Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 70, NO.

10, OCTOBER 2022 9795

A Hybrid Method for Fast and Efficient Evaluation


of Electromagnetic Performance of the
Radome-Enclosed Antennas
Haiwen Ke , Pengyuan Wang, Jinbo Liu , Ji Li, and Mang He , Senior Member, IEEE

Abstract— In this article, a fast and efficient hybrid method strong specular lobe, and input impedance mismatch [1], [2].
that combines the volume-surface integral equation (VSIE) and In some scenarios, natural or man-made deformation/ablation
the modified surface integration (MSI) method is proposed to of the radome also changes the radiation pattern of the anten-
evaluate the electromagnetic (EM) performance of the large-
/medium-sized radome-enclosed antennas (REAs). An approxi- nas significantly, which in consequence severely degrades the
mate equivalence principle is presented to formulate the physical overall performance of the radar/communication system [3].
model for the EM radiation of the REA, and then, the full-wave In order to minimize or counteract the adverse effects of
VSIE for the antennas and the high-frequency MSI method for radome to recover the EM performance of the enclosed
the dielectric radome are coupled with each other in an iterative antennas, it is always desirable to know the accurate alteration
manner. Since mutual interactions up to the second order among
the antennas and the radome are considered, the method shows of the antennas’ radiation characteristics due to the presence
good accuracy in both the near- and far-field regions of the REA of radome, e.g., the boresight error and its slope with respect
system. In addition, all the computation processes involved in to scanning angle of the antenna arrays, the change of the
the method, including the method of moments (MoM) solution input impedance of the antenna element at the feeding port,
of the VSIE, surface integrations (SIs) in the MSI, and ray tracing and so on [3]. Although all the required quantities, i.e., the
for determining the local transmission matrix for the dielectric
radome, are accelerated by the multilevel fast multipole algorithm active radiation pattern and S-parameters of each radome-
(MLFMA), which makes the method highly efficient in terms of enclosed antenna (REA) element [4], can be obtained through
both computational time and memory requirement. Numerical repetitive measurements, it is of high cost and is very time-
results show that the proposed hybrid method can predict the EM consuming, particularly for large-scale arrays. Therefore, if the
performance, e.g., the input impedance and radiation pattern, experimental measurements can be replaced by fast numerical
of the REA system with good accuracy compared with the
full-wave simulations. simulations with high accuracy, it will be very helpful to
expedite the radiation characteristics correction process for the
Index Terms— Hybrid method, modified surface integration REAs systems [3], [5].
(MSI), multilevel fast multipole algorithm (MLFMA), radome-
enclosed antenna (REA) array, ray tracing, volume-surface Many methods have been proposed to solve the EM radi-
integral equation (VSIE). ation problem of the REA system, including the full-wave
numerical methods [6]–[10], the high-frequency approxima-
I. I NTRODUCTION tion methods [1], [11]–[13], and the hybrid methods [14]–[17].
The full-wave methods, such as the finite element method
R ADOMES are widely used for protecting communication
and radar antennas from rain, snow, wind, and other
negative environmental effects. However, the existence of the
(FEM) [6], the method of moments (MoM) solution of various
types of integral equations [7]–[9], and the time-domain finite
radome inevitably deteriorates the electromagnetic (EM) per- integration method [10], can yield accurate results but at
formance of the enclosed antennas in both near- and far-field the price of high computational cost even when fast algo-
regions, such as main lobe deflection, raise of sidelobe level, rithms are utilized [11], so they are usually restricted to
the REA system of small/medium sizes. In contrast, high-
Manuscript received 29 January 2022; revised 17 April 2022; accepted frequency methods, such as the aperture integration-surface
30 May 2022. Date of publication 24 June 2022; date of current version integration (AI-SI) method [12] and the physical optics (PO)
9 November 2022. This work was supported by the National Natural Sci-
ence Foundation of China under Grant 62171026. (Corresponding author: method [13], [14], can efficiently evaluate the EM radiation of
Mang He.) REA systems with smooth shapes and large sizes. However,
Haiwen Ke, Pengyuan Wang, and Mang He are with the School of Integrated mutual interactions between the radome and antennas are
Circuits and Electronics, Beijing Institute of Technology, Beijing 100081,
China (e-mail: 3120195374@bit.edu.cn; 3120205355@bit.edu.cn; hemang@ totally ignored in these approximate methods, so they cannot
bit.edu.cn). provide results with adequate accuracy, particularly in the near-
Jinbo Liu is with the School of Information and Communication Engi- field region. The hybrid approaches combine the advantages
neering, Communication University of China, Beijing 100024, China (e-mail:
liuj@cuc.edu.cn). of high-frequency and full-wave methods and use the concept
Ji Li is with the Beijing System Design Institute of Mechanical-Electrical of domain decomposition to select the appropriate solution
Engineering, Beijing 100854, China (e-mail: 2120130690@bit.edu.cn). in different regions to ensure acceptable accuracy and high
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TAP.2022.3184532. efficiency. In [15]–[17], the hybrid PO-MoM method is used
Digital Object Identifier 10.1109/TAP.2022.3184532 to analyze the EM radiation of the REA, but mutual coupling
0018-926X © 2022 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on April 12,2024 at 09:15:31 UTC from IEEE Xplore. Restrictions apply.
9796 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 70, NO. 10, OCTOBER 2022

between the radome and antennas is neglected or only partially


considered. In [18], a hybrid scheme combining the AI-SI
and PO is presented to compute the radiation of ground-based
radome-antenna system, in which the induced currents on the
metal frame are computed by PO and the transmitted fields
on the outer surface of the radome are obtained by the AI-SI.
The far-field radiation of the REA system is calculated as
superposition of the fields from these two equivalent currents.
Apparently, this method also decouples the analyses of antenna
and radome, which limits its accuracy. In order to improve the
computational accuracy of induced voltage at the feeding ports
of the REA elements when the array is working in receiving
mode, the mutual interactions among the radome and array are
considered iteratively in the hybrid method proposed in [19].
This method is highly efficient and possesses good accuracy;
however, the underlying physical model is only valid for the
REA system in receiving mode and cannot be applied to Fig. 1. Schematic of the REA in the transmitting mode.
the analysis of EM radiation when the antenna array is in
transmitting status. radome, respectively. For a radome with an open base, we use
In this article, we use an approximate equivalent principle a virtual surface Sv (shown in dashed line in Fig. 1), together
to formulate the physical model for analysis of the EM radi- with Se and Sb , to close the bottom of the radome (clearly,
ation from the REA in transmitting mode. By introducing an if the radome itself is a closed structure, both Sv and Sb will
auxiliary vacuum radome with the same shape as the realistic vanish). According to the surface equivalence theorem [11],
radome, the radiating fields from the REA are represented the radiating fields E and H outside the closure can be given
by the superposition of the EM fields from the antennas by superposition of the partial fields due to the equivalent EM
and the difference field due to the induced EM currents on currents on the surfaces Se , Sb , and Sv
the surfaces of the realistic and auxiliary radomes. Then, 
the full-wave volume-surface integral equation (VSIE) model E = Ee + E b + Ev
(1)
for the antennas and the modified surface integration (MSI) H = He + Hb + Hv
model for the realistic and auxiliary radomes are hybridized
in an iterative manner, and the mutual interaction between the where El and Hl (l = e, b, v) are the partial EM fields
radome and antennas is considered up to the second order. produced by equivalent EM currents on Se , Sb , and Sv ,
During the numerical implementation of the method, both respectively.
the solution of VSIE and the computation of surface/volume If we replace all the dielectric layers of the radome by
integrations in the MSI are accelerated by the multilevel fast vacuum while keeping their geometrical structures (including
multipole algorithm (MLFMA), and the ray-tracing process the shape and thickness of each layer and the interface between
for calculating the local transmission matrix of the multi- two consecutive layers) unchanged, then the radiating EM
layer dielectric radome is also enhanced through using the fields outside the closure are

E = E e + Ebair + Evair
air
MLFMA-tree structure. Numerical results show that the pro-
posed method not only reduces computational cost massively (2)
H = H e + Hbair + Hvair
air
compared with existing full-wave solvers but also improves the
computational accuracy compared with the traditional high- where Elair and Hlair (l = e, b, v) are the partial EM fields due
frequency methods. to equivalent EM currents on Sb , Se , and Sv of the auxiliary
vacuum radome, respectively. Apparently, for the vacuum
II. M ETHODOLOGY radome, the radiating EM fields in the outside space are just
A. Approximate Equivalent Model for Analysis of Radiation the same as those radiated by the free-standing antenna array,
of REA in Transmitting Mode i.e.,

The general configuration of the REA working in transmit- E = E Ant
(3)
ting mode is shown in Fig. 1. The radome has N dielectric H = H Ant .
layers indexed by v 1 , v 2 , . . . , v N . Both the regions interior and
exterior to the radome are free space, which are denoted by Since the partial fields ( E v , Hv ) and ( Evair , Hvair ) are induced
v 0 and v N +1 , respectively. The relative permittivity, relative by EM currents on the virtual surface Sv that is not a real
permeability, and loss tangent of the lth layer are εrl , μrl , part of the REA, as a first-order approximation, we assume
and tanδl (l = 0, 1, 2, . . . , N + 1), respectively. The volume that once the geometrical structure of the radome is preserved,
N
occupied by the radome can be expressed as V = i=1 Vi with these fields remain unchanged even if the materials of radome
its boundary given by S = Se + Sb + Si , where Se , Sb , and Si are altered, that is to say, ( Ev , Hv ) and ( Evair , Hvair ) are
represent the outermost, bottom, and innermost surfaces of the assumed irrelevant to the dielectric types of the radome, and

Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on April 12,2024 at 09:15:31 UTC from IEEE Xplore. Restrictions apply.
KE et al.: HYBRID METHOD FOR FAST AND EFFICIENT EVALUATION OF EM PERFORMANCE OF THE REAs 9797

in which high computational accuracy is needed. Thus, the


full-wave VSIE solver [7] is used to evaluate the EM current
distributions on the antennas. For the radome with large
or even medium size, if we still use the VSIE solver, the
computational overhead will be prohibitive [15]–[17]. There-
fore, in the radome region, the high-frequency method with
adequate accuracy is desired to achieve high efficiency. In this
article, we embed the rigorous ray-tracing process into the
MSI method [19] to improve the accuracy of the EM wave
refraction, transmission, and multiple reflection in the radome
region.
To be more specific, in the antenna region, the PEC surfaces
Sc can be generally modeled by the combined field integral
equation (CFIE) as follows:
   
−χ L SE JS + L VE JV
Fig. 2. Schematic of the interaction between antenna and radome.
tan    
+(1 − χ)η0 JS (r ) − n̂ × L SH JS + L SH JV
they are approximately identical when different radomes are = χ E exc (
r) + (1 − χ)η0 n̂ × H exc (r ), r ∈ Sc (6)
used to enclose the antenna array. Therefore, in (1) and (2), tan

we have and the inhomogeneous dielectrics Vd can be modeled by the


 volume integral equation (VIE)
E v ≈ Evair    
(4)
Hv ≈ Hvair . E V (r ) − L SE JS − L VE JV = E exc (r ) , r ∈ Vd . (7)
Replacing (2)–(4) into (1), we obtain an approximate repre- Equations (6) and (7) are combined to build a coupled VSIE
sentation of the radiating EM fields from the realistic REA formula for composite conductor-dielectric antenna structures.
system as follows: The integral–differential operators L αP (α is E or H and P is
⎧     S or V ) are defined as
⎨ E ≈ E Ant + E e + Eb − E air + E air
  e b
 ⎧ E 
(5) ⎪
⎪ L J , M
⎩ H ≈ H Ant + He + Hb − H air + H air . ⎪ p
 
e b ⎪
⎪ 

⎪ 1  · ∇  ∇  + k 2 J − j ωε0 M
 × ∇

⎪ = J
Thus, for a given antenna array, if the partial fields due to ⎪
⎪ j ωε0


p
 
 × G(r , r )d p
the EM currents on Se and Sb of both the realistic and the

auxiliary vacuum radomes are known, the radiating EM fields ⎪

⎪ L H J, M 
in the outer space of the REA system can be conveniently ⎪ p
⎪   

⎪ 1
calculated by (5). ⎪
⎪ = M  · ∇ ∇ + k2 M  + j ωμ0 J × ∇ 

⎪ j ωμ

⎩ 0 p
× G(r , r )d p
B. Physical Process of Mutual Interactions in the REA
System and Coupled VSIE and MSI Solver (8)

Fig. 2 shows the physical process of mutual interactions where G(r , r ) = e− j k|r −r | /4π|r − r | is the Green function in
between the antennas and the radome in a typical REA free space and J and M  are equivalent electric and magnetic
system. First, the antennas are initially excited by the applied currents, respectively. The definitions of the parameters Sc ,
sources ( E δ and H δ ) to produce the radiating EM fields Vd , χ, n̂, and η0 , and the variables JS and JV are the same
( E1iAnt and H1iAnt ), and then, these fields impinge on the inner as those in [19]. The VSIE can be numerically solved by the
surface of the radome. Part of the incident fields is reflected MoM method, and the final matrix equation is written as [19]
back into the space enclosed by the radome, part of them      
Z ss Z sv I Vs
dissipates in the radome, and the remaining part transmits [Z ] · [I ] = · s = [V ] = . (9)
Z vs Z vv Iv Vv
through the radome and radiates into the outer space. The
first-order reflected fields ( E 1rAnt and H1rAnt ) not only affect the Considering the mutual interaction of radome, EM excita-
antennas in turn but also produce the second-order reflection tion E exc and H exc at the feeding port of the enclosed antenna
( E2rAnt and H2rAnt ) at the inner surface Si of the radome. Thus, can be seen as the superposition by two parts
the EM current distributions on the antennas and the surfaces ⎧  
⎨ E exc = E δ + L E Jre f , M
 re f
of the radome (Se , Si , and Sb ) will dynamically change until s
  (10)
final stable status is achieved. ⎩ H exc = H δ + L H Jre f , M re f
s
Using the idea of domain decomposition, the entire
REA structure is partitioned into two computational regions, where Jre f and M
 re f are the equivalent electric and magnetic
as shown in Fig. 2. The first one is the antenna region, currents on Si and Sb induced by reflected fields, respectively,

Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on April 12,2024 at 09:15:31 UTC from IEEE Xplore. Restrictions apply.
9798 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 70, NO. 10, OCTOBER 2022

which are obtained by Jre f = n̂ × H re f and M  re f = E re f × n̂. As stated in the last subsection, the high-frequency approx-
 re f  re f
( E , H ) are the total reflected fields of the incident fields imate method is utilized to improve the computational effi-
from the antenna, which can be expressed as the sum of the ciency, and the reflected and transmitted fields by the radome
first- and the second-order reflections ( E1rAnt , H1rAnt ) and ( E2rAnt , are calculated through the transmission matrix (ABCD matrix)
H2rAnt ). It is noticed that (10) depicts a dynamic process as method [1]. For a dielectric radome with N layers, both of
stated previously, i.e., the EM currents on antenna and radome them can be computed by using the cascade ABCD matrices.
interact with each other continuously until the final stability Specifically, the reflected fields are expressed as
is reached.    
ElrAnt = Eli R  + n̂ inc · n̂re f EliAnt⊥ − n̂ re f · EliAnt⊥ n̂ inc
Ant
Combining (9) and (10), an iterative VSIE-MSI formula is
built as follows:
 × R⊥ (18)
⎧   re f  (k−1)    Ant  
⎨ E exc(k) = E δ + L SE Jre f I (k−1) , M  I HlrAnt  Ant⊥ ⊥
= Hli   Ant
R + n̂ inc · n̂re f Hli − n̂re f · Hli n̂ inc
  (11)
⎩ H exc(k) = H δ + L H Jre f  I (k−1) , M 
 re f I (k−1) × R (19)
S
 
[Z ] · [I ](k) = V E exc(k) , H exc(k) , k = 1, 2, . . . , Niter . where n̂ inc and n̂re f are the unit wave vectors of the incident
and reflected wave at Si and Sb , respectively. The total
(12) transmitted fields ( Et and H t ) at Se are superposition of the
The iteration will stop until the relative error is adequately transmitted fields ( E lt and H lt )(l = 1, 2) of the first and
small second incident fields at Si , which are given as
 (N )  ⎧
 I it er − I (Nit er −1)  ⎪ 2  2  
  < ε I ter . (13) ⎪
⎪  =  = liAnt⊥ T ⊥ + tˆ tˆ · EliAnt T 
 I (Nit er )  ⎪
⎨ E t E lt E
l=1 l=1
  (20)
In (11)–(13), the superscript k means that the kth ⎪
⎪ 2  2
⎪ Ht =
⎪ Hlt = Hli T + tˆ tˆ · Hli
Ant  Ant⊥
T ⊥
iteration,ε I ter , is the target relative residual error, Niter is the ⎩
number of iterations, and · denotes the Euclid norm. As seen l=1 l=1

in (11) and (12), when we use the MLFMA to accelerate the where tˆ is the unit vector tangential to Se . The transmission
solution of VSIE in the antenna region, these two equations and reflection coefficients T ⊥,|| and R ⊥,|| for the N-layer
are actually a nested outer–inner iterative algorithm, i.e., the radome are defined as
outer iteration is used to update the excitation (11) for the ⎧

⎪ 2
antenna, and the inner one is the iterative solution of (12) by ⎨T ⊥, = ⊥,
A + B ⊥, + C ⊥, + D ⊥,
the MLFMA. (21)

⎪ A⊥, + B ⊥, − C ⊥, − D ⊥,
⎩ R ⊥, = ⊥,
A + B ⊥, + C ⊥, + D ⊥,
C. Modified ABCD Method for the Multilayer Radome
where A⊥, , B ⊥, , C ⊥, , and D ⊥, are entries of the total
It is seen that in the iterative solution of (11) and (12),
ABCD matrix for the N-layer radome, which is the multi-
we need to compute the incident fields ( EliAnt and HliAnt ) and
plication of the local ABCD matrices for each dielectric layer
reflected ones ( ElrAnt and HlrAnt ) (l = 1, 2) at Si and Sb  ⊥, 
 ⊥,  
repeatedly and, then, the final transmitted fields ( E t and Ht ) at A B ⊥,
N
Al
⊥,
Bl
=
Se to obtain the radiating EM fields outside the REA system. C ⊥, D ⊥, Cl
⊥,
Dl
⊥,
As shown in Fig. 2, the first incident fields at Si and Sb are l=1
⎡ ⊥,

radiated by the antenna N cosξl j η̄l si nξl
⎧     ⎢ j si nξ ⎥
= ⎣ l ⎦ (22)
⎨ E1iAnt = L SE JS + L VE JV ⊥,
cosξl
    (14)
l=1 η̄l
⎩ H Ant = L H J + L H J
1i S S V V with

and the second incident fields at Si and Sb are the reradiation of ξl = k0 dl εrl μrl cosβl (23)
the equivalent EM currents induced by the first-order reflected η̄l cosθl
fields, which can be calculated by the SIs as η̄l⊥ = (24)
  cosβl

⎨ E 2iAnt = L sE J1rAnt , M
 1r
Ant
 η̄l cosβl
η̄l = (25)
  (15) cosθl
⎩ H Ant = L H J Ant , M  1r
Ant

2i s 1r where k0 is the wavenumber in free space, η̄l = μrl /εrl is
J1rAnt = n̂ × H1rAnt ,  1rAnt = E Ant × n̂.
M 1r (16) the normalized characteristic impedance of the lth layer, and
θl and βl are the incident and refraction angles at the interface
Decompose the incident fields into two components of vertical of the lth and (l + 1)th layer, respectively.
and parallel polarization as For radomes of variable wall thickness, the computational
 Ant
Eli = EliAnt⊥ + Eli accuracy of (22)–(25) can be further improved by embedding
Ant
(17) the ray-tracing process. The implementation detail is illustrated
HliAnt = HliAnt⊥ + HliAnt. in the Appendix.

Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on April 12,2024 at 09:15:31 UTC from IEEE Xplore. Restrictions apply.
KE et al.: HYBRID METHOD FOR FAST AND EFFICIENT EVALUATION OF EM PERFORMANCE OF THE REAs 9799

D. MLFMA Acceleration of the Involved Computation points and integration points can be given as

Not only the ray-tracing process benefits from the idea of M = p · N S A + q · NV A
grouping provided in the MLFMA but also both the matrix- (26)
N = l · NS R
vector multiplication (MVM) {[Z ] · [I ] in (9)} in the VSIE
solution of the antenna and the computation of SIs and volume where N S A is the number of triangles in the antenna region,
integrations (VIs) in (15) and (16) for evaluating the mutual NV A is the number of tetrahedrons in the antenna region, and
interactions between the antenna and radome regions can be N S R is the number of triangles on Si and Sb in the radome
accelerated by the MLFMA. By doing so, the heavy overhead region.
of direct computations will be reduced significantly since these If we let Nt = M + N and M = α Nt , then α is a constant
computations need to be implemented repeatedly in each inner for a given REA system and meshing scheme. The time cost
or outer iteration. by direct computation of SIs and VIs as was done in the
1) MLFMA Acceleration of the VSIE Solution: A relatively traditional AI-SI method is estimated as
small oct-tree is designed to cover the antenna region only
and the MLFMA is applied in the inner iteration to accelerate T1 ≈ c1 α(1 − α)Nt2 (27)
the VSIE solution. The acceleration of computation of self- where c1 is a real constant.
interactions in the antenna region is implemented in the In contrast, since SIs’ and VIs’ computation is accelerated
same way as in the conventional MLFMA. Apparently, the by the MLFMA, the computational overhead is massively
computational complexity of this process is on the order of reduced. In the MLFMA implementation for (11), the aggre-
O(Na log Na ) by virtue of the MLFMA, where Na is the gation procedure is first carried out only in the radome boxes
number of unknowns in the antenna region. that contain the surface triangular elements on Si and Sb , then,
2) MLFMA Acceleration of the Computation of Mutual translation is operated from these radome boxes to the antenna
Interactions Between the Antenna and Radome Regions: In boxes that contain the enclosed antenna structures, and finally,
the computation of mutual coupling between the antenna and disaggregation is implemented only in the antenna boxes to
radome regions, a larger oct-tree is set up to cover the entire complete the fast computation. In addition, the MLFMA can
structure and the MLFMA is used to compute the SIs and VIs be used in a similar way to calculate the secondary incident
involved in the outer iteration. Both (11) and (14) and (15) are fields at Si and Sb due to the equivalent EM currents induced
computed in the framework of the MLFMA. Different from by the first reflected fields as in (15), but only the radome
the traditional MLFMA, it is not necessary to implement the boxes are involved in the MLFLA implementation. Therefore,
aggregation, translation, and disaggregation in each nonempty according to the computational complexity of each step in the
box although the second oct-tree covers the entire model. MLFMA [11], the total time cost by SIs’ and VIs’ computation
In fact, to accelerate the computation of (11), the aggregation in (11) and (15) is reduced to
procedure is first carried out only in the nonempty boxes in the
radome region, then, translation is operated from these radome T2 ≈ c21 NlogN + c22 NlogN + c23 MlogM
  
 
boxes to the antenna boxes that contain the RWG/SWG basis ≈ c21 + c22 (1 − α)Nt logNt + c23 α Nt logNt
functions in the antenna region, and finally, disaggregation   

= c21 (1 − α) + c22 α Nt logNt (28)
is implemented only in the antenna boxes to complete the
    
fast computation. Similarly, in the computation of (14), the where c21 , c22 , c23 , c21 , c22 , c23 , c21 , and c22 are all constants
aggregation is first carried out only in the antenna boxes, depending on the implementation of the MLFMA. In (28),
then, translation is operated from these boxes to the radome the first term on the right-hand side represents the time cost
boxes, and finally, disaggregation is done only in the radome for calculating the secondary incident fields in the radome
boxes to complete the procedure. In (15), the MLFMA can boxes, the second term is the aggregation time consumed
be implemented in a similar way but only in the radome in radome boxes, and the third term denotes the time used
boxes. in translation and disaggregation procedures in the antenna
boxes.
E. Overall Computational Complexity Analysis of the Method Similarly, in the fast computation of (14), the aggregation
procedure is first carried out only in the antenna boxes; then,
It is clear that the computational complexity of direct SIs translation is operated from the antenna boxes to radome
and VIs is on the order of O(MN), if M and N are the numbers boxes; and finally, disaggregation is implemented in the
of the observation points and integration sampling points. radome boxes to complete the fast computation of the radiated
In the numerical implementations of the proposed method, fields from antennas. Therefore, its total time cost is
Si and Sb are discretized by triangular grids for the MSI
computations, while for the solution of VSIE, the metallic T3 ≈ c31 MlogM + c32 NlogN
 
surfaces and dielectric volumes in the antenna region are ≈ c31 α Nt logNt + c32 (1 − α)Nt logNt
meshed into triangular and tetrahedral elements, respectively.   

= c31 α + c32 (1 − α) Nt logNt (29)
If we use the l-point integration scheme to compute the SIs and
 
VIs in (11), (14), and (15) on Si and Sb , and p- and q-point where c31 , c32 , c31
and , c32
are all constants. In (29), the
integration schemes for the RWG and SWG basis functions in first term on the right-hand side represents the aggregation
the VSIE model, respectively, then the numbers of observation time consumed in antenna boxes, and the second term is the

Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on April 12,2024 at 09:15:31 UTC from IEEE Xplore. Restrictions apply.
9800 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 70, NO. 10, OCTOBER 2022

time cost of translation and disaggregation procedures in the


radome boxes.
As shown in (11) and (12), the iterative VSIE-MSI method
calculates the excitation fields N I ter times. The total CPU time
cost of SIs and VIs computation in the iterative MSI is
TI ter ≈ N I ter T2 + (N I ter − 1)T3
= [c2 N I ter + c3 (N I ter − 1)]Nt log Nt . (30)
In (30), the first term on the right-hand side represents the
time cost of SIs computation on radome surfaces by (11)
and (15), and the second term denotes the time consumed
in the computation of the antennas’ radiation by (14).
Because the time cost of the MLFMA solution of VSIE in
the antenna region is on the order of O(Na logNa ), the total
computational time of the proposed method is proportional to
O(Nt logNt ). Compared with the conventional AI-SI method,
the time cost is significantly reduced with respect to T1
in (27). It should be pointed out that the time cost strongly
depends on the parameter α and the number of outer iterations.
Fortunately, the outer iteration has very fast convergence, and
usually, N I ter is less than 3 or 4.
Fig. 3. Configuration of the radome-enclosed 2 × 2 λ0/2 dipole array. (a) Top
Concerning memory requirement of the method, it is clear view of the dipole array. (b) Side view of the radome. (c) 3-D view of the
that the unknowns in the radome region need very few memory REA system.
due to the use of high-frequency method. Thus, the memory
cost of this method is mainly from that consumed by the
elements are excited by applied voltage source with the same
MLFMA solution of VSIE in the antenna region, which is on
amplitude and phase.
the order of O(Na logNa ). It is evidently much less than that of
The radiation pattern of the REA system is computed by
the full-wave solutions of which the total memory requirement
three different numerical schemes for the purpose of com-
is at least O(Nt logNt ) even if these solutions utilized the fast
parison, and the results are shown in Fig. 4. The first and
algorithm.
second schemes use the proposed hybrid method, but the
difference of them is the former scheme that does not include
III. N UMERICAL R ESULTS the mutual interaction among the array and the radome, that
In this section, numerical examples will be presented to is to say, the first scheme does not include EM reflection
show the accuracy and efficiency of the proposed method, from the radome and its effect on the array in the hybrid
and all the simulations are carried out on a workstation with model, whereas the second scheme considers the mutual
2.4 GHz CPU and 384 GB core memory. Unless otherwise interactions among the radome and the array as depicted
specified, the operating frequency is 10 GHz, and the aver- in (11)–(13). Therefore, it is expected that the first scheme
age mesh size of PEC and dielectric for discretizing the is reduced to the conventional hybrid methods such as those
geometrical structures in the antenna region is about 0.1λ0 presented in [15]–[17] and has similar computational accuracy
and 0.1λd (λ0 and λd are the wavelengths in free space and as the existing hybrid methods. The third scheme uses our
dielectrics, respectively). The GMRES solver [20] with SAI in-house full-wave VSIE solver, which is also accelerated by
preconditioners [21] is used to solve the matrix equations (12) the MLFMA, to provide the benchmark results for check-
in the MLFMA, and the convergence criteria of both the inner ing the accuracy of the proposed hybrid method. It is noted
iteration (12) and outer iteration (11) are set to 0.001. that the VSIE is reduced to the surface integral equation (SIE)
in this example since no dielectric material exists in the
antenna region. In Fig. 4, the three types of results are indi-
A. Dipole Array Enclosed by a Single-Layer Radome cated as proposed/no-reflection, proposed/reflection, and full
Fig. 3 shows the configuration of a 2 × 2 λ0 /2 dipole wave.
array enclosed by a single-layer paraboloid radome. The It is seen from Fig. 4 that the results of the second scheme
diameter and height of the radome are 334.8 and 324.5 mm, agree well with those of the full-wave VSIE solver in the upper
respectively, and the wall thickness varies from 14.5 mm at half-space and are only slightly less accurate in the back-lobe
the top to 7.5 mm at the root of the radome. The relative region. In contrast, the first scheme renders much less accurate
permittivity of the radome is 3.0, and the loss tangent is 0.008. results in both the upper and lower half-spaces. Therefore, the
The array is x-polarized and the interelement distances are results indicate the importance of including multiple reflection
20 and 15 mm along the x- and y-directions, respectively. among the REA system in the hybrid model, especially for the
The array locates 0.25λ0 (7.5 mm) above the round PEC radome in which the mutual interaction between the array and
ground plane that has the diameter of 300 mm, and all antenna radome is quite strong.

Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on April 12,2024 at 09:15:31 UTC from IEEE Xplore. Restrictions apply.
KE et al.: HYBRID METHOD FOR FAST AND EFFICIENT EVALUATION OF EM PERFORMANCE OF THE REAs 9801

Fig. 5. Radome-enclosed two-element microstrip antenna array. (a) Top view


of the MSA element. (b) Side view of the radome. (c) 3-D view of the REA
system.

TABLE I
C ALCULATED I NPUT I MPEDANCE OF THE
A NTENNA E LEMENTS IN E XAMPLE 2

Fig. 4. Radiation pattern of the REA shown in Fig. 3 in (a) xoz plane and
(b) yoz plane.

B. Two-Element Microstrip Antenna Array Covered by a


Single-Layer Radome With Closed Base solver for comparison. All the simulation details are given in
The configuration of a two-element microstrip antenna array Table III.
(MSA) enclosed by a medium-sized variable-thickness radome It is seen in Fig. 6 that the results of VSIE-MSI (2, 3,
with closed base is shown in Fig. 5. Fig. 5(a) shows the and 4) are in good agreement with the ones of the full-
geometry of the antenna element, and Fig. 5(b) shows the wave VSIE solver in all radiation directions, but the VSIE-
shape and overall size of the radome. The relative permittivity MSI(1), i.e., the proposed method without outer iteration, has
of the dielectric substrate for the MSA is 2.2, and its loss the lowest accuracy in the forward direction (θ is from −30◦
tangent is 0.01. The overall dimensions of the radome are to 30◦ ) and backward direction (θ is from 100◦ to 260◦ ). With
122.4 mm (4.08λ0 )×122.4 mm (4.08λ0 )×117.8 mm (3.93λ0 ), regard to the results of the near-field quantities, the relative
and the relative permittivity and loss tangent are 3.8 and 0.008, error of input impedance calculated by VSIE-MSI(1) is about
respectively. The maximum wall thickness is 2.16 mm at the 9.62% compared with the accurate method (full-wave VSIE
top of the radome, and the minimum thickness is 1.2 mm at solver). After only one or very few outer iterations (N I ter =
its bottom. 2, 3, and 4), the iterative VSIE-MSI(n) scheme apparently
First, the two MSA elements are fed by voltage source with improves the computational accuracy, and the relative error
the same amplitude and phase, and the proposed method is is reduced to less than 1% in VSIE-MSI(4). It is also found
utilized to calculate the radiation patterns and input impedance that since each outer iteration starts using the result of former
of the MSA. Table I lists the computed input impedance of the iteration as the initial value, the corresponding number of inner
two antenna elements, and Fig. 6 shows the radiation patterns iterations decreases quickly from 33 to 24, then to 14, and
in the xoz and yoz planes. In Fig. 6, the results of the proposed finally to 10 to reach the target relative error. Moreover, the
method with a different number of outer iterations, indicated difference between the results of VSIE-MSI(n) is small if n
by VSIE-MSI(n) (n = 1, 2, 3, and 4), are shown together is greater than 2, which indicates the fast convergence of the
with the result of the MLFMA-accelerated full-wave VSIE outer iteration process.

Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on April 12,2024 at 09:15:31 UTC from IEEE Xplore. Restrictions apply.
9802 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 70, NO. 10, OCTOBER 2022

Fig. 7. Transmissivity of the radome with various scanning angles.

Fig. 6. Radiation pattern of the REA shown in Fig. 5 in (a) xoz plane and
(b) yoz plane.

TABLE II
PARAMETERS OF THE A-T YPE S ANDWICHED R ADOME

Fig. 8. A-type radome-enclosed 6 × 8 dipole array. (a) Top view of dipole


array. (b) Side view of the radome. (c) 3-D view of the REA system.

Table III shows that the proposed method only requires


nearly 300 MB core memory, while the full-wave VSIE ratio of 20.28 (2147.35/105.85 ≈ 20.28). In addition, results
solver costs 22.7 GB. More than 98% core memory is saved with good accuracy can be attained by only three outer
because the number of near-zone interactions between SWG iterations [VSIE + MSI(3)] although the target relative error
functions reduces from 1 877 444 084 to 1 638 314. Thus, both 0.001 has not been reached. This means that the convergence
the sparse matrix storage for the near-zone interactions and criterion for the outer iteration process is not necessarily set
the memory consumption of the SAI preconditioner will be as the same as that of the inner iteration. Thus, we can set
massively reduced. Accordingly, the CPU time also decreases a strict convergence criterion on the inner iteration, while the
in these two parts, as shown in Table III, and the time one for the outer iteration can be relaxed.
cost in SAI preconditioning reduces from 1589 to 2.31 s. In addition, when the two MSA elements are fed with
More importantly, the time consumed in each inner iteration different phases but with the same amplitudes, the radiation
reduces from 3.83 to less than 0.067 s because the number of pattern of the REA scans in the H-plane. Fig. 7 shows the
unknowns in the antenna region is only about 1.3% [(5557 + transmissivity of the radome versus scanning angles from 0◦
2504)/(609 994 + 5557) ≈ 1.3%] of that in the full-wave to 20◦ . The results of the proposed method agree well with
method. Therefore, the proposed method achieves a speedup the ones of the full-wave VSIE solver, and the maximum and

Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on April 12,2024 at 09:15:31 UTC from IEEE Xplore. Restrictions apply.
KE et al.: HYBRID METHOD FOR FAST AND EFFICIENT EVALUATION OF EM PERFORMANCE OF THE REAs 9803

TABLE III
D ETAILS OF THE U NKNOWNS , CPU T IME , AND M EMORY R EQUIREMENT FOR E XAMPLES 2 AND 3

C. Dipole Antenna Array Enclosed by an A-Type Sandwiched


Radome
The radiation of a 6 × 8 dipole antenna array enclosed
by an A-type sandwiched radome is investigated in the third
example. The geometry of the REA system is shown in Fig. 8,
and the detailed parameters are listed in Table II. As done in
the second example, we calculate the radiation patterns of the
REA system in the xoz and yoz planes with different outer
iteration numbers. Fig. 9 shows the radiation patterns, and all
the simulation details are given in Table III.
As shown in Table III, the hybrid method saves more than
95% core memory usage and reduces the total CPU time
from 8045 to 492 s. The speedup ratio of the proposed method
is about 16.35 compared with the full-wave VSIE solver. It is
seen that the speedup ratio of this example is lower than that
of the second example. If we take a closer look, it is found
that the ratio of the number of the unknowns in the antenna
region to the total number of unknowns of the full-wave solver
[(30 446)/(30 446 + 1 904 448) ≈ 1.57%] is larger than that of
the second example (1.3%), which means that for a given REA
system, the efficiency of the new method will increase if the
proportion of the number of unknowns in the antenna region
to the total number of unknowns becomes small.

IV. C ONCLUSION
An approximate equivalence principle is proposed to model
EM radiation of the REA arrays in the transmitting mode. The
radiating fields of the REA system are represented as superpo-
sition of the fields of the free-standing antenna array and the
difference fields of the equivalent EM currents on the surfaces
of the realistic and the auxiliary vacuum radomes. By dividing
the entire REA structure into the antenna region and radome
region, a fast and efficient hybrid method called VSIE-MSI
Fig. 9. Radiation pattern of the 6 × 8 REA system in (a) xoz plane and is proposed, which includes the mutual interactions among
(b) yoz plane.
the antenna array and the radome in an iterative manner, and
average relative errors of the transmissivity are 7.1% and 4.1%, the second-order reflection between them is considered. The
respectively. ray-tracing process is embedded in the ABCD matrix, and the

Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on April 12,2024 at 09:15:31 UTC from IEEE Xplore. Restrictions apply.
9804 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 70, NO. 10, OCTOBER 2022

EM field transmission through the variable-thickness wall of


the radome can be computed with enhanced accuracy. The
computation of the surface and VIs involved in the MSI,
as well as the ray tracing through multilayered dielectrics,
is accelerated by the idea of the MLFMA. The computational
time of the method is proven to be on the order of O(Nt logNt ),
and numerical results show that the constant in the time
complexity seems very small. The memory cost of the method
is on the order of O(Na logNa ), which is significantly less than
the full-wave solution.
It should point out that the main approximations introduced
in the proposed method include: 1) the partial fields due to
the EM currents on the virtual surface Sv are independent of
the type of radome, i.e., (4); 2) the equivalent EM currents
on the radome’s surface are calculated using the ABCD
matrix method that is only rigorously valid for planar stratified
dielectric layers; 3) multiple reflections between antenna and
radome are only considered up to the second order and the
high-order effect is neglected; and 4) approximation existed
in the GMRES solver for matrix equation [20] and the SAI
preconditioners [21] in the MLFMA. However, although the
existence of the above approximations, numerical experiments
indicate that the present hybrid method has good accuracy
Fig. 10. Ray tracing through the wall of the multilayered radome.
and reduces the CPU time and core memory usage at least (a) Determination of the incident and refraction angles at each interface.
by one order of magnitude compared with the MLFMA- (b) Determination of the positiondependent dielectric thickness.
accelerated VSIE solver. As well, the proposed method has
a fast convergence rate and usually reaches the target residual where the subscript l runs from 1 to N + 1 for θl and r̂l and
error within very few outer iterations. Therefore, the criterion runs from 0 to N for βl . The initial incident angle r̂ N +1 at the
of convergence for the outer iteration process can be further inner surface of radome is obtained from the SIs in (14)–(16).
relaxed in implementation of the method. Moreover, according to the ABCD matrix method and
Fig. 10(b), dl is the thickness in layer l along the reverse
A PPENDIX direction of normal vector n̂l . For a radome with variable
R AY-T RACING P ROCESS FOR THE M ULTILAYER thickness, it is not a constant rather than a position-dependent
R ADOME W ITH VARIABLE T HICKNESS variable. For instance, Fig. 10(b) shows the computation
process of d N . Assume that each interface in the radome has
As shown in Fig. 10(a), for a multilayered dielectric radome been discretized by triangular meshes, and the intercept point
with variable thickness, the refraction angle βl at the interface of the incident ray at Si , i.e., the interface of layer N + 1 and
between layers l + 1 and l is not necessarily identical to the layer N, is P that resides in the A B  C  . The unit normal
incident angle θl−1 at the interface between layers l and l − 1 vector of the triangle is n̂ P , and the reverse extension line of
because these two interfaces may be not parallel to each other. n̂ P intersects the next interface between layer N and N − 1 at
Therefore, unlike the traditional ABCD matrix method [1], point Q in the triangle ABC. If the unit normal vector of
[19], each ray impinged on Si and Sb should be traced in ABC is n̂ q , then d N can be calculated using the following
order to accurately compute the incident and refraction angles equation:
θl and βl at each interface. Fig. 10(a) shows the ray-tracing
process; if we specify the normalized Poynting vector of (14) n̂ q · MP
dN = n̂ P (34)
or (15) and (16) as the direction of incident wave at each n̂ q · n̂ P
incident point on Si and Sb , then the unit incident vector r̂l
from the lth layer to the (l − 1)th layer can be derived using where M is an arbitrary point in ABC. The other
the following recursive relationship: N − 1 subsequent thicknesses for the same initial point of
incidence P can be computed in the same way.
sinβl Considering the fact that the wall thickness of a realistic
r̂l = −|cosβl | · n̂l + · (r̂l+1 + |cosθl | · n̂l ) (31)
sinθl radome is much smaller than its transverse and longitudinal
sizes, the exit point of an incident ray will be not very far
and the corresponding incident and refraction angles at each
from the point of incidence, that is to say, the distance of a ray
interface are given as
traveling through the wall of a radome is much shorter than
!
the overall size of the radome. Therefore, using an oct-tree
sinθl = 1 − |r̂l+1 · n̂l |2 (32) structure to enclose the MSI region and using the idea of

si nθl · μr(l+1) εr(l+1) neighboring box in the MLFMA, redundant computation can
sinβl = √ (33)
μrl εrl be removed and the ray-tracing process can be much acceler-
Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on April 12,2024 at 09:15:31 UTC from IEEE Xplore. Restrictions apply.
KE et al.: HYBRID METHOD FOR FAST AND EFFICIENT EVALUATION OF EM PERFORMANCE OF THE REAs 9805

ated. It is noted that the size of leaf box of the oct-tree should Haiwen Ke received the B.S. degree in applied
be larger than the maximal thickness of a single dielectric electronics from Nanchang University, Nanchang,
China, in 2018. He is currently pursuing the Ph.D.
layer in the multilayered radome. Clearly, by embedding the degree with the Beijing Institute of Technology,
ray-tracing process [see (31)–(34)] in the traditional ABCD Beijing, China.
matrix [see (22)–(25)], high-frequency modeling of EM wave His current research interests include computa-
tional electromagnetics, radome design, and numer-
transmission in the variable-thickness radome can be made ical analysis of antenna arrays.
closer to physical reality compared with the conventional
method used in [1] and [15]–[19].

R EFERENCES
[1] D. J. Kozakoff, Analysis of Radome-Enclosed Antennas. Norwood, MA,
USA: Artech House, 1997.
[2] M. He, “On the characteristics of radome enclosed Archimedean spiral Pengyuan Wang received the B.S. degree in com-
antennas,” IEEE Trans. Antennas Propag., vol. 56, no. 7, pp. 1867–1874, munication engineering from North China Electric
Jul. 2008. Power University, Baoding, China, in 2019. She is
[3] P. Zhou, Z. Zhang, and M. He, “Radiation pattern recovery of the currently pursuing the Ph.D. degree with the Beijing
impaired-radome-enclosed antenna array,” IEEE Antennas Wireless Institute of Technology, Beijing, China.
Propag. Lett., vol. 19, no. 9, pp. 1639–1643, Sep. 2020. Her current research interests include computa-
[4] D. M. Pozar, “The active element pattern,” IEEE Trans. Antennas tional electromagnetics and parallel computation.
Propag., vol. 42, no. 8, pp. 1176–1178, Aug. 1994.
[5] L. Liu and Z. Nie, “Performance improvement of antenna array-radome
system based on efficient compensation and optimization scheme,” IEEE
Antennas Wireless Propag. Lett., vol. 18, no. 5, pp. 866–870, May 2019.
[6] R. K. Gordon and R. Mittra, “Finite element analysis of axisymmetric
radomes,” IEEE Trans. Antennas Propag., vol. 41, no. 7, pp. 975–981,
Jul. 1993.
[7] C. C. Lu and W. C. Chew, “A coupled surface-volume integral equation
approach for the calculation of electromagnetic scattering from com-
posite metallic and material targets,” IEEE Trans. Antennas Propag., Jinbo Liu received the B.S. degree in electronic
vol. 48, no. 12, pp. 1866–1868, Dec. 2000. information engineering from Zhengzhou University,
[8] P. Ylä-Oijala, M. Taskinen, and J. Sarvas, “Surface integral equation Zhengzhou, China, in 2010, and the Ph.D. degree
method for general composite metallic and dielectric structures with from the Beijing Institute of Technology, Beijing,
junctions,” Prog. Electromagn. Res., vol. 52, pp. 81–108, 2005. China, in 2016.
[9] M. He, X. Xu, B. Hu, and Y. Zheng, “Accurate analysis of arbitrar- He is currently an Associate Professor with
ily shaped wire antenna-dielectric radome structures,” IEEE Antennas the School of Information and Communication
Wireless Propag. Lett., vol. 6, pp. 408–410, 2007. Engineering, Communication University of China,
[10] J. W. You, S. R. Tan, X. Y. Zhou, W. M. Yu, and T. J. Cui, Beijing. His current research interests include com-
“A new method to analyze broadband antenna-radome interactions putational electromagnetics, parallel computation,
in time-domain,” IEEE Trans. Antennas Propag., vol. 62, no. 1, and the numerical analysis of antenna arrays.
pp. 334–344, Jan. 2014.
[11] W. C. Chew, J. M. Jin, E. Michielssen, and J. M. Song, Fast and Efficient
Algorithms in Computational Electromagnetics. Norwood, MA, USA:
Artech House, Feb. 2001.
[12] D. Paris, “Computer-aided radome analysis,” IEEE Trans. Antennas
Propag., vol. AP-18, no. 1, pp. 7–15, Jan. 1970. Ji Li received the B.S. degree in physics from North-
[13] J. A. Shifflett, “CADDRAD: A physical optics radar/radome analysis east Forestry University, Harbin, China, in 2013,
code for arbitrary 3D geometries,” IEEE Antennas Propag. Mag., vol. 39, and the Ph.D. degree from the Beijing Institute of
no. 6, pp. 73–79, Dec. 1997. Technology, Beijing, China, in 2019.
[14] T. Schuster and M. Sabielny, “REACH/PREACH—A physical optics Since 2019, he has been with the Beijing System
based tool for simulation of radome effects on antenna patterns,” in Proc. Design Institute of Mechanical-Electrical Engineer-
6th Eur. Conf. Antennas Propag. (EUCAP), Mar. 2012, pp. 3225–3229. ing, Beijing. His current research interests include
[15] M. A. A. Moneum, Z. Shen, J. L. Volakis, and O. Graham, “Hybrid computational electromagnetics and computational
PO-MoM analysis of large axi-symmetric radomes,” IEEE Trans. Anten- method of multiphysics problems.
nas Propag., vol. 49, no. 12, pp. 1657–1666, Dec. 2001.
[16] X.-C. Nie, Y.-B. Gan, N. Yuan, C.-F. Wang, and L.-W. Li, “An efficient
hybrid method for analysis of slot arrays enclosed by a large radome,”
J. Electromagn. Waves Appl., vol. 20, no. 2, pp. 249–264, Jan. 2006.
[17] H. F. Meng and W.-B. Dou, “A hybrid method for the analysis
of radome-enclosed horn antenna,” Prog. Electromagn. Res., vol. 90,
pp. 219–233, 2009. Mang He (Senior Member, IEEE) received the B.S.
[18] Y. An and R. Chen, “A fast hybrid method for EM analysis of electrically and Ph.D. degrees from the Department of Elec-
large metal space frame radomes,” IEEE Antennas Wireless Propag. trical Engineering, Beijing Institute of Technology,
Lett., vol. 13, pp. 1124–1127, 2014. Beijing, China, in 1998 and 2003, respectively.
[19] B. Wang, M. He, J. Liu, C. Zhang, and H. Sun, “Fast and efficient analy- From 2003 to 2004, he was a Research Asso-
sis of radome-enclosed antennas in receiving mode by an iterative-based ciate with the Department of Electronic Engineer-
hybrid integral equation/modified surface integration method,” IEEE ing, City University of Hong Kong, Hong Kong.
Trans. Antennas Propag., vol. 65, no. 5, pp. 2436–2445, May 2017. From 2008 to 2009, he was a Post-Doctoral Research
[20] Y. Saad, Iterative Methods for Sparse Linear Systems. Philadelphia, PA, Fellow with the Department of Electrical and Com-
USA: SIAM, 2003. munication Engineering, Tohoku University, Sendai,
[21] J. Liu, Z. Li, M. He, and J. Su, “An effective sparse approximate Japan. He is currently a Full Professor with the
inverse preconditioner for the MLFMA solution of the volume-surface Beijing Institute of Technology. His current research interests include com-
integral equation,” Appl. Comput. Electromagn. Soc. J., vol. 38, no. 8, putational electromagnetics and its applications, antenna theory and design,
pp. 1119–1127, 2019. radome, and frequency-selective surface design.

Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on April 12,2024 at 09:15:31 UTC from IEEE Xplore. Restrictions apply.

You might also like