Application of adaptive TFSF method for solving EM wave scattering

from complex conducting targets

Vaibhav Shah*, K. Anandhanarayanan
Directorate of Computational Dynamics, DRDL, Hyderabad, INDIA
Avijit Chatterjee
Department of Aerospace Engineering, IIT Bombay, Mumbai, INDIA
*correspondence email: webhaw@gmail.com

Abstract
A novel automated dynamically Adaptive Total Field Scattered Field (ATFSF) method designed to
optimize the total field and scattered field regions based on the solution is considered for accurate prediction of
RCS in two-dimensional problems. The electromagnetic wave scattering from complex Perfect Electric
Conducting (PEC) configurations which involves various scattering mechanisms, e.g. specular reflection,
multiple bounce, diffraction and surface wave scattering is assessed. The spatial approximation is carried out
using higher-order accurate spectral volume method, and temporal discretization is achieved using higher-order
Runge - Kutta procedure for solving time-domain Maxwell's equations on unstructured triangular grids. The outer
boundary reflections are minimized by the Perfectly Matched Layer (PML) boundary treatment. Initially, the
Method of Moment (MOM) solution is employed to access the accuracy of the algorithm for a square cylinder
configuration. Then the numerical performance of the algorithm is demonstrated for a variety of challenging
examples. In the first series of application, airfoil, square cavity and dihedral are simulated, while in the second
series interacting multiple cylinders are simulated. The algorithm can predict the bistatic RCS accurately for
variety of complex configurations, and it is finally applied to a practical airborne structure.

Keywords: Electromagnetic wave scattering, FVTD, adaptive TFSF method, higher-order method, spectral
volume method, and radar cross-section (RCS).

1. Introduction:
Electromagnetic scattering solutions for complex targets are challenging to treat with either analytical or
numerical methods because of the complicating effects of corners, curvatures, apertures and material of structures
[1, 2]. The radar signature of a complex target is based on the scattering of electromagnetic energy which occurs
when a radar beam induces currents on the structure surface. This surface current density on the scatterer surface
stimulus the near field around it and the resultant distribution of the energy levels at the far-field contribute for
the Radar Cross Section (RCS) [3]. To accurately compute the bistatic RCS of realistic models, the analytical
and numerical approaches are first applied and tested for the canonical structures. A modest complex target

1
example is an aircraft, which involve several scattering mechanisms such as specular reflection, diffraction,
interaction echo, cavity return, travelling wave and creping wave [4]. Multiple wave reflection, interaction and
inlet/exhaust cavity contribute significantly to the overall radar signature of the aircraft, and thus particular
emphasis must be done to predict their contribution accurately.
The physical and geometrical optics have proven their importance for accurate RCS prediction to most of
the aircraft surface. At the same time, for resonant structures like inlet/exhaust cavity, they are less suitable [3].
For such structures, the numerical simulation must proceed with full-wave electromagnetic exact integral or
partial differential equation methods. These techniques solve the discretized version of partial differential
equations in the time domain and then it is changed into the frequency domain by a Fourier transforms to get the
spectral information. The Finite-Difference Time-Domain (FDTD) developed by Yee [5] is the most popular
method for the numerical simulations and analysis of electromagnetic wave scattering problems. This method is
based on Cartesian non-body conforming grids; it is easy to implement and computationally efficient. The
requirement of the Cartesian grid is a restrictive limitation of the method for the application to increasingly
complex configurations. In order to remove these restrictions, the unstructured grid-based Finite Volume Method
(FVM) is borrowed from computational fluid dynamics (CFD), due to the similarity between the hyperbolic
natures of governing Maxwell's equations to the Euler equation. This similarity permits the use of algorithms
developed in CFD, such as Finite volume time-domain (FVTD), for solving the governing equations in CEM [6,
7] to obtain the radar cross-section of complex configuration. However, in order to maintain the accuracy of
numerical wave propagation, these low order methods require significant computational resources for large
electrical length scatterers due to the points per wavelength (PPW) criteria to avoid dispersion and dissipation
errors. To overcome that a higher-order method is preferred for this class of problems, where increasing the order
of approximation is more computationally efficient for achieving the desired level of accuracy than merely
reducing the mesh size. For triangular mesh, a higher-order conservative and efficient Spectral volume (SV)
method developed by Wang et al. [8, 9] is widespread in CFD for Large Eddy Simulations and aero-acoustics
studies. In Spectral Volume in Time Domain (SVDT), the reconstruction of field variables is carried out
analytically, and the reconstruction stencil is fixed, non-singular and less CPU intensive. SVDT is also amenable
in increasing the spatial order of accuracy at all cells, including cells near boundaries. Further, the outer boundary
reflections are minimized by a Perfectly Matched Layer (PML) boundary treatment [10, 11, 12].
A Total Field Scattered Field (TFSF) method proposed by Merewether et al. [13] is prevalent in FDTD,
and it became an essential feature in FDTD based commercial software [14]. Here we placed the source of the
incident field at a fictitious surface (called TFSF interface) surrounding the target and only Total field (TF) is
solves inside this interface while Scattered field (SF) is solved outside this interface. This predefined TFSF
interface based TFSF method is improved by the authors to identify TF and SF regions dynamically and is called
2
Adaptive Total Field Scattered Field (ATFSF) method [15]. In the present work, the novel automated ATFSF
method is applied to computing the electromagnetic scattering of complex shaped metal objects excited by an
external plane wave.
The paper briefly describes the numerical method used for the solutions of Maxwell equations and the
formulation of ATFSF procedure in the framework of the Finite Volume Time Domain (FVTD) method. Both
Transverse Electric (TE) and Transverse Magnetic (TM) modes of electromagnetic wave illumination are
considered. Initially, a square cylinder configuration is simulated and obtained surface current is compared with
the standard SF, TF and conventional TFSF methods against a Method of Moment (MOM) solution. Then the
numerical performance of the algorithm is demonstrated for a variety of challenging examples, and the test results
are validated against the results available in the literature. In the first series of application, airfoil, square cavity
and dihedral are simulated, while in the second series interacting fields from multiple staggered cylinders are
simulated. The algorithm is able to predict the bistatic RCS accurately for all the complex configurations. Finally,
the automated prediction capability is demonstrated for a practical airborne structure with various illumination
direction that automatically identifies the shadow region without any manual interaction. The proposed
methodology is easily extendable to three-dimensional FDTD / FVTD based software.

2. Numerical Method:
A Maxwell's equation solver is developed using higher-order accurate Spectral volume time-domain
(SVTD) technique, and it is validated for standard test cases [9, 12, 15]. The briefed formulation of the solver is
discussed below.

2.1 Maxwell's equations:

The time-domain Maxwell's equations in two-dimensions can be written in conservative vector form in
terms of total field variables as
𝜕 𝜕 𝜕
𝛼 𝜕𝑡 𝑈 + 𝜕𝑥 𝐹𝑥 + 𝜕𝑦 𝐹𝑦 = 𝑆 (1)

The conserved variables and fluxes for two-dimensional Transverse Magnetic (TM) waves are:
𝜇 0 0 𝐻𝑥 0 𝐸𝑧 0
𝛼 = [0 𝜇 0 ] , 𝑈 = [ 𝐻𝑦 ], 𝐹𝑥 = [ −𝐸 𝑧 ], 𝐹𝑦 = [ 0 ] and 𝑆 = [ 0 ]
0 0 𝜀 𝐸𝑍 −𝐻𝑦 𝐻𝑥 −𝐽𝑖𝑧

and for Transverse Electric (TE) waves are:

3
 𝜀 0 0 𝐸𝑥 0 −𝐻𝑧 −𝐽𝑖𝑥
𝛼 = [0 𝜀 0], 𝐸 𝐻
𝑈 = [ 𝑦 ], 𝐹𝑥 = [ 𝑧 ], 𝐹𝑦 = [ 0 ] and 𝑆 = [−𝐽𝑖𝑦 ]
0 0 𝜇 𝐻𝑍 𝐸𝑦 −𝐸𝑥 0

In the absence of source term in free space, the equation (1) is same for Total Field (TF) formulation and
Scattered Field (SF) formulation; only it solves the respectively total / scattered field variable.
Here, 𝜶 is a diagonal matrix of permittivity/permeability, 𝑼 is a vector of conserved variables, 𝑭𝒙 and 𝑭𝒚
are the fluxes along with x and y directions respectively, 𝑺 is the source term, 𝑬 is the electric field strength
vector, 𝑯 is the magnetic field strength vector, 𝑱 is the impressed current density vector,  is the free space electric
permittivity (F/m) and  is the free-space magnetic permeability (H/m).

2.2 Finite volume framework:

The governing partial differential equations (1) can be expressed in a vector form by integrating over a
finite volume  (without the source term) and by virtue of Gauss divergence theorem, the integral volume
term, can be written in terms of surface integral over the bounding volume  as

𝜕
𝛼∫ 𝑈 𝑑 + ∫ 𝐹(𝑈). 𝑛̂ 𝑑𝑠 = 0
 𝜕𝑡 𝑠

In two dimensions, the above governing equation is discretized in a domain composed of a number of
non-overlapping triangular cells. A formula for the flux update of ith cell in the cell-centred formulation is

𝜕𝑢𝑖 1
𝛼 =− ∑3 [(𝑓(𝑢). 𝑛̂ 𝑠)𝑘 ]𝑖 (2)
𝜕𝑡 |Ω𝑖 | 𝑘=1

Perfect Electric Conductor (PEC) boundaries are applied by setting the total tangential electric field zero
at the boundary points of the conducting scatterers. It is numerically treated as
TF formulation: 𝑛 × 𝐸 = 0
SF formulation: 𝑛 × (𝐸 𝐼𝑛𝑐 + 𝐸 𝑆𝑐𝑎 ) = 0; ∴ 𝑛 × 𝐸 𝑆𝑐𝑎 = −𝑛 × 𝐸 𝐼𝑛𝑐

̂ is the outward unit normal vector of the surface, 𝒖𝒊 the volume average of
Where F is the flux vector, 𝒏
u over cell i, cell volume 𝜴i, f(u) is the flux across the face and k is the face number of cell i. 𝑬𝑰𝒏𝒄 is the incident
electric field, and 𝑬𝑺𝒄𝒂 is the scattered electric field.
In the finite volume framework, the above equation can be solved using various numerical methods. In
present work, it is solved using a spectral volume method for spatial derivatives. The flux vector splitting of

4
Steger and Warming [16] is used in upwind flux computations at the cell boundary, while the third-order Runge-
Kutta method is used for temporal accuracy [12].

3. Dynamically Adaptive Total Field Scattered Field (ATFSF) Method:

The scattering problem around a target (scattering body) is solved numerically in a computational domain
compared to the infinite region in the real world. Similar to the conventional TFSF formulation [13], in ATFSF
method, the computational domain is divided into TF and SF regions, and these regions are separated by an
interface called TFSF interface. The total EM field and scattered EM field variables are solved in the respective
TF/SF domains. The region around the scattering body is TF domain, and the part outside the TF domain is the
SF domain. A PML layer surrounds these domains to reduce the wave reflection from the outer boundary, as
shown in Fig.1. In the conventional TFSF formulation, the TFSF interface is symmetrically placed around the
scattering body without any clue of the illumination orientation, and it is fixed thought out the numerical
simulation as shown by the red boundary in the Fig.1a. While in ATFSF method it is dynamically adapted based
on the progress of the solution at a particular time period for a given illumination direction and it evolves with
each time period shown by the red curve in Fig.1b. Thus at each time period, the optimum number of total field
cells and scattered field cells are insured.

(a) Conventional TFSF method (b) ATFSF Method

Figure 1: Comparison of the TFSF interface

In Adaptive Total Field Scattered Field (ATFSF) method, the TF and SF regions are dynamically selected
based on the solution. This automation is carried out using the intensity of Electromagnetic (EM) wave as a sensor

5
for the domain adaptation, which leads to improved accuracy in the solutions. In time-harmonic simulations, the
intensity of the EM wave is defined as the time-averaged of energy density over a time period.

𝑐 𝑇
The intensity of the EM wave in a cell (i): 𝐼𝑖 = 𝑇 ∫0 𝐸𝑖 𝑑𝑡

where Ei is the energy of cell i, c is the speed of EM wave, and T is the time period.
The solution is initialized using the scattered field formulation, and at the end of each time period, the EM
wave intensity is calculated. In the process of dynamically decomposing the regions of TF and SF cells, a certain
threshold of the intensity of EM wave is specified, and further simulations are carried out using their respective
TF / SF field variables. The threshold value was obtained numerically as 5% for TE and 15% for TM in our earlier
work [15]. If the intensity in a cell is less than the threshold, the cell is flagged as TF region; otherwise, it is
marked as SF region. The TFSF interface is marked by the SV faces that share TF and SF cells. The incident field
is analytically defined and impressed into the domain by modifying the flux terms around the TFSF interface to
satisfy the flux consistency condition [17, 18].

4. Results:
The computational performance of ATFSF method is now demonstrated by considering its application to
the solution of several test cases involving the scattering of plan TM/TE wave, of wavelength . SVTD solver
with second-order spatial and third-order temporal accuracy (S2T3) is used for most of the problems; otherwise,
it is mentioned. The computational domain is extended to a minimum of three times the wavelength, including
one wavelength of PML in all the cases. In the time-harmonic simulation, the EM wave intensity averaging is
carried out at each time period (TP), the dynamic adaptation of the TF/SF cell is carried out at each time period.
The solution is advanced in time for a sufficient number of cycles of the incident wave until a steady periodic
solution is achieved. The radar cross-section (RCS) is calculated in the post-processing stage to assess the efficacy
of the method using near field to far-field transformation. The incidents waves are defined as

(a) TM case: A plane cosine wave [20] used in the simulation :

𝑐𝑜𝑠𝜑 𝑠𝑖𝑛𝜑
𝐸𝑧𝐼𝑛𝑐 (𝑡) = 𝑐𝑜𝑠{|𝑘|(𝑥 𝑐𝑜𝑠𝜑 + 𝑦 𝑠𝑖𝑛𝜑 − 𝑐𝑡)}; 𝐻𝑦𝐼𝑛𝑐 = − 𝐸𝑧𝐼𝑛𝑐 ; 𝐻𝑥𝐼𝑛𝑐 = 𝐸𝑧𝐼𝑛𝑐 ;
𝜇𝑐 𝜇𝑐

(b) TE case: A plan sine wave [16] used in the simulation :

𝑐𝑜𝑠𝜑 𝑠𝑖𝑛𝜑
𝐻𝑧𝐼𝑛𝑐 (𝑡) = 𝑠𝑖𝑛{|𝑘|(𝑥 𝑐𝑜𝑠𝜑 + 𝑦 𝑠𝑖𝑛𝜑 − 𝑐𝑡)}; 𝐸𝑦𝐼𝑛𝑐 = 𝜀𝑐
𝐻𝑧𝐼𝑛𝑐 ; 𝐸𝑥𝐼𝑛𝑐 = − 𝜀𝑐
𝐻𝑧𝐼𝑛𝑐

6
 𝟐𝝅
Where 𝒌 = ;  is the angle of incidence, k is the wavenumber, and a is the characteristic length of the scatterer.
𝝀

4.1 Square cylinder:

The EM wave scattering solution from TM polarized plane wave impinging at =180 on an infinite
cylinder of a square cross-section is simulated here, which is a well-documented canonical problem [19]. The
scatterer is of moderate electrical length ka =5, where a is the half-edge length of the square cylinder and k is the
wavenumber, and the example enables the validation of the proposed algorithm for geometry with sharp corners
against the standard Method of Moment (MOM) solution available [19]. The schematic of the square cylinder
test case is shown in figure 2, and the mesh used is shown in figure 3. The mesh is having 8,780 nodes and 17,232
triangular cells. Each side of the square cylinder has twenty triangular cells which are equivalent to 12.5 points
per wavelength (PPW). The solution is propagated in time for six time periods. The TFSF interface is placed at a
wavelength away from the surface of the scatterer all around, forming a symmetric square interface for
conventional TFSF method.

Figure 2: Schematic of square cylinder illumination Figure 3: Triangular grid around a square cylinder

The EM wave intensity is computed at each time period, and the TF and SF region is adapted at each time
period using the intensity of the wave, as shown in figure 4. It is observed that the shadow region around the
target is captured well, and the number of TF cells is increasing in the shadow area with the time periods. The
EM wave intensity contours after the solution reaching the steady-state are shown in figure 5. The scattered field
contours are shown in figure 6, and the corresponding computed scaled induced surface current density along a-
b-c-d points is plotted and compared with the reference solution in figure 7. The current amplitude obtained using
SF, TFSF and ATFSF methods is qualitatively comparing well with the literature MOM [19] data.

7
 Figure 4: Adapted TFSF Boundary Figure 5: EM wave intensity contours

Figure 6: Scattered field contours Figure 7: Comparison of current amplitude

The accuracy of predicted induced surface current density is quantitatively compared against the available
MOM results in the literature [19] using 𝑳𝟏 , 𝑳𝟐 and 𝑳∞ norms. The error norms are defined as below.

∑𝐿𝑙=1|𝑋(𝜃𝑙 )−𝑋𝑒 (𝜃𝑙 )| ∑𝐿𝑙=1[𝑋(𝜃𝑙 )−𝑋𝑒 (𝜃𝑙 )]2 𝑚𝑎𝑥𝑙 |𝑋(𝜃𝑙 )−𝑋𝑒 (𝜃𝑙 )|
𝐿1 (𝑋) = ∑𝐿𝑙=1|𝑋𝑒 (𝜃𝑙 |
, 𝐿2 (𝑋) = √ ∑𝐿𝑙=1[𝑋𝑒 (𝜃𝑙 )]2
, 𝐿∞ (𝑋) = 𝑚𝑎𝑥𝑙 |𝑋𝑒 (𝜃𝑙 |

Where, X (l) is the induced surface current from the simulated results at an angle  and Xe (l) is the MOM
current at an angle .
The error norms of the surface current are estimated and are given in Table 1. The error norms are least in
the ATFSF formulation, highest in the SF formulation and the conventional TFSF is between them. The 𝑳∞ error
is reduced by 13.5% in ATFSF formulation compared to TFSF formulation.

8
 Table 1: Error norms for induced surface current density for phi=1800 illumination

Error SF TFSF ATFSF

L1 0.0716 0.0677 0.0529
L2 0.0790 0.0769 0.0644
L 0.1751 0.1712 0.1480

Another electrical size of ka=1 is simulated to assess the algorithm for the lower frequency with inclined
illumination of =225. Here, the wavelength of incident EM wave is greater than the linear dimension of the
square cylinder by a factor of ; thus the coarse PPW is not followed, and each side of the square cylinder has
meshed with ten triangular cells. The schematic of the test case is shown in figure 8. The triangular mesh consists
of 3,496 nodes and 6,776 cells. The TFSF interface is obtained using EM wave intensity at the end of a time
period, as shown in figure 9. The scattered field contours show the outgoing scattered field is propagating out of
the domain smoothly without any reflection, as seen in figure 10.

Figure 8: Schematic of square cylinder illumination with TM wave at phi=225o

Figure 9: Adapted TFSF boundary Figure 10: Scattered field contours

9
 The computed induced surface current density's magnitude and phase along the points a-b-c are plotted and are
compared with the reference's MOM [1] solution in figure 11. The surface current amplitude and phase are agreed well with
the literature MOM [1] data. The error in the surface current amplitude and phase are calculated and are shown in
Table 2. The error norms in amplitude and phase of the surface current are least in the ATFSF formulation. As
expected, the errors are maximum in SF formulation. The accuracy of TFSF is better than SF. The error of 𝑳∞
norm is reduced by 10% and 21% in amplitude and phase of current respectively in ATFSF formulation compared
to TFSF.

(a) Current magnitude on the surface (b) Current phase on the surface
Figure 11: Comparison of induced surface current density with various formulation on a square cylinder

Table 2: Error norms for induced surface current density for phi=2250 illumination

Amplitude of current Phase of current

Error
SF TFSF ATFFSF SF TFSF ATFFSF
L1 0.0623 0.0608 0.0589 0.0213 0.0191 0.0177
L2 0.1076 0.1061 0.0977 0.0241 0.0225 0.0221
L 0.2047 0.2007 0.1806 0.0461 0.0343 0.0271

4.2 NACA-0012 airfoil:

The second example is an exciting problem of EM wave scattering around a conducting NACA-0012
airfoil. The scattering from airfoil includes the specular reflections from its leading-edge / surfaces and diffraction
from its trailing edge. The airfoil is illuminated with a TM polarized plane wave perpendicular to the chord line
and illuminates directly on the broadside of the airfoil, as shown in figure 12. The chord length of the airfoil is

10
ten times the wavelength of the incident wave leading to an electrical size ka=10 [20]. The circular
computational domain is selected around the airfoil, and it is extended five times the wavelength from the leading
and trailing point of the airfoil. The unstructured triangular mesh is generated using commercial software, and the
mesh has 3,682 nodes and 7,076 cells.

Figure 12: Schematic of PEC airfoil illumination with TM wave at phi=90o

The EM wave intensity contours and adapted TFSF interface contours are shown in figures 13a and 13b
respectively. Similar to the square cylinder case the intensity is low in the shadow region, and a number of
adaptation increases, the size of the TF region also increases as per the expectation. The computed bistatic RCS
of the airfoil is compared with the literature solution [20] in figure 14. The RCS mono-static point ( = -90),
shadow region ( = 90), leading-edge ( = 180), and trailing edge ( = 0) comparing very well with the
reference solution, while small difference is observed around the leading and trailing edge region of airfoil. It
may due to the present simulation using a coarse mesh with second-order spectral solver, while the reference
solution being fourth-order Essential Non-Oscillatory (ENO) on a very fine mesh [20].

(a) Intensity Contours (b) Adapted TFSF Boundary

Figure 13: EM wave intensity contours and adapted TFSF boundary

11
 Figure 14: Comparison of bistatic RCS for NACA-0012 airfoil

4.3 Small PEC square cavity:

Inlet/exhaust cavity is the dominant benefactor to the overall RCS of any fighter aircraft [3]; thus, a small
PEC square cavity is taken up as for the third test case. The square cavity is having inner walls equal to the
wavelength of the incident TM polarized wave, and its thickness is 1/20th of the wavelength of the EM wave, as
shown in figure 16. The cavity is illuminated axially from the cavity open direction (=0) [3]. Similar to the
airfoil, a circular computational domain is selected around the square cavity, and it is extended five times the
wavelength from the front and base axial point of the cavity. The radar cross-section is based on resultant
scattering from the inner wall and external wall of the cavity. The unstructured triangular mesh is generated using
commercial software, and the mesh has 10,519 nodes and 20,766 cells which is equivalent to 20 PPW on the
cavity surface.

Figure 15: Schematic of PEC small square cavity

12
 The EM wave intensity contours and adapted TFSF interface contours are shown in figures 16a and 16b
respectively. The cavity is having large EM wave intensity in the centre line, while the intensity is low at the
inside wall and the base shadow region. Based on this intensity, the number of adapted TF cell is increasing with
the number of time periods. Contours of the converged scattered field and total field after ten time periods are
shown in figures 17a and 17b respectively. The scattering is higher on the inside vertical surface due to the direct
illumination, while the horizontal surfaces do not have any specular reflection due to the parallel EM wave
illumination and hence less scattering is observed.

(b) Intensity Contours (b) Adapted TFSF Boundary

Figure 16: EM wave intensity contours and adapted TFSF boundary

(a) Scattered field contours (b) Total field contours

Figure 17: Contours for NACA-0012 airfoil

The bistatic RCS computed for the square cavity is compared with the reference solution available in the
literature [3] as shown in figure 18. The RCS is comparing well with the reference solution at most of the places

13
except a small difference is observed in front and shadow regions. Another semi-open cavity illumination at
various illumination direction as already simulated and compared against the published results by the author's in
their earlier work [15].

Figure 18: Comparison of bistatic RCS for small PEC square cavity

4.4 Right-angle Dihedral:

One major contributor to the higher scattering cross-section for any structure is due to the multiple
scattering and multiple shadowing phenomena [21]. A two-dimensional right angle dihedral is the perfect example
to simulate multiple scattering and shadowing. After shown the efficacy of ATFSF for TM polarized EM waves,
TE polarized plane wave is simulated for the PEC dihedral. It comprises of two legs, joined perpendicular to each
other, at one end. The legs are aligned along x and y-axis, and the lengths of legs are ten times wavelength from
inside, and the thickness of these legs are 1/5th of the wavelength of the incident EM wave [22]. The schematic
of the dihedral geometry is shown in figure 19. The incident wave is propagating in the x-axis (=0) for the
illumination of dihedral. Here a square computational domain is selected around the dihedral, and the outer
boundary is extended up to 5 times the wavelength from the edges of dihedral leg. The unstructured triangular
mesh with 17,114 nodes and 33,582 cells which is equivalent to 10 points per wavelength on the dihedral surface
is used in the simulation.

14
 Figure 19: Schematic of PEC dihedral illumination

The simulation is carried out, and the steady-state solution is observed after 50 time periods. The scattering
pattern is a vital function of the illumination direction. Various illumination directions () 00, -300, -450 and -600
are simulated using third-order spatial and temporal accurate solver (S3T3), and the monochrome scattered field
contours of the converged simulation are plotted. The results are compared against the literature [22] results in
figure 20. The predicted results are comparing well with the reference solution; however, the intensity is higher
in the reference solution that was simulated using higher-order accurate with mesh adaption.

(a) Literature [22] (b) ATFSF

Figure 20: Contours of the scattered field with illumination at () 00, -300, -450 and -600

15
 The bistatic RCS of the right angle dihedral is computed for axial illumination (=0) and compared with
the literature solution in figure 21. The RCS peaks at mono-static and shadow regions are captured accurately,
while all the other peaks are comparing well with a small difference is observed at large. The present third-order
accurate SVTD solver is able to predict the bistatic RCS fairly compared to the fifth-order FEM results of
literature [22].

Literature
30 ATFSF

20

10
Bistatic RCS

-10

-20

-30

-40
-180 -135 -90 -45 0 45 90 135 180
Viewing angle (deg)

Figure 21: Comparison of bistatic RCS for PEC dihedral

4.5 Two adjacent cylinders:

Another essential phenomenon for the scattering solution around complex configurations is the multiple
reflections and mutual interaction between the scattering bodies. The scattering of an incident wave by two PEC
cylinders is a canonical problem of particular interest in multiple-reflection and mutual interaction [23]. The
general solution of the scattering around arbitrator cross-section cylinders was first published by Twersky [24].
The scattered field from one cylinder is the incident wave to another cylinder; hence the multiple reflections and
the mutual interaction continues [22]. Thus scattering around two adjacent PEC cylinder is simulated to verify
the efficacy of ATFSF method for interacting fields. The diameter of each cylinder is four times the wavelength
of the incident plane sine wave, thus optical scattering region. They are placed horizontally along the x-axis, and
the horizontal distance between the centres of the cylinder is 6.4 times the wavelength of wave [22]. These two

16
adjacent PEC cylinders are illuminated with a TE polarized wave propagating in the negative y-axis (= -900).
The schematic of the two adjacent illuminations is shown in figure 22.

Figure 22: Schematic of two adjacent cylinder illumination

Similar to the dihedral case, a square computational domain is selected around the adjacent cylinders, and
the outer boundary is extended up to 4.8 times the wavelength in x-axis from the cylinder surface and up to 6
times the wavelength in y-axis from the cylinder surface. The unstructured triangular mesh with 16,206 nodes
and 31,710 cells that is equivalent to 15 PPW on the cylinder surface is used in the simulation. The simulation is
carried out, and the steady-state solution is observed after 50 time periods. The converged scattered field contours
are compared with the literature [22] results, as shown in figures 23. The contours are comparing very well,
especially near the cylinder surfaces and in between the two-cylinder. Similarly, the converged total field contours
are compared with the literature [22] results, as shown in figures 24, and fair comparison is observed.

(a) Literature [22] (b) ATFSF

Figure 23: Contours of the scattered field for two adjacent cylinder

17
 (a) Literature [22] (b) ATFSF
Figure 24: Contours of the total field for two adjacent cylinder

The EM wave intensity contours and adapted TFSF interface contours are shown in figures 25a and 25b
respectively. The interaction between the cylinders is observed, and the low-intensity regions are present in
between the cylinders. Based on this intensity, the TF cells are adapted at the shadow region of each cylinder, and
they are also adapted in between the cylinders at the interaction region. The TF cells at the shadow region are
increasing with the number of time periods, while they vary a little at the interaction region.

(a) Intensity Contours (b) Adapted TFSF Boundary

Figure 25: EM wave intensity contours and adapted TFSF boundary

The bistatic RCS of the two adjacent cylinders is computed and compared with the literature solution in
figure 26. The complete bistatic RCS variation with viewing angle is captured very well. The present ATFSF

18
based second-order accurate SVTD solver is sufficient to predict the bistatic RCS accurately compared to the
results of sixth-order FEM results of literature [22].

Figure 26: Comparison of bistatic RCS for two adjacent cylinders

4.6 Three adjacent cylinders:

After testing the two adjacent cylinder problem for TE polarization, the similar interacting problem of
scattering around three adjacent PEC cylinder is simulated for TM polarization. The electrical size of each
cylinder is 0.75, and they are placed horizontally along the x-axis. The horizontal distance between the centres of
the cylinder is equal to the wavelength of incident TM wave [25, 26]. These three adjacent PEC cylinders are
illuminated with a TM polarized wave propagating in the positive y-axis (= 900), and the schematic of
illumination is shown in figure 27. The axial distance between the three cylinders is equal to the wavelength of
the incident wave compared to the 6.4 times the wavelength for the two-cylinder test case. Thus unlike the two-
cylinder cases, the three-cylinder test cases are in the Rayleigh scattering region.

Figure 27: Schematic of three adjacent cylinder illumination

19
 Similar to the two-cylinder illumination, a square computational domain is selected around the three
adjacent cylinders, and the outer boundary is extended up to three times the wavelength in the x-axis from the
cylinder surface and up to four times the wavelength in the y-axis from the cylinder surface. The unstructured
triangular mesh with 32cells on each cylinder surface which is equivalent to 8,833 nodes and 17,421 cells is used
in the simulation. The simulation is carried out, and the steady-state solution is achieved in 50 time periods. The
converged scattered field contours and total field contours are plotted in figures 28. Interaction of scattered fields
between the cylinders can be observed, and the contour plot is varied compare to the two-cylinder case due to the
smaller electrical size and small horizontal distance between the cylinders.

(a) Scattered field contour (b) Total field contours

Figure 28: Contours for three adjacent PEC cylinder

The converged EM wave intensity contours and adapted TFSF interface contours are shown in figures 29a
and 29b respectively. Similar to the two adjacent cylinder case, the interaction of fields between the cylinders is
observed. The low-intensity region due to interaction of fields is not present between the cylinders but is observed
slightly downstream of the shadow region. Based on this intensity, the TF cells are adapted at the shadow region
of each cylinder, and they are also adapted in the interaction region in the shadow region. The bistatic RCS of the
three adjacent cylinders is computed and compared with the literature's exact solution in figure 30. The predicted
bistatic RCS is in excellent agreement with the solution of equation invariance (MEI) [25] and boundary value
method [26].

20
 (b) Intensity Contours (b) Adapted TFSF Boundary
Figure 29: EM wave intensity contours and adapted TFSF boundary

Figure 30: Comparison of normalized bistatic RCS for three adjacent cylinders

4.7 Airborne Structure:

A modest complex target example is a practical airborne structure, which involves several scattering
mechanisms, i.e. specular surface return, tip diffraction, corner diffraction, interaction echo, cavity return,
gap/seam echo, travelling wave and creping wave [4]. Scattering of EM waves from an airborne structure presents
an interesting problem to the RCS expert. It offers the specialist with the opportunity to assess the accuracy of
his/her predictions for first-order scattering terms, such as specular reflections and diffraction. It also offers the

21
chance to assess the impact of second-order scattering terms, that is, creeping waves, travelling waves, or tip-tip
interactions on the overall RCS predictions [21].

Thus the ATFSF algorithm based SVTD solver is applied to a practical airborne structure with a TM polarized plane
wave at various illumination directions (=0, 45, 90, 135 and 180) and the scattering around the object is observed.
The fuselage of the structure is taken as ellipsoid having minor radius 1/10 of the major radius. The wing is modelled as a
clipped rectangle having span twice the major radius and width 2/5 times the major radius. The wing is clipped from the
tips at half wing width to give a swept-back shape and the wing trailing edge is located the centre of the fuselage. The tail
wing is 1/3 scaled-down of the wing and is located at 4/5 times the major radius towards the aft side from the centre. The
wavelength of the incident wave is taken as 1/10 of the length of the airborne structure. The configuration is shown in figure
31. Similar to the airfoil illumination, a circular computational domain is selected around the aircraft, and the
outer boundary is extended up to 5 times the wavelength in x-axis from the nose and tail of the aircraft and up to
6 times the wavelength in y-axis from the wingtip of the aircraft. The unstructured triangular mesh with 9,798
nodes and 19,014 cells which is equivalent to 10PPW, as shown in figure 32.

Figure 31: Practical airborne structure Figure 32: Grid and computational domain

The simulation is carried out, and the steady periodic state solution is achieved in 50 time periods. The
TFSF interfaces converged scattered field contours, and bistatic RCS for various illumination directions are
shown in figures 33. The ATFSF algorithm automatically selects the TF cells in the shadow region according to
the incident field direction without any human intervention in the selection of TFSF interface as in conventional
TFSF formulation. It is advantageous in automated RCS prediction of an object at various incident angles as
demonstrated here.

22
 Illumination TFSF Scattered field
Bistatic RCS
angle () Interface contours

0

45

90

135

180

Figure 33: TFSF interface, scattered field contours and bistatic RCS at various illumination direction

5. Summary:
The performance of automated dynamically ATFSF method for solving two-dimensional electromagnetic
wave scattering for complex targets in the time domain is assessed. Higher-order accurate spectral volume method
based on unstructured triangular grids is employed for spatial accuracy, and Runge-Kutta method is incorporated
for higher-order temporal accuracy. The ATFSF algorithm dynamically selects the TF/SF cells in the
computational domain based on the EM wave intensity computed at the end of each time period to resolve the
fields in the shadow region and to provide an accurate solution. The algorithm is tested for square cylinder

23
illumination, and its efficiency is assessed with respect to the available SF and conventional TFSF method, and
it is predicting the amplitude and phase of induced surface current density's more accurately (minimum 10%)
compare to the conventional TFSF method. The algorithm is rigorously tested for a number of complex problems
involving airfoil, dihedral, cavity, interacting cylinders. Multiple scattering phenomena from these complex
geometries illuminated with time-harmonic TM or TE polarized plane waves is captured correctly and the
scattering width obtained is compared very well with the available literature solution. Finally, the flexibility of
the algorithm shows the capability of accurate RCS prediction for practical airborne structure at various
illumination direction without any manual consideration of the shadow region. The ATFSF algorithm can be
easily extendable to three dimensions and other FDTD based codes for accurate RCS predictions.

Bibliography:
1. Umashankar K., and Taflove A., "A Novel Method to Analyze Electromagnetic Scattering of Complex
Objects", IEEE Transactions on Electromagnetic Compatibility, Vol. EMC-24, No. 4 November 1982.
2. Taflove A. and Umashankar K., "A Hybrid Moment Method / Finite-Difference Time-Domain Approach To
Electromagnetic Coupling And Aperture Penetration Into Complex Geometries", IEEE Transactions on
Antennas and Propagation, Vol. AP-30, No. 4 July 1982.
3. Weber Y. S., Hill K. and Young J.," The Application of Finite Volume Time Domain Techniques to EM
Scattering From Cavities and Inlets", A9636727, AIAA Paper 96-2336, AIAA meetings Papers on Disc,
June 1996.
4. Knott, E.F., "Radar Observables," Tactical Missile Aerodynamics: General Topics, Vol. 141, M. Hemsch,
ed., AIAA, Washington, DC, 1992, Chap. 4.
5. Yee K. S., "Numerical Solution of Initial Boundary Vaile Problems Involving Maxwell's equations in
Isotropic Media", IEEE Transactions on Antenna and Propagation, Volume 14, pp 302-307, May 1966.
6. Shang J.S., "Time-Domain Electromagnetic Scattering Simulations on Multicomputer", Journal of
Computational Physics 128, 381–390 (1996), Article no. 0218.
7. Shankar V., Hall W. and Mohammadian A. H., "A CFD based finite volume procedure for computational
electromagnetics – interdisciplinary applications of CFD methods", AIAA, 89-1987-CP.
8. Wang Z. J. and Liu Y., "Spectral (Finite) Volume Method for Conservation Laws on Unstructured Grids II:
Extension to two-dimensional scalar equations", Journal of Computational Physics 179, 665-697 (2002).
9. Shah V., Anandhanarayanan K. and Chatterjee A., "Spectral / Finite Volume Method for Solving Maxwell's
Equations 30th National Convention of Aerospace Engineering & National Seminar on Aerospace
Technology, The Institution of Engineers (India), Hosur, November 2016.

24
10. Sankaran K., "Cell-Centered Finite-Volume-Based Perfectly Matched Layer for Time-Domain Maxwell
System", IEEE Transactions on Microwave Theory and Techniques, Vol. 54, No. 3, March 2006.
11. Bonnet F. and Poupaud F., "Berenger Absorbing Boundary Condition with Time Finite Volume Scheme for
Triangular Meshes", Applied Numerical Mathematics 25 (1997) 33-354.
12. Shah V., Anandhanarayanan K. and Chatterjee A., "Higher-order Electromagnetics Wave Simulation with
Perfectly Matched Layer", Proceedings of 21st Annual CFD Sym. AeSI, Bangalore, August 2019.
13. Merewether D.E., Fisher R. and Smith F.W., "On Implementing A Numeric Huygen's Source Scheme In A
Finite Difference Program To Illuminate Scattering Bodies" IEEE Transactions on Nuclear Science, Vol. NS-
27, No. 6 December 1980.
14. Potter M. and Berenger J.P., "A Review of the Total Field/Scattered Field Technique for the FDTD Method",
Forum for Electromagnetic Research Methods and Applications Technologies (FERMAT), Vol. 19, 2017.
15. Shah V., Anandhanarayanan K. and Chatterjee A., "Automated adaptive TFSF method for solving time-
domain Maxwell's equations", communicated to Int. J. of Numerical Modelling, October 2020.
16. Rao S.M., “Time Domain Electromagnetics”, Academic Press, San Diego/San Francisco / New York /
Boston / London / Sydney / Tokyo.
17. Taflove A. and Hagness S., "Computational Electrodynamics: The Finite Difference Time Domain Method",
2nd edition, Norwood, MA, Artech House, 2000.
18. Schneider J.B., "Understanding the Finite Difference Time Domain Method", 8, May 2011.
19. Weber Y.S., "Investigations On The Properties Of A Finite-volume, Time-domain Method For
Computational Electromagnetics", AIAA 95-1964.
20. Chatterjee A. and Joshi S.," A p-variable Higher-Order Finite Volume Time-Domain Method for
Electromagnetic Scattering", Progress in Electromagnetics Research M, Vol. 64, 147-156, 2018.
21. Youssef N. H., "Radar Cross Section of Complex Targets", Proceedings of the IEEE, Vol. 77, No. 5, May
1989.
22. Davies R. W., Morgan K. and Hassan O., "A High Order Hybrid Finite Element Method Applied to the
Solution of Electromagnetic Wave Scattering Problems In The Time Domain", Computational Mechanics
(2009) 44:321–331; DOI 10.1007/s00466-009-0377-4.
23. Young J. W. and Bertran J. C., "Multiple scattering by two cylinders", J. Acoust. Soc. Amer., Vol. 58, No.
6, December 1975.
24. V. Twersky, "Multiple scattering of radiation by an arbitrary configuration of parallel cylinders", J. Acoust.
Sco. Amer. Vol. 24, pp. 42-46, 1952.

25
25. Chen J. and Hong W. "An Iterative Algorithm Based on the Measured Equation of Invariance for the
Scattering Analysis of Arbitrary Multi-cylinders", IEEE Transactions On Antennas And Propagation, Vol.
47, No. 9, September 1999.
26. Ragheb H. A. and Hamid M., "Scattering by N parallel conducting circular cylinders", Int. J. Electronics,
1985, Vol. 59, No. 4, 407-421.

26