 and z+ interface is considered as port 2,
decompose each plane wave into orthogonal vertical (v) and in (7):
horizontal (h) polarized components. For this decomposition, • the 2N× 2N matrix [S11 ] relates the waves exiting the
the unit vectors for vertical polarization (
v) and horizontal slab at z− interface to the waves entering the slab at
polarization (
h) are defined by: z− interface,
⎧ ± • the 2N× 2N matrix [S21 ] relates the waves exiting the
⎪ ±
⎨h±
z×kn
⎪ n = ± slab at z+ interface to the waves entering the slab at
±z×kn
 ± . (5) z− interface,
⎪
⎪ v±
±
hn ×kn
⎩ n = ± ± • the 2N× 2N matrix [S12 ] relates the waves exiting the
hn ×kn
slab at z− interface to the waves entering the slab at
In (5), 
h+
n and  v+
n are unit vectors for forward propagating
z+ interface, and
−
waves and hn and  v−
n are for backward propagating waves.
• the 2N× 2N matrix [S22 ] relates the waves exiting the
The electric field for the nth plane wave is decomposed into slab at z+ interface to the waves entering the slab at
vertical and horizontal polarizations as: z+ interface.
⎧ ± ± ±
In the next section the elements of S matrix is derived for
⎪
⎪ An = An · v±
n  v± ± ±
n + An · hn hn a slab.
⎪
⎪
⎨ = A±  ± ± ±
n,v vn + An,h hn
In the field expansion in (1), the z-position of the origin
± ± ± (6) of the coordinate system is not important. Assuming that
⎪ v± v± ± ±
⎪ Bn = Bn ·
⎪
⎪ n  n + Bn · hn hn
the fields are known at z = z0 (source plane), the fields at
⎩ ± ±
= B± n,vv±
n + Bn,h hn . z = z0 +zd (observation plane) are obtained by using (1) and
± ± (3) as:
In (6), A± ±
n,v , An,h , Bn,v , and Bn,h are complex scalars. In
this case, each of forward propagating wave and backward Ex,y (x, y, z0 + zd )
 +∞  +∞   +∞  +∞ 
propagating wave is expanded by 2N plane waves at each 1  
Ex,y x , y , z0 ej(kx x +ky y ) dx dy
 
= 2
ejkz z0
of the two interfaces. For a fixed number of plane waves 4π −∞ −∞ −∞ −∞
throughout the analyses, the wave vector and directions of × e−j(kx x+ky y) e−jkz (z0 +zd ) dkx dky . (8)

VOLUME 4, 2023 591

YEKTAKHAH AND SARABANDI: PHYSICS-BASED COHERENT MODELING

In (8), the terms including z0 are cancelled out and the exiting the last slab. On the other hand, using the first 2N
term including zd (e−jkz zd ) remains. The same is true for elements results the total electric field for the wave exiting
z-component of the field. This shows that the evaluation of the first slab.
fields does not depend on the choice of the z-position for the
origin of coordinate system, and it only depends on the dif- III. ELEMENTS OF THE S-MATRIX FOR A SLAB
ference between the z-positions of the source plane and the CONTAINING SCATTERERS
observation plane, i.e., zd . The term e−jkz zd shows the prop- The wave exiting a slab at an interface can be decom-
agation of planes waves in + z or −
z direction (depending posed into waves entered the slab at the other interface and
on kz ) from the source plane to the observation plane. This directly propagated to the output, which is referred to as the
term is included in the elements of S matrix as discussed in direct wave, and waves entered the slab from both interfaces
the next section. and scattered by the scatterers, which is referred to as the
If plane wave coefficients are known for v and h polariza- scattered waves. This is illustrated in Fig. 2(c).
tions, denoted by Cn,v and Cn,h , respectively, at a plane, the The contribution of the direct wave is represented by the
corresponding total field at the same plane can be calculated S-matrix Sdirect and that of the scattered waves is denoted
from: by the S-matrix Sscattered . The S-matrix of the slab is equal
to Sdirect + Sscattered .

N
 
hn e−j(kx,n x+ky,n y) .
The method described in Section II is accurate if the lateral
E(x, y) = vn + Cn,v
Cn,v (9)
extent of the slabs is large. The accuracy degrades when
1
the propagation distance (the length of the cascaded slabs)
As an example, if the wave exiting the slab is calculated becomes large. This mainly affect the mean-field component
vn , and 
at z+ interface, then Cn,v , Cn,h ,  hn in (9) are B+
n,v , of the propagating wave. To circumvent this difficulty and
+
Bn,h , + +
vn , and hn , respectively. enhance the accuracy of field computation with moderate size
If the S-matrices are known for slab A and slab B (SA of the lateral dimensions of the slabs, a correction approach
and SB , respectively), then the equivalent S-matrix for the is also developed.
cascaded slab C in Fig. 2(b), SC , can be obtained by the
formulas presented in [24] for general multi-port networks. A. CALCULATION OF THE DIRECT WAVE USING
The submatrices in (7) for Sc are related to those of SA and GREEN’S FUNCTION
SB by: To calculate the direct wave at the output plane, Green’s
⎧         −1  
⎪
⎪ SC,11 = SA,11 + SA,12 SB,11 I − SA,22 SB,11 SA,21
function in conjunction with the field equivalence theo-
⎪
⎪         
⎪
⎨ S −1 rem and image theory are used. In this method, first,
C,21 = SB,21 I − SA,22 SB,11 SA,21
          −1    for each plane wave entering the slab, the electric field
⎪
⎪ S = S S + S S B,11 I − SA,22 SB,11 SA,22 SB,12
⎪
⎪ C,12 A,12 B,12 A,12
⎪
⎩     
SC,22 = SB,22 + SB,21 I − SA,22 SB,11
  −1  
SA,22 SB,12 .
 (E(x, y)) is calculated at the input interface over the region
π
− k ≤ x, y ≤ π/k (extent of the slab) using (9). Assuming
(10)
the fields are zero in the region outside the extent of the
Here, I is a 2N× 2N identity matrix. slab, using the equivalence theorem and image theory, the
To analyze a random medium, first, the S-matrices for equivalent magnetic currents can be calculated as:
all slabs are calculated separately and then the overall S- n
jm = −2
n × En (x, y), (11)
matrix is obtained by cascading all slabs using (10). Then
the fields due to a source (or multiple sources) are evalu- where n is + z for forward propagating waves and − z for
ated and expanded as incoming waves to form the vector backward propagating waves. Using the tangential compo-
on the right-hand-side of (7), i.e., the source vector. If the nents of the magnetic current (Jm,y and Jm,x ), the tangential
source is close to the z− interface of the first slab, then components of the electric field can be calculated at the
the electric field of source is evaluated and expanded at z− output interface using the free-space Green’s function as:
⎧
interface as incoming wave. The resulting wave coefficients ⎪ En (r) = − k2  k  k e−jk|r−r |
π π 

form A+ + ⎪
⎪ j + 1
n,v and An,h (n = 1, 2, . . . , N) in the source vector ⎪ π π
⎪ x 4π − − k|r−r | k|r−r |
 
− ⎪
⎪     (r−rk ).z k
in (7) and A− n,v and An,h for this source are zero. If the ⎨ Jm,y
n x , y × |r−r | dxdy
source is close to the z+ interface of the last slab, then  k  k (12)
e−jk|r−r |
π π 
⎪
⎪ n (r) = + k2 +
the electric field of source is evaluated and expanded at z+
1
⎪
⎪ E y − π
− π j | | | |
 
⎪
⎪
4π
    (r−rk ).z k 
k r−r k r−r
interface as incoming wave. The resulting wave coefficients ⎪
⎩ Jm,x
−
n x , y × |r−r | dx dy .
form A− +
n,v and An,h in the source vector in (7) and An,v and
+
An,h for this source are zero. If multiple sources are present, In (12), r is the position vector on the output plane with
the same coefficients in the source vector are added. By the origin at the center of the input plane and is equal to
multiplying the source vector by the S-matrix in (7), the left- x
x+y y+ds z for forward propagating waves and x x+yy−ds z
hand-side vector is obtained. Using the last 2N elements of for backward propagating waves. r = x x+yy is the position
this vector in (9) results the total electric field for the wave vector on the input plane.

592 VOLUME 4, 2023

 1) A local coordinate system is defined for the slab with
the origin at the center of the input plane.
2) For each input plane wave, the phase term modeling
the propagation of the plane wave from the origin of the
local coordinate to the center of the scatterer is calculated.
Assuming that the center of the scatterer is at (xc , yc , zc )
in the local coordinate system, for the nth plane wave this
term is:
FIGURE 3. Ey of a y-directed Hertzian dipole at different distances from the dipole, Pn = e−jkx,n xc e−jkx,n yc e−jkz,n zc . (13)
(a) amplitude and (b) phase (solid line: actual values, dotted line: S-matrix and Green’s
function method).
3) Assuming the response of the scatterer to any input
plane wave is known, the scattered fields due to the scatterer
The tangential fields at the output plane obtained is calculated at the input and output planes considering its
from (12) are expanded in terms of plane waves to find location.
the wave coefficients for the exiting wave from the slab 4) The scattered fields at the input and output planes are
(B coefficients). In this method, each input plane wave at expanded and the resulting coefficients are added to Sscattered .
an interface, affects all plane waves exiting at the other 5) Steps 1 to 4 are repeated for all input plane waves (at
interface and therefore, all elements in Sdirect,21 and Sdirect,12 both interfaces) and also for all scatterers in the slab.
are non-zero. Sdirect,11 and Sdirect,22 are zero matrices. In this paper, for simplicity, the scatterers are assumed to
The power density of a plane wave exiting a slab is be lossy dielectric spheres which have analytical solution for
lower than its value at the input since a part of its power is the scattering due to the incident plane wave (Appendix B).
coupled to another plane waves indicated by the non-zero
non-diagonal elements of Sdirect,21 and Sdirect,12 . This intro-
IV. SIMULATION RESULTS
duces some error in the fields as the waves propagate. Using
In this section, the S-matrix method is employed for cal-
larger periods will result in smaller non-diagonal components
culation of wave propagation and scattering through rain
and less error for a given range. Figure 3 shows amplitude
for different rain rates and within different frequencies at
and phase of the electric field (Ey ) of a y-directed Hertzian
mm-wave band. The simulated specific attenuation is com-
dipole propagating along z-direction as a function of range
pared to the measured values/models for rain reported in the
using the exact solution and the Green’s function/S-matrix
literature.
method. The dipole is 50 mm away from the first slab. The
thickness of each slab (ds ) is 5.13λ0 . The lateral dimension
of slab is 35λ0 × 35λ0 . It can be observed that the calcu- A. STATISTICAL CHARACTERISTICS OF RAIN
lated fields using Green’s function and S-matrix method are Rain as a random medium is defined by drop size distribution
accurate up to a range of about 120λ0 . (DSD). DSD which is denoted by  N(D) is used to find the
To resolve the issue with the limited range of validity for total number of rain droplets per unit volume (ND ):
the Green’s function method, a correction scheme is intro-  D2
duced in Appendix A. In this method, the random medium ND = 
N(D)dD. (14)
is divided into super slabs with a length slightly smaller D1
than the range of validity of the Green’s function method In (14), for 
N(D) with the unit (m-3 cm-1 ), ND is the total
(about 120λ0 for the example considered here). Then the number of rain droplets with diameters (in cm) between D1
analysis is pursued with cascaded super slabs and the cor- and D2 per 1 m3 .
rection algorithm enforced. In this approach the fluctuation 
N(D) has been parametrized by exponential [15], log-
and mean part of the fields entering each super slab are normal, or gamma distribution [16]. Among them, gamma
first estimated and separated and then the mean field com- distribution can accurately model wide range of naturally
ponent is replaced by a mean field directly calculated using occurring DSDs [25]. DSD based on gamma distribution is
the effective medium index of refraction (similar to Foldy’s represented by [16]:
method [17]) obtained from the previous step. With this cor-
rection the accumulation of error in the mean field as the 
N(D) = N0 Dμ e−D , 0 ≤ D ≤ Dmax , (15)
waves travel in the random medium is avoided.
where D (in cm) is the diameter of a sphere with the same
volume as a rain droplet. μ is the distribution parameter and
B. THE SCATTERED WAVE CALCULATION can be a negative, positive, or zero. N0 is a coefficient with
A slab, depending on the concentration of particles, may the unit (m–3 cm–1–μ ).  has the unit (cm–1 ) and is described
contain one or more scatterers. To include the effect of scat- by [16]:
tering in the S-matrix of the slab (Sscattered ), the following 3.67 + μ
steps are followed: = . (16)
D0
VOLUME 4, 2023 593
YEKTAKHAH AND SARABANDI: PHYSICS-BASED COHERENT MODELING

TABLE 1. Rain gamma distribution parameters for different rain rates.

FIGURE 5. Complex dielectric constant of rain droplets (pure water) at 20◦ C.

C. SIMULATION PARAMETERS
For simulations in this paper, the lateral dimensions for
the slabs are chosen to be 35λ0 × 35λ0 where λ0 is the
free-space wavelength at the frequency of simulation. The
selected frequencies are 28, 40, 60, 77, 100, and 230 GHz.
As mentioned in Section II, the spacing for field samples
FIGURE 4. DSD for the rain types in Table 1.
in spatial domain is selected to be λ0 /4 and a raised cosine
TABLE 2. Number of particles per unit volume for different particle diameter ranges window is applied to the fields before expansion. The spac-
at different rain rates.
ing between samples of kx and ky (k) is selected to be
k0 /35. In this case, N is 3841.
In S-matrix method presented here the multiple scatter-
ing within a slab is not considered, and thus the maximum
number of particles per slab must be limited by properly
selecting the slab thickness. With the specified particle dis-
tributions in Table 2 and lateral dimension for slabs, the slab
thickness is selected to be 0.02 m in the simulations. With
this thickness, in the generated random media based on the
specified distributions, the number of scatterers within each
Here D0 is the median diameter in (cm) and is a function slab never exceeds 3. This is important in the accuracy of
of the rain rate, R. the S-matrix method as it does not consider multiple scat-
In this paper, five different rate rates are considered for tering between scatterers within a given slab. For very dense
simulations and their distribution parameters are summarized random media with strong scatterers, the thickness of each
in Table 1 [25] and their DSDs are plotted in Fig. 4. slab must be decreased to limit the maximum number of
In S-matrix method, to speed up the calculations, the scat- scatterers per slab.
tered fields of a particle due to different input plane waves For simulation at 28, 40, 60, and 77 GHz, the number of
are precalculated and stored and used when the S-matrix slabs in each super slab is chosen to be 20 resulting in a total
for any slab containing such particles are calculated. Since thickness of 0.4 m for the super slab. In this case the largest
the scattered fields cannot be stored for large number of super slab thickness in terms of λ0 is 103λ0 (at 100 GHz)
particle types, a limited number of particles with different which is below the range of validity 120λ0 shown in Fig. 3.
diameters are selected. For this purpose, six different diam- At 100 GHz and 230 GHz, the number of slabs per super
eters are considered for spherical rain particles which are slab is 15 (super slab length is 100λ0 ) and 7 (super slab
listed in Table 2. In Table 2, for all selected rain rates, each length is 107λ0 ), respectively.
selected diameter is approximately the mean diameter within A y-polarized electric field distribution (spherical wave-
the indicated range in the second column. For a rain rate, front) with farfield 3-dB beamwidth of 1.65◦ is placed at
ND for each rain droplet type is obtained by integrating the the z− interface of the first slab of the first super slab.

N(D) of that rain rate in the indicated range in the second All the rain particles in the simulation are considered to
column of Table 2. ND s in Table 2 are used to generate the be spheres with dielectric constants indicated in Fig. 5 and
random media for simulations. diameters listed in the first column of Table 2. For each
rain type, five different random media with length of 400 m
B. RAIN DROPLET DIELECTRIC CONSTANT and particle distribution following Table 2 are generated to
The most accurate model for complex dielectric constant perform Monte Carlo simulations.
of water is double-Debye dielectric model (D3M) which is
valid up to 1000 GHz [26]. The complex dielectric constant D. SIMULATION VALIDATION
of pure water versus frequency at 20◦ C is derived using D3M In this section, the accuracy of S-matrix method in a
and shown in Fig. 5. These values are used for simulations. scenario involving a few numbers of scatterers is investigated

594 VOLUME 4, 2023

 frequency is 77 GHz. Assuming the input wave at the bottom
of the first slab is a Gaussian beam with beam waist radius
of 7.5λ0 and Ey = 0 and Ex = 0 at z = 0, the ampli-
tude and phase of forward propagating Ey at z+ interface of
the second slab is shown for S-matrix method in Fig. 6(b)
and 6(d) and for COMSOL in Fig. 6(c) and 6(e), respec-
tively. The amplitude and phase of backward propagating
Ey (due to scattering by the particles) at z− interface of the
first slab is shown for S-matrix method in Fig. 6(f) and 6(h)
and for COMSOL in Fig. 6(g) and 6(i), respectively. The
results confirm the accuracy of S-matrix method in analyses
of large problems including sparse scatterers.

E. SIMULATION RESULTS FOR WAVE PROPAGATION IN

RAIN
In this section, S-matrix simulation results for wave propaga-
tion through rain at different frequencies within the mm-wave
band are presented. The simulation parameters are described
in Section IV-C.
In the simulations, after correction of the electric fields at
the output of a super slab, the electric field is calculated on
the z+ interface of that super slab and the total power of the
y- component of the field, PEy , is calculated by integrating
|Ey |2 /η0 over the slab interface where η0 is the free-space
wave impedance. Because of multiple scattering among par-
ticles within a random medium, the electric field phase and
amplitude fluctuations increase as r increases. This is shown
in Fig. 7(a) and (b) at 77 GHz for rain rate 50 mm/hr. As an
example, Fig. 7(c) shows that the phase fluctuation increases
as the range increases from 20 m to 75 and 150 m. Fig. 7
also shows the ability of the S-matrix method in capturing
the phase and amplitude variation of the waves propagating
in the random medium.
At far distances from the source (r), considering only free
space path loss, the power decreases by 1/r2 . In this case,
multiplying PEy by r2 must remove the effect of free-space
path loss and the resulting quantity must remain constant
over range. However, in the presence of rain, the quantity
PEy × r2 is reduced with r. The rate of this reduction indi-
cates the attention of waves by the rain medium. Figure 8
shows PEy and PEy × r2 versus r for rain rates 2 mm/hr
FIGURE 6. Validation of the S-matrix method. (a) Geometry of the problem. and 50 mm/hr at 77 GHz (the random medium is generated
Amplitude of Ey (in dB) at z + interface of the second slab (z = 0.04 m): (b) S-matrix
method and (c) COMSOL. Phase of Ey (in deg) at z + interface of the second slab: using the particle distribution in Table 2). The slope of the
(d) S-matrix method and (e) COMSOL. Amplitude of Ey (in dB) at z − interface of the
first slab (z = 0 m): (f) S-matrix method and (g) COMSOL. Phase of Ey (in deg) at z −
dashed line in Fig. 8 which is fitted to PEy ×r2 at ranges less
interface of the first slab: (h) S-matrix method and (i) COMSOL. than 100 m, shows the attenuation by rain. The fluctuations
in phase and amplitude of fields at high ranges (as shown in
using COMSOL 3-D full-wave simulations. Since the lateral Fig. 7) result in fluctuations in the power shown in Fig. 8(b).
dimension of slabs is large compared to the wavelength, After a certain range, which is called the knee point in [23],
only a few numbers of cascaded slabs can be simulated by the fluctuating part of fields dominates the mean compo-
COMSOL. Figure 6(a) shows the geometry of the problem nent, resulting in a lower specific attenuation compared to
which includes two cascaded slabs. One spherical particle is the near ranges. This is shown in Fig. 8(b) where moving
present within each slab. The diameter, complex dielectric average of power over ranges of 30 m is shown by a dotted
constant, and location of the center of particles are mentioned line. It can be observed that after 150 m (the knee point),
in Fig. 6(a). The simulation parameters for S-matrix method the dotted line deviates from the dashed line (which means
are the same as those in Section IV-C and the simulation a lower specific attenuation after the knee point). It is noted

VOLUME 4, 2023 595

YEKTAKHAH AND SARABANDI: PHYSICS-BASED COHERENT MODELING

TABLE 3. Specific attenuation in dB/km (before the knee point) for different rain
rates at different frequencies (S-matrix/Foldy’s method).

FIGURE 7. (a) amplitude (in dB) and (b) phase (in deg) in yz-plane and (c) phase (in
yz-plane) at different ranges for the forward propagating electric field (Ey ) in rain
(50 mm/hr) at the outputs of super slabs at 77 GHz.

FIGURE 9. Comparison of the specific attenuation predicted by the S-matrix method

with the ITU-R P836-3 model [27] and the measured results in [28], [29], [30] for rain
rate (a) 2 mm/hr, (b) 5 mm/hr, (c) 10 mm/hr, (d) 25 mm/hr, and (e) 50 mm/hr.

Figure 9 shows the calculated specific attenuation over

FIGURE 8. Power of Ey at the outputs of super slabs at different distances from the
source (r) at 77 GHz for rain rate (a) 2 mm/hr and (b) 50 mm/hr. Slope of the dashed frequency for different rate rates. Figure 9 also shows the
line is the specific attenuation in dB/m. ITU-R P836-3 model [27] and the measured data in [28],
[29], [30]. At the lower rain rates (2, 5, and 10 mm/hr),
that at lower rain rates where the scattering is lower, the the measured specific attenuation has a wide variation range
knee point occurs at larger ranges. which indicates that the specific attenuation is highly sen-
The specific attenuation (before the knee point) for sitive to the particle distribution at low rain rates. Since
different rain rates at different frequencies obtained by the particle distribution for ITU-R model is not specified,
S-matrix method and Monte Carlo simulations is summarized the specific attenuations calculated by S-matrix (based on
in Table 3. The attenuation predicted by Foldy’s method the selected particle distribution in Table 2) deviate from
considering single scattering assumption is also included in ITU-R model, but the results still fall within the range
Table 3. It can be observed that in all cases Foldy’s method of the variation of the measured results. At high rain
overestimates the rain specific attenuation. rates (25 and 50 mm/hr) where the range of variation of

596 VOLUME 4, 2023

specific attenuation at each frequency reduces in measure- the reverse direction for the backward propagating waves).
ment results, ITU-R model and measured results as well Because of the limited range of validity of Green’s func-
as the S-matrix method predict almost the same specific tion method, instead of cascading S-matrices of all slabs
attenuation values. in the random medium, the random medium is divided
Table 4 compares the measured specific attenuation for into super slabs with maximum length of L and only the
different rain rates and frequencies and the simulation results slabs belonging to the same super slab are cascaded as
by S-matrix method with ITU-R P836-3 model. In major- depicted in Fig. 10(a). The length L is selected a value
ity of cases, the measured values show a larger variation less than the maximum range of validity of the Green’s
range than the range predicted by ITU-R model. Even for function method (with the S-matrix method parameters
a fixed rain rate, the reported measured specific attenua- described in the last sections and according to Fig. 3, L
tion has a relatively large range of variations which can be is 120λ0 ).
attributed to different rain particle distributions for a given The correction scheme works as follows. At the z−
rain rate. The S-matrix method results with rain particle dis- interface of the nth super slab, the total corrected electric
t, corrected
tribution in Table 2 fall within the range of measured results. field from the last super slab (E(n−1) ) is expanded and
A comprehensive list of measured results is provided in [13]. used as a source in (7). As shown in the Fig. 10(a), the
components of the resulting total forward propagating wave
V. CONCLUSION at the output of the nth super slab (Ent ) are:
A method for accurate analyses of scattering and propaga-
1) Eni : the incident electric field propagating in free space
tion through random media is presented. In this method, the
from the source (Fig. 10(a)) to the z+ interface of nth
random medium is divided into finite-size slabs and each
super slab without any interaction with scatterers.
slab containing a few numbers of scatterers is modeled as
2) <Ens >: the mean scattered field due to the input
a network with multiple input/output ports described by a
wave that passed through the nth super slab without
S-matrix. The entire random medium is modeled by cascad-
interacting with scatterers.
ing the S-matrices. Unlike available methods, the proposed
3) <Ens,c >: the mean scattered wave due to the interaction
method considers multiple scattering between the scatter-
of the input wave by the scatterers within the nth super
ers and therefore it can be applied in analyses of both
slab.
strong and weak scatterers with any shape. The method
4) 
Ens : the fluctuating part of the scattered wave due to
tracks the polarization and phase information of waves along
the input wave that passed through the nth super slab
the path of propagation. Hence it can provide the phase
without interacting with scatterers.
front aberration and depolarization of the waves propagating
5) 
Ens,c : the fluctuating part of the scattered field due to
through the random medium. This is a useful information
the interaction of the input wave by scatterers within
in performance assessment of coherent imaging radars and
the nth super slab.
communication systems.
6)  s,m
En,n+1 : the scattered field due to multiple scattering
between super slabs n and (n + 1).
APPENDIX A
CORRECTION SCHEME FOR THE S-MATRIX METHOD Each of the first five components in the output wave (Ent )
As mentioned earlier, for estimation of the mean field have a level of error which will be accumulated if it is not
(coherent component) over long distance propagation, the corrected. To do the correction, the fifth component,  Ens,c ,
lateral dimension of the slab must be very large. But must be estimated and separated from the other components.
that will make the computation very inefficient. In this As depicted in Fig. 10(b), the summation of the first five
Appendix, a correction scheme to the S-matrix is described components, Enb , can be obtained by calculating the forward
for forward propagating wave. Total forward propagat- propagating wave at the z+ interface of the nth super slab
ing electric field at an observation point is composed of using the same source as Fig. 10(a) and S-matrix of the nth
three components: 1) the incident electric field due to super slab ([Sn ]). As depicted in Fig. 10(c), the summation of
direct propagation of waves from source to the observa- the first four components (Enc ) can be obtained by calculating
tion point without any interaction with scatterers, 2) the the forward propagating wave at the z+ interface of the nth
mean scattered electric field, and 3) the fluctuating part super slab due to the same source as Fig. 10(a) while the
of the scattered field. In a sparse random medium, the super slab with random particles is replaced by a super slab
summation of incident and mean scattered waves can be with the same lateral dimension and thickness but filled with
obtained by modeling the random medium as a homoge- a homogeneous material with refractive index of neff . As
neous medium with an effective complex refractive index, shown in Fig. 10(c), for this calculation, S is the S-matrix
neff , providing the same attenuation (before the knee point) of the substitute super slab and Enc is the total output field.
for the waves. In this case, Ens,c = Enb − Enc .
The correction scheme is illustrated in Fig. 10 for forward As depicted in Fig. 10(d), at ranges close to the source
propagating waves (the same procedure can be followed in where the region within the beamwidth of the source is

VOLUME 4, 2023 597

YEKTAKHAH AND SARABANDI: PHYSICS-BASED COHERENT MODELING

TABLE 4. Comparison of the measured rain specific attenuation for terrestrial links and simulation results by S-matrix method with ITU-R P836-3 model.

smaller than the slab extent,  Ens,c is correct. However, at Finally, the corrected fields at the output of the nth super
far ranges, where slab extent is smaller than area within the slab is Encorrected = Eni,corrected + < Ens,corrected > + <
beamwidth,  Ens,c is underestimated and must be corrected. As Ens,c,corrected + 
Ens,corrected + Ens,c,corrected .
illustrated in Fig. 10(d), at far ranges, the scatterers outside In the presented method, since the super slabs are not
the slab extent will contribute to the scattered field within the cascaded, interactions between the super slabs are not con-
slab, but these scatters are not included in S-matrix method. sidered. To include a second order interaction between two
Noting that the fluctuating part of scattered fields due to adjacent super slabs for forward propagation wave, the
scatters are not coherent, the power of scattered fields due method illustrated in Fig. 10(g) is used. The summation
to scatterers outside the slab must be added to the scattered of corrected fields at the output of the nth super slab is
fields due to scatterers within the slab. To correct for  Ens,c , expanded at the z− interface of (n + 1)th super slab and

a factor of A (R)/A is used, that is: the resulting source vector is multiplied by the S-matrix of
the (n + 1)th super slab. The resulting backward propagating

Es,c , A > A (R) waves at z− interface of the (n+1)th super slab represent the

Ens,c,corrected = ns,c   , (17) first order backscattered waves. The first order backscattered
En × A (R)/A, A < A (R)
waves are then used as source at z+ interface of nth super
slab. The resulting forward propagating wave at z+ interface
where A is the area of the slab interface and A (R) is the of the nth super slab is the second order scattered wave and
area within the beamwidth of the source at the range R is denoted by  s,m
En,n+1 . In this case, the total corrected field
(Fig. 10(d)). at the output of the nth super slab (input of (n + 1)th super
To correct for the first three components of Ent , as depicted slab) is Ent,corrected = Encorrected +  s,m
En,n+1 . The total fluctuat-
in Fig. 10(e), (12) can be used to directly calculate fields
ing part,  En , is equal to 
s,t s,corrected
En +  Ens,c,corrected + s,m
En,n+1
(due to the source) at the output of the nth super slab. Here,
and will be used to calculate corrected  s
E(n+1) for the next
the Green’s function for a homogeneous medium with refrac-
super slab.
tive index of neff is used. The obtained field is Eni,corrected +
< Ens,corrected > + < Ens,c,corrected >.
Assuming the total fluctuating component of the scattered
part of the input wave to the nth super slab,  s,t
E(n−1) , is known, APPENDIX B
as illustrated in Fig. 10(f), (12) with the Green’s function SCATTERING FROM A SPHERICAL SCATTERER
of a homogeneous medium with refractive index of neff can In this section, scattered field from a lossy dielectric sphere
be used to find the fields at the output of the nth super slab is calculated for an incident plane wave with arbitrary
due to  s,t
E(n−1) . The result is the corrected 
Ens (= Ens,corrected ). polarization and angle of incidence.

598 VOLUME 4, 2023

 FIGURE 11. Sphere illuminated by a plane wave. Original coordinate system is
denoted by (x, y, z) and rotated coordinate system in denoted by (x  , y  , z  ). The plane
wave used in (18) is shown in blue and the arbitrary plane wave is shown in pink. The
arbitrary plane wave is defined by  k and 
e in the original coordinate system and
k = z and e = x in the rotated coordinate system.

(ε = ε0 ). The radius of the sphere is a and its center is

located at the origin of a coordinate system. Assuming the
sphere is illuminated by an incident plane wave with k =
k0
z (k0 is the propagation constant in free space at angular
frequency ω) and E = 1e−jk0 z x, the scattered electric field
components at a desired coordinate (r > a, θ , φ) are [33]:
⎧  
⎪ !∞ (2)
⎪
⎪ E s = k0 cos(φ) n(n + 1)a hn (k0 r) 1
P (cos(θ ))
⎪ r
⎪ ωr n=1 n k0 r n
⎪
⎪ ⎛ ⎞
⎪
⎪  (2)
(k 
⎪
⎨ s ! ja h n 0 r)
sin(θ)P 1 (cos(θ))
Eθ = cos(φ) ∞ n=1
⎝ n k0 r (2)

n ⎠
hn (k0 r) P1n (cos(θ )) (18)
⎪
⎪ −b
⎪ ⎛ n k0 r sin(θ ) ⎞
⎪
⎪  (2)
hn (k0 r) P1n (cos(θ ))
⎪
⎪ s ! ja
⎪
⎪ E = sin(φ) ∞ ⎝ n k0 r sin(θ ) ⎠,
⎪
⎩ φ n=1 (2) 
−b hn (k0 r) sin(θ)P1 (cos(θ))
n k0 r n

where  h(2)
n is Schelkunoff spherical Hankel function of
second kind and order n, P1n is the associated Legendre
function of degree n and order 1, the primes indicate the
first derivative of the function with respect to its argument,
and an and bn are given by:
⎧
an = (−j)
n (2n+1)
⎪
⎪
⎪
⎪
n(n+1)
√
⎪
⎪ − ε /ε  j (k a) jn (k0 a)
j (k a)+ jn (ks a)
⎨ × √ s (1)0 n 0 n s (1)
hn (k0 a)
εs /ε0 hn (k0 a)
jn (ks a)+  jn (ks a)
(19)
bn = (−j) (2n+1)
n
⎪
⎪
⎪
⎪ √
n(n+1)
⎪
⎪   
jn (k0 a)
⎩ × √− εs /ε(1)0 jn (k0 a)jn (ks a)+(1) 
jn (ks a)
.
hn (k0 a)
εs /ε0 jn (ks a)−
 hn (k0 a)
jn (ks a)

In (19), 
jn is Schelkunoff spherical Bessel function of first
√
kind and order n and ks = ω μ0 εs .
To find the scattered fields due to a general plane wave
FIGURE 10. Correction scheme for S-matrix and Green’s function method.
propagating in  k (= kxx+kyy+kz z) direction with the electric
(a) Components of forward propagating wave at the output of the nth super slab. field E = 1 e and 
ee−jk0 r ( k are unit vectors and e ·
k = 0) at
(b) Extracting the first five components of (a). (c) Extracting the first four components
of (a). (d) Scattering from the particles outside the slab. (e) Corrected Eni , < Ens >, and
a point P defined by (xp , yp , zp ), the following steps are
< Ens,c >. (f) Corrected E s . (g) Calculating E
n
 s,m .
n,n+1
followed:
1) As shown in Fig. 11, the coordinate system must be
Figure 11 shows the geometry of the problem. A dielectric rotated such that in the new coordinate (defined by
and non-magnetic (permeability = μ0 ) sphere with complex orthonormal vectors x , 
y , and 
z ) 
z is aligned with
dielectric constant of ε = εs is considered in free space 
k and  
x is aligned with  e. The point P in the new

VOLUME 4, 2023 599

YEKTAKHAH AND SARABANDI: PHYSICS-BASED COHERENT MODELING

rotated coordinate, (xp , yp , zp ), is defined by: [11] T. S. Rappaport, G. R. MacCartney, M. K. Samimi, and S. Sun,
“Wideband millimeter-wave propagation measurements and channel
⎛ ⎞ ⎛ ⎞⎛ ⎞
xp ê · x̂ ê · ŷ ê · ẑ xp models for future wireless communication system design,” IEEE
Trans. Commun., vol. 63, no. 9, pp. 3029–3056, Sep. 2015.
⎝ yp ⎠ = ⎝ (k̂ × ê) · x̂ (k̂ × ê) · ŷ (k̂ × ê) · ẑ ⎠⎝ yp ⎠. [12] H. Zhao et al., “28 GHz millimeter wave cellular communication
zp k̂ · x̂ k̂ · ŷ k̂ · ẑ zp measurements for reflection and penetration loss in and around build-
   ings in New York city,” in Proc. IEEE Int. Conf. Commun. (ICC),
2013, pp. 5163–5167.
R [13] I. Shayea, T. A. Rahman, M. Hadri Azmi, and A. Arsad, “Rain
(20) attenuation of millimetre wave above 10 GHz for terrestrial links in
tropical regions,” Wiley Trans. Emerg. Telecommun. Technol., vol. 29,
In (20), R is the rotation matrix and all the parameters on no. 8, Aug. 2018, Art. no. e3450.
[14] S. Shrestha and D.-Y. Choi, “Rain attenuation statistics over millimeter
the right-hand side are in the original coordinate system. wave bands in South Korea,” J. Atmos. Solar-Terr. Phys., vol. 152,
2) Using (18), the scattered fields can be calculated at pp. 1–10, Jan. 2017.
(xp , yp , zp ) in the rotated coordinate. The obtained [15] J. S. Marshall and W. M. Palmer, “The distribution of raindrops with
size,” J. Meteorol., vol. 5, no. 4, pp. 165–166, 1948.
scattered field components in the rotated coordinate [16] C. W. Ulbrich, “Natural variations in the analytical form of the
  
system are denoted by (Exs , Eys , Ezs ). raindrop size distribution,” J. Climate Appl. Meteorol., vol. 22, no. 10,
pp. 1764–1775, Oct. 1983.
The scattered field components in the original coordinate [17] L. L. Foldy, “The multiple scattering of waves,” Phys. Rev., vol. 67,
system, (Exs , Eys , Ezs ), are obtained by: no. 3, pp. 107–119, Feb. 1945.
[18] L. Tsang, J. A. Kong, and K.-H. Ding, Scattering of Electromagnetic
⎛ s⎞ ⎛ ⎞ Waves: Theories and Applications, New York, NY, USA: Wiley, 2000.
Ex Exs [19] K. Sarabandi, A. Tavakoli, and F. T. Ulaby, “Propagation in a
⎝Exs ⎠ = R ⎜ ⎟
T
⎝Eys ⎠, (21) two-dimensional periodic random medium with inhomogeneous par-
 ticle distribution,” IEEE Trans. Antennas Propag., vol. 40, no. 10,
Exs Ezs pp. 1175–1186, Oct. 1992.
[20] P. R. Siqueira and K. Sarabandi, “T-matrix determination of effective
T permittivity for three-dimensional dense random media,” IEEE Trans.
where R is the transpose of R. Antennas Propag., vol. 48, no. 2, pp. 317–327, Feb. 2000.
[21] W. Zhensen, Y. Yi, and C. Lihong, “Monte Carlo simulation for
millimeter wave propagation and scattering in rain medium,” Int. J.
REFERENCES Infrared Millimeter Waves, vol. 13, no. 7, pp. 981–994, 1992.
[1] T. S. Rappaport, Y. Xing, G. R. MacCartney, A. F. Molisch, E. Mellios, [22] F. Wang and K. Sarabandi, “A physics-based statistical model for
and J. Zhang, “Overview of millimeter wave communications for wave propagation through foliage,” IEEE Trans. Antennas Propag.,
fifth-generation (5G) wireless networks—With a focus on propa- vol. 55, no. 3, pp. 958–968, Mar. 2007.
gation models,” IEEE Trans. Antennas Propag., vol. 65, no. 12, [23] A. A. Ibrahim and K. Sarabandi, “Simulation of long distance wave
pp. 6213–6230, Dec. 2017. propagation in 2-D sparse random media: A statistical S-matrix
[2] L. Chettri and R. Bera, “A comprehensive survey on Internet of Things approach in spectral domain,” IEEE Trans. Antennas Propag., vol. 62,
(IoT) toward 5G wireless systems,” IEEE Internet Things J., vol. 7, no. 5, pp. 2708–2720, May 2014.
no. 1, pp. 16–32, Jan. 2020. [24] G. R. Simpson, “A generalized n-port cascade connection,” in Proc.
[3] S. Sun, A. P. Petropulu, and H. V. Poor, “MIMO radar for advanced IEEE MTT-S Int. Microw. Symp. Dig., 1981, pp. 507–509.
driver-assistance systems and autonomous driving: Advantages and [25] S. Ermis, “A microwave backscattering model for precipitation,” Ph.D.
challenges,” IEEE Signal Process. Mag., vol. 37, no. 4, pp. 98–117, dissertation, Dept. Elect. Eng., Univ. Texas at Arlington, Arlington,
Jul. 2020. TX, USA, 2015.
[4] K. Sarabandi, A. Jam, M. Vahidpour, and J. East, “A novel [26] F. T. Ulaby and D. G. Long, “Microwave dielectric properties of
frequency beam-steering antenna array for submillimeter-wave appli- natural earth materials,” in Microwave Radar and Radiometric Remote
cations,” IEEE Trans. THz Sci. Technol., vol. 8, no. 6, pp. 654–665, Sensing. Ann Arbor, MI, USA: Univ. Michigan Press, 2017, ch. 4.
Nov. 2018. [27] Attenuation by Hydrometeors, in Particular Precipitation, and Other
[5] A. Alaqeel, A. A. Ibrahim, A. Nashashibi, H. Shaman, and Atmospheric Particles, CCIR document 721-3, ITU, Geneva,
K. Sarabandi, “Experimental characterization of multi-polarization Switzerland, 1990.
radar backscatter response of vehicles at J-band,” IEEE Trans. Intell. [28] T. Manabe, T. Ihara, J. Awaka, and Y. Furuhama, “The relationship
Transp. Syst., vol. 20, no. 9, pp. 3337–3350, Sep. 2019. of raindrop-size distribution to attenuations experience at 50, 80,
[6] F. Wang and K. Sarabandi, “An enhanced millimeter-wave foliage 140, and 240 GHz,” IEEE Trans. Antennas Propag., vol. 35, no. 11,
propagation model,” IEEE Trans. Antennas Propag., vol. 53, no. 7, pp. 1326–1330, Nov. 1987.
pp. 2138–2145, Jul. 2005. [29] J. R. Norbury and W. J. K. White, “Microwave attenuation at 35 8 GHz
[7] S. Joshi and S. Sancheti, “Foliage loss measurements of tropical trees due to rainfall,” Electron. Lett., vol. 8, no. 4, pp. 91–92, Feb. 1972.
at 35 GHz,” in Proc. Int. Conf. Recent Adv. Microw. Theory Appl., [30] J. Sander, “Rain attenuation of millimeter waves at λ = 5.77, 3.3, and
2008, pp. 531–532. 2 mm,” IEEE Trans. Antennas Propag., vol. 23, no. 2, pp. 213–220,
[8] T. E. Bogale, X. Wang, and L. B. Le, “mmWave communication Mar. 1975.
enabling techniques for 5G wireless systems: A link level perspective,” [31] Y. Furuhama and T. Ihara, “Remote sensing of path-averaged raindrop
in mmWave Massive MIMO, Cambridge, MA, USA: Academic Press, size distributions from microwave scattering measurements,” IEEE
2017, ch. 6. Trans. Antennas Propag., vol. 29, no. 2, pp. 275–281, Mar. 1981.
[9] M. Rangwala, F. Wang, and K. Sarabandi, “Study of millimeter-wave [32] A. Hirata et al., “Effect of rain attenuation for a 10-Gb/s 120-GHz-
radar for helicopter assisted-landing system,” IEEE Antennas Propag. band millimeter-wave wireless link,” IEEE Trans. Microw. Theory
Mag., vol. 50, no. 2, pp. 13–25, Apr. 2008. Techn., vol. 57, no. 12, pp. 3099–3105, Dec. 2009.
[10] M. R. Akdeniz et al., “Millimeter wave channel modeling and cellular [33] K. Sarabandi, “Spherical wave functions and their applications,”
capacity evaluation,” IEEE J. Sel. Areas Commun., vol. 32, no. 6, in Foundations of Applied Electromagnetics, Ann Arbor, MI, USA:
pp. 1164–1179, Jun. 2014. Michigan Publ., 2022.

600 VOLUME 4, 2023

 BEHZAD YEKTAKHAH (Member, IEEE) received KAMAL SARABANDI (Life Fellow, IEEE) is
the B.S. and M.S. degrees in electrical engineer- currently the Fawwaz T. Ulaby Distinguished
ing from the University of Tehran, Tehran, Iran, in University Professor and the Rufus S. Teesdale
2008 and 2011, respectively, and the Ph.D. degree Endowed Professor of Engineering with the
in electrical engineering from the University of University of Michigan at Ann Arbor, Ann Arbor,
Michigan at Ann Arbor in 2019. MI, USA. He led the Center for Microelectronics
He is currently a Postdoctoral Research Fellow and Sensors funded by the Army Research
with the University of Michigan at Ann Arbor. Laboratory from 2008 to 2018 and has been
His research focuses on radar imaging, electro- leading the Center of Excellence in Microwave
magnetic scattering, antennas, and phased arrays. Sensor Technology. He has supervised 63 Ph.D.
and numerous master’s students and postdoctoral
fellows. He has published many book chapters, more than 325 articles
in refereed journals, and more than 770 conference papers. His research
interests include microwave and millimeter-wave radar remote sensing,
metamaterials, electromagnetic wave propagation, antenna miniaturization,
and bioelectromagnetics. He, together with his students, was a recipient of
35 paper awards. His contributions to the field of electromagnetics have been
recognized by many awards, including the Humboldt Research Award, the
IEEE GRSS Distinguished Achievement Award, the IEEE Judith A. Resnik
Medal, the IEEE GRSS Education Award, the NASA Group Achievement
Award, and many other awards from the University of Michigan at Ann
Arbor. He was the Chair of Commission F of USNC/URSI and has been
serving as a member of the AdCom for the IEEE Antennas and Propagation
Society. He is a member of the National Academy of Engineering. He has
served as a member of the NASA Advisory Council for two consecutive
terms from 2006 to 2010 and the President of the IEEE Geoscience and
Remote Sensing Society from 2015 to 2016. He is a Fellow of the American
Association for the Advancement of Science and the National Academy of
Inventors.

VOLUME 4, 2023 601