Fundamentals of RCS Prediction Methodology using Parallelized Numerical Electromagnetics Code NEC and Finite Element Pre processor Vineetha Joy pdf download
Fundamentals of RCS Prediction Methodology using Parallelized Numerical Electromagnetics Code NEC and Finite Element Pre processor Vineetha Joy pdf download
https://textbookfull.com/product/electromagnetics-through-the-
finite-element-method-a-simplified-approach-using-maxwells-
equations-1st-edition-jose-roberto-cardoso/
https://textbookfull.com/product/numerical-solution-of-
differential-equations-introduction-to-finite-difference-and-
finite-element-methods-1st-edition-zhilin-li/
https://textbookfull.com/product/harmonic-balance-finite-element-
method-applications-in-nonlinear-electromagnetics-and-power-
systems-1st-edition-junwei-lu/
https://textbookfull.com/product/fundamentals-of-the-finite-
element-method-for-heat-and-mass-transfer-2nd-edition-perumal-
nithiarasu/
Finite Element Analysis of Weld Thermal Cycles Using
ANSYS 1st Edition G. Ravichandran (Author)
https://textbookfull.com/product/finite-element-analysis-of-weld-
thermal-cycles-using-ansys-1st-edition-g-ravichandran-author/
https://textbookfull.com/product/the-finite-element-method-
fundamentals-and-applications-in-civil-hydraulic-mechanical-and-
aeronautical-engineering-zhu-b/
https://textbookfull.com/product/extended-finite-element-and-
meshfree-methods-timon-rabczuk/
https://textbookfull.com/product/finite-element-analysis-a-
primer-2nd-edition-musa/
Vineetha Joy
G. L. Rajeshwari
Hema Singh
Raveendranath U. Nair
Fundamentals
of RCS Prediction
Methodology using
Parallelized Numerical
Electromagnetics
Code (NEC) and Finite
Element Pre-processor
123
SpringerBriefs in Electrical and Computer
Engineering
Series Editors
K. J. Vinoy, Indian Institute of Science, Professor, Electrical Communication
Engineering, Indian Institute of Science, Bangalore, Karnataka, India
Rakesh Mohan Jha (Late), Centre for Electromagnetics, CSIR-National Aerospace
Laboratories, Bangalore, Karnataka, India
The phenomenon of electric and magnetic field vector (wave) propagation through the
free-space, or any other medium is considered within the ambit of electromagnetics. The
media themselves, in general, could be of diverse type, such as linear/non-linear,
isotropic/non-isotropic, homogeneous/inhomogeneous, reciprocal/non-reciprocal, etc. Such
electromagnetic wave propagation problems are formulated with the set of Maxwell’s
equations. Computational Electromagnetics endeavors to provide the solution to the
Maxwell’s equations for a given formulation. It is often difficult to find closed forms
solutions to the Maxwell’s equation formulations. The advent of computers, and particularly
the initial developments of efficient coding for numerical analysis, encouraged the
development of numerical electromagnetics. A second motivation came from the interaction
of the electromagnetic wave with the matter. This could be visualized as scattering bodies,
which required incorporation of the phenomena of reflection, refraction, diffraction and
polarization. The finite/large nature of the scatterer required that problem of electromagnetics
is considered with respect to the operation wavelength leading to the classification of
low-frequency, high-frequency and resonance region problems. This also inspired various
asymptotic and grid-based finite-method techniques, for solving specific electromagnetic
problems. Surface modeling and ray tracing are also considered for such electromagnetic
problems. Further, design optimization towards hardware realization have led to the recourse
to various soft computing algorithms. Computational Electromagnetics is deemed to
encompass the numerical electromagnetics along with all other above developments. With
the wide availability of massively parallel high performance parallel computing platforms,
new possibilities have emerged for reducing the computation time and developing macro
models that can even be employed for several practical multi-physics scenarios. Both volume
and surface discretization methods have been given a new boost, and several acceleration
techniques including GPU based computation, learning based approaches, and model order
reduction have been attempted. Limitations of generating meshes and modifying these for
parametric estimation have been addressed by statistical approaches and smart solvers. Many
nature-inspired algorithms and other soft computing approaches have been employed for
electromagnetic synthesis problems. One of the recent additions is Game Theoretic
optimization.
Finally, the emergence of Computational Electromagnetics has been motivated by myriad
applications. These diverse application include but are not restricted to those in Electronics
and Communication, Wireless Propagation, Computer Hardware, Aerospace Engineering,
Biomedical Engineering, Radio-astronomy, Terahertz Technology, Photonics, etc for
modelling of devices, components, systems and even large structures. Some of the
well-known applications are in Analysis and design of radio frequency (RF) circuit, antennas
and systems, Analysis of antenna on structures, radar imaging, radar cross section
(RCS) computation and reduction, and analysis of electromagnetic wave-matter interactions
at discrete, random and periodic geometries including metamaterials. Authors are encouraged
to submit original research work in the area of Computational Electromagnetics. The content
could be either theoretical development, or specific to particular applications. This Series also
encourages state-of-the-art reviews and easy to comprehend tutorials.
Fundamentals of RCS
Prediction Methodology
using Parallelized Numerical
Electromagnetics Code
(NEC) and Finite Element
Pre-processor
123
Vineetha Joy G. L. Rajeshwari
Centre for Electromagnetics (CEM) Centre for Electromagnetics (CEM)
CSIR-National Aerospace Laboratories CSIR-National Aerospace Laboratories
Bengaluru, Karnataka, India Bengaluru, Karnataka, India
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
To
Our Families
Preface
With the advent of stealth technology, precise computation of radar cross section
(RCS) has become an inevitable component in the efficient design and development
of military vehicles. However, the scattering characterization of stealth platforms is
undoubtedly the most demanding problem in modern applied electromagnetics.
When it comes to the computation of scattered fields from conducting structures,
the accuracy and efficacy of a method of moments (MoM)-based integral equation
formulation are difficult to be surpassed. Among MoM-based solvers, Numerical
Electromagnetics Code (NEC) is a versatile open-source computer program used
for the electromagnetic analysis of metallic structures in the presence of sources or
incident fields. Since its development, NEC has continued to be one of the most
widely used electromagnetic simulation codes even in the presence of commercial
MoM-based solvers. This can be attributed to the availability of the well-
documented computational engine. However, a serious drawback of NEC is the
absence of an appropriate module for the wire-grid-based meshing of geometries
and the consequent generation of segmentation data in an NEC compatible format.
This is one of the most intricate and time-consuming steps in the computation of
RCS using NEC. In this regard, this brief presents a detailed methodology for the
computation of RCS of metallic structures using a parallelized version of NEC in
conjunction with a finite element pre-processor, which has been strategically
incorporated for geometry modelling catering to NEC guidelines. It includes a
thorough overview of the theoretical background of NEC including all relevant
aspects of formulation and modelling. This book will serve as a valuable resource
for students, researchers, scientists, and engineers working in the field of RCS
predictions and measurements.
vii
Acknowledgements
First and foremost, we would like to thank the God Almighty for being the beacon
of hope during the course of this research work.
Further, we would like to thank Shri. Jitendra J. Jadhav, Director, CSIR-National
Aerospace Laboratories, Bengaluru, for the permission to write this SpringerBrief.
We would also like to acknowledge the valuable suggestions from our col-
leagues at the Centre for Electromagnetics during the course of writing this book.
But for the concerted support and encouragement of Ms. Swati Meherishi,
Editorial Director, Applied Sciences and Engineering, and Ms. Muskan Jaiswal of
Springer Nature India Private Limited, it would not have been possible to bring out
this book within such a short span of time.
We are also forever beholden to our family members for their incessant inspi-
ration which supported us to stay at rough tracks. Vineetha Joy would like to
specially thank her husband Johnu George for his constant support and encour-
agment during the course of this work.
ix
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Theoretical Background for the Computation of Radar
Cross-Section (RCS) Using NEC . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Formulation of Integral Equations (IE) . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Electric-Field Integral Equation (EF-IE) . . . . . . . . . . . . . . 5
2.1.2 Magnetic-Field Integral Equation (MF-IE) . . . . . . . . . . . . . 6
2.1.3 EF-IE/MF-IE Combined Equation . . . . . . . . . . . . . . . . . . . 7
2.2 Computation of Scattered Fields . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Numerical Solution of Integral Equations . . . . . . . . . . . . . . . . . . . 9
2.3.1 Selection of Basis Functions and Weight
Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3.2 Solution of Matrix Equation . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Strategy for Parallelization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 Modelling Guidelines and Input/Output Formats in NEC . . . . . . . . 17
3.1 Modelling of Wires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Modelling of Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Format of NEC Input File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4 Format of NEC Output File . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4 Methodology for the Computation of RCS using Parallelized NEC
and Finite Element Pre-processor . . . . . . . . . . . . . . . . . . . . . . . . . .. 23
4.1 Overview of Finite Element Pre-processor . . . . . . . . . . . . . . . . .. 23
4.2 Revised Methodology for the Computation of RCS using
Parallelized NEC and Finite Element Pre-processor . . . . . . . . . .. 25
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 27
xi
xii Contents
Dr. Hema Singh is working as Senior Principal Scientist in Centre for Electro-
magnetics, National Aerospace Laboratories (CSIR-NAL), Bangalore, India. She
has received Ph.D. degree in Electronics Engineering from IIT-BHU, Varanasi
India in Feb. 2000. For the period 1999-2001, she was Lecturer in Physics at P.G.
College, Kashipur, Uttaranchal, India. She was a Lecturer in EEE of Birla Institute
of Technology & Science (BITS), Pilani, Rajasthan, India, for the period
2001-2004. She joined CSIR-NAL as Scientist in January 2005. Her active areas of
research are Computational Electromagnetics for Aerospace Applications, EM
analysis of propagation in an indoor environment, Phased Arrays, Conformal
Antennas, Radar Cross Section (RCS) Studies including Active RCS Reduction.
xiii
xiv About the Authors
She has contributed in the projects not only sponsored by DRDO on low RCS
phased array, Active RCS reduction, RAS development, but also in the project
sponsored by Boeing USA on EM analysis of RF field build-up within Boeing 787
Dreamliner, She received Best Woman Scientist Award in CSIR-NAL, Bangalore
for period of 2007-2008 for her contribution in the area of active RCS reduction.
Dr. Singh co-authored 14 books, 2 book chapters, 7 software copyrights, 370
scientific research papers and technical reports.
xv
xvi List of Figures
xvii
Chapter 1
Introduction
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 1
V. Joy et al., Fundamentals of RCS Prediction Methodology using Parallelized Numerical
Electromagnetics Code (NEC) and Finite Element Pre-processor,
SpringerBriefs in Computational Electromagnetics,
https://doi.org/10.1007/978-981-15-7164-0_1
2 1 Introduction
References
Burke, G.J., and A.J. Poggio. 1981 January. Numerical Electromagnetics Code (NEC)—Method
of Moments, Lawrence Livermore National Laboratory, Livermore, California, Technical
Document, Rep. UCID-18834, 719.
Burke, G.J., E.K. Miller, and A.J. Poggio. 2004 June. “The Numerical Electromagnetics Code
(NEC)—A Brief history.” In Proceedings of IEEE Antennas and Propagation Society Symposium,
vol. 3, 2871–2874, Monterey, California.
References 3
Hubing, T.H., C.H.H. Lim, and J. Drewniak. 1994 March.“A Geometry Description Language for
3D Electromagnetic Analysis Codes.” In Proceedings of the 10th Annual Review of Progress in
Applied Computational Electromagnetics, Monterey, CA, 417–422.
Rubinstein, A., F. Rachidi, M. Rubinstein, and B. Reusser. 2003. A Parallel Implementation of NEC
for the Analysis of Large Structures. IEEE Transactions on Electromagnetic Compatibility 45
(2): 177–188.
Ross, J.E., L.L. Nagy, and J. Szostka. 1999. “CAD tools for vehicular antennas.” In Proceedings of
the 1999 IX National Symposium of Radio Science, Poznan, Poland, March 16–17.
Rubinstein, A. 2004 January. Simulation of Electrically Large Structures in EMC Studies: Applica-
tion to Automotive EMC, Ph.D. dissertation, École polytechnique fédérale de Lausanne (EPFL),
https://doi.org/10.5075, 156.
Toit, C.D., and D.B. Davidson. 1995. Wiregrid: A NEC2 Pre-processor. Journal of Applied
Computational Electromagnetics Society 10: 31–39.
Chapter 2
Theoretical Background
for the Computation of Radar
Cross-Section (RCS) Using NEC
The entire theoretical framework of NEC starting from the definition of integral
equations to the solution of matrix equation is briefly introduced in this chapter. The
basis functions used for representing the unknown current densities and the weight
functions used for testing are also described in detail. Further, the strategy used for
parallelization of NEC (Rubinstein et al. 2003) is also included here.
The integral representation for the electric field due to a surface current distribution
−
→
J S on a perfectly conducting body (S) is given by,
−
−
→ →
r ) = (− jη/4π k)
E( JS r · Ḡ¯ r, r dA (2.1)
S
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 5
V. Joy et al., Fundamentals of RCS Prediction Methodology using Parallelized Numerical
Electromagnetics Code (NEC) and Finite Element Pre-processor,
SpringerBriefs in Computational Electromagnetics,
https://doi.org/10.1007/978-981-15-7164-0_2
6 2 Theoretical Background for the Computation …
where,
G r, r = k 2 I + ∇∇ g r, r
e− j k r −r
g r, r =
r − r
√ μ0
k = ω μ0 ε0 ; η = ; I = x̂ x̂ + ŷ ŷ + ẑ ẑ
ε0
−
→
r ) is the unit normal vector to the surface at r and E Sca is the scattered field
where n̂(
−
→ −
→Sca
due to the induced current J s . Substituting (2.1) for E in (2.2) yields the final
EF-IE as,
−
→ → 2
−
r ) × E Inc (
−n̂( r ) = (− jη/4π k) × n̂(
r) × J s (r ) · k I + ∇∇ g r, r dA
S
(2.3)
Since NEC employs thin cylindrical wires for representing the conducting surface
(S), the vector integral equation in (2.3) will be transformed to a scalar integral
equation. The assumptions applicable for the thin-wire approximation used in most
of the cases are given below:
(i) Currents in the transverse direction can be ignored with respect to the currents
along axial direction of the wire.
(ii) Circumferential variation in the currents along axial direction can be ignored.
(iii) The unknown current may be expressed as a filament along the wire axis.
(iv) The boundary condition pertaining to the electric field is required to be imposed
only in the axial direction.
The integral representation for the magnetic field due to a surface current distribution
−
→
J s on a perfectly conducting body is given by,
2.1 Formulation of Integral Equations (IE) 7
−
→ −→
H (
r ) = (1/4π ) J s (r ) · ∇ g r, r dA (2.4)
S
−
→ −
→
where H Inc is the incident field without the structure and H Sca is the scattered field
−
→
due to the induced current J s given by (2.4).
−
→
Substituting H Sca in (2.5) yields the final MF-IE as,
−
→ −
→
r0 ) × H Inc (
−n̂( r0 ) = n̂(
r0 ) × (1/4π ) lim J s r × ∇ g r, r dA (2.6)
r→
r0
S
r0 ) × tˆ2 (
tˆ1 ( r0 ) = n̂(
r0 ) (2.7)
The EF-IE/MF-IE combined equation is used for a structure consisting of both wires
and surfaces. In such cases, r in (2.5) is restricted to wires and the integral for
−
→Sca
E ( r ) extends over the complete structure. Similarly, r0 in (2.6) is restricted to
−
→
surfaces and the integrals for H Sca (
r ) extends over the entire structure. The integral
can be simplified by using the thin-wire approximation on wires.
The final coupled integral equations (Burke and Poggio 1981) are,
2
−
→ ∂ dS
−ŝ · E Inc (
r ) = (− jη/4π k) I (s )
k 2 ŝ · s − g
r , r
∂s ∂s
L
→ 2
− ∂
− ( jη/4π k) Js r k ŝ − ∇ g r, r dA ;
∂s
S1
−
→
tˆ2 ( r ) = (−1/4π )tˆ2 (
r ) · H Inc ( r) · I s sˆ × ∇ g r, r dS
L
− (1/2)tˆ1 (
r ) · JS (
r ) − (1/4π ) tˆ2 (
r)
S1
−
→
· J S (r ) × ∇ g r, r dA ;
for r on surfaces excluding wires (2.9)
−
→
−tˆ1 ( r ) = (1/4π )tˆ1 (
r ) · H Inc ( r) · I (s ) sˆ × ∇ g r, r dS
L
− (1/2)tˆ2 (
r ) · JS (
r ) + (1/4π ) tˆ1 (
r)
S1
· JS (r ) × ∇ g r, r dA ;
The scattered fields due to induced currents can be computed using a far-field approx-
imation. This is applicable when the observation points are far apart from the induced
current distribution by distances much larger compared to the wavelength as well
as the dimensions of the current distribution. The far-zone scattered fields from a
conducting body consisting of a wire region with contour L having linear current
−
→ −
→
distribution I (s) and a surface region S with surface current density J s ( r ) are
given by,
⎧
−
→ ⎨ − →
r0 ) = ( jkη/4π ) × (e− jkro /ro ) ×
E ( (k · I (s))k − I(s) e j k·r
⎩
L
⎫
⎬
−→ −
→
+ (k · J s (
r ))k − J s (
r ) e j k·r dA (2.11)
⎭
S
2.2 Computation of Scattered Fields 9
where k = |rroo | , k = 2π
λ
and ro is the position of the observation point.
The contour integral can be computed in closed form over each wire segment
for the cosine, sine, and constant components of the basis functions and this in turn
becomes a summation over the wire segments. Since the surface current on the patch
is represented by a delta function at the centre of the patch, the second integral
transforms to a summation over the patch elements.
−
→
For an incident plane wave with electric field E Inc , the NEC program constants are
defined such that the radar cross-section (RCS) σ/λ2 is displayed under the column
gain in the output file.
−
→Sca 2
σ E
= 4π
→ 2
(2.12)
λ2 −
E Inc
−
→
where E Sca is given by (2.11).
The integral equations described in the previous sections are solved numerically in
NEC by using the method of moments (MoM). Method of moments starts with a
conventional linear-operator equation as,
L( f ) = u (2.13)
N
f = βj f j (2.14)
j=1
The unknown coefficients (β j ) are obtained by taking the inner product of (2.13)
with a set of weighting functions wi , i.e.
N
β j wi , L f j = wi , u ; i = 1, 2, . . . , N (2.16)
j=1
The critical factors determining the accuracy and convergence of MoM are the weight
functions and the basis functions. NEC employs different basis and weight functions
unlike Galerkin’s method. The weighing functions (wi ) are chosen as a set of delta
functions in accordance with the collocation method, where the integral equation is
enforced at specific sample points on the geometry, i.e.
wi (
r ) = δ(
r − ri )
where {ri } are the set of points on the conducting surface. In NEC, conducting wires
are sub-divided into short linear segments where a sample point is defined at the
centre of each segment. On the other hand, surfaces are represented by a set of flat
patches with a sample point defined at the patch centre.
The linear current density on wires and surface current density on surfaces are
the unknown quantities in NEC which are to be represented by a set of known basis
functions. NEC uses sub-domain basis functions to simplify the calculation of inner
product integrals and also to ensure that the matrix G is well conditioned.
In the case of wires, the current on each segment is approximated as a sum of
three terms, viz., a sine, a cosine, and a constant. The total current on ith segment
can thus be expressed as,
−
→ Mp
J s (
r) = (J1i t 1i + J2i t 2i )vi (
r) (2.20)
i=1
where t 1i = t 1 (
ri ), t 2i = t 2 (
ri ), ri is the position of the centre of patch i, vi (
r) = 1
for r on patch i and 0 otherwise. Thus, every patch has two unknown quantities and
the equation is enforced for each vector component at the sample point in a given
patch.
The matrix equation obtained by appropriately substituting the basis functions and
weight functions in (2.16) has to be finally solved for the unknown currents. The
equation can be alternately expressed as,
where [G] is the interaction matrix, [I ] is the column vector with unknown currents,
and [E] is the excitation vector. NEC employs Gauss elimination method for solving
(2.21). Here, the matrix G is first factorized into the product of upper and lower
12 2 Theoretical Background for the Computation …
Let,
where,
[U ] [I ] = [F] (2.25)
Equation (2.24) is first solved for F by forward substitution. Then (2.25) is solved
for I by backward substitution. The order of the matrix Eq. (2.21) for a structure
having N w wires segments and N p patches is given by,
N = N p + Nw
The final matrix Eq. (2.21) can be written in terms of sub-matrices as,
P Q IW EW
= (2.26)
R S IP HP
where I w is the column vector of segment basis function amplitudes, I p is the column
vector of patch-current amplitudes, E w is the LHS of (2.8) evaluated at segment
centres, and H p is the LHS of (2.9) and (2.10) evaluated at patch centres.
The sub-matrix element Pij denotes the electric field at the centre of segment i
due to the basis function centred on segment j. On the other hand, the sub-matrix
element S ij denotes the tangential magnetic-field component at patch u due to the
surface current pulse on patch v where
i −1
u = 1 + Int ; Int[ ]indicates truncation
2
j −1
v = 1 + Int
2
The elements in sub-matrix Q denote the electric fields due to surface current
pulses. On the other hand, the elements in R denote magnetic fields due to segment
basis functions.
2.4 Strategy for Parallelization 13
The structure of NEC can be broadly divided into two parts, namely the input section
and the calculation section. The input section reads in the geometrical data pertaining
to the structure under consideration as well as the program control commands. The
calculation section on the other hand computes the elements of the interaction matrix
[G] which holds the geometrical and electromagnetic information about the model.
This section also deals with the solution of (2.21) by Gauss elimination method. This
final step is carried out in two routines. The first one performs the computationally
intensive task of LU decomposition of [G]. Later, it is used along with the excitation
vector [E] in another routine to obtain the solution. The computation of the constituent
elements of [G] and evaluation of the final solution to the matrix equation accounts
for more than 90% of the computation time (Burke and Poggio 1981). These parts
of the code need to be parallelized for the same reason.
The parallelized version of NEC used in this book is in accordance with (Rubin-
stein et al. 2003) where the subroutines responsible for LU decomposition (FACTR)
and solution of matrix equation (SOLVE) have been revised to work in a parallel envi-
ronment (Rubinstein 2004). This version distributes the interaction matrix among the
available processors and the matrix elements are locally and individually calculated
by their respective processors. The final system of linear equations is also solved
using a parallel version of the Gauss-Doolittle algorithm. In this way, the combined
processing power and memory of several processors working together as a cluster can
be efficiently utilized. The code can automatically detect the number of processors
and hence exploit the available memory and computing resources. The parallelized
code (Rubinstein et al. 2003) is designed to run on distributed memory parallel super-
computers and it can work on any system/cluster supporting conventional message
passing parallel environments such as Message Passing Interface (MPI) and Parallel
Virtual Machine (PVM). Since it is completely based on the original NEC code, the
input files need not be changed and the output format remains the same too.
The efficient distribution of the information stored in the matrices among the
available processors is very much critical for the successful parallelization of LU
decomposition and the parallel solution of the system of equations. For this reason,
the interaction matrix and the excitation vector have to be divided into smaller sub-
matrices which will be local to the respective processors. In (Rubinstein et al. 2003),
two-dimensional block-cyclic distribution has been used to uniformly distribute the
computation effort among all available processors and it is discussed in the following
paragraphs.
To illustrate two-dimensional block-cyclic distribution, consider a parallel
computer with four processors numbered as 1, 2, 3, and 4, as shown in Fig. 2.1. A
matrix is to be distributed among them. In a conventional way, as shown in Fig. 2.1a,
the matrix is distributed column wise to each processor. For ease of understanding,
the contents of processor 1 are shaded blue. Here, once all the elements that corre-
spond to processor 1 are assigned, it will become idle for the rest of the time. The
14 2 Theoretical Background for the Computation …
1 2 3 4 (a)
1 2 1 2 1 2 1 2
3 4 3 4 3 4 3 4
1 2 1 2 1 2 1 2
3 4 3 4 3 4 3 4
(b)
1 2 1 2 1 2 1 2
3 4 3 4 3 4 3 4
1 2 1 2 1 2 1 2
3 4 3 4 3 4 3 4
same situation repeats for the rest of the processors too. On the other hand, two-
dimensional block-cyclic distribution as shown in Fig. 2.1b guarantees that, on an
average, all the processors remain active from the start to the end of the algorithm.
A square processor grid is a desirable attribute for better efficiency as it maintains
equitable distribution of the computational load. But, in a practical scenario, the
number of processors varies from one machine to another. Therefore, the squarest
possible processor grid on the running platform is automatically calculated at the
beginning of the algorithm. ScaLAPACK library has been used in (Rubinstein et al.
2003) for implementing the block-cyclic decomposition in NEC. A brief description
of ScaLAPACK and MPI is given in Appendix A.
It is noted that the formation of the interaction matrix on a single processor
and then the distribution among other processors is not an efficient solution for
memory problem. The interaction matrix computation is therefore parallelized in
such a manner so as to produce an already distributed matrix as it is being computed.
Each processor on the grid will have adequate information required to determine
the elements of [G] once the input section is completed. An additional function
is employed to notify each processor whether the element (i, j) being calculated
corresponds to its local sub-matrix. In that case, another sub-routine will compute
the local values of (i, j) for the same element so as to be assigned to the corresponding
local matrix. Once the matrix generation is completed, each processor will hold its
local version of the interaction matrix and there would not be any further need for
2.4 Strategy for Parallelization 15
communication among the processors. Once the matrix equation is solved, data is
compiled by a block-cyclic composition routine and all the processors become aware
of the solution.
An approximation of the memory requirements in bytes at a local node for the
parallelized model of NEC is given by (Rubinstein et al. 2003),
16 × Nseg
2
Memory =
P
where P is the number of processors being used during the execution of the program
and Nseg is the number of segments. On the other hand, the original version of NEC
which is not parallelized requires (16 × Nseg
2
) bytes of memory. Figure 2.2 presents
the difference in memory requirements at a particular node between the original
version of NEC and its parallelized counterpart.
Considering a structure meshed with a particular number of segments, the vari-
ation in the run time of NEC with respect to the number of processors has been
studied in (Rubinstein et al. 2003), keeping all other parameters fixed. When the
number of processors was increased from one to four, it has been found that the
total run time reduced by almost 98%, thereby emphasizing the efficiency of paral-
lelization. However, the number of processors has to be judiciously chosen based on
the electrical dimensions of the structure under consideration, as employing large
number of processors may also increase the run time due to communication delay.
700
300
200
100
0
1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000
Number of segments
Fig. 2.2 Variation in memory requirements of NEC at a particular node with respect to number of
segments
Another Random Document on
Scribd Without Any Related Topics
Wirthshaus tritt. Zu Hause, auf Reisen, wo es auch sei!
Schreiben Sie ihnen, und Sie haben den ersten Posttag
Antwort. Die Gräfin Augusta — vom Morgen bis im Abend
laufen die Depeschen bei ihr ein, wie bei einem
Staatsminister, und werden sorgfältiger abgefertigt als in
einer Kanzlei.
Letzthin allegorisirte ich darüber mit Tellow (der
Liebesname seines und meines Freundes Cramer).
Wo ist nun die Gräfin wieder? fragte ich (Klopstock).
Cramer: Oben; schreibt Briefe.
Klopstock: Das ist wahr! Die Stolbergs! Sie liegen am
Briefschreiben recht krank danieder.
Cramer: Freilich, es ist eine Krankheit zum Tode.
Klopstock: O! sie sind schon gestorben.
Cramer: Und begraben dazu.
Klopstock: Was? Sie sind schon auferstanden.
Cramer: Ei! sie sind schon selig.
Klopstock: Ja, nun — kann ich nicht weiter.
Hierüber kommt die Gräfin herunter.
Wir sprachen, sagt ihr Klopstock, eben zusammen von
Ihrer Krankheit, Ihrem Begräbniß, Ihrer Auferstehung,
Ihrer Seligkeit!
Wie so?
Ja, gestehen Sie es nur, schöne Gräfin, Ihr Briefschreiben
ist doch eine wahre Krankheit, eine Schwachheit, eine
Seuche!
Sie mögen aber doch wol selbst gern Briefe haben?
Das mag ich wohl; — o, das Briefel e s e n ist eine ganz
vortreffliche Sache; aber das S c h r e i b e n! Es ist eine
Schwachheit, ein Fehler, sage ich, aber eine nicht eben
unliebenswürdige Schwachheit! Wenn sich die Briefe, die
Antworten wenigstens, nur selbst schrieben!
Meine Anekdote ist zu Ende. Die mußte man freilich von
Klopstock selbst erzählen hören!
Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.
textbookfull.com