[PPT] - Tools for Synthesis of Antennas and Sensors Project of Technology PowerPoint Presentation

SLIDE 1

Tools for Synthesis of Antennas and Sensors

Project of Technology Agency of the Czech Republic Miloslav ˇ Capek & Project team1

Department of Electromagnetic Field CTU in Prague, Czech Republic miloslav.capek@fel.cvut.cz

Prague, Czech Republic March 15, 2016

1Results you will see in this presentation are collective work of the whole project team. ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 1 / 70

SLIDE 2

1 Motivation

2 Source Concept What is the source concept? Selected applications of the source concept

3 Characteristic mode decomposition

4 About the project

5 Project infrastructure

6 AToM architecture AToM – Closer investigation AToM’s – Features

7 Integration into Visual CEM (ESI Group)

8 Conclusions

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 2 / 70

SLIDE 3

Motivation

analysis synthesis

20
15
10
5

f0 s11 [dB]

Qmax = 7

Perfect Electric Conductor Feeding Point  Antenna characteristics electric current Antenna analysis × antenna synthesis.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 3 / 70

SLIDE 4

Source Concept What is the source concept?

Source Concept

What is actually the Source Concept?

Source Concept

Integral and variational methods Modal de- compositions Perspective topol-

gy and

geometry HPC, algorithm efficiency Heuristic

r convex
ptimization

Sketch of main fields of the source concept.

It can be observed that . . . ◮ an antenna is completely represented by a source current, ◮ all parameters can be inferred from a source current, ◮ any proper int.-diff.

perator can be

decomposed into modes

r can be inverted,

◮ spatial decomposition

f current is possible.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 4 / 70

SLIDE 5

Source Concept Selected applications of the source concept

Source Concept

Applications: Characteristic Modes

W J1 W J2

Modes J1 and J2 are depicted.

J =

M

m=1

Jm, E 1 + λm Jm ◮ characteristic modes (CMs) decomposition2 XJ = λRJ (1)

other useful decompositions

J =

m

αmJm (2) ◮ CMs are excellent for pattern synthesis

r feeding network synthesis3
2R. F. Harrington and J. R. Mautz. “Theory of Characteristic Modes for Conducting Bodies”.

In: IEEE Trans. Antennas Propag. 19.5 (1971), pp. 622–628. doi: 10.1109/TAP.1971.1139999

3R. F. Harrington and J. R. Mautz. “Pattern Synthesis for Loaded N-Port Scatterers”.

In: IEEE

Trans. Antennas Propag. 22.2 (1974), pp. 184–190. doi: 10.1109/TAP.1974.1140785

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 5 / 70

SLIDE 6

Source Concept Selected applications of the source concept

Source Concept

Applications: Structural Decomposition

WB WA J(rA) J(rB)

Division of Ω into two parts.

J =

K

k=1

Jk ◮ similar to structural decomposition in mechanical engineering ◮ to decide what part of a radiator stores significant portion of energy / radiates well4 ◮ excellent for synthesis of reflect arrays5 ◮ combination with CM: sub-structure modes6

4M. Capek et al. “The Measurable Q Factor and Observable Energies of Radiating Structures”.

In: IEEE Trans. Antennas Propag. 62.1 (2014), pp. 311–318. doi: 10.1109/TAP.2013.2287519

5J. Ethier. “Antenna Shape Synthesis Using Characteristic Mode Concepts”.

PhD thesis. University

f Ottawa, 2012
6J. L. T. Ethier and D.A. McNamara. “Sub-structure characteristic mode concept for antenna shape

synthesis”. In: Electronics Letters 48.9 (2012), pp. 471–472. issn: 0013-5194. doi: 10.1049/el.2012.0392

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 6 / 70

SLIDE 7

Source Concept Selected applications of the source concept

Source Concept

Applications: Optimization

W0 Wmax Wfinal Wmax

Optimization of antenna’s shape.

single-objective optim.: {yj} = min

{xi} F (J)

multi-objective optim.: {yj} = min

{xi} {Fj (J)}

◮ both single- and multi-objective

ptimization can be utilized in order to
btain best antenna performance

◮ many objectives can be subjects of convex optimization7

F (J, J) has to be positive

semi-definite8

convex optimization does not result in

specific design, only minimizes given convex function

7M. Gustafsson and S. Nordebo. “Optimal antenna currents for Q, superdirectivity, and radiation

patterns using convex optimization”. In: IEEE Trans. Antennas Propag. 61.3 (2013), pp. 1109–1118. doi: 10.1109/TAP.2012.2227656

8S. Boyd and L. Vandenberghe. Convex Optimization.

Cambridge University Press, 2004

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 7 / 70

SLIDE 8

Source Concept Selected applications of the source concept

Source Concept

Applications: Advanced Post-processing VG1 VG2A VG2B

Feeding network synthesis.

βm,n = ℜ {αmα∗

n}

where: αm = Jm, E 1 + λm ◮ any antenna parameter can be defined by functional containing current(s) ◮ recently derived:

radiation efficiency without IBC9
measurable QZ factor10
energies for sub-wavelength

radiators11 (ka < 1)

no matter if modal / structural / total

current is substituted

9M. Capek, J. Eichler, and P. Hazdra. “Evaluation of Radiation Efficiency from Characteristic

Currents”. In: IET Microw. Antennas Propag. 9.1 (2015), pp. 10–15. doi: 10.1049/iet-map.2013.0473

10M. Capek et al. “The Measurable Q Factor and Observable Energies of Radiating Structures”.

In: IEEE Trans. Antennas Propag. 62.1 (2014), pp. 311–318. doi: 10.1109/TAP.2013.2287519

11G. A. E. Vandenbosch. “Reactive Energies, Impedance, and Q Factor of Radiating Structures”.

In: IEEE Trans. Antennas Propag. 58.4 (2010), pp. 1112–1127. doi: 10.1109/TAP.2010.2041166

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 8 / 70

SLIDE 9

Source Concept Selected applications of the source concept

Source Concept

Applications: Fundamental Bounds and Optimal Currents

Optimal Q on a spherical shell.

2X′In = QnRIn ◮ optimal current can be found12 AnJn = ξnBJn (3) ◮ optimal (minimal / maximal)

quality factor Q
D/Q ratio
radiation efficiency ηrad
antenna gain G
other parameters. . .

◮ additional constraint of current resonance can be enforced

12M. Capek and L. Jelinek. “Optimal Composition of Modal Currents For Minimal Quality Factor Q”. .

In: (2016). arXiv:1602.04808,L. Jelinek and M. Capek. “Optimal Currents on Arbitrarily Shaped Surfaces”. In: (2016). arXiv:1602.05520

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 9 / 70

SLIDE 10

Source Concept Selected applications of the source concept

Design of optimal antenna

Source Concept Design of Op- timal Antenna Antenna Synthesis

AToM

◮ The source concept was recently utilized for so-called optimal antenna design.

see e.g. recent papers by M. Cismasu

and M. Gustafsson13 or by J. Ethier and D. McNamara14

◮ To this purpose, it is beneficial to have a fast prototyping environment with partially open-source code.

13M. Cismasu and M. Gustafsson. “Antenna Bandwidth Optimization With Single Freuquency

Simulation”. In: IEEE Trans. Antennas Propag. 62.3 (2014), pp. 1304–1311

14J. L. T. Ethier and D. A. McNamara. “Antenna Shape Synthesis without Prior Specification of the

Feedpoint Locations”. In: IEEE Trans. Antennas Propag. 62.10 (2014), pp. 4919–4934. doi: 0.1109/TAP.2014.2344107

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 10 / 70

SLIDE 11

Source Concept Selected applications of the source concept

Design of optimal antenna

Source Concept Design of Op- timal Antenna Antenna Synthesis

AToM

◮ The source concept was recently utilized for so-called optimal antenna design.

see e.g. recent papers by M. Cismasu

and M. Gustafsson13 or by J. Ethier and D. McNamara14

◮ To this purpose, it is beneficial to have a fast prototyping environment with partially open-source code. The optimal antenna design leads at least to a partial antenna synthesis!

13M. Cismasu and M. Gustafsson. “Antenna Bandwidth Optimization With Single Freuquency

Simulation”. In: IEEE Trans. Antennas Propag. 62.3 (2014), pp. 1304–1311

14J. L. T. Ethier and D. A. McNamara. “Antenna Shape Synthesis without Prior Specification of the

Feedpoint Locations”. In: IEEE Trans. Antennas Propag. 62.10 (2014), pp. 4919–4934. doi: 0.1109/TAP.2014.2344107

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 10 / 70

SLIDE 12

Characteristic mode decomposition

1 Motivation 2 Source Concept

What is the source concept? Selected applications of the source concept

3 Characteristic mode decomposition 4 About the project 5 Project infrastructure 6 AToM architecture

AToM – Closer investigation AToM’s – Features

7 Integration into Visual CEM (ESI Group) 8 Conclusions

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 11 / 70

SLIDE 13

Characteristic mode decomposition 1968

CM was born

(Garbacz)

1 Schelkunoff’s work on modes

1930s 1940s — 1950s

intogro-dif.

perators in

Hilbert space

~1967

MoM

(Harrington)

2 CM theory (re)formulated

(Harrington & Mautz)

1971 1974

Synthesis of N-port scat.

(Harrington & Mautz)

~1975

First Fortran CM code

(Matz & Harrington)

1976

Surface formulation

f CM

(Chang & Harrington)

Small antenna location synthesis

(Newman)

1979

First chapter on CM in book

(Mittra, ed.)

CM events Important milestones Ph.D. theses

1 9 8

Inagaki modes

(Inagaki & Garbacz)

1985

CM for apertures

(Harrington & Mautz)

1982

Antenna synthesis and optimization

(Liu, Garbacz, Pozar)

3

1 9 9 2000 2010

CM in book

(Van Bladel)

2007 2003

CM theory revived at UPV

(Bataller et al.)

Review paper

n CM

(Cabedo et al.)

CM on Chassis of Mobile Phones

(Schroeder & Famdie)

2005

CM for arrays, CM as basis functions

(Cabedo & Daviu & Bataller)

2004

Modal Significance

(Ethier & McNamara)

2009

CM on UAV

(Obeidat & Raines)

CM theory at CTU

(Hazdra et al.)

2014 2015 2012 2013

CM special session at APS2014 CM special session at EuCAP2015 IEEE AP-Trans. Special Issue

n CM

CMC project started

(Safin et al.)

AToM project started

(Capek et al.)

Initiative

f prof. Lau

(Lund Univ.) Geometry synthesis at single frequency

(Ethier & McNamara)

FEKO implemented CM Complete feeding synthesis

(Capek & Hazdra & Eichler)

New tracking

(Capek & Eichler & Hazdra)

CM theory at KIEL

(Manteuffel et al.)

Reconfigurable and multiport antennas

(Obeidat & Raines & Rojas)

CM theory at Ohio State Univ.

(Volakis, Rochas et al.)

Synthesis

f antenna

array

(Chen & Wang)

CM reconstruction from FF

(Safin & Manteuffel)

Form Factor Reduction

(Chen & Martens & Valkonen & Manteuffel)

CM on Rectangular Plates

(Wu & Su)

MIMO antennas

(Li & Miers & Lau)

Book on CM

(Chen & Wang)

CM special sessions at EuCAP2016 and APS 2016

now

4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 CM & MLFMA

(Chew et al.)

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 12 / 70

SLIDE 14

Characteristic mode decomposition

Are CMs nowadays a hype?

XJ = λRJ 1970 1980 1990 2000 2010 5 10 15

now number of published papers* fitted curve†

bisquare polynomial fit of 3rd order

* †

papers on CM that are utilized at CTU IEEE AP-Trans. special issue

Number of journal papers which are used at CTU to CM development.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 13 / 70

SLIDE 15

Characteristic mode decomposition

CM theory

discretization segment 20 40 60 80 100 Iz(z) [A]

3
1

1 3

2

2

char. mode #1
char. mode #2

analytical, sin(z) analytical, sin(2 z)

Comparison of CMs and sine basis.

◮ natural basis for radiating problems ◮ forms complete basis for any planar radiator ◮ ill-posed GEP (generalized eigen-value problem)

mainly because of R
some eigen-values

are negative!

◮ quite sensitive to non-symmetry of R, X

e.g. problem for Makarov’s code
A = 1

2

A + AT

, A ∈ RN×N seems useless

◮ tracking problems (discussed below)

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 14 / 70

SLIDE 16

Characteristic mode decomposition

CMs and numerical precision

◮ Characteristic modes pose ill-conditioned problem15. mode n 20 40 60 80 100

100
50

50

~167 dB (dynamical range

f double precision)

10 log10áJn,RJn

ñ

10 log10áJn,XJn

ñ

10 log10|áJn,RJn

ñ|

(negative power) negative radiated power 10 log10áJn,RJn

ñ, 10 log10áJn,XJn ñ

CM decomposition – numerical stability.

15M. Capek, P. Hazdra, and M. Masek. “On Some Theoretical and Numerical Aspects of Characteristic

Mode Decomposition”. In: (2015). url: http://arxiv.org/abs/1509.02825

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 15 / 70

SLIDE 17

Characteristic mode decomposition

CM practice

◮ alternatives

SEM modes
XJ = κJ
Inagaki modes16

◮ commercial software

FEKO
CST (2015, officially 2016)
WIPL-D (?)
CNC (in-house)

◮ challenges

potential utilization of periodic bound. cond. for CM
electrically large structures17, CM for materials
theory related to the CM
tracking
16D. Liu. “Some Relationships Between Characteristic Modes and Inagaki Modes for Use in Scattering

and Radiation Problems”. PhD thesis. The Ohio State Univ., 1986

17Q. I. Dai et al. “Multilevel Fast Multipole Algorithm for Characteristic Mode Analysis”.

In: (2014). url: http://arxiv.org/abs/1412.1756v2

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 16 / 70

SLIDE 18

About the project

About Project

1 Motivation 2 Source Concept

What is the source concept? Selected applications of the source concept

3 Characteristic mode decomposition 4 About the project 5 Project infrastructure 6 AToM architecture

AToM – Closer investigation AToM’s – Features

7 Integration into Visual CEM (ESI Group) 8 Conclusions

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 17 / 70

SLIDE 19

About the project

Tools for Synthesis of Antennas and Sensors

Project started from September 2014. ◮ We would like to develop unique tool allowing further research,

base point for attack on the problem of antenna synthesis.

◮ We would like to share our know-how with commercial partner.

Previously, our know-how in CMA offered to CST.

Main outputs: ◮ AToM (Antenna Toolbox For Matlab) (R, software), ◮ FOPS (Fast Optimization Procedures) (R, software), ◮ Characteristic mode implementation into CEM One (R, software), ◮ canonical radiators (G, functional prototype), ◮ circularly polarized antenna for GNSS (F, industrial design). RIV: 40+40+40+40+40 ◮ expected outputs are at the level of one Czech patent (200) and far below one EPS/USPTO patent (500)

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 18 / 70

SLIDE 20

About the project

Project Review and Approval

◮ application: 48 pages, appendices: 24 pages, 6 letters of interest (well-recognized persons in antennas) ◮ completed and detailed workflow

before the project has been submitted, all parts more or less existed

in working order!

Reviews (R1/R2/R2/Reporter): 80+86+86+87 (=339/400 pts, 85%) Modifications given by TACR commission: ◮ AR/ER changed from 100/0 to 50/50 (CTU/BUT) and to 90/10 (MECAS ESI s.r.o.) → higher participation needed, Criticism in reviews: ◮ major flaws about almost completely missing possible risks, ◮ detailed market analysis not included.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 19 / 70

SLIDE 21

About the project

Budget

◮ support of Technology Agency of the Czech Republic

07/2014 – 12/2017 (under project TA04010457)
approx. 600 ke(see below)

◮ participation: 20% (CTU, BUT), 65% (MECAS ESI s.r.o.)

all partners pay their own participation

◮ legal counselling completely covered by faculty from overhead costs ◮ work loads (∅): 1.8 (CTU), 3 (BUT), 2.6 (MECAS ESI s.r.o.) ◮ same salary for all team members ◮ only 2.2% of budget changed during project solution (till now) Participation 6 035 Financial support 11 141 Total budget 17 175 Claimed economic return 10 940

Financial balance-sheet (in kCZK).

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 20 / 70

SLIDE 22

About the project

Team

◮ 3 participants:

CTU in Prague, BUT, MECAS ESI (subsidiary of ESI Group)

◮ project’s staff (from left to right, no Czech diacritic marks)

Miloslav Capek, Pavel Hazdra, Petr Kadlec, Vladimir Sedenka,

Viktor Adler, Filip Kozak, Jaroslav Rymus, Milos Mazanek, Zbynek Raida

◮ students (from left to right, no Czech diacritic marks)

Martin Marek, Ondrej Kratky, Vit Losenicky, Michal Masek,

Miroslav Cupal

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 21 / 70

SLIDE 23

About the project

AToM: Antenna Toolbox For Matlab

,,Antenna source concept” – New approach to antenna design.

AToM (Antenna Toolbox For Matlab) is that part which remains the property of CTU.

Logo of the AToM project.

The main idea behind the AToM toolbox is to develop new package that will be able to: ◮ utilize the source concept features ◮ handle with data from third party software ◮ accept other codes from the community ◮ make it possible the fast-prototyping of advanced antenna designs

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 22 / 70

SLIDE 24

About the project

Project Details #1 – Propagation

◮ antennatoolbox.com

source concept18 (charact.

modes, optimization, post-processing)

white papers, news,. . .

◮ all in Matlab ◮ YouTube channel

AToM’s core is almost

complete

numerical methods are now

implemented

◮ complete graphical manual

18See also the presentations from COST VISTA meetings in Madrid and Nice. ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 23 / 70

SLIDE 25

About the project

Project Details #2 – Conception

◮ application to become Matlab Pre-product Partner19 submitted

however not yet approved. . . since MathWorks’ Antenna toolbox

◮ partially open-source code

key parts will be compiled (.p-code or .mex)
new functionality can easily be added by the users

◮ detailed documentation of all features

it is a good practice to know how the things work

19See http://www.mathworks.com/products/connections/join/prod_join.html. ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 24 / 70

SLIDE 26

About the project

Project Details #2 – Conception

◮ application to become Matlab Pre-product Partner19 submitted

however not yet approved. . . since MathWorks’ Antenna toolbox

◮ partially open-source code

key parts will be compiled (.p-code or .mex)
new functionality can easily be added by the users

◮ detailed documentation of all features

it is a good practice to know how the things work

Our conception is a trade-off between popular (and necessary)

pen-source and conventional (and necessary) commercial code.

19See http://www.mathworks.com/products/connections/join/prod_join.html. ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 24 / 70

SLIDE 27

About the project

Project Details #3 – PoI

Agreement between CTU-FEE and the authors: AToM cannot be sold without authors’ approval. ◮ 4 undergraduate (well motivated & hard-working) students ◮ 3 Ph.D. (independent & skilled) colleagues ◮ 3 post-docs (core of the team, full-job position) ◮ 4 senior researchers (project supervision, consulting) Average age of project member: 29.8 (valid data on 10/2015) ◮ 1 undergraduate student got a position in Siemens (CTU→) ◮ 2 colleagues began their doctoral studies (→CTU/BUT) ◮ 1 undergraduate is about to begin doctoral studies (→CTU) Ph.D. / (assoc.) prof. degrees are obviously not needed to get TACR!

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 25 / 70

SLIDE 28

About the project

Benefits for Department / Alma Mater

Benefits for the project team are obvious, but what about others?

Project enables ◮ concentration of know-how, ◮ new collaborations (young team in BUT, technology company in Pilsen), ◮ development of powerful software (research opportunities, licences), ◮ part-time jobs for students, ◮ better position within upcoming GACR/TACR calls for proposals, ◮ profit (maybe). . . . but, at least, not to make a dishonour to our university and department!

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 26 / 70

SLIDE 29

About the project

AToM development – Where we are now?

Mar15 Sep15 Mar16 0×104 2×104 4×104 6×104

start of AToM development

400 800 1500

Matlab mfiles functions / methods

Now June15 Dec15

number of code lines AToM – focused on well-balanced development.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 27 / 70

SLIDE 30

About the project

AToM development – Where we are now?

Mar15 Sep15 Mar16 0×104 2×104 4×104 6×104

HDF save/load rewritten start of AToM development

400 800 1500

Matlab mfiles functions / methods

Now June15 Dec15

number of code lines AToM – focused on well-balanced development.

directories 61 packages 68 classes 121 m-files 887 functions 1742 unitTests 1147 lines of code 60803 comments 6138

Valid data on 14/03/2016, 10:19AM.

◮ data analysed daily at GIT server by Jenkins

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 27 / 70

SLIDE 31

Project infrastructure

Infrastructure

◮ modern projects should take advantage of many powerful tools ◮ tenting to agile development and process automation MECAS servers ◮ FTP storage (incl. backup) ◮ web server (antennatoolbox.com) ◮ email server (user@antennatoolbox.com) ◮ (dedicated) GIT server (version control system) CTU server (Sokrates) ◮ iceScrum (SCRUM) ◮ Bugzilla (bug reports) ◮ Jenkins (continuous integration system) BUT server ◮ BigBlueButton (communication) ◮ Blade server (benchmarks) yED, Skype,. . .

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 28 / 70

SLIDE 32

Project infrastructure

Periodic Meetings (once every 2 months)

#tacr #connectingpeople :)

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 29 / 70

SLIDE 33

Project infrastructure

Project’s infrastructure – SCRUM

SCRUM – iterative and incremental agile SW development.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 30 / 70

SLIDE 34

Project infrastructure

Project’s infrastructure – iceSCRUM

Open-source iceScrum sofware for Scrum methodology.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 31 / 70

SLIDE 35

Requirements: Fast-prototyping Environment

MathWorks MATLAB logo.

◮ up to now, there is no commercial package that completely implements techniques mentioned above ◮ however, scientists develop and utilize their own codes

the codes are mainly written in

Matlab20

20The MathWorks. The Matlab.

2015. url: www.mathworks.com

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 34 / 70

SLIDE 41

AToM architecture

Why Matlab?

Pros ◮ high-definition language

excellent for

fast-prototyping

many built-in functions

are embedded

◮ new functionality can easily be published21 ◮ maybe other. . . Cons ◮ still not fast as e.g. C

and to be efficient,

Matlab needs very good programming skill

◮ not open-source ◮ to make standalone application is a nightmare ◮ maybe other. . . ◮ What is your opinion??

21www.mathworks.com/matlabcentral/fileexchange ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 35 / 70

SLIDE 42

AToM architecture

Features of Matlab (R2015b)

Did you know?

◮ Run-time Type Analysis (>Matlab 6.5)

data types in m-file are noticed during the first run

◮ Just-In-Time-Accelerator22

parts of code that satisfy certain conditions are precompiled

◮ Object-oriented programming (>R2008)

surprisingly rich OOP (all classical OOP patterns feasible)
still under development
starts to be integrated everywhere (see e.g. in graphics in >R2014a)

◮ unitTest framework (>2014b) ◮ Source Control Integration

GIT, SVN
e.g. Jenkins can be utilized

◮ profiling via profile

JIT however deactivated during the profile measurement

22See http://www.ee.columbia.edu/ marios/matlab/accel matlab.pdf. ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 36 / 70

SLIDE 43

AToM architecture

OOP & vectorization

1 % ... 2 % check whos weight is even number 3 function val = isWeightEven(objs) 4 val = mod([objs.weight],2) == 0; 5 end 6 7 % return "object" who is the oldest one... 8 function oldest obj = whoIsTheOldest(objs) 9 allAges = [objs.age]; % for acceleration purposes 10

ldest obj = objs(allAges == max(allAges));

11 end 12 13 % increase age of all objects 14 function objs = increaseAge(objs, incr age) 15 [objs.age] = indexing.listEntries([objs.age] + incr age); 16 end 17 % ... 1 % increase age (modification) − FOR approach 2 for thisObj = 1:N(thisN) 3 ppl(thisObj).increaseAge(10); 4 end 1 % increase age (modification) − vectorized approach 2 ppl.increaseAge(10);

OOP and vectorization (highest level of abstraction in Matlab).

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 37 / 70

SLIDE 44

AToM architecture

OOP: for × vectorization #1

bjects [-]

101 102 103 104 speed-up [-] 100 101 102

bjs creation
bjs manipulation
bjs comparison
bjs log. indexing

R2014b Full OOP code – comparison of for and vectorization, warp-up runs skipped, R2014b.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 38 / 70

SLIDE 45

AToM architecture

OOP: for × vectorization #2

R2015b

bjects [-]

101 102 103 104 speed-up [-]

bjs creation
bjs manipulation
bjs comparison
bjs log. indexing

100 101 102

Full OOP code – comparison of for and vectorization, warp-up runs skipped, R2015b.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 39 / 70

SLIDE 46

AToM architecture

GEP (CM): (naive) parallel solver

◮ e.g. CM cannot easily be accelerated on GPU

100 80 60 40 20 1 2 3 4 5 6 7 8 total time [s] parfor nodes

self-time total time (incl. pool alloc.) CM decomposition parallelized on CPU in Matlab.

◮ beware of Amdahl’s law23 S (p, τs) ≤ 1 τs + 1−τs

p

23T. Larsen. Parallel High Performance Computing (With Emphasis Jacket Based GPU Computing).

Aalborg University. 2011

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 40 / 70

SLIDE 47

AToM architecture

Bilinear forms: CPU × GPU

◮ Matlab R2013a, Jacket + bsxfun, GPU card: GTX580

N 5×103 10×103 speed-up GPU(N)/CPU(N) 5 10 15

speed-up GPU/CPU speed-up > 1 maximal speed-up GRAM threshold GRAM breakdown Integration of radiated power as a function of current density discretized into N segments.

Pr = 1 8πωǫ ˆ

V1

ˆ

V2

k2J (r1) · J∗ (r2) − ∇1 · J (r1) ∇2 · J∗ (r2)

sin (kR) R dV2 dV1

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 41 / 70

SLIDE 48

AToM architecture

Ways how to accelerate hardware

Requirements: Computational Resources

CPU

×

GPU

maybe FPGA in the future? ◮ advanced post-processing and

ptimization need high-performance

computers24

HPC techniques

◮ depending on the nature of the problem

CPU can be employed in parallel /

distibutive mode

GPU can be employed

◮ Matlab fully supports CPU and GPU acceleration

implicit acceleration (matrix

multiplication, fft,. . . )

24M. Capek et al. “Acceleration Techniques in Matlab for EM Community”.

In: Proceedings of the 7th European Conference on Antennas and Propagation (EUCAP). Gothenburg, Sweden, 2013

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 42 / 70

SLIDE 49

AToM architecture AToM – Closer investigation

AToM – Current state

(Relatively) actual UML scheme of AToM – Seems like a mess but it is not!

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 43 / 70

SLIDE 50

AToM architecture AToM – Closer investigation

AToM History and Workspace

◮ all actions in AToM are captured (subsref overloaded)

can be saved as m-file (and modified)
full control, can be evaluated as batch

◮ AToM has own Workspace

numerical values can be entered as variables
external function can be called to set up particular value

◮ code-GUI approach like in FEMlab25

25Former Matlab toolbox, now Comsol Multiphysics. ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 44 / 70

SLIDE 51

AToM architecture AToM – Closer investigation

AToM History and Workspace

One of videos from our YouTube channel.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 45 / 70

SLIDE 52

AToM architecture AToM – Closer investigation

Geometry and Discretization

Just two examples:

W3W4 W5W6 W1\W2 W9\W10 W7\W8 W1 W2 W3 W4 W5 W6 W7 W8 W9 W10

A set of geometrical primitives. Polygon depicting Czech Republic.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 46 / 70

SLIDE 53

AToM architecture AToM – Closer investigation

Geometry and Discretization

Just two examples:

W3W4 W5W6 W1\W2 W9\W10 W7\W8

Boolean operations made by Geom module. Equidistant grid from Mesh generator.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 46 / 70

SLIDE 54

AToM architecture AToM – Closer investigation

AToM 2D-3D EFIE MoM (Matlab)

ka [-] Zin [kW]

2
1

1 2.1 4.2 6.3 8.4 10.5

Z Î 256´256, nQuad = 4

2a a 50

P1 P2 Rin(P2) Xin(P2) Xin(P1) Rin(P1)

AToM FEKO CEM One

Preliminary results for simple structure (comparison with FEKO).

◮ multiple feeders

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 47 / 70

SLIDE 55

AToM architecture AToM – Closer investigation

AToM DesignViewer

Presentation of already generated spherical helix in AToM DesignViewer.

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 48 / 70

SLIDE 56

AToM architecture AToM – Closer investigation

Four voltage gaps & RWG junctions

2
1

1 2 3 1 2 4 3 5 ka [-] Zin [kW]

Z Î 1608´1608, nQuad = 1

Xin Rin

a a 100 AToM FEKO CEM One

Preliminary results for more complex structure (comparison with FEKO).

◮ multiple feeders, 3D surface, junctions

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 49 / 70

SLIDE 57

AToM architecture AToM – Closer investigation

Computational time – Comparison

◮ the same helix has been calculated both in 4 various simulators

same discretization
same number of frequency samples (500)
same feeding model

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 50 / 70

SLIDE 58

AToM architecture AToM – Closer investigation

Computational time – Comparison

◮ the same helix has been calculated both in 4 various simulators

same discretization
same number of frequency samples (500)
same feeding model

total time [s] FEKO26 13041 Makarov27 861 CEM One 895 AToM28 (nQuad = 1) AToM28 (nQuad = 2) . . . and the computational time of AToM?

26Parallel FEKO has been enabled.

27S. N. Makarov. Antenna and EM Modeling with Matlab.

John Wiley, 2002

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 50 / 70

SLIDE 59

AToM architecture AToM – Closer investigation

Computational time – Comparison

◮ the same helix has been calculated both in 4 various simulators

same discretization
same number of frequency samples (500)
same feeding model

total time [s] FEKO26 13041 Makarov27 861 CEM One 895 AToM28 (nQuad = 1) 805 AToM28 (nQuad = 2) 1581

HW configuration: Intel Core i7-4970 CPU@4GHz 16 GB RAM

26Parallel FEKO has been enabled.

27S. N. Makarov. Antenna and EM Modeling with Matlab.

John Wiley, 2002

28Implicit Matlab parallelization heavily utilized. ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 50 / 70

SLIDE 60

AToM architecture AToM – Closer investigation

MoM Features We Have

11 5 7 8 8 4 4 6 2 3 12 3 10 2 7 3 5 6 9 z 1 1 y 2 1 x

Benchmark for symmetrization.

◮ PEC / PMC symmetry planes29, ◮ higher-order quadrature rules30, ◮ multi-edge feeding30, ◮ adaptive integration scheme30.

29J. J. H. Wang. Generalized Moment Methods in Electromagnetics.

John Wiley, 1991

30A. F. Peterson, S. L. Ray, and R. Mittra. Computational Methods for Electromagnetics.

John Wiley – IEEE Press, 1998

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 51 / 70

SLIDE 61

AToM architecture AToM – Closer investigation

MoM Features We Are About To Have

VG

#1

VG

#2

VG

#3

VG

#4 Multiple feeders

W

PEC PEC symmetry plane

EFIE MFIE CFIE

Combined FIE

W

PMC PMC symmetry plane Periodic boundaries

W

RWG junctions Plane-wave feeding k H E Multi-edge feeding E Lumped elements

R

Scheduled features.

◮ CFIE (EFIE + MFIE)30, ◮ lumpled elements30, ◮ periodic structures30, ◮ analytic expressions31 of ∂Z/∂ω, ◮ static × dynamic part32 of Z.

31M. Cismasu and M. Gustafsson. “Antenna Bandwidth Optimization With Single Freuquency

Simulation”. In: IEEE Trans. Antennas Propag. 62.3 (2014), pp. 1304–1311

32T. L. Simpson, J. C. Logan, and J. W. Rockway. “Decomposition Of The Moment Method

Impedance Matrix Into Quasi-Static And Residual Components”. In: Proceedings of IEEE Energy and Information Technologies in the Southeast. 1989, pp. 291–295. doi: 10.1109/SECON.1989.132382

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 52 / 70

SLIDE 62

AToM architecture AToM – Closer investigation

CM decomposition

◮ Schur decomposition33 (eig) or ◮ iteratively restarted Arnoldi33 (eigs) can be used ◮ powerful tracking34 (heuristic approach)

still can be improved (Pearson formula35, far-field correlation36)

◮ modal quantities37 (W u,v

m , W u,v e

, W u,v

rad, P u,v r

, Du,v, ηu,v

rad)

all quantities, except radiated power, have cross-terms!

βu,v = Ju, EJv, E (1 + λuλv) (1 + λ2

u) (1 + λ2 v)

, A = β : Amodal (4)

33J. H. Wilkinson. The Algebraic Eigenvalue Problem.

Oxford University Press, 1988

34M. Capek et al. “A Method for Tracking Characteristic Numbers and Vectors”.

In: Progress In Electromagnetics Research B 33 (2011), pp. 115–134. doi: 10.2528/PIERB11060209

35D. J. Ludick, U. Jakobus, and M. Vogel. “A Tracking Algorithm for the Eigenvectors Calculated with

Characteristic Mode Analysis”. In: Proceedings of the 8th European Conference on Antennas and Propagation (EUCAP). 2014, pp. 629–632

36Z. Miers and B. K. Lau. “Wide Band Characteristic Mode Tracking Utilizing Far-Field Patterns”.

In: IEEE Antennas Wireless Propag. Lett. 14 (2015), pp. 1658–1661. doi: 10.1109/LAWP.2015.2417351

37M. Capek, P. Hazdra, and J. Eichler. “A Method for the Evaluation of Radiation Q Based On Modal

Approach”. In: IEEE Trans. Antennas Propag. 60.10 (2012), pp. 4556–4567. doi: 10.1109/TAP.2012.2207329

ˇ Capek et al. Tools for Synthesis of Antennas and Sensors 53 / 70