GRECO: Graphical Electromagnetic Computing

for RCS Prediction in Real Time*

Juan M. Rius, M. Ferrando, and L. Jofre
Antennas, Microwave and Radar Group
Dpt. Teoria del Senyal i Comunicacions
Universitat Politecnica de Catalunya
Apdo. 30002
08080 Barcelona
Spain

Abstract tion, diffraction at edges, multiple scattering, shadowing effects,

surface-wave scattering at discontinuities, creeping waves, etc.
his paper presents a new and original approach to computing Numerical techniques based on rigorous formulation take into
T the high-frequency radar cross section (RCS) of complex radar
targets in real time, using a 3D graphics workstation. The target
account all these effects, giving very accurate results, but, on the
other hand, the computational cost is prohibitive, for very large tar-
(typically, an aircraft) is modeled with the I-DEAS solid-modeling gets.
software, using a parametric-surface approach. The high-frequency
RCS is obtained through Physical Optics (PO), Method of
Fortunately, it is well known that high-frequency scattering is
Equivalent Currents (MEC), Physical Theory of Diffraction (PTD),
a local phenomenon, and that the main contributions come from the
and Impedance Boundary Condition (IBC) techniques.
specular points at surfaces or edges [11. Accordingly, second-order
effects are not usually taken into account, and the RCS of complex
This method is based on a new and original implementation of
targets can be predicted, with reasonable accuracy, by high-
high-frequency techniques, which we have called “Graphical
frequency asymptotic approximations. Surface reflection is analyzed
Electromagnetic Computing (GRECO).” A graphical-processing
by Physical Optics. Edge diffraction is analyzed by the Method of
approach to an image of the target on the workstation screen is
Equivalent Currents, in which the far field is computed as radiated
used to identify the surfaces of the target, visible from the radar
by equivalent surface or edge currents, respectively.
viewpoint, and to obtain the unit normal at each point of these sur-
faces. High-frequency approximations to RCS prediction are then
easily computed from the knowledge of the unit normal at the illu- The main difficulty for predicting the RCS of complex targets,
minated surfaces of the target. using high-frequency techniques, is the computation of surface and
line integrals over an arbitrary shape. This shape is defined by CAD
The image of the target on the workstation screen, to be geometric-modeling software, using either a facets-and-wedges
processed by GRECO, is obtained, in real time, from an I-DEAS approximation, or a parametric-surface approach. In both cases, the
geometric model, using the 3D graphics hardware accelerator of the unit normal to the surface at each point of the geometric model
workstation. Therefore, the CPU time for the RCS prediction is must be obtained, in order to perform the electromagnetic compu-
spent only on the electromagnetic part of the computation, while tations.
the more time-consuming geometric-model manipulations are left to
the grqphics hardware. This hybrid, graphic-electromagnetic com- An additional problem i s that, according to high-frequency
puting (GRECO) results in real-time RCS prediction for complex theory, these equivalent currents are assumed to be zero over the
radar targets. surfaces or edges not illuminated by the incident wave (i.e., features
shadowed by other parts of the aircraft), so that the surface- or
edge-radiation integrals extend only over the region of the target
1. Introduction illuminated by the incident field.
The aim of this paper is focused on the prediction of the
monostatic radar cross section (RCS) of large and complex radar The classical high-frequency techniques for RCS prediction
targets. The main objective of the algorithms presented here is are based on a target modeled in terms of facets and wedges [2-41.
obtaining real-time results for arbitrary and very general target High-frequency approximations can be applied separately to each
shapes, using a graphics workstation. Thus, the RCS prediction facet and wedge, so that a closed formula is obtained for Physical
software can be easily integrated with the computer-aided-design Optics [ 5 ] and for the Method ofEquivalent Currents [I]. The RCS
(CAD) package used to model the target, providing an efficient tool of the whole target is then computed, adding coherently the sepa-
for interactive modeling, design, and analysis of aircraft with RCS rate contributions of all the illuminated facets and wedges. Multiple
specifications. interactions between facets and/or wedges can also be considered
by the classical methods.
The computation of the RCS of large and complex targets
involves different scattering mechanisms, such as specular reflec- Further processing is thus required in order to remove the
contribution of shadowed parts. The jdentification of illuminated
and shadowed regions, on the geometrical model of the target, is a
*[Editor’s note: This paper is a companion to a paper entitled, difficult and very time-consuming problem. Therefore, the classical
“High-Frequency RCS of Complex Radar Targets in Real Time,” high-frequency codes must dedicate great effort to the manipulation
which will appear in ZEEE Trum. Ant. Prop. See the Editor’s of the aircraft geometric model, prior to the electromagnetic com-
Comments in this issue for more information.] putations.

/€E€ Antennas and Propagation Magazine, Vol. 35, No. 2, April 1993 7

- ~
~ _ _ ~ __ -
-~ __
 The computational cost required for the identification of Target modeling Parametric surface Electromagnetic
shadowed regions and for multiple-scattering computations with CAD package data-base analysis
increases very rapidly with the number of facets required to model a
complex radar target. For that reason, the classical techniques are 1 1
usually implemented on powerful computers, different from the Real-time target image Hardware graphics Cavities and
at workstation screen accelerator general edges
graphics workstations used to model the target. Accordingly, a dif-
ferent approach to compute high-frequency approximations is
needed, in order to obtain real-time results on the graphics work-
stations. Graphical computation Impedance Fresnel Surface reflection:
unit normal to surface reflection coeff. Physical Optics

2. Graphical electromagnetic computing (GRECO)

Graphical identification Incremental ienght Edge diffraction:
These difficulties are overcome by the new graphical process- of edges and angles diffraction coeff Equivalent currents
ing technique: using a graphics workstation, we can obtain a 3D
image of the target (for example, the image generated by the CAD Figure 1. A block diagram with the different steps for RCS
geometric modeling software). If the viewpoint of the target is prediction by the Graphics Electromagnetic Computing
located at the position of the monostatic radar, then the picture on (GRECO) code.
the workstation screen contains only the illuminated surfaces and
edges: the shadowed ones are not visible from the observer’s view-
point, because they have been removed by the 3D visualization 3 . The x, y , z ,nX,nY,n, coordinates of each point of the illu-
hardware. minated surface are obtained by graphical processing of the
image
Furthermore, if the image is obtained using 3D illumination
and rendering, we can define the illumination-source parameters in 4. The following high-frequency approximations are then
such a way that the red, green, and blue (R,G,B) components of the computed:
color of each pixel of the image are equal to the (nt,,nY,tiz) com-
ponents of the unit vector normal to the surface at this point The a. Reflection at perfectly-conducting surfaces by the
screen memory of the workstation then has six-dimensional infor- physical-optics approximation [ 11
mation at each pixel x,y,z coordinates, and the R,G,B color com-
b. Reflection at coated surfaces by physical-optics and
ponents, which are equal to (nx,,n,,,nz) Thus, the screen memory impedance-boundary-condition approximations
has available all the information needed in order to compute high- [I, 20,211
frequency approximations This information is provided in real time
by the hardware graphics accelerator of the workstation c. Diffraction at edges by the method of equivalent
currents, using PTD [I] incremental length diffraction
Accordingly, the RCS prediction problem has been reduced to coefficients [22]. In the near future, Mitzner ILDCs will
only the electromagnetic part of the computation, which involves be implemented for bistatic RCS prediction [ 11.
the high-frequency surface and line integrals This is performed
processing the x, y , z , n,, n,,, n, information of the illuminated sur- Steps 1 to 3 involve the geometric and graphical part of the
computation, and will be explained in detail in Section 3, while Step
faces of the target on the workstation screen 4, the electromagnetic part, will be treated more briefly in Section
4.
Fresnel reflection coefficients for each pixel can be obtained,
because the unit normal to the surface is known [20, 211 The PO
surface integral is then computed by adding coherently the contri- 3. Graphical processing of an image of the target
bution of each pixel In order to perform the MEC line integral, the
edges are identified as discontinuities of the surface unit normal 3. 1 Target geometric modeling
We can compute the wedge interior angle, and the edge orientation
relative to the observer, because we know the unit normal at both A computer-aided-design package for geometric modeling of
sides of the edge Thus, we can obtain incremental length diffrac- solids [12] has been used for modeling the target geometry. The
tion coefficients (ILDCs) for each pixel of the image that lays on an aircraft is described either as a collection of facets and wedges, or
edge, and we can compute the MEC line integral by adding coher- by parametric surfaces. These last are defined using two-dimen-
ently the ILDC contributions sional non-uniform rational B-splines (NURBS) [ 13, 141.

If the workstation has a hardware graphics accelerator, the As stated before, classical RCS analysis packages usually
image of the target and the x,y,z,n,,n,,,n, information for each describe the target in terms of facets and wedges [2-41. However,
pixel can be obtained in real time. In conclusion, the CPU running parametric surfaces present the following advantages for both
time involves only the electromagnetic part of the computation, complex-object modeling and RCS prediction and optimization:
leaving the geometric part to the graphics hardware.
Complex objects require a very large number of facets,
As shown in Figure 1, the RCS is obtained by the GRECO while only a few parametric surfaces. Thus, the
code in the following steps [6-1 I]: parametric approach requires a smaller quantity of
information to define the model, which results in less
1. Geometric modeling of the aircraft with a CAD package mass-storage memory and faster processing. Another
important point is that the number of degrees of free-
2. The image of the target on the workstation screen is dom for RCS optimization algorithms is also smaller
obtained in real-time by the graphics hardware accelerator with parametric surfaces.

8 /E€€ Antennas and Propagation Magazine, Vol. 35, No.2 , April 1993
 The faceting approach introduces artificial edges and
vertices between facets, so that the surface presents a Z
“faceted” appearance, while the parametric surface is
smooth and conforms precisely to the real one. For that
reason, the RCS computed from the faceted model
contains the so-called “facet noise,” with is not present
t
when the parametric surface model is used instead.

For these reasons, parametric surfaces have been used in the

GRECO code, and are also being implemented in the TOTAL code
[IS, 161, developed at the University of Cantabria, Spain.

It must be noted that the parametric-surface database is com-

patible with an electromagnetic analysis code for predicting the
Y
RCS of arbitrary edges. This code, which has been recently devel-
oped by our group, is based on a hybrid-mode boundary-element
approach [17].
/ SCREEN
3.2 Real-time image of the target J
Hardware graphics accelerators of high-performance work- X
stations are able to render a 3D visualization of a parametric surface
model in real time. Shadowed parts of the scene are removed from Figure 3. The three-dimensional coordinate-axis convention for
the image by the graphics hardware, so that the picture on the graphical processing.
workstation screen contains only the surfaces visible from the
observer’s viewpoint.
components; the 2D location on the screen; and the distance to the
observer. For example, in the old Hewlett-Packard Turbo SRX
As shown in Figure 2, the input information for the graphics
accelerator must be, at a minimum [IS], the following: graphics accelerator used by GRECO, the coding of this informa-
tion is as follows: 2D screen 1280 x 1024 pixels; distance to the
observer in 16 bits; and R,G,B color components in 8 bits each x 3
Geometric model: NURBS surface parameters
= 24 bits. However, much better resolution can be provided by the
latest, more-powehl graphics accelerators.
Rendering: Surface-reflection coefficients for specu-
lar and diffise reflection, pattern of the reflected beam,
etc.
3.3 Graphical processing
Illumination: Position, orientation, radiation pattern In this section we will use the following convention for the
and color of the light sources 3D coordinate axis (see Figure 3): xy are the 2D coordinates of the
workstation screen, while z is along the normal direction to the
Observer: 3D viewpoint and direction of observation screen. If the observer’s viewpoint is located at the monostatic
radar position, the result is that the z coordinate of each pixel is
With these basic input parameters and hrther advanced ren- equal to the distance between the observer and each surface ele-
dering information, the result may be an extremely realistic real- ment.
time image of the aircraft, if the proper parameters are used.
From the electromagnetic point of view, this z information is
It must be noted that all the rendering computations are per-
of substantial importance for coherently adding the local high-
formed separately for each one of the three R,G,B color compo-
frequency contribution of each surface element. Using this conven-
nents. As shown in Figure 2, the six-dimensional output of the
tion, the six-dimensional outputs of the hardware graphics accelera-
graphics accelerator for each pixel includes the three R,G,B color
tor are the x, y,z,R,G,Bcoordinates and color components for
each pixel.

3.3.1 Shadowed and eclipsed surfaces identification

modeling photorealistic
database image
As stated before, one of the main difficulties in computing the
physical-optics surface integral by classical techniques (faceting
approach) [2-41 is the detection of shadowed regions. However,
GRECO has no need to face this problem, because hidden surfaces
surlace
parameters
- Graphics
hardware
accelerator
-
b) 2-D s c r e e n
of the image have been previously removed by the hardware
graphics accelerator.
c ) distance to
observer
A very simple validation of the hidden-surface elimination and
of the correctness of the z coordinate information, supplied by the
graphics accelerator, is the two-sphere system, shown in Figure 4.
Figure 2. The inputs and outputs of the hardware graphics The RCS computed by GRECO has been normalized with respect
accelerator. to the RCS of only one sphere. It can be noticed that when one of

IEEE Antennas and Propagation Magazine, Vol. 35, No. 2, April 1993 9
 (see Figure 5 ) . For three light sources of purely green, red, and blue
colors, respectively, located over each one of the three coordinate
axis, the three color components for this pixel are equal to the
(n,,,ny,nz) components of the unit normal to surface:

Figure 6 shows an image of F-117 stealth aircraft, illuminated

according to Equation (1). Thus, a purely red color means that the
unit normal to the surface is horizontal (x), a purely blue color is
vertical (y), and a purely green color is perpendicular ( z ) , to the
screen. When the unit normal is not parallel to any of the three
coordinate axis, the blending of the three color components (red,
blue, green) is equal to the (n,,,ny,nz) components of the unit
normal.
R=0.1),, d=1),
2 As the color components are always positive quantities, there
? is an ambiguity in the sign of the (nx,,ny,nz)components of the
.c unit normal. To obtain only the positive values, the graphics accel-
2 1.5
r erator must display only the illuminated surfaces, with cos@,> 0,
U)
0 and must remove the rear-facing ones, with cose, < 0. In order to
[r
. 1 obtain positive and negative values for ( n x , , t i y , n 2 ) ,it is necessary
U)
E to illuminate the target from both the positive- and negative-axis
.c
directions, using different light sources. As we need a total of six
U) 0.5
cu light sources in order to obtain the six positive and negative values
U)
0 of (nx,,ny,nz),and as there are only three independent color com-
C ponents (R,G,B), two different three-color images must be dis-
played separately. Figure 7 shows the two images of the generic
Aspect angle (degrees) missile, defined by N. Youssef in [2].
Figure 4. (a) A two-sphere system for visualizing the shadow-
ing effects in GRECO, using the hidden-surface removal
4. Electromagnetic computing
capabilities of the graphics accelerator. The distance between
the spheres is ;1and the radius is 0.12. (b) The RCS of the two-
sphere system is normalized with respect to the RCS of only Electromagnetic computing, the second main step in GRECO,
one sphere. The shadowing effect is important when the aspect uses as input information the x , y , z coordinates and the ~ ~ . ~ , t i ~ , , n ,
angle is close to 90": the normalized RCS of the system unit normal of each illuminated pixel of the target. From the knowl-
approaches one as the first sphere eclipses the second one.

the two spheres is shadowing the other one, with aspect angle close
to go", the RCS of the two-sphere system is equal to the RCS of
only the visible one.

3.3.2 Computation of unit normal to surface

If the scene is rendered using the Phong local-illumination

model [19], the color of each pixel depends only on the normal to
the surface element associated with this pixel, and on the locations
of the observer and of the light sources. As the positions of both the
observer and of the light sources are known, it is possible to obtain
the normal to the surface for each pixel of the image from the color
information.

According to the Phong illumination model, when the surface

reflection is diffise, not specular, the brightness of a pixel is com-
puted separately for each R,G,B color as the projection, No';., of Figure 5. The geometry for diffuse reflection according to the
the unit normal to the surface, Ti, on the direction of illumination, ri Phong illumination model.

10 /E€€ Antennas and Propagation Magazine, Vol. 35, No. 2, April 1993

. ~
 Figure 6. A rendered image of an F-117 aircraft on the workstation screen. Three red, blue,
and green light sources are located on the x,y,z axes, respectively. The blending of the three
colors at each pixel is equal to the ( , ? , , U , , ) I : ) components of the unit normal.

Figure 7. Rendered images of the generic missile model defined in [2]. Six red, blue, and
green light sources are located on the positive and negative x,y axis (a) and y,z axis (b). The
blending of the three colors at each pixel of the two images is equal to the positive and
negative ( t i , , ) ? ,, t i z ) components of the unit normal.

/€E€Antennas and Propagation Magazine, Vol. 35, No. 2,April 1993 11

edge of this input information, a number of different high-frequency
approximations can be implemented, in a way completely independ-
ent of the target geometry. As high-frequency theory for RCS pre-
diction is well-known [I, 201, we will concentrate on the efficient
implementation of high-frequency techniques in the GRECO code.

4.1 Physical Optics

I
According to the Physical Optics (PO) technique, the n
monostatic RCS of a perfectly-conducting surface can be approxi-
mated, at high frequencies, by the expression [20]

Figure 8. The surface element and its projection on the work-

station screen.
where 8 is the angle between the normal to the surface and the
direction of incidence, and z is the distance from the differential of
surface, ds,to the observer, projected on the direction of incidence.
The surface integral extends only over the region illuminated by the Equation (5) can be implemented very efficiently in GRECO,
incident wave. because the sinc function depends only on the angle, 8. Using the
illumination sources described in Section 3.3.2 (see the F-117 in
However, the image of the target processed by GRECO is Figure 6), the green color component of each pixel is equal to
actually a projection, on the workstation screen, of the real, 3D
surface (see Figure 8), so that the differential of the surface on the Green = nz = cos8 (6)
screen, ds', equivalent to one pixel, is equal to ds' = COSMS,and
the PO surface integral (2) can be written as As the green color is codified into 8 bits, the sinc function can
be tabulated in a 256-entry table, indexed by the green-color com-
ponent of each pixel. If the phase exponential is also tabu-
lated in a 216-entry table, indexed by coordinate z, we can add the
(3) contribution from each pixel to the PO integral with only two
floating-point real-number multiplications and additions.

Discrete computation of surface integral (3) leads to 4.2 Impedance Boundary Condition

Radar-absorbent coatings are considered through impedance-

(4) boundary-condition and physical-optics approximations, which lead
to a very simple formulation of the PO surface integral. It must be
noticed that the PO + IBC approach is valid only for surface reflec-
which is equal to the coherent addition of the phase contribution tion, when a specular reflection point exists. Impedance wedges
from all the pixels in the target image. This phase contribution is should be treated by a higher-order approximate boundary condi-
due to the distance, z, from each pixel to the observer. tion.

It must be noted that Equation (4) is correct only if a pixel According to the IBC, the contribution from each pixel in
radiates as an infinitesimal aperture, i.e., it is equivalent to the pro- Equation (5) must be multiplied by the Fresnel reflection coeffi-
jection on the screen of a differential of surface, ds, much smaller cients for polarizations parallel (q,)and perpendicular (r,) to the
than a wavelength. If the number of pixels on the screen is large plane of incidence:
enough, this condition is usually accomplished.However, when the
incidence is grazing over the surface, 8 -+ 90°, the projection, drs',
on the screen (one pixel) is very small, but the surface, ds, may be
very large.

Accordingly, in general we cannot assume that each pixel

radiates as an infinitesimal aperture, but as one which is electrically where E,f and EL are, respectively, the components of the incident
large. Assuming that one pixel is equivalent to a rectangular aper- field in the directions parallel and perpendicular to the plane of inci-
ture with uniform illumination, its contribution to the far field can dence, t# and 2;.
be approximated by a sinc function of the angle, 8, and the PO sur-
face integral becomes, in the discrete domain, Formulation of the Fresnel reflection coefficients as functions
of equivalent surface impedance is well known, and can be found in
[I, 20, 21, etc.]. The surface impedance of a dielectric coating over
a perfect-conducting surface is obtained in the usual way, through a
transmission-line equivalent circuit.

In order to implement Equation (7), we must first separate the

where !is the size of a square pixel on the screen, and e / cos8 is incident field into its E;; and ELcomponents, which is done from
the size of the drs projected on this pixel. the knowledge of the unit normal to the surface at each pixel. As

12 I€€€ Antennas and Propagation Magazine, Vol. 35, No. 2, April 1993

__
reflection coefficients are dependent only on the angle, 0, which, A
according to Equation (6), is related to the green color component
of each pixel in the image (see Figure 6), the parallel and perpen-
t
A A
dicular reflection coefficients can be tabulated in a 256-entry table,
indexed by the green color component, thus avoiding the computa- A
r=z
tion of the reflection coefficients for each pixel.

4.3 Method of Equivalent Currents (MEC)

According to high-frequency theory, the far field scattered

from a wedge can be assumed to be that radiated by an equivalent
line current, located on the edge [I, 201. This equivalent current
depends on both the directions of incidence and of observation,
relative to the orientation of the edge, so that its value is not con-
stant along the edge. Equivalent currents are usually expressed as a
1
hnction of some incremental-length diffraction coefficients [22].
The monostatic far-field scattering from the wedge, resulting from
the radiation of equivalent currents, is [ l ]

where the line integral extends along the edges illuminated by the
incident wave, ii and Ciare, respectively, the unit vectors parallel
/ ’.
and perpendicular to the plane of incidence (which is defined by the A
incident and edge directions) and y is the angle between the incident
electric field and ii.

In Equation (8), Q1,Dl, and D., stand for the monostatic

ILDCs, which depend on the angles a, 4, and p, , defined in Figure Figure 9. The wedge geometry. The directions of incidence and
9. The formulation and references to the original papers for the observation are along the z-axis, as defined in Figure 3.
GTD, the PTD, and Michaeli’s or Mitzner’s ILDC can be found in
[l, 201. In order to compute the MEC line integral (8), the GRECO
code must obtain the ILDCs for each pixel of the image laying
along an edge, and must coherently add the contributions from all
the pixels.
incident field is perpendicular to the plane of incidence, i.e. D., = 0.
As the unit normal to both faces of the edge is known, the This is actually a very good approximation when the incidence
angles a,4, and /?, can be obtained using the equations [ 1 11 direction is near the plane perpendicular to the edge, and it is well-
known that normal incidence produces the RCS flashes from the
a = c0s-1(-ri1 ri2) edge scattering.
(9)
In summary, the GRECO code computes high-frequency
scattering from edges by the Method of Equivalent Currents in the
following steps:

1. An image of the target is made on the workstation screen.

Hidden edges are removed by the graphics accelerator
hardware, so that only visible ones are displayed.

where f is the unit vector along the edge direction 2. The surface unit normal at each pixel of the image is com-
puted by graphical processing of the image (see 3.3.2).

3 . Edges are detected on the target image as discontinuities of

the unit normal to the surface.
As the reflection at the (curved or flat) faces of the wedge has 4. For each pixel along the detected edges, a, 4, and p , are
been already obtained in Section 4.1 using Physical Optics, we now computed from the unit normal to each face of the wedge,
must compute only the contribution to far-field scattering from the using Equations (9)-(12). Monostatic PTD diffraction coef-
edge alone. Although the exact ILDCs for scattering from edges ficients are then obtained, using a very simple linear
without surface reflection at faces are Mitzner ILDCs, in the approximation, which saves running time by avoiding the
GRECO code we have implemented Physical Theory of Diffraction computation of trigonometric fbnctions.
coefficients. The reason for this is that in monostatic scattering,
PTD coefficients are equal to Mitzner ILDCs, except for the 5. The line integral (Equation 8) of the MEC is evaluated,
assumption that there are no cross-polarization effects when the coherently adding the PTD coefficients for each pixel.

/€€€Antennasand Propagation Magazine, Vol. 35, No. 2, April 1993 13

5. Results for complex objects

The graphical electromagnetic computing technique presented

in this paper has been validated by comparing the results for simple
and canonical objects with analytical solutions [ 1I]. Results for
complex radar targets can be found in [6-111, and show good
agreement with both measurements and faceting-approach codes.

In this section, we will only present the results obtained by

GRECO for the generic missile, defined by N. Youssef in his classic
paper [2], together with the result for the airfoil section defined at
the JINA'90 workshop [23].

Using an old Hewlett-Packard 380 workstation (a Motorola

68040 CPU with a speed of only 2.5 MFlops) and a Turbo SRX
graphics accelerator, we can compute the RCS of a complex air- degrees
craft in 0.2 seconddangle (PO), or about 5 to 10 seconds/angle ....................................
(MEC). This speed should be improved by a factor of 5 or 10 using TOTAL. P.O. t PTD t Mulliple Scalering
Heal Time P.O.Graphical Processing
the new and faster HP-700 RISC workstations with the Turbo Verlical Polarizalion
VGX graphics accelerator.
Figure 11. The results of GRECO (only Physical Optics) com-
pared with the predictions of TOTAL [3,4] at 12 GHz.
5.1 Generic Missile

Figure 10 shows the geometry of the generic missile defined

in [2]. The results of the faceting-approach code REC072, devel- 1 Measured I RECOTA 1 TOTAL 1 GRECO
oped by Boeing Aerospace, can be found in [2]. Figure 11 presents Leading edge I 16" I 13' I I 11'
the results of GRECO (only Physical Optics), compared with the horiz. siabic -19 dBsm -17 dBsm -21 dBsm
prediction of the TOTAL code [3, 41, developed at the University of 29' 29' 37" 29"
Leading edge
Cantabria, Spain. The TOTAL code is based on a facets-and- of wing -13 dBsm -12 dBsm -12 dBsm -11 dBsm
NURBS model, and takes into account surface reflection, edge dif- Fuselage and 89" 90" 89" 91'
fraction, and surface-surface and surface-edge interactions. 8dBsm 9dBsm 8dBsm 8dBsm .
vert, sLbil.
The following table compares the monostatic RCS results of RCS null 124" 131' 122"-125" 119'
GRECO, TOTAL, and RECOTA, with measurements performed by RCS angles -18 dBsm -20 dBsm -15 dBsm -17 dBsm
Boeing Aerospace [2] at a frequency of 12 GHz, and with vertical 140"-150"
polarization. The agreement between the three predictions and the
measurements is good, except for the flash produced by the leading

edge of the wing in the TOTAL prediction. The reason for this dis-
crepancy is an error in the missile geometric model used by the
TOTAL code.

Physical Optics results of GRECO agree well with measure-

\ ments because the vertical-polarization diffraction at the trailing
39" edges is negligible. In general, we have noticed that the first-order
PO approximation usually predicts the RCS of non-stealth radar
targets (for example, the Boeing 727 [6-111) with reasonable accu-
racy, so that the computation of edge diffraction and multiple inter-
actions is not always necessary.

5.2 Airfoil section

Figure 12 shows the two-dimensional airfoil-section geome-

try, as defined in the workshop, "RCS of Perfectly- Conducting or
Coated Bodies," [23] held at Nice in November, 1990. In Figures
13 and 14, the GRECO high-frequency prediction is compared with
a numerical-method solution, presented at the workshop by the
Centre Commun de Rech. Louis-Bleriot of AEROSPATIALE.

It can be noticed, in Figures 13 and 14, that PO results cor-

rectly predict surface reflection, but not edge diffraction, when the
incident polarization is parallel to the edge (TM). However, if the
Figure 10. The generic missile model defined in (21. The fuse- Method of Equivalent Currents with PTD coefficients is added to
lage length is about 4 0 1 at 12 GHz. PO, the result agrees very well with the numerical solution

14 IEEEAntennas and Propagation Magazine,Vol. 35, No. 2, April 1993

__
 P.O.+PTD P.O.
7 x -
MOM
----- ..............
1

7 x g x

Figure 12. The airfoil section defined in the JINA’90 workshop

1231.

P.O.+PTD P.O.
-
MOM
.................... :
----- ..............
.........................................................
1

-150 -100 -50 0 50 100 150

Azimuth aspect angle

Figure 14. The GRECO results for the Airfoil in Figure 12, for
TE polarization (Physical Optics and Physical Optics +
Physical Theory of Diffraction), compared with a numerical
solution [23].

regions is avoided, which leads to a great advantage

over the classical codes for RCS prediction.

The sirface and line integrals (PO and MEC) are

evaluated by graphical processing of an image of the
-150 -100 -50 0 50 100 l50
target on the workstation screen. As this is independ-
Azimuth aspect angle ent of target complexity and electrical size, the CPU
time and RAM requirements do not increase with tar-
Figure 13. The GRhCO results for the Airfoil in Figure 12, for get size or complexity.
TM polarization (Physical Optics and Physical Optics +
Physical Theory of Diffraction), compared with a numerical The graphical processing approach obtains a 2D
solution [231. matrix, containing the unit normal to the surface over
the illuminated regions of the target. As this is inde-
pendent to the electromagnetic part of the computa-
6. Conclusions tion, it is easy to develop and integrate code for com-
puting the different high-frequency approximations
In this paper we have presented a new implementation of This electromagnetic code is relatively small in size,
well-known high-frequency techniques. Using a 3D workstation and absolutely independent of the target geometry.
with a graphics hardware accelerator, monostatic RCS prediction is
obtained in real time for large and complex radar targets. The approach permits real-time computation with a
high-performance workstation and hardware graphics
Graphical electromagnetic computing (GRECO) has the fol- accelerator, while the classical techniques require
lowing advantages over classical techniques [2-41: powefil super computers in order to obtain real-time
results.

The target can be modeled by parametric NURB sur- The GRECO code can be integrated with a CAD geo-
faces, requiring less mass-storage memory that the metric modeling package [ 121, thus providing an effi-
faceting approach, and enabling more accurate cient tool for interactive modeling, design and analysis
adjustment to the real target surface, thus avoiding the of aircraft with RCS specifications (see Section 7).
“facet noise” usually present in classical facet-
modeling codes. It must be noted that GRECO is able to analyze targets of
electrical size as large as 2“ / 162, with a maximum phase error of
The hardware graphics accelerator removes hidden 1 8 , where n is the number of bits in which the distance, z, to the
surfaces and edges, so that they do not contribute to observer is discretized. This means that we can analyze 40002 with
surface or line integrals. The difficult and time- the usual 16-bit discretization. The resolution in the discretization
consuming software identification of shadowed o f x y 2D screen coordinates, usually 1024x 2048 pixels, only limits

/€€E Antennas and Propagation Magazine, Vol. 35, No. 2, April 1993 15

- ~- _ . -
the maximum complexity of the target (details must be larger than a 3. R. Abad, J. I. Casado, P. L. Primo del Val, R. Torres, M.
pixel), but does not limit the target electrical size, because XJJ lay Domingo, F. Rivas, M. F. Catedra, “Modelos matem ticos para el
on a plane perpendicular to the directions of incidence and obser- estudio de la RCS debida a doble reflexion, difraccion en aristas, y
vation. sombras por eclipse en parches poligonales planos,” in C’ Synipo-
sium Nacional del Comite Espaiiol de la U I W , pp. 190-194, Vigo,
In conclusion, graphical processing is probably the optimum 26-28th Septiember de 1990.
approach for analyzing very large and complex aircraft, using high-
frequency approximations. In Section 7, the application to interac- 4. M. Domingo, F. Rivas, M. F. Catedra, R. Abad, J. I. Casado, P.
tive design of aircraft with RCS specifications will be discussed. L. Primo del Val, R. Torres, “Programa base de GTD-PTD para el
calculo de la RCS de cuerpos conductores modelados por parches
However, there are some scattering sources that cannot be planos y teniendo en cuenta simple y doble reflexion, difraccion en
analyzed by GRECO: cavities at engine inlets, creeping waves, dis- aristas y eliminacion de zonas oclutas,” in V Symposirrni Nricional
continuities and slots over the aircraft surface, etc. These effects del Comite Espaiiol de la URSI, pp. 185-189, Vigo, 26-28th Sep-
should be analyzed by different codes, and the results added to tiember 1990.
GRECO.
5 . W. B. Gordon, “Far Field Approximation of the Kirchhoff-
Helmholtz Representation of Scattered Fields,” 1EEE limn. Ant.
7. Potential application to RCS optimization Prop., AP-23, No. 5, July 1975, pp. 864-876.

As the GRECO code is based on graphical processing of an 6. J. M. Rius, M. Ferrando, “Real Time Radar Cross-section of
image of the target, it can be easily integrated with the CAD soft- Complex Targets by Physical Optics Graphical Processing,” in
ware package [12] used for modeling the aircraft. This makes pos- Digest 1990 IEEE International Symposiimi on Atitenims and
sible the interactive modeling, design, and analysis of aircraft with Propagution,” pp. 1280-1283, Dallas, 7-1 1 May 1990.
RCS specifications.

Moreover, the GRECO code has some other important

advantages for automatic RCS minimization, using non-linear con- 7. J. M. Rius, M. Vall-llossera, M. Ferrando, “Fast Algorithms for
strained optimization methods: Radar Cross Section Computation of Complex Objects,” in
Journees Internationalesde Nice sur les Antennes JINA ’90, Nice,
Parametric surface modeling with NURBS allows a France, 13-16th Noviember 1990.
very large reduction in the number of parameters
necessary to define the surface, which may accelerate 8. J. M. Rius, M. Vall-llossera, “High Frequency Radar Cross Sec-
the optimization run-time by several orders of magni- tion of Complex Objects in Real Time,” in Digest 1991 IEEE AP-S
tude. InternationalSymposium,London, Ontario, Canada, June 1991.

The sensitivity analysis of the RCS optimization algo- 9. J. M. Rius, M. Vall-llossera, A. Cardama, “High Frequency RCS
rithm is simplified, because specular-reflection points of Perfectly Conducting or Coated Complex Objects in Real Time,”
on the target surface are visible on the workstation in 21th EuropeanMicrowave Coiference, 9-12th September, 1991,
screen. This means that the parameters to which the Stuttgart, Germany.
RCS is more sensitive are the local control points of
the specular surface, so that the number of degrees of
freedom to optimize is reduced from of the order of 10. J. M. Rius, M. Vall-llossera, A. Cardama, “Real Time RCS of
thousands to of the order of tens. Perfectly Conducting or Coated Radar Targets,” in hu’ Interna-
tional Conference on Electromagnetics in Aetmpace Apjilications
3D visualization of the target allows the monitoring of (ICEAA 91), 17-20th September 1991, Torino, Italy.
shape evolution in real time.
11. J. M. %us, “Seccion Recta de Blancos Radar Coniplejos en
Tiempo Real,” Tesis Doctoral, Universidat Politecnica de Cata-
lunya, July 1991.
8. Acknowledgments 12. “1-DEASGeomod Users Guide,” Structural Dynamics Research
Corporation (SDRC), Milford, OH 45150.
This work has been supported by the Spanish “Comision
Interministerial de Ciencia y Tecnologia” (CICYT) under the pro-
ject ACCION ESPECIAL PRONTIC “Programa de investigacion 13. C. de Boor, A Practical Guide to S/dines,” New York,
aplicada para el desarrollo y validacion de metodos de calculo Springer-Verlag, 1978.
numeric0 para la prediccion y analisis de las caracteristicas de 10s
ecos radar (RCS) y su reduccion,” TIC 88-288E. Management: 14. G. Farin, Curves and Surfaces for Conipztter Aided Geonietric
“Direccion General de Telecomunicaciones” (D.G.Tel) Coordina- Design: A practical Guide,” New York, Academic Press, 1988.
tion: C.A.S.A.
15. M. Pizarroso, J. Perez, M. F. Catedra, “Software geometric0
para el calculo de RCS de solidos modelados por NURBS,” in V
9. References Symposium Nacional del Comite Espaiiol de la U I W , pp. 195-199,
Vigo, 26-28th September 1990.
1. E. F. Knott, J. F. Shaeffer, M. T. Tuley, Radar Cross Section,
Artech House, 1985. 16. X. Fernandez Hermida, A. Garcia Pino, “Spline Modeling to
Compute the RCS of Arbitrarily Shaped Cavities by the GO/&
2. N. N.Youssef, “Radar Cross Section of Complex Targets,” Method,” in Digest of IEEEIURSI Meeting, Dallas, 7-1 1 May
Proc. IEEE, Vol. 77, No. 5 , May 1989. 1990, p. 280.
16 IEEEAntennas and Propagation Magazine, Vol. 35, No. 2, April 1993

~~ ~
17. J. M. Alvarez, J. C. Cruellas, M. Ferrando, “A Hybrid Modal- 21. D. Klement, J. Peissner, V. Stein, “Special Problems in Apply-
Boundary Element Method for Electromagnetic Scattering from ing the Physical Optics Method for Backscatter Computations of
Arbitrary Conducting Wedges,” in Digest of 1990 E L L Itilertia- Complicated Objects,” IEEE Trans. Ant. Prop., AP-36, No. 2, Feb-
tional Symposium on Anfennas and Propogafion,” pp 1288- 1291, ruary 1988, pp. 228-237.
Dallas, 7-11 May 1990.
22. R. A. Shore, A. D. Yaghjian, “Incremental Diffraction Coeffi-
18. B. R. Dewey, “Computer Graphics for Engineers,” New York: cients for Planar Surfaces,” IEEE Trans. Ani. Prop., AP-36, No. 1,
Harper & Row, 1988. January 1988.

19. B. T. Phong, “Illumination for Computer Generated Images,” 23, Workshop organized by Dassault Aviation, Societe Mothesim,
PhD Dissertation, University of Utah, 1973. and CNET-PAB La Turbie, Nice, France, 16th Noviember 1990.

20. A. L. Maffet, Topics for a Statistical Descripiion of Radar

Cross Section,” New York: John Wiley & Sons, 1989.

Nosich Conference Grant THE NOSICH CONFERENCE GRANT

Awardees Announced Electromagnetic Theory and Applications
Ex-USSR applicants are welcome to apply for the annual Confer-
The first two recipients of the Nosich Conference Grant have
ence Grant for attending international conferences held outside the
been announced: Alexander Y. Svezhentsev, and Andrey Andrenko.
ex-USSR. The scope of the Conference must be within the topic
The Noisch Conference Grant is awarded to young PhD candidates
areas of Electromagnetic Wave Theory and Applications, such as
or recipients who are from the ex-USSR, for the purpose of attend-
Antennas, Waveguides, Scattering, etc. Applicants should 1) be
ing international conferences held outside the ex-USSR. An
under 40, 2) have completed or be in the process of completing a
announcement of the Grant, and hrther details on its requirements,
PhD (Candidate of Science) thesis in radiophysics, and 3) have
appears in this issue of theMagazine.
good oral and written English skills. Preference is given to young
researchers of the Ukraine traveling to the West for the first time.
A Grant was given to Svezhentsev for attending the Interna-
However, applicants from other countries are eligible.
tional Conference on Lightwave Technology and Communications
(BILCON’92), held at Bilkent University, Ankara, Turkey, July
The amount of Grant depends on circumstances, but does not
26-28, 1992. He presented a paper entitled, “Microstrip/Slot Lines
exceed $500 USD. The Grant is not restricted to travel expenses,
with Optically Switched Characteristics.”He also received a partial
but can cover registration fees and accommodations. There is no
award to attend the International Symposium on Antennas and
application deadline. However, application should be made well be-
Propagation (ISAP’92), held in Sapporo, Japan, September 22-25,
fore a proposed event. It should be written in English, in free for-
where he presented a paper entitled, “Coupling Effects for Complex
mat, and should contain references. A document certifiing the
Waves in Multilayer Cylindrical Strip and Slot Lines.”
acceptance of the conference paper must be enclosed. Normally, the
successhi applicant is supposed to pass a personal interview.
A. Y. Svezhentsev was born in 1957 in Tambov, Russia. He
Correspondence should be sent to both addresses:
obtained both the MS and PhD degrees in Radiophysics from the
Kharkov University, Ukraine, in 1979 and 1987, respectively. He is Prof. A. I. Nosich, c/o
Prof. A. I. Nosich
currently a Senior Researcher with the Institute of Radiophysics & Dept. Sci. Techn. Information
Visiting Scientist
Electronics, Ukrainian Academy of Sciences (IRE USE), Kharkov. Institute of Radiophysics and
Dept. of Electrical Engineering
His main research interests are microstrip and slot lines, open Electronics
waveguides, and complex-mode behavior. and Computer Science
Kumamoto University Ukrainian Academy of Sciences
Kumamoto 860, Japan 12, UI. Proskury, Kharkov,
Andrenko also received a Grant to attend BILCON’92, where 3 10085, Ukraine
he presented a paper entitled, “Wave Transformation by Integrated Tel: (81) 96 344 21 11 ext. 3635
Fax: (81) 96 345 1553 3635 Tel: (0572)-448486
Finite-Periodic Metal-Rod Coupler in a Thin-Film Guide.” He also Fax: (0572)-441105
received a partial Grant to attend the URSI Symposium on EM Fax: (81) 96 345 1553
Theory in Sydney, Australia, August 17-20, where he presented a
paper entitled, “Rigorous Solution of the Problem of Dielectric Slab
Mode Scattering by Finite-PeriodicInhomogeneities.”

A. S. Andrenko was born in 1964 in Kharkov, Ukraine. He

obtained both the MS and PhD degrees in Radiophysics from the
Kharkov University in 1986 and 1992, respectively. He is currently
a Junior Researcher with the IRE UAS. His main research interests
are scattering from inhomogeneities in dielectric-slab waveguides.

I€€€ Antennas and Propagation Magazine, Vol. 35, No. 2, April 1993 17