- Universitas Brawijaya

Transcription

Electrical Power, Electronics, Communications,
Controls & Informatics International Seminar
(EECCIS) 2012
Hall of Engineering Faculty, Brawijaya University
Malang, May 30-31, 2012
Proceedings
International Session
Organized by:
Department of Electrical Engineering
Brawijaya University
Indonesia
PUBLISHED BY:
Department of Electrical Engineering
Faculty of Engineering
Brawijaya University
eeccis@ub.ac.id
LAYOUT EDITOR
COORDINATOR
Wijono
MEMBERS
Angger Abdul Razak
Eka Maulana
Renie Febriyanti
Marina Dicarara
Firman Triyanto
Fahad Arwani
Erny Anugrahany
All papers in this book have been selected by the reviewers and technical committee.
All authors have signed the copyright declaration of their papers.
All rights reserved. No part of this book may be reproduced, downloaded,
disseminated, published, or transferred in any form or by any means, except with the
prior written permission of, and with express attribution to the authors.
The publisher makes no representation, express or implied, with regard to the
accuracy of the information contained in this book and cannot accept any legal
responsibility or liability for any errors that may be made.
Copyright © by Department of Electrical Engineering, Brawijaya University
2012
ii
ORGANIZING INSTITUTION
DEPARTMENT OF ELECTRICAL ENGINEERING
BRAWIJAYA UNIVERSITY
MALANG, INDONESIA
STEERING COMMITTEE
Prof. Ir. Harnen Sulistio, M.Sc., Ph.D.
Dr. Ir. Sholeh Hadi Pramono, M.S..
REVIEWER
Asc.Prof. Dr. Mamdouh (Aswan University, Egypt)
Asc. Prof. Dr. Mahrus (Aswan University, Egypt)
Dr. Corina Martineac (Rumania)
Ishtiaq R. Khan, Ph.D (Singapore)
Hazlie Muslikh, Ph.D (UM, Malaysia)
Dr. Hamzah Arouf (Malaysia)
Prof. Dr. Kaharudin Dimyati (Malaysia)
Md. Atiqur Rahman Ahad, B.Sc.,M.S.,M.S.,PhD (Bangladesh)
Prof. Adi Susanto, MSc. Ph.D (UGM, Indonesia)
Prof. Thomas Sri Widodo, DEA (UGM, Indonesia)
Prof. Dr. Ir. Arif Djunaidy, MSc (ITS, Indonesia)
Dr. Aris Triwiyatno (UNDIP, Indonesia)
Dr. Ir. Son Kuswadi (ITS, Indonesia)
Purnomo Sidi Priambodo, Ph. D (UI, Indonesia)
Dr. Ir. Muhammad Nurdin (ITB, Indonesia)
Dr.-Ing. Ir. M. Sukrisno (STEI-ITB, Indonesia)
Dr. Ferry Hadary, ST, M. Eng (UNTAN, Indonesia)
Dr. Mashury Wahab (PPET-LIPI, Indonesia)
Dr. Rini Nurhasanah, M. Sc (UB, Indonesia)
Ir. Wijono, MT. Ph.D (UB, Indonesia)
Hadi Suyono, Ph.D (UB, Indonesia)
Dr. Sholeh Hadi Pramono (UB, Indonesia)
iii
TECHNICAL PROGRAM COMMITTEE
Muhammad Ary Murti (IEEE Indonesia Section)
Kuncoro Watuwibowo (IEEE Indonesia Section)
Arief Hamdani (IEEE Indonesia Section)
Ford Lumban Gaol (IEEE Indonesia Section)
Panca Mudjiraharjo (KIT - Japan)
Onny Setyawati (Universitat Kassel - Jerman)
M. Rusli (University of Wollongong - Australia)
Sholeh Hadi Pramono (UB - Indonesia)
Agung Darmawansyah (UB - Indonesia)
M. Aziz Muslim (UB - Indonesia)
Hadi Suyono (UB - Indonesia)
Rini Nurhasanah (UB - Indonesia)
Wijono (UB - Indonesia)
iv
SEMINAR PROGRAM
WENESDAY, MAY 30, 2012
HALL OF ENGINEERING FACULTY, BRAWIJAYA UNIVERSITY
07.00 - 08.25
REGISTRATION
08.25 - 08.30
OPENING CEREMONY
08.30 - 08.45
SPEECH BY CHAIRMAN OF THE ORGANIZING COMMITTEE
08.45 - 09.10
WELCOME SPEECH BY THE DEAN OF ENGINEERING FACULTY
09.10 - 09.30
BREAK
09.30 - 10.45
1ST KEYNOTE SPEECH BY DR. IR. UNGGUL PRIYANTO, M.SC (DEPUTY
CHAIRMAN FOR TECHNOLOGY OF INFORMATION AND COMMUNICATION,
ENERGY, AND MATERIALS OF THE AGENCY FOR THE ASSESMENT AND
APPLICATION OF TECHNOLOGY)
10. 45 - 12.00
2ND KEYNOTE SPEECH BY DR. EKO FAJAR PRASETYO (FOUNDER OF
VERSATILE SILICON TECHNOLOGY, FIRST IC DESIGN COMPANY IN
INDONESIA)
“INTRODUCING SEMICONDUCTOR TECHNOLOGY AND CMOS LSI DESIGN
& FABRICATION”
12.00 - 13.00
BREAK: PRAYING AND LUNCH
DEPARTMENT OF ELECTRICAL ENGINEERING BUILDING
13.00 - 15.00
COMMISSION SEMINAR: ORAL PRESENTATION SESSION I
15.00 - 15.25
BREAK: PRAYING AND COFFEE BREAK
15.25 - 17.25
COMMISSION SEMINAR: ORAL PRESENTATION SESSION II
17.25
CLOSING
v
FOREWORD BY THE DEAN OF
FACULTY OF ENGINEERING,
BRAWIJAYA UNIVERSITY
Assalamu’alaikum warahmatullahi wabarakatuh
F
irst of all, I would like to express my acknowledgement to the whole parties,
lecturers, students, and all other people impossible to cite individually, for having
involved in the good achievement of the organization of the EECCIS 2012 Seminar.
I also would like to express my gratitude to Dr. Ir. Unggul Priyanto, M.Sc and Dr. Eko
Fajar Prasetyo for having accepted to become the keynote speakers of this EECCIS 2012
Seminar.
The EECCIS 2012 Seminar follows the success of the previously held seminars of
EECCIS 2000, 2004, 2006, 2008 and 2010. It becomes a part of scientific activity
programmes
in the Faculty of Engineering to contribute to the creation of Brawijaya
University as a research university, and furthermore as an entrepreneurial university.
As a part of the Brawijaya University, civitas academica of the Faculty of Engineering
play a very strategic and active role in producing a tight link to industry and society in
general. It is hoped that through the EECCIS 2012 Seminar the tight link could be
maintained and improved either nationally or internationally, so that the scientific culture
among the research and education institutions as well as its link-and-match to industry
could bring out the welfare of the Indonesian society, and humanity in general.
Wassalamu’alaikum warahmatullahi wabarakatuh
Dean of Faculty of Engineering
Brawijaya University,
Prof. Ir. Harnen Sulistio, M.Sc., Ph.D
vi
PREFACE
BY THE CHAIRMAN OF
THE ORGANIZING COMMITEE
Assalamu’alaikum warahmatullahi wabarakatuh
he EECCIS 2012, which stands for The
Electrical Power, Electronics,
Communications, Controls and Informatics Seminar 2012, is held following the
success of the previous EECCIS seminars organized biennially by the Department of
Electrical Engineering, Brawijaya University. The EECCIS 2012 Seminar takes place on
May 30-31, 2012 at the Faculty of Engineering Hall, Brawijaya University.
T
The EECCIS seminar is purposed to establish an interdisciplinary discussion forum in the
fields commonly covered in Electrical Engineering, i.e. Electrical Power Engineering,
Electronic Engineering, Telecommunication Engineering, Control System Engineering and
Information Technology. Despite the energy and economic crises which are still being
endured by our country, it is hoped that the hardwork of researchers from many
universities, research institutions, and also industry, could contribute to the acceleration of
our national recovery process from the crises. The academic and industry dynamics in
these efforts can be seen from their enthusiasm for participating and attending this
EECCIS 2012 Seminar.
The hardwork of our technical program committee for the success of this seminar has
been indicated by the large number of the scientific paper received. There have been
received about 189 papers coming from Indonesia, Malaysia, Japan, and Australia. After a
very rigorous process of reviewing by reviewers coming from Switzerland, Egypt,
Malaysia, Singapore, Bangladesh, and Indonesia, only about 83% of the received papers
have been accepted to be presented in a series of oral-presentation sessions during the
seminar, and also to be documented and published in the Proceedings of EECCIS 2012.
Sincere thanks go to all members of the Steering Committee and reviewers, who have
worked hard to guarantee the good quality of papers presented in this seminar.
On the part of the chairman of the Organizing Committee, I also would like to convey my
very high appreciation on the enthusiasm and hardwork shown by the whole technical
program committee, and also to many other people who are involved directly or indirectly
in contributing to the good achievement of this seminar.
Finally, I would like to thank and welcome all researchers, lecturers, students, industry,
and all other participants to the EECCIS 2012 Seminar. We admit that there are still
numerous lacks in the organization of this seminar, however any suggestions are always
welcome for our improvement in the future.
Wassalamu’alaikum warahmatullahi wabarakatuh
,
Organizing Committe of the EECCIS 2012 Seminar
Chairman,
M. Aziz Muslim, ST., MT., Ph.D
vii
TABLE
OF
CONTENT
Cover
Organizing Institution
Seminar Program
Preface by the Dean of the Faculty of Engineering
Preface by the Chairman of the Organizing Committee
Table of Content
i
iii
v
vi
vii
viii
A. ELECTRICAL POWER
[149-EEA_28] Fractional Open Voltage Maximum Power Point Tracking Using ATMega8535
For Photovoltaic System
Gunawan Wibisono, Sholeh Hadi Pramono, M. Aziz Muslim
A1
Student of Master Degree Program Lecturer of Brawijaya University
[185-EEA_31] Reducing Cogging Force in A Cage-secondary Linear Induction Motor (LIM)
by One-Side Shifting
Mochammad Rusli
Electrical Department Faculty of Engineering, Unversity of Brawijaya
A2
B. ELECTRONICS
[111-EEB_24] Automated Measurement of Haemozoin (Malarial Pigment) Area in Liver Histology Using
Image J 1.6
Dwi Ramadhani, Tur Rahardjo, and Siti Nurhayati
B1
Center for Technology of Radiation Safety and Metrology, National Nuclear Energy Agency of Indonesia
[147-EEB_32] High-Input-Range Low-Offset-Voltage Flipped Voltage Followers Using FG-MOSFETs
Zainul Abidin, Koichi Tanno, Agung Darmawansyah
Electrical Engineeering of Brawijaya University, Electrical and Electronic Engineering of University of Miyazaki
B2
[151-EEB_33] Design of 3-phase Fully-controlled Rectifier using ATMega8535
Mochammad Rif’an, ST., MT., Ir. Hari Santoso, MS., Nandan Pratama Putra, ST.
Electrical Engineering of Brawijaya University
B3
[168-EEB_35] Design of Boost Inverter for Setting Motor Induction 3 Phase
Ir. Dedid Cahya Happyanto, Agus Indra Gunawan, Bregas Wiratsongko P.
Politeknik Elektronika Negeri Surabaya
B4
C. COMMUNICATION
[034-EEC_05] Android Smartphone Based for the Local Directory Application of Public Utility
Arini, MT., Viva Arifin, MMSi., Chery Dia Putra, S.Kom.
Informatics Engineering Program State University (UIN) Syarief Hidayatullah, Jakarta
vii
C1
[062-EEC_11] Tropical rain effects on Free-Space Optical and 30 GHz wireless systems
M. Derainjafisoa and G. Hendrantoro
Department of Electrical Engineering, Institut Teknologi Sepuluh Nopember
C2
[150-EEC_27] First Aid Application Based Android Smartphone
Qurrotul Aini, Husni Teja Sukmana, and Imamul Huda
Faculty of Science and Technology Syarief Hidayatullah State Islamic University, Jakarta
C3
[172-EEC_35] Singled-fed Circularly Polarized Triangular Microstrip Antenna with Truncated Tip Using
Annular Sector Slot for Mobile Satellite Communications
Muhammad Fauzan Edy Purnomo, Sapriesty Nainy Sari
Department of Electrical Engineering, University of Brawijaya
C4
[173-EEC_36] Improvement in Performance of WLAN 802.11e Using Genetic Fuzzy
Admission Control
Setiyo Budiyanto
Electrical Engineering Department, Faculty of Engineering, Mercu Buana University
C5
[176-EEC_38] Characterization of Tilted Fiber Bragg Grating as a Sensor of Liquid Refractive Index
Eka Maulana, Sholeh Hadi Pramono, A. Yokotani
Department of Electrical Engineering, University of Brawijaya and University of Miyazaki
C6
[177-EEC_39] Video Streaming Analysis on Worldwide Network Interoperability for Microwave
Access (WiMAX) 802.16d
Dwi Fadila Kurniawan, Muhammad Fauzan E.P. dan Widya Rahma M.
Department of Electrical Engineering Faculty of Engineering UB
C7
[187-EEC_42] Statistical Propagation of Terrestrial Free-Space Optical Communication
Using Gamma-Gamma
Ucuk Drusalam, Purnomo Sidi Priambodo, Harry Sudibyo, Eko Tjipto Rahardjo
Department of Electrical Engineering, Faculty of Engineering, Universitas Indonesia
C8
D. CONTROLS
[051-EED_12] Maximum Power Point Tracking using Fuzzy Logic Control for Buck Converter in
Photovoltaic System
Mahendra Widyartono, Sholeh Hadi Pramono, M. Aziz Muslim
Student of Master Degree Program and Lecturers, Department of Electrical Engineering, Brawijaya University
D1
[093-EED_21] A Computer Fluid Dynamics Study of 6.5 Micron AA 1235 Annealing Treatment in
Sided Blow Inlet-Outlet Furnace
Ruri A. Wahyuono, Wiratno A. Asmoro, Edy Sugiantoro, Muhamad Faisal
Department of Engineering Physics, Institut Teknologi Sepuluh Nopember, Surabaya
viii
D2
[132-EED_29] The Height Control Systems of Hydraulic Jack Using Takagi Sugeno Fuzzy
Logic Controller
Fitriana Suhartati, Ahmad Fahmi
Department of Electrical Engineering University of Brawijaya, Electrical Engineering of State University of Malang
D3
[146-EED_32] An Application of Adaptive Neuro Fuzzy Inference System (ANFIS) with Substractive
Clustering for Lung Cancer Early Detection System
Mohamad Yusuf Santoso, Syamsul Arifin
Faculty of Industrial Technology, Institut Teknology Sepuluh Nopember, Surabaya
D4
[163-EED_36] PID Design for 3-Phase Induction Motor Speed Control Based on Neural Network
Levenberg Marquardt
Dedid Cahya H., Agus Indra G., Ali Husein A., Ahmad Arif A.
D5
[182-EED_39] Zelio PLC-Based Automation of Coffe Roasting Processs
M. Aziz Muslim, Goegoes Dwi N., Ali Mahkrus
Department of Electrical Engineering, Faculty of Engineering, Brawijaya University
D6
[188-EED_40] Prediction of CO and HC on Multiple Injection Diesel Engine Using Multiple Linear
Regression
Bambang Wahono, Harutoshi Ogai
Graduate School of Information, Production and Systems, Waseda University
D7
E. INFORMATICS
[048-EEE_07] Acceptance of Mobile Peyment Application in Indonesia
Hendra Pradipta
E1
Informatics Management, State Polytechnics of Malang
[057-EEE_10] Attitude Consensus of Multiple Spacecraft with Three-Axis Reaction Wheels
Harry Septanto, Bambang Riyanto Trilaksono, Arief Syaichu-Rohman and Ridanto Eko Poetro
Center of Satellite Technology, Indonesian Institute of Aeronautics and Space
School of Electrical Engineering and Informatics, Insitut Teknologi Bandung
Faculty of Mechanical and Aerospace Engineering, Insitut Teknologi Bandung
E2
[058-EEE_11] Implementing Naive Bayes Classifier and Chi Square o the Abstract to Classify
Research Publication Topics
Imam Fahrur Rozi, Rudi Ariyanto
Magister Program of Department of Electrical Engineering, University of Brawijaya, Politeknik Negeri Malang
E3
[072-EEE_16] Analysis and Implementation of Combined Triple Vigenere Cipher and ElGamal Cryptography
using Digital Image as a Cryptographic Key
Komang Rinartha, Agung Darmawansyah, Rudy Yuwono
STMIK Stikom Bali, Department of Electrical Engineering, Faculty of Engineering, Brawijaya University
ix
E4
[083-EEE_19] Heart Rate Variability Analysis on Sudden Cardiac Death Risk RR Interval by
Using Poincare Plot Method
Ponco Siwindarto, I.N.G. Wardana, M. Aris Widodo, M. Rasjad Indra
Faculty of Engineering and Medical Faculty of Brawijaya University
E5
[094-EEE_21] Lung Cancer Prediction in Imaging Test Based on Gray Level Co-occurrence Matrix
Sunngging Haryo W., Agus M. Hatta., Syamsul Arifin
Department of Engineering Physics, Faculty of Industrial Technology, Institut Teknologi Sepuluh Nopember
E6
[102-EEE_23] Optimal EDR Methods for Sleep Apnea Classification
Mungki Astiningrum, Sani M. Isa, Aniati Murni Arimuthy
E7
Faculty of Computer Science, University of Indonesia
[112-EEE_26] Automated Detection of Congested Central Vein Liver Histology of Mice
Infected with Plasmodium berghei using CellProfiller 2.0
Tur Rahardjo, Dwi Ramadhani, Siti Nurhayati
Center for Technology of Radiation Safey and Metrology, National Nuclear Energy Agency of Indonesia
E8
[156-EEE_33] Telemonitoring Application in Health Safety and Environment at PT. Pertamina
Refinery Unit IV Cilacap using Android Smartphone
Budi Santosa, Bambang Yuwono, Mariza Feary
Informatics Engineering Department, Universitas Pembangunan Nasional Babarsari Tambakbayan
E9
[169-EEE_36] RIFASKES Geographic Information System Based on Web
Istikmal, Yuliant S., Ratna M., Tody AW., Ridha MN., Kemas ML., Tengku AR.
Electro and Communication Faculty, Telkom Institute Technology
E10
[184-EEE_38] Fast and Accurate Interest Points Detection Algorithm on Brycentric Coordinates
using Fitted Quadratic Surface Combinaed with Hilbert Scanning Distance
Tibyani Tibyani, Sei-ichiro Kamata
Graduation School of Iformation , Production, and System, Waseda University
E11
[186-EEE_39] Generating Security Keys From Combination of multiple Biometric Sources
Primantara Hari Trisnawan
Camputer Sciences, University of Malaysia
E12
x
The 6th – Electrical Power, Electronics, Communications, and Informatics International Seminar 2012
May 30-31, Brawijaya University, Malang, Indonesia
Fractional Open Voltage Maximum Power Point
Tracking Using ATMega8535 for Photovoltaic
System
Gunawan Wibisono, Sholeh Hadi Pramono, M. Aziz Muslim
Student of Master Degree Program, Electrical Engineering, Engineering Faculty,
Brawijaya University,Malang, Indonesia
ndhoel@gmail.com, sholeh_hp@ub.ac.id, muh_aziz@ub.ac.id
Abstract—Photovoltaic (PV) systems are power source
systems that have non-linear current – voltage
characteristics (I-V) under different environments condition.
The system consists of PV generator (cells, modules, PV
array), energy storage (batteries), buck converter, and
resistive load. The proposed maximum power point control
is based on fractional open voltage The controller must
maintain PV voltage at VMPP by changing the converter duty
cycle so that maximum power can be generated in varying
operating condition. Using proposed maximum power point
tracking (MPPT) method, average power is 11,8% higher
than direct load, and 21.9% higher than constant output
voltage scheme. The system have better accuracy and
stability even in dynamic operating conditions.
Index Terms—Photovoltaic system, Maximum Power
Point Tracking, Fractional Open Voltage.
controlled by the duration of the switch on and off (ton and
toff). This method is known as pulse-width modulation
(PWM) switching [3].
I. INTRODUCTION
Photovoltaic systems (PV) is a system to convert the
sun's energy directly into electrical energy. PV system is
one of the renewable energy alternatives. The power of
sunlight received by the earth outside the atmosphere is
about 1300 watt/m2 [1].
Simple PV system consists of a PV cell unit, a solar
charger, and a battery. The battery is an electro-chemical
devices. Various types of batteries have different
characteristics, and different ways of charging. The battery
charger must adjust according to these characteristics so
that the battery can last long. On the other hand, the PV
cell output also has certain characteristics that may not be
suitable to charge the battery.
A solar charger should be able to bridge between the
PV cell output voltage and current is varied according to
the level of solar lighting with a battery that must be
charged with a certain voltage for optimum power transfer
and not damage the battery.
Conversion efficiency of solar energy into electric
energy via PV cells are low, only around 15-20%. One of
the effort to improve energy conversion efficiency of
photovoltaic cells is using the Maximum Power Point
Tracking (MPPT) method [2].
Figure 1. Ideal characteristic of PV cell showing MPP
Maximum Power Point Tracking (MPPT) is a subsystem designed to extract maximum power from power
source [2]. The maximum power point is shown in Figure
(1) above. In the case of solar power source, the
maximum point varies due to the influence of changes in
electrical characteristics as function of temperature, solar
iradiation, heating and others. With the change of
temperature and solar iradiation, the voltage and current
output of the PV modules are also changing and reducing
efficiency of PV systems. MPPT maximizes power output
of the panels in different conditions to detect the best
working point of the power characteristics and then
controls the current or voltage on the panel.
General requirement for MPPT is simple and low cost,
DC-DC converter is used to convert the DC input
voltage that varies into controlled DC output voltage at the fast tracking the changing conditions, and fluctuations of
desired voltage level. The basic form of DC-DC converter small output. Due to the nature of PV system is nonis buck converter, are also called step-down converter. As linear, i.e. current and voltage that varies depending on
the name implies, step-down converters produce a dc environmental conditions, it isvery important to operate
output voltage of the average lower than the input dc
voltage. In DC-DC converter, average output voltage is
A1- 1
the PV system at the condition of maximum power point.
This will improve the efficiency of PV systems.
There are various methods and ways to implement the
MPPT control, e.g., by perturb and observe, incremental
conductance, fractional open voltage, parasitic capacitance
[2].
Figure 2. Proposed system
The Buck converter used here is a simple common
Buck converter as is Figure (3)
II. FRACTIONAL OPEN VOLTAGE MPPT
The approximately linear relationship between VMPP
and VOC of the PV array, under varying irradiance and
temperature levels, is the basis for the fractional VOC
method:
VMPP ≈ k1 VOC,
(1)
where k1 is a constant dependent on the characteristics
of the PV array. However, it has to be computed
beforehand by empirically determining VMPP and VOC
for the specific PV array at different irradiance and
temperature levels. The factor k1 is usually between 0.71
and 0.78.
Using Equation (1) and measuring VOC from a noloaded PV array, VMPP can be calculated with the known
k1. The output terminals of the PV array should be
disconnected from the power converter. This results in a
temporary loss of power, which is the main drawback of
this technique. To overcome this drawback, pilot cells can
be used to measure VOC. These pilot cells should have the
same irradiation and temperature as well as the same
characteristics with the main PV array for better
approximation of the open-circuit voltage. P–N junction
diodes generate a voltage that is approximately 75% of
VOC. Thus, there is no need to measure VOC. A closedloop voltage control can be implemented after the MPPT
DC/DC converter for voltage regulation of the inverter
input.
Figure 3. Buck converter
The switch used is an IRFP 450 power MOSFET that
has VDSon about 2 V. The MOSFET gate is drived via an
AN2222 switching transistor. Switching frequency is 4
kHz to minimize audible noise and yet keeping the
switching losses low.
The ATMega8535 is chosen because it is wide
availability, Easy to be programmed, and built in ADC.
PWM signal at 25% from the microcontroller can be seen
in Figure (4).
The VOC is sampled by turning off the switch for 10 ms
so that the PV to return to its open voltage condition. The
VOC is sampled every about 5 second. Therefore energy
loss due to Voc sampling is negligible.
The PV module used is Wuhan Rixin MBF75 PV
module. Table 1 summarized specifications of the PV
module.
TABLE I.
Brand
Model
Material
Power output (max)
Voltage output (max)
Current output (max)
Open circuit voltage
Short circuit current
temperature coefficient
Working temperature
The PV array technically never operates at the MPP
since Equation 1.32 is an approximation. This
approximation can be adequate, depending on the
application of the PV system. The technique is easy to
implement and cheap because it does not require a
complicated control system; however, it is not a real
MPPT technique. Also, k1 is not valid under partial
shading conditions and it should be updated by sweeping
the PV array voltage. Thus, to use this method under
shaded conditions, the implementation becomes
complicated and incurs more power loss [2].
III. PROPOSED SYSTEM
The proposed system can be desribed as Figure (2)
below.
PV MODULE SPECIFICATION
Wuhan Rixin
MBF75
Polycrystalline Silicon
75 W
17,5 V
4,29 A
21,5 V
4,72 A
-0,35% / °C
+0,036% / °C
- 40 ~ 90°C
The control algorithm is a simple IF-THEN control. If
the input voltage is higher than VMPP then the duty cycle is
increased to increase apparent load. If the input voltage is
lower than VMPP then the duty cycle is decreased to
decrease apparent load.
IV. EXPERIMENT SETUP
PWM control pulse and output voltage ripple is
measured using digital oscilloscope.
A1- 2
The system is tested at 700 W/m2 insolation with
k1=0.7. The resistive load is varies from 40 Ω to 10 Ω with
5 Ω interval. Then from 10 Ω to 0 Ω with 1 Ω interval.
Theoretical optimum load as indicated by specification
is 4,1 Ω. Therefore we defined 40 Ω to 9 Ω as low load,
and 8 Ω to 0 Ω as high load.
The system is tested and compared with direct load (no
controller), and a constant output voltage controller.
Figure 6. output voltage ripple at 88% duty cycle.
Figure 4. Experiment setup
V. RESULT AND ANALYSIS
PWM output at 25% from the microcontroller is shown
in Figure (4) below. Output voltage is 5 volt with duty
cycle 25%.
The PWM signal is quite satisfying 5 V pulse at 4 kHz
(note that the oscilloscope gave false frequency analysis).
Figure 7. PV Voltage over resistance
PV Power
45
40
35
PV Power
30
Figure 5. PWM output at 25%
25
20
15
10
Output voltage ripple is shown in Figure (6) below.
The ripple is satisfactory low at 125 mV.
5
0
OC
The input voltage graph can be seen in Figure (7)
below. As seen from the figure, the PV voltage can be
maintain around VMPP at high load. Average voltage using
MPPT at high load is 14.99 V, while direct load gave
10.22 V, and constant output voltage gave 12.43 V. To
maintain VMPP at low load, we must use buck-boost
converter to increase apparent load. With only a buck
converter, the algorithm only work at high load.
As seen from Figure (8) below, the algorithm also
maintain power at high load. Average power using MPPT
at high load is 28,47 W, evidently higher than direct load
at 25.45 W, and constant output voltage gave 23,35 W.
40
35
30
25
20
15
10
9
8
7
6
5
4
3
2
1
0
Resistance (Ohm)
Constant output
MPPT
direct load
Figure 8. PV Power over resistance
VI. CONCLUSION
From the results and analysis, it can be concluded that
PV system using fractional open voltage can maintain
voltage at VMPP and PV power at high load. Using MPPT,
panel voltage is maintained closer to VMPP at 14.99 V. The
average power also 11,8% higher than direct load, and
21.9% higher than constant output voltage scheme.
A1- 3
REFERENCES
[1] Markvart, T dan Castaner, L. Solar Cells: Materials, Manufacture
and Operation. Elsevier Amsterdam (2005)
[2] Khaligh, A dan Onar, O.C. Energy Harvesting: Solar, Wind and
Ocean Energy Conversion System. CRC Press Florida (2010)
[3] Mohan N, Undeland T, M, and Robbins, W, P. 1995. Power
Electronics. Converters, Applications, and Design. (2nd Edition).
John Wiley & Sons, Inc.
Gunawan Wibisono Bachelor Degree from
Universitas Brawijaya, Malang, Indonesia, in
2004, in electrical engineering. Currently, he is
working toward Master Degree in power system
engineering at Brawijaya University, Malang,
Indonesia. His current research interest is solar
power system and renewable energy.
A1- 4
Sholeh Hadi Pramono received
Bachelor
Degree
from
Electritrical
Engineering
Department, Brawijaya University in 1986. He
received his Master Degree and Doctoral Degree
both from University of Indonesia, in 1995 and
2010, respectively. Since 1987 he is with
Electrical Engineering Department, Brawijaya
University. His current research interest including
fiber optics, telecommunication and renewable
energy.
M. Aziz Muslim received Bachelor Degree and
Master Degree from Electritrical Engineering
Department of Institut Teknologi Sepuluh
Nopember, Surabaya, Indonesia, in 1998 and
2001, respectively. In 2008 he received Ph.D
degree from Kyushu Institute of Technology,
Japan. Since 2000 he is with Electrical
Engineering Department, Brawijaya
Reducing Cogging Force in A Cage-secondary
Linear Induction Motor (LIM) by One-Side
Shifting
Mochammad Rusli 1) 2)
School of Mechanical, Materials and Mechatronic Engineering,
Faculty of Engineering, University of Wollongong, NSW, Australia
2)
Electrical Department Faculty of Engineering, University of Brawijaya, Indonesia
mochammad.rusli@gmail.com
1)
Abstract— High precision linear machining tools is one of
interesting research field in related to high qualitative
products which is also becoming one of competitive
factor. The movement precision can be affected by
existence of ripple force, unpredicted external load and
frictional force. The existence of cogging force is the one
limited factor of the linear precision. The reduction of
cogging force of its linear movement precision of
machining tools using rotary motor drive can be obtained
by the skewed rotor or implement the feedback control
system. Many researchers have conducted the reduction
of cogging torque of its machining tools drive supported
by using a feedback control algorithm variation concepts.
Because of the great opportunity of construction variation
in linear motors, this paper proposes to investigate an
innovation of the cage secondary Double sided linear
induction motor construction aimed to obtain the zero
cogging Force. The cogging force can be predicted by
investigated of the variation of stored energy magnetic in
the air gap. Therefore, at first in this paper the
implementation of estimation flux path in multi-tooth
model in which is built as similar construction to the cage
single sided linear induction motor, and will be verified by
building experimental multi-tooth test-bed. Based on that
multi-tooth experiment and the justification of estimation
flux path method, the double-sided linear induction motor
with offset position between both sided will be developed
with the assumption that the cogging force will be able to
cancel each other. In this paper will be described the
arrangement of LIM model using FEM software and
simulated. This motors consist of two layers, moving and
stationary part. The stationary part are arranged as the
cage-ladder structure.
completely eliminate many of the performance limiting
factors associated with rotary-linear translation
methods[2]. The most common linear motor used in
precision machine tools is the permanent magnet linear
motor, particularly in high speed applications[3].
However, permanent magnet linear motors have a
major disadvantage in precision metal cutting as the
metallic dust and swarf associated with these processes
can be attracted by the permanent magnets, which are
typically along the entire length of the axis. Therefore,
alternative linear motor technologies, such as the Linear
Induction Motor (LIM), offer great potential as a
solution for precision linear metal cutting axes.
One design aspect of linear motors that is important
from a precision machining perspective is the
minimization of cogging. Cogging is represented in
linear machines as a variation in the magnetic forces
along the machine axis, and can have a severe impact
on the overall precision of the axis. For rotary motors,
many researchers have reduced the cogging effect by
using the feedback controller design or skewed rotor of
motors. In linear induction motor, cogging effect can be
reduced by modification of its construction, because
linear motors have the great opportunities to modify the
construction forms. This paper will present the
investigation of the modification of A Cage-secondary
Double Sided Linear induction motor with cogging
effect nearly zero value.
Keywords— Cogging Forces, Electromagnetic Field,
Double sided-Linear Induction Motors
I. INTRODUCTION
The demands for high precision machining are
rapidly increasing, especially in industrial processes
such as semiconductor manufacturing or metal cutting
machine tools[1]. For machine tools in particular, the
current international competitive levels of precision are
below 1mm Linear Motors can offer significant
advantages over rotary motors for driving linear
machine tool axes, in that they either reduce or
A2-1
Figure 1: cage-secondary single-sided LIM (Photo
Courtesy of Krauss Maffei Automationstechnik GmbH,
Minimization of the cogging effect in LIMs requires
knowledge of the variation of stored magnetic energy in
the air-gap. The calculation of magnetic circuits, where
the air-gap permeance the magnetic flux and the flux
density distribution are determined, is one of the most
difficult problems in electrical machines[3]. Due to
slotted cores, many researchers approximate the air-gap
permeance in relation to magnetic energy stored in the
air-gap. This paper proposes also to describe
developing of the estimation of flux path in linear
Induction motor construction in understanding the
variation of cogging force in the air gap. The cogging
force analysis based on the prediction of variation of
stored magnetic energy in the air gap of motor. The
calculation of the cogging forces will be conducted by
using FEM approximation and EFP method. The both
method will be compared and provide the relationship
between difference relative position of the side in the
DSLIM and the opportunity for reducing cogging force.
II. PROPOSED MODEL
Figure 2 shows the common a cage-secondary
DSLIM which it will be described in this paper. It
consist of two main parts, moving part and stationary
part. The three phases AC electrical signal are
impressed into coils placed in the slots of stationary
part. The winding system based on the common
structure in rotary AC machines. The stationary part is
divided into two layer, the left side layer and right
layer. Each layer have been designed with same number
of slots, 9 slots in three poles pitch of winding.
Figure 2: proposed model of LIM
This paper proposes to describe the cogging
reduction can be obtained by one-side shifting of this
motor. The proposed model will be built in the FEMsoftware for investigation the distribution of flux
density quantity if the one of side are shifted. The flux
density and cogging force prediction will be calculated
by using the Estimation Flux Path (EFP) method. Based
on that prediction equation, the one-side shifting length
will be determined numerically.
III. COGGING FORCES
According to Arger [4] the term “cogging” can be
defined as the “variation in the motor torque as it turns
slowly”. Based on this definition, “cogging in an LIM
can also be described as the variation of
electromagnetic forces. The existence of cogging forces
can be detected by energy variation or the magnetic
energy gradient[4]. The direction of cogging forces is
perpendicular to the air gap or called as tangential
forces.
Each devices that consist of the some magnetic
circuits especially that is implemented in electrical
machines, including linear induction motors, the
interaction between magnetic material, produced by
nature – permanent magnet – or by electrical current
source – electromagnet, with the iron core will effect
generating of the attractive forces. When the rotor of
motors exhibit a movement from one position to the
other position, it can be change the direction of the
attractive force in both surfaces of materials.
The position of iron core to the magnetic materials
determines the direction of the attractive force. The
cogging force can be manifested as the projection of
the attractive force in the x-axis of movement. The
cogging force in linear induction motors can be also
referred to cogging force in permanent magnet motors.
In permanent magnet motor, cogging torque arises
from the interaction of the rotor magnets with the steel
teeth on the stator[5]. Yoshimura et al.[6] predict the
existence of cogging force associated with the
interaction between magnet end and the steel teeth of
the primary winding.
The cogging force is a function of position and
independent of the load angle. Due to the slotted nature
of the primary core, the cogging force is periodic and
repeat itself over every slot pitch[7]. Cogging torque is
produced by the interaction between permanent magnet
(PM’s) and slotted iron structure and manifests itself by
the tendency of a rotor align in a number of stable
positions even when motor not energized[8]. However
in Linear Induction Motors (LIMs), energy magnetic
variation in the air gap can be used for prediction of the
cogg8ng foces [9].
Because of the electromagnetic interaction between
the exterior teeth of the armature core and the
permanent magnets, the cogging force is inevitable in
both a short primary type and short secondary type
PMLM [10]. As in rotary PM machines, linear PM
motors can exhibit significant cogging forces due to
interaction of the permanent magnet in the stator and
with the iron in the stator[11].
Based on the above explanation, it can be concluded
that cogging forces are: (a) that effected by the
interaction between edge of certainty magnet and the
A2-2
slotting iron core; (b) a function of the position and
independent to load angle; (c) is periodic and repeat
itself over every slot pitch.
Analogy to the cogging force in permanent magnet
motors, it can be defined that the cogging forces in a
linear induction motor might be caused by the
interaction between edge of electromagnet on primary
section with the slotted iron core in secondary layer. It
should be independent to the load angle and periodic
according to slot pitch.
IV. GEOMETRIC PARAMETERS STANDARD DESIGN
This LIM-model design will be initialized by
calculating of geometric parameters of DSLIM. design
parameters of upper side of Double-sided Linear
Induction motor using standard procedures. Because the
DSLIM consist of principally two sides that have
symmetry dimensions each other. Therefore design
concept would be developed only in one side. The
upper sided design procedure are referred by standard
design of main dimension calculations and electrical
dimension. The number of slot and winding system will
be given as the first step of design.
The first important parameter in designing linear
induction motor is pole pitch. The pole pitch is defined
as the distance between slots where some three phase
windings for one pole are connected. Due to significant
influence of pole pitch to the synchronous velocity of
such as linear induction motor, thus pole pitch could be
calculated by using equation that describe relationship
between synchronous velocity and pole pitch.
v s = (2. f )τ
(1)
Where:
vs : synchronous velocity
f : three phases signal input frequency
τ : Pole pitch
'
totally number of slot ( Z1 ) and number full filled slot
( Z 1 ) . The winding system in this design are given as
double winding with 3 slots are half filled. Slot pitch
can be calculated using the equation (see fig. 3)
ωτ = ω th + S slt =
(2)
ωτ
(3)
The pole number is given that of as 2, so slot pitch
can be obtained by:
ωτ =
2 * 0.06 0.1
=
= 13.33mm ≈ 15mm
9
9
(4)
For improving the distribution of magnetic flux
density and reduce the resistance and reactance, this
linear induction motor use chorded winding system.
Based on the construction of chorded winding system,
there is parameter called coil pitch parameter. It can be
determined based on the slot pitch. Because was given
the number of slot in which consist only half filled coil,
so the coil pitch should be:
ω c = 2 *13.33 = 26.66mm ≈ 30mm
(5)
Totally, the length of primary layer can be obtained
with addition of multiplication of number pole with
pole pitch and coil pitch and end distance. The end
width of primary in this design will be defined as that
of 10 mm. Thus the length of primary layer is:
The rated thrust of small and large linear induction
motor depend on the area of primary layer. According
the previous designer, that for small linear induction
motor for rated thrust which have thrust bigger than 100
N, the ratio between rated thrust and the area of primary
layer approximately is:
Fx
= 5000 ( N / m 2 )
A
hsl
ωth
2. p.τ
Z1'
Lτ = 2 p *τ +ωc + c1 = 2*60+ 30+15=165mm (6)
If the synchronous velocity is given 6 m/s, so pole
pitch for that Linear Induction motor should be:
6
= 0.06m = 60mm
τ=
100
slot pitch are tooth width ( ωt th ) , slot width ( S slt ) ,
(7)
Then we can calculate area of primary layer:
Slt
A=
Figure 3: sketch of two slots in the moving part
The next main dimension of moving part of linear
induction motor is slot pitch. This parameter reveals the
distance between slots in moving part of LIM. The slot
pitch could be related to the dimension of slot- and
tooth- width of motor. Parameters which refereed to
Fx
100
=
= 0.017m 2 = 17000mm2 (8)
6000 6000
The area of primary layer is multiplication between
depth and width of primary layer. Because primary
width has already calculated, so The primary depth may
be obtained easily.
A2-3
Li =
A 17000
=
= 103.03 ≈ 100mm
Lτ
165
d1 = Li + 0.1*τ = 0.075+ 0.1*0.06=
0.030m = 30mm
(9)
The three phases AC current signal flowing in
primary coil generated the travelling magnetic flux in
air-gap. This magnetic flux influence or induced
voltage signal into the secondary layer of LIM.
Typically, the induced voltage in secondary layer is
approximated about a half of the rms value of input
voltage signal.
Ei ≈ 0.5 * Vi = 110Volt (rms values)
The slot high may be determined by using ratio
typical flux density normal and tangential and pole
pitch. Detail equation are:
hsl = 0.3 *τ *
The thickness of aluminum can be defined as
TABLE 1:
DESIGN RESULTS
The number of turn per phase is calculated by using
variation of electromagnetic power (EP). The AC
current signal flowing in primary coil generate the
travelling magnetic flux. The output coefficient for
primary can also determined with modification of
carter coefficient.
No.
Parameter
1
2
Pole pitch
Current
Per
phases
Primary depth
Aluminum
thickness
Back
iron
thick-ness
Slot width
Teeth width
End section
Turn number
Number of slot
of full
Height of slot
3
5
6
Pelm = m ph * E1 * I i ≈ 7.59VA
(12)
P
7.591520
σ p = elm ≈
≈15000VA/ m2
Vsc * A 5*4.6
(13)
Based on the output coefficient line current density
and the output coefficient, with assuming that
B z = 0.4T , approximately line current density can be
calculated by using f this methods)
mph 2 * Ii N1
Bz
=
32000
≈ 88900A/ m (14)
0.4
Therefore number of turn per phase is:
N1 =
88900* 0.05* 2
= 449 ≈ 450Turn
3 2 * 2.3
(15)
The width of air-gap is assumed = 0.5mm And the
width of teeth will be taken as 5.5mm
ω th = 5.5
S lt = ωτ − ωτ = 5.5mm
(19)
d 2 = 5mm .
Fx * vi
100 * 3.5
≥
≥ 8 Ampere (11)
m phV1η cosφ 3 * 220 * 0.115
Jy =
Bn
1.6
= 0.3 * 0.050 *
Bt
0.7
≈ 0.0342 ≈ 35mm
(10)
And input current in coil of primary layer:
Ii =
(18)
(16)
(17)
For single sided linear induction motor, the thickness
of back Iron can be calculated using the following
equation. However for double-sided linear induction
motor, the thickness of secondary layer can be obtained
by the subtraction the back iron thickness with air-gap
and primary high.
7
8
9
10
11
12
Symbol
Value
Unit
60
8
Mm
A
100
5
Mm
Mm
d1
30
Mm
S slt
10
10
10
951
9
Mm
Mm
Mm
35
Mm
τ
Ii
Li
d2
W th
ω end
N1
Z1
hy
V. OFF-SET POSITION OF MOVING PART LIM-MODEL
The double sided linear induction motor consist of
two parts, secondary and primary layer. The secondary
part are placed in between the double primary layers in
which the electrical current flowing into the coils
placed to their slots. The secondary layer compose of
some cages or ladders circuits that made of aluminum.
Figure 1 shows the simulated DSLIM model in the flux
software version 10.2.1 produced by Cedrat [12].
In each primary part compose of 9 slots in which AC
three phase signal current are flowed into them. The
arrangement of the three phase winding in similarity to
the rotary induction motors. The moving part is the
secondary layer and fixed part is the primary layer. The
The cage width (secondary tooth) of secondary are
specified as 6 mm (less than the tooth width of primary
– 10mm), for fulfilling the starting requirement. The
secondary and primary slot are defined with the same
values, each is 10 mm.
The measurement of cogging action in this model are
conducted by using the multi-static option. The cage-
A2-4
secondary are made in open circuit condition, so that
there is no electrical current flowing in them. After the
coils are supplied by electrical current, the primary
layer are moved in 1 mm step in positive direction (up
direction). Because of there is no current flowing in the
ladder, in the secondary occur the electromagnetic
forces (attractive and repel forces).
The
electromagnetic force that have the tangential direction
in this model the magnetic forces varies up and down
called cogging action.
Figure 6: Norma Flux density in the left and right air-gap
TABLE 2:
COMPARISON BETWEEN REC. MODEL AND OFFSET 12 MM
t-width
(mm)
4
5
6
7
8
9
10
11
12
13
14
15
Figure 4: the cogging forces by the offset position variation
The investigation for variation of offset position
between left and right side of model have been
conducted. By using the finite element method, the
simulation results show that the offset-position of both
sides could provide the minimum cogging forces.
Figure 4 shows that the minimization of cogging can be
obtained in the 8 mm offset model.
Figure 5 shows the tangential flux densities in the left
and the right side of the air gap of LIM model. The
tangential flux density in both sides could be cancelled
each other. The normal flux densities also looks
symmetry, so the reduction of electromagnetic forces in
the tooth will be minimized.
thrust
No offset
72.84
77.32
83.53
88.94
93.43
96.5
101.63
104.67
107.25
113.64
119.58
124.87
Thrust
12_ofset
65.10
69.11
74.66
79.49
83.51
86.25
90.84
93.55
101.92
101.57
106.88
111.61
The designed LIM-model have simulated by using
Cedrat Flux Software. By the slot pitch variation from 4
until 15 mm, The investigation of the useful thrust
between no offset model and 12mm offset-model
shown that provide the slight difference between them.
Table 2 show the comparison results of shifted one side
of LIM-model in 12 mm length
Figure 7 shows the distribution magnetic field in all
of regions of the 8 mm ffset model. The circle of
magnetic field between both sides are similar, therefore
the useful thrust in the both air gap have a similar
direction. the 8mm-offset model. In this paper will be
shown the cogging forces investigation results of that
model, if the right side of model would be shifted up in
step 1 mm. Table 3 shows simulation results of useful
thrust of offset model.
The offset position length variation in the simulations
results influences the useful thrust significantly
compared with the no offset-model.
Figure 5: Tangential Flux density in the left and right air-gap
A2-5
TABLE 3:
COMPARISON THRUST FOR OFFSET VARIATION
TABLE 4:
PERCENT OF COGGING OVER THRUST FOR 12MM AND 9 MM OFFSET
MODEL
speed
ofset 5
ofset 6
ofset 7
ofset 8
m/s
Thrust(N)
Thrust(N)
Thrust(N)
Thrust(N)
offset
cogging
cogging
thrust
thrust
percent
percent
(mm)
12 mm
9 mm
12mm
9 mm
12 mm
9 mm
4
8.29
8.16
89.65
84.53
9.24707
9.65338
5
9.09
7.92
91.01
85.03
9.98791
9.31436
6
0
0
0
0
5.75
15.34
15.44
19.08
19.42
5.5
34.65
34.8
34.24
34.58
5.25
43.66
43.78
43.86
44.2
5
53.7
53.75
53.82
54.16
6
9.5
9.28
91.09
85.12
10.4292
10.9023
4.75
63.02
63.17
63.23
63.57
7
9.55
6.52
91.17
85.23
10.4749
7.64989
4.5
71.63
71.75
71.78
72.12
8
9.75
8.6
91.41
85.54
10.6662
10.0538
4.25
77.45
77.67
77.76
78.1
4
83.52
83.7
83.85
84.19
3.75
87.35
87.32
87.43
87.77
3.5
89.55
89.77
89.82
90.16
3.25
90.67
90.86
90.92
91.26
3
91.01
91.09
91.17
91.51
2.75
90.21
90.4
90.4
90.74
2.5
88.56
88.99
89.04
89.38
2.25
86.43
86.54
86.65
86.99
2
84.66
84.71
84.88
85.22
1.75
83.07
83.18
83.23
83.57
79.21
1.5
78.23
78.39
78.87
1.25
76.33
76.45
76.56
76.9
1
72.01
72.06
72.23
72.57
0.75
69.23
69.37
69.54
69.88
0.5
65.44
65.67
65.88
66.22
0.25
69.45
69.57
69.65
69.99
0
57.13
57.24
58.33
58.67
VI. CONCLUSION
The cage secondary LIM model with double layer
moving part might generate the high useful thrust and
also would used for reducing the cogging force. The
reducing cogging would be developed by compensation
way in which the one side will be up-shifted in order to
the flux magnetic in both sides would be cancelled each
other. By using finite element, the reducing cogging
forces on this model could be forced down until under
10% compared the useful thrust values Although the
offset-model can reduced cogging forces, however the
useful thrust value are in 12% lower than no-offset
LIM-model.
REFERENCES
Table 3 sows that the offset variation from 5 until 8
mm could only provide the useful thrust much smaller
compared the no-offset model. The offset in 12 mm, the
model could be made bigger thrust ( see table 2).
However the cogging force by the 12mm offset length
is still higher than the other model. Therefore the
cogging force will be investigated only in the 9 mm and
12 mm offset. Table 4 shows that cogging forces on
both offset model can be reduced into under 10%
compared the useful thrust.
Isovalues Results
Quantity : Equi flux Weber
Time (s.) : 49.5E-3 Pos (mm): 73.313
Line /Value
1 / -670.48 61E-6
2 / -656.08 422E-6
3 / -641.68 143E-6
4 / -627.27 864E-6
5 / -612.87 58E-6
6 / -598.47 301E-6
7 / -584.07 022E-6
8 / -569.66 743E-6
9 / -555.26 465E-6
10 / -5 40.86186 E-6
11 / -5 26.45901 E-6
12 / -5 12.05623 E-6
13 / -4 97.65344 E-6
14 / -4 83.25065 E-6
15 / -4 68.84787 E-6
16 / -4 54.44505 E-6
17 / -4 40.04226 E-6
18 / -4 25.63947 E-6
19 / -4 11.23666 E-6
20 / -3 96.83387 E-6
21 / -3 82.43108 E-6
22 / -3 68.02827 E-6
23 / -3 53.62548 E-6
24 / -3 39.22269 E-6
25 / -3 24.81988 E-6
26 / -3 10.41709 E-6
27 / -2 96.0143E-6
28 / -2 81.61149 E-6
29 / -2 67.2087E-6
30 / -2 52.80591 E-6
31 / -2 38.40311 E-6
32 / -2 24.00031 E-6
33 / -2 09.59752 E-6
34 / -1 95.19472 E-6
35 / -1 80.79192 E-6
36 / -1 66.38913 E-6
37 / -1 51.98633 E-6
38 / -1 37.58353 E-6
39 / -1 23.18074 E-6
40 / -1 08.77794 E-6
41 / -9 4.37514E-6
42 / -7 9.97235E-6
43 / -6 5.56955E-6
44 / -5 1.16675E-6
45 / -3 6.76396E-6
46 / -2 2.36116E-6
47 / -7 .95836E-6
48 / 6.4444 3E-6
49 / 20.847 23E-6
50 / 35.250 02E-6
51 / 49.652 82E-6
52 / 64.055 62E-6
53 / 78.458 42E-6
54 / 92.861 21E-6
55 / 107.26 401E-6
56 / 121.66 68E-6
57 / 136.06 96E-6
58 / 150.47 241E-6
59 / 164.87 519E-6
60 / 179.27 799E-6
61 / 193.68 078E-6
62 / 208.08 358E-6
63 / 222.48 639E-6
64 / 236.88 917E-6
65 / 251.29 199E-6
66 / 265.69 478E-6
67 / 280.09 756E-6
68 / 294.50 035E-6
69 / 308.90 317E-6
70 / 323.30 595E-6
71 / 337.70 874E-6
72 / 352.11 156E-6
73 / 366.51 434E-6
74 / 380.91 713E-6
75 / 395.31 995E-6
76 / 409.72 274E-6
77 / 424.12 552E-6
78 / 438.52 834E-6
79 / 452.93 113E-6
80 / 467.33 391E-6
81 / 481.73 673E-6
82 / 496.13 952E-6
83 / 510.54 23E-6
84 / 524.94 509E-6
85 / 539.34 788E-6
86 / 553.75 072E-6
87 / 568.15 351E-6
88 / 582.55 63E-6
89 / 596.95 909E-6
90 / 611.36 187E-6
91 / 625.76 466E-6
92 / 640.16 751E-6
93 / 654.57 029E-6
94 / 668.97 308E-6
95 / 683.37 587E-6
96 / 697.77 865E-6
97 / 712.18 144E-6
98 / 726.58 429E-6
99 / 740.98 707E-6
100 / 7 55.38986 E-6
101 / 7 69.79265 E-6
[1] Kok Kiong Tan, Hui Fang Doi, Yang quan Chen and Tong
Heng Lee, High Precision Linear Motor Control Via Relay-Tuning
and Iterative Learning Based on Zero-Phase Filtering, IEEE
Transactions on Control Systems Technology, Vol. 9, No. 2,pp244253, March, 2001.
[2] S. B. Yoon, J. Hur, and D.S. Hyun, A Method of Design of
Single-Sided Linear Induction Motor for Transit,
IEEE
Transactions on Magnetics, Vol. 33, No. 5, September 1997
[3] Gieras, J. F., Linear Induction Drives, Clarendon Press, Oxford,
1994.
[4] Philip L. Arger, Induction Machines, Gordon and Breach
Science Publishers, New York, 1961.
[5]
Moscorop, J., Commins, P., Cook, C., Torque Perturbations
and Dynamic Stiffness of Linear Motors for Grinding Machines,
University of Wollongong, Australia, 2003.
[6]
Lee, B.J., Koo, D H, Cho Y H, Investigation of Secondary
Conductor type of linear Induction Motor Using the Finite Element
Method, Proceeding of IEEE the 2008 International Conference on
Electrical Machines, 2008.
[7] G. Brandenburg, S. Brueckl, J. Dormann, J. Heinzl, C. Schmidt,
Comparative Investigation of Rotary and Linear Motor Feed Drive
Systems for High Precision Machine Tools}, AMC2000 - IEEE,
Nagoya, 2000.
[8] Sang-Moon Hwang, Geun-Bae Hwang, Weui-Bong Jeong and
Yoong-Ho Jung, Cogging Torque and Noise reduction in Permanent
Magnet Motors by Teeth Pairing, IEEE, 2000.
[9] Rusli, M., Moscrop J.,Cook C. and Platt D., An Analytical
Method for Predicting Cogging Forces in Linear Induction Motors,
Proceeding of 8th LDIA conference, Netherland, 3-6 july 2011.
[10] Sung Whan Youn, Jong Jin Lee, Hee Sung Yoon, and Chang
Seop Koh, A New Cogging-free Permanent-magnet Linear Motor,
IEEE Transactions on Magnetics, Vol. 44, No. 7, July, 2008.
[11] Xu, W., Zhu, J., Tan, L., Guo, Y., Wang, S., Wang, Y.,
Optimal Design of a Linear Induction Motor Applied in
Transportation, IEEE, 2009.
Figure 7 : proposed 8mm-off-set model
A2-6
[12] __,Cedrat Inc., CAD Package for electromagnetic and Thermal
Analysis using Finite Elements, CEDRAT Copyright, July 2007.
[13]
Kumin, L, Stumberger, G., Dolinar, D., and
Jezernik, K., Modeling and Control Design of a Linear Induction
Motor, ISIE’99, IEEE, Bled, Slovenia, 1999.
[14]
Zhu, Z.Q., Xia, Z.P., Howe, D., Mellor, P.H.,
Reduction of Cogging force in slotless linear permanent magnet
motors, IEE Proc.-electr. Power Appl., Vol 144, 4 July 1997.
[15]
E. R. Laithwaite, and S.A. Nasar, Linear-Motion Electrical
Machines, Proceedings of IEEE, Vol. 58, No. 4., April 1970.
[16]
B. T. Ooi, A Generalized Machine Theory of Linear
Induction Motor, Presentation paper on PES Winter Meeting, New
York, 1973.
[17]
Cruise, R.J., and Landy, C. F., Reduction of Cogging Forces
in Linear Syncronous Motors, IEEE, 1999.
[18]
Liu, J., Lin, F., Yang, Z., Zheng, T. Q., Field Oriented
Control of Linear Induction Motor Considering Attraction Force &
End-Effects, IEEE, 2006.
[19] Philip L. Arger, Induction Machines, Gordon and Breach
Science Publishers, New York, 1961.
[20] S. Nonaka, and T. Higuchi, Elements of Linear Induction
Motor Design, IEEE Transactions on Magnetic, Vol. MAG-23, No. 5,
September 1987.
ACKNOWLEDGMENT
The author would like to thank to the Directorate
general
of high education of the Cultural and
educational ministry of Republic of Indonesia in related
to financial supporting for completing this research
project and the School of Mechanical, Materials and
Mechatronic Engineering, Faculty of Engineering,
University of Wollongong which has provided the
research facilities and finite element software and The
University of Brawijaya which has encouraged for
completing this research project.
Mochammad Rusli was Completing the
under graduate at the Institut technology
Sepuluh Nopember Surabaya on 1986, and
Master program for Dipl.-Ing. At the
Techniche
Universitaet
Braunschweig
Germany on year 1996. Now he is as PhD
postgraduate student at the University of
Wollongong Australia with research interest
reducing cogging effects by improving design
procedure and implement control strategies in linear induction motor.
A2-7
High-Input-Range Low-Offset-Voltage Flipped
Voltage Followers Using FG-MOSFETs
Zainul Abidin1), Koichi Tanno2), Agung Darmawansyah1)
1) Electrical Engineering of Brawijaya University
2) Electrical and Electronic Engineering of University of Miyazaki
zainulelektro@gmail.com, tanno@cc.miyazaki-u.ac.jp, agungdarmawansyah@yahoo.com
Abstract— In this paper, high-input-range and
low-offset-voltage flipped voltage followers are presented.
The conventional voltage follower has two disadvantages;
narrow input range and offset voltage. In this paper, these
problems are solved. The former problem is modified by
adding two MOSFETs in the diode-connecting path. The
latter problem is modified by using floating-gate MOSFETs
(FG-MOSFETs). Using this device, the threshold voltage can
be eliminated. The circuits are simulated in a 0.25 µm CMOS
process. Simulation results demonstrate that the
conventional circuit has narrow input range (0.46V) and
offset voltage (0.7V). The enhancement MOSFETs scheme
has 48.9% higher input range and offset voltage (0.7V). The
FG-MOSFET and doubler scheme has 42.5% wider input
range with the gain ≈ 1 and low offset voltage (<0.01V). It
can realize the actual condition of MOSFET implementation
because of the capacitors.
Index Terms— Voltage Follower, Flipped Voltage
Follower, High Input Range, Offset Voltage, Analog Circuits
I. INTRODUCTION
Analog building blocks using CMOS technology are
the key components in mixed-mode digital and analog
LSI’s1), 2). In these building blocks, the transmission of
signals is extremely important. In order to make this
transmission be possible, a conventional CMOS voltage
follower which is called Flipped Voltage Follower was
proposed3). The circuit is very simple and consists of three
enhancement-mode MOS transistors and one current
mirror. Furthermore, it is very useful for heavy load drive
without increasing power consumption. Actually, the
voltage follower is used to build Operational
Transconductance Amplifier (OTA)4). However, it has the
disadvantage of narrow input range and offset voltage.
These problems restrict application to various analog
circuits.
In this paper, some voltage followers are proposed.
The proposed circuits are based on the voltage follower
presented in Reference 3). Only two MOS transistors are
added to support input range enlargement. Ideal Vout and
Iout also will be given. Next, the offset voltage can be
decreased by using depletion MOSFETs or FG-MOSFETs
schemes. The performances of the proposed voltage
followers are characterized through HSPICE simulations.
In this paper, the simulation results are reported in detail.
Fig. 1 Conventional Voltage Follower.
II. FLIPPED VOLTAGE FOLLOWER
Fig. 1 shows the conventional CMOS voltage
follower3). The circuit operation can be described as
follows.
M3 operates as a current source, which is controlled
by the bias voltage (VB). Because the drain-to-source
current of M1 (Ids1) is equal to Ids3 through the current
mirror consisting of M4 and M5, the gate-to-source voltage
of M1 (Vgs1) is equal to Vgs3. Therefore, the output voltage
(Vo) is determined by the input voltage (Vin) and VB and it is
independent of the load. As a result, the circuit can drive
not only capacitive load but also a resistive load. Next, Vout
will be derived using Ids-Vgs characteristics of an MOS
transistor.
Assuming the back gate is connected to the source of
M1, both M1 and M3 operate in the saturation region, and
that the transconductance parameters of M1 and M3 are
same, that is to say, the channel width W and the channel
length L of M1 are equal to those of M3. Then, Ids1 and Ids3
are given by6)
B1-1
= (
= (
=
−
− )
−0− )
(1)
(2)
(3)
where K is the transconductance parameter proportional to
W/L, µ is carrier effective mobility, COX is capacitance of
the oxide layer per unit area, VT is the threshold voltage of
the NMOS transistor. Because Ids1 is equal to Ids3 by the
current mirror, we obtain the following equation.
=
(4)
Using (1)-(4), Vout is given by
=
−
(5)
Vout is insensitive to device parameters in the fabrication
process because Vout has no dependence on VT and K.
Next, the role of M2 is discussed. From Fig. 1, Ids2 is
given by
=
+
(6)
where IRL is the current through the load resistance RL.
The added path from the drain of M1 to gate of M2 makes
M1-M2 behave like a diode-connected single device, and
hence Vp is varied by Ids2, which is determined by Vin and
RL. The current of Ids1 + IRL flows through M2 under the
condition that Vo is larger than 0, i.e., M2 operates as the
current absorber. However, this circuit has a disadvantage
that the input range of Fig. 1 is narrow. Because Vp is
restricted by Vgs of M2. Moreover, this voltage follower has
the offset voltage of VB from Eq. (5). These problems
restrict the circuit application.
III. PROPOSED CIRCUITS
In this section, the problems of input range and offset
voltage are solved. There are some modifications based on
the conventional circuit. The proposed circuits are
enhancement
scheme,
FG-MOSFETs
scheme,
FG-MOSFETs and doubler scheme.
A. Enhancement MOSFETs scheme
The enhancement scheme is shown in Fig. 2. The
circuit operates similar with the conventional circuit.
However, gate of M2 in the proposed circuit is not directly
connected to the drain of M1. In the connection line, there
are two transistors (M6 and M7). M2 and M6 are combined
as current mirror circuit. It makes Ids6 can be controlled by
W of M6 and M2 (under the condition of current mirror). If
those W and L of M7 are similar with those M6, Ids6 will be
equal to Ids7. It means that Vgs6 also similar with Vgs7. The
gate of M7 is directly connected to Vp. It means that Vp
depends on Vgs7. In order to make M1 and M4 to be in
saturation region, Vp must be adjusted. Therefore, Vp can
be optimized not only by M2 but also M6 and M7. In this
way, the addition of two MOSFETs greatly expands the
possibility voltage follower design with the high input
range.
Next, offset voltage will be decreased. Offset voltage
makes the input range start from more than 0. Because of
that condition, the proposed circuit will be modified by
using and floating-gate MOSFETs (FG-MOSFETs). This
kind of transistor will decreases the offset voltage by
replacing some transistors.
B. FG-MOSFETs scheme
Fig. 3(a) shows the basic structure of the
FG-MOSFET proposed by Shibata and Ohmi as a
functional MOS transistor featuring a gate-level weighted
sum and threshold operation7). It consists of an n-channel
MOS transistor with a floating gate first poly layer) over
the channel and, in some cases, extends to the field-oxide
area. Multiple input gates are formed by the second poly
layer over the floating gate. The capacitive coupling
between the multiple input gates and the floating gate is
shown in Fig. 3(b). C0 shown in Figs. 3(a) and 3(b) is the
capacitance between the floating gate and the substrate.
Fig. 3(c) is the symbolic representation of a FG-MOSFET.
Now, in the case of a k-input FG-MOSFET, capacitances
between the multiple input gates and the floating gate are
defined as C1, C2,…,Ck, in order, from the drain side as
shown in Fig. 3(c). (Fig. 3(c) is an example of the case of k
= 2). Fig. 3(d) shows an example of the physical layout of
an FG-MOSFET.
When the floating gate to source voltage (Vfs) is
larger than the threshold voltage of the FG-MOSFET, as
seen from the floating gate (VT), the drain to source voltage
(Vds) is larger than Vfs − VT and the initial charge of the
floating-gate equals 0, the FG-MOSFET operates in the
saturation region. Ids of the k-input FG-MOSFET under the
saturation region is
=
Fig. 2 Proposed Circuit.
B1-2
−
−
! (7)
May 30-31,
31, Brawijaya University, Malang, Indonesia
#$
where Ci is the capacitance between the floating gate and
i-th input gate, C0 is the oxide capacitance between the
floating gate and the substrate,
=
%&'()* +, .%' +'/
%&'()*
&'() .%'
(8)
Fig. 5 (b) shows detail circuit of FG-MOSFET
FG
M3. Vfg3 can
be defined as follow.
%
+
= % &'()-.%,
#$
&'()-
(9)
-
Because Vfs1=Vfs3 by the current mirror and all
capacitances have same value, substitution Eq (7)-(9) will
obtain Vout as follow.
+'/
=
(10)
VDD
M5
Ids7
M4
Ids1
Vq
M7
Vp
M1
Vin
Vout
IRL
RL
Ids2
Ids6
Ids3
VB
M3
M2
M6
Fig. 4 FG-MOSFET
MOSFETs Scheme.
Fig. 3 FG-MOSFET (a) An illustration of the cross-sectional
cross
structure of
an FG-MOSFET with two inputs, (b) capacitive model of the
FG-MOSFET, (c) symbolic representation of the FG-MOSFET
and (d) physical layout of the FG-MOSFET.
MOSFET.
Vs is the source voltage of the FG-MOSFET
MOSFET and VT is the
threshold voltage of the FG-MOSFET,
MOSFET, as seen from the
floating-gate. In Eq. (7), K and Wi are defined and referred
to a transconductance parameter and capacitive weight,
respectively5).
Fig. 4 shows the modification circuit. In the circuit,
M1 and M3 are FG-MOSFETs
MOSFETs instead of enhancement
MOSFETs. M1 is connected to VB and Vin. M3 is connected
to VB and ground. Fig. 5 (a) shows detail circuit of
FG-MOSFET M1. If C0 is much smaller than the other
weight capacitances, Vfg1 can be defined as follow.
(a)
VB
Cbias1
C1
Vfg3
fg
M3
(b)
Fig. 5 (a) Detail of Transistor M1, (b) Detail of Transistor M3
B1-3
C. FG-MOSFETs and Doubler scheme
Based on the theoretical analysis, the output voltage
of FG-MOSFETs scheme is not equal to input voltage
(gain < 1). It means that the circuit did not work properly
as voltage follower. In this case, voltage doubler is
required to make the Vin become twice (see Fig. 6).
IV. SIMULATION RESULTS
The circuits are simulated in a 0.25 µm CMOS
process by using HSPICE. The simulations compare the
conventional voltage follower and the proposed one based
on single supply voltage. M6 and M7 are the differences of
both circuits. According to offset voltage, simulations also
show improvement of the proposed circuit. The simulation
results are demonstrated by the input range and offset
voltage.
A. Evaluation of the proposed high-input-range voltage
follower
The simulations of conventional and proposed circuit
are based on the condition shown in Table 1. The
simulation results are shown in Fig. 9. This figure shows
Vin-Vout characteristic. It is also compared to the ideal
Vin-Vout characteristic that calculated by using (5). This
figure also shows that the input range of the proposed
voltage follower is 0.90V and the conventional one is
0.46V. It means that the input range of the proposed circuit
(Fig. 2) is 48.9% higher than that of the conventional one.
Fig. 6 Voltage Doubler and The Voltage Follower Shown in Fig. 4
Table 1. Simulation Condition for Input Range
Fig. 1
Fig. 7 Voltage Doubler
Fig. 7 shows one of the realization circuit of the voltage
doubler. Because Ids of M11 becomes equal to Ids of M10 by
current mirror (M8 and M9), then Vgs of M10 becomes equal
to Vfs of M11. Therefore, the output voltage of the circuit
can be given by Vout=2Vin under the condition that C0 is
much smaller than the other weight capacitances. However
this voltage doubler has narrow input range. The design of
the voltage doubler suitable for the proposed voltage
follower is the future work. FG-MOSFETs and doubler
scheme is shown in Fig. 8
Fig. 2
Rl
VDD
10kΩ
3.0 V
10kΩ
3.0 V
Vin
1.5 V
1.5V
VB
0.7 V
0.7V
W /L of M1
2µm/5µm
2µm/5µm
W /L of M2
50µm/5µm
50µm/5µm
W /L of M3
2µm/5µm
2µm/5µm
W /L of M4
6µm/5µm
6µm/5µm
W /L of M5
6µm/5µm
6µm/5µm
W /L of M6
-
50µm/5µm
W /L of M7
-
50µm/5µm
Fig. 8 FG-MOSFETs and Doubler Scheme
Fig. 9 Simulation Results Figs. 1 and 2 (Input Range)
B1-4
B. Evaluation of the voltage follower using the
FG-MOSFETs
In order to decrease the offset voltage, the
FG-MOSFETs scheme (Fig. 4) will be implemented. The
simulation is based on condition in Table 2.
The simulation result of FG-MOSFETs scheme
without the voltage doubler is shown in Fig. 10. It shows
the output voltage is almost similar with the theoretical one
and the gain is ≈ 0.5. From this figure, we can find that the
offset voltage can be eliminated. The simulated offset
voltage is less than 0.001V. The input range of the voltage
follower is 1.68V. It is 140.6% higher than the
conventional one.
The simulation of FG-MOSFETs and doubler
scheme is based on condition in Table 3. Fig. 11 shows the
input range of the voltage follower is 0.8V. It is 42.5%
higher than the conventional one. Fig. 12 shows the output
voltage is almost similar with the input voltage and the
gain is ≈ 1. From this figure, we can find that the offset
voltage can be eliminated.
V. CONCLUSIONS
In
this
paper,
high-input-range
and
low-offset-voltage flipped voltage followers are presented.
These followers are evaluated through HSPICE. The
design and analysis results can be summarized as follow:
1) The conventional circuit has narrow input range
(0.46V) and offset voltage (0.7V).
2) The enhancement MOSFETs scheme has 48.9%
higher input range and offset voltage (0.7V).
3) The FG-MOSFET and doubler scheme has 42.5%
higher input range and low offset voltage (<0.01V). It
can realize the actual condition of MOSFET
implementation because of the capacitors.
In the future, we try to design new circuit using back
gate control technique and actually implement the
proposed voltage followers.
Table 2. Simulation Condition for Offset Voltage
Fig.4
Table 3. Simulation Condition for Offset Voltage
Rl
Fig.13
100 kΩ
VDD
3.0 V
Vin
0.7 V
2V
Rl
VDD
100kΩ
3.0 V
VB
W /L of M1
20µm/5µm
Vin
0.2V
W /L of M2
80µm/5µm
VB
1.5V
W /L of M3
20µm/5µm
W /L of M1
20µm/5µm
W /L of M4
50µm/5µm
W /L of M2
80µm/5µm
W /L of M5
50µm/5µm
W /L of M3
20µm/5µm
W /L of M6
80µm/5µm
W /L of M4
50µm/5µm
W /L of M7
80µm/5µm
W /L of M5
50µm/5µm
W /L of M8
50µm/5µm
W /L of M6
80µm/5µm
W /L of M9
50µm/5µm
80µm/5µm
W /L of M10
20µm/5µm
W /L of M11
20µm/5µm
W /L of M7
Fig. 10 Simulation Result of Fig. 4
Fig. 11 Simulation Result of Fig. 8 (Input Range)
B1-5
authors would like to acknowledge anonymous reviewers
for valuable comments.
REFERENCES
Fig. 12 Simulation Result of Fig. 8 (Offset Voltage)
ACKNOWLEDGMENT
This work is supported by VLSI Design and Education
Center (VDEC), the University of Tokyo in collaboration
with Synopsis, Inc. and Cadence Design Systems, Inc. The
[8]
B1-6
[1] S. C. Fan, R. Gregorian, G.C. Temes, and M. Zomorrodi, “Switched
capacitor filters using unit-gain buffers,” in Proc. IEEE Int. Symp.
Circuit Syst., 1980, pp. 334-337.
[2] A. Hyogo and K. Sekine, “SC immitance simulation circuits using
UGB’s and their applications to filters,” IEICE Trans., vol. J-72A, no.
3, pp. 535–540, 1989.
[3] K. Tanno, H. Matsumoto, O. Ishizuka, and Z. Tang, “Simple CMOS
Voltage Follower with Resistive-Load Drivability,” IEEE
Transactions on Circuits and Systems –II: Analog and Digital Signal
Processing, vol.46, No.2, February 1999.
[4] R. G. Carvajal et al., “The Flipped Voltage Follower: A Useful Cell
for Low-Voltage Low-Power Circuit Design,” IEEE Transactions on
Circuits and Systems –I: Regular Papers, Vol. 52, No. 7, July 2005.
[5] K. Tanno, O. Ishizuka, and Z. Tang, “A 1-V, 1-Vp-p Input Range,
Four Quadrant Analog Multiplier Using Neuron Transistors,” IEICE
Trans. Electron., vol.E82-C, No.5, May 1999.
[6] R. Gregorian and G. C. Temes, Analog MOS Integrated Circuits for
Signal Processing. New York, NY: Wiley, 1986.
[7] T. Shibata and T. Ohmi, “A functional MOS transistor featuring
gate-level weighted sum and threshold operation,” IEEE Trans.
Electron Devices, vol.39, no.6, pp.1444–1455, June 1992.
Automated Measurement of Haemozoin
(Malarial Pigment) Area in Liver Histology
Using Image J 1.6
1
Dwi Ramadhani1 , Tur Rahardjo1, and Siti Nurhayati1
Center for Technology of Radiation Safety and Metrology,
National Nuclear Energy Agency of Indonesia
dhani02@batan.go.id
Abstract— Common histopathological changes in the
liver due to Plasmodium infection is the presence of
haemozoin (malarial pigment) in liver histology section.
Identification of haemozoin generally done manually under
microscope. Measurement of haemozoin area rarely done
because it is quite difficult to separate the haemozoin area
from other element in liver histology. Identification and
measurement haemozoin area can be done by image
analysis using ImageJ. ImageJ is a public domain Java image
processing program that enables a plugin development.
Plugins are small Java modules for extending the
functionality of ImageJ by using a simple standardized
interface. Aim of this research is developed ImageJ plugin
to measure the haemozoin area in liver histology. Totally
60 random liver histology images were analyzed using our
plugin. Algorithm of plugin contain several sequential
stages, such as splitting channels, thresholding the image
for detection haemozoin area in blue channel and measure
haemozoin area. Average haemozoin area from 60 images
defined with our plugin was 3884.5 µm2. Our plugin
succeeded in detecting and measuring the haemozoin area
in liver images at approximately 3.91 seconds.
Keywords.: Haemozoin, ImageJ,
Malaria, Plasmodium berghei, Plugin
Liver
Histology
I. INTRODUCTION
Malaria is considered as one of the most important
infectious diseases in the worldwide. It affects 350 to
500 million people and causes more than one million
deaths every year. Malaria is caused by protozoan
parasites belongs to the genus Plasmodium, which are
transmitted by blood-feeding Anopheline mosquitoes.
The disease is characterized by a range of clinical
features from asymptomatic infection to a fatal disease
[1,2].
Malarial involvement of liver is now a known entity
with specific histopathological changes. The commonly
histopathological changes in the liver due to
Plasmodium infection is the present of haemozoin [3].
Haemozoin or malaria pigment has a history in the
scientific literature older than the malaria parasite itself,
having first been described in the early 18th century by
the noted Italian physician Lancisi [4]. Eventually, this
pigment played a role in the discovery of the parasite and
the elucidation of its life cycle [1,5]. Hemozoin is a
polymer of heme produced by the parasite during
hemoglobin breakdown inside the host red blood cells
(RBC). Red blood cells lysis during infection results in
release of merozoites with this pigment, which are
phagocytized by circulating monocytes, neutrophils and
resident macrophages [6,7]. The amount of haemozoin
in tissues increases throughout infection, so the greater
amount of pigment, greater degree of chronicity of lesion
[1].
Liver histology is congested with a brown or black
pigmentation as a result of accumulation of haemozoin
[3]. Haemozoin identification in liver histology
commonly does manually under the microscope.
Measurement of haemozoin in liver histology can be
done by measuring the brown area. Measurement of
haemozoin can be done by image analysis using ImageJ.
ImageJ is a public domain Java image processing
program inspired by NIH Image for the Macintosh. It
runs on any computer with a Java 1.1 or later virtual
machine, either as an online applet or as a downloadable
application. ImageJ has a large number of native
functions supplemented by an ever increasing number of
“plugins” (optional extras needing installation). A plugin
is a file (named *.class) which needs to be in the
“plugins” sub-folder of the ImageJ folder, otherwise
ImageJ will not load it [8].
Aim of this research is to build plugin for ImageJ that
can be use for measure the haemozoin area in liver
histology of laboratory mice that already infected with
Plasmodium berghei. The advantages of using
laboratory mice as a model for malaria include a well
studied immune system of the host, the opportunity to
assess pathologic changes at all stages of the disease, and
the availability of genetic variants [1].
II. MATERIALS AND METHODS
2.1. Mice
Male Swiss mice age 8 to 12 weeks was purchased
from Pusat Penelitian dan Pengembangan Gizi dan
Makanan, Kementerian Kesehatan Indonesia.
2.2. Parasites and infections
Mice were inoculated intraperitoneally with 106
B2-1
erythrocytes infected by P. berghei. Mice were subjected
to euthanasia at one week after inoculation. Fragments of
the liver were fixed by immersion in 10% buffered
formalin during 24 hours. These samples were then
dehydrated, and processed for paraffin embedding. Five
µm sections were cut and stained with hematoxylin-eosin
(H&E).
because with this value all the haemozoin area can be
convert to black area in a binary image. Commonly 1
value will showed as a black color in a binary image.
After thresholding process we selected the black area in
binary image as a region of interest (ROI) using
CreateSelection command so the black area can be
measure using Measure command
2.3. Image acquisition
A Nikon Biophot microscope attached with Nikon
D3000 digital single lens reflects (DSLR) camera system
was used to capture images of the smears. The slides
were examined under 40× objective lens. Images were
captured at a resolution of 1936×1296 and saved as
JPEG files.
2.4.3. Measuring haemozoin area
To measure the haemozoin area, we used the Measure
command in ImageJ Analyze menu. Measure command
will calculates and displays area statistics, line lengths
and angles, or point coordinates the ROI. ROI defined as
a black area in binary image (Fig 2).
2.4. Image analysis
A plugin for measuring the haemozoin area in liver
histology was developed. Totally 60 random liver
histology images were analyzed using our plugin in
ImageJ 1.60. The algorithm of plugin can be divided into
the following four sequential stages (Fig 1): (1) Splitting
channels, (2) Detecting haemozoin area in blue channel,
(3) Measuring haemozoin area, (4) Showing outlining
haemozoin area in images, (5) Detecting total tissue area
in green channel, and (6) Measuring tissue area.
2.4.1. Splitting channels
The purpose of this method is the separation of the
red, green and blue channels of the RGB image.
Haemozoin area is easy to identify in blue channel
compared to red and green channels. In blue channel, the
haemozoin area color is dark and the other component is
light. Splitting channel also used for converting the RGB
image to monochrome image for thresholding process.
After splitting the channel we look at each channel
individually to determine which one of the channel
creates better contrast than another. The channel
containing the highest contrast is the best one to choose
for use for thresholding later on [9].
2.4.2. Detecting haemozoin area in blue channel
This method is performed by thresholding the image
and making the binary image with ImageJ. Thresholding
is quick method to identify areas of an image to include
and areas of an image to ignore.With sufficient contrast,
objects of interest may then be “detected,” resulting in
masking binary image components, where each pixel is
either “on” or “off” [9].
Thresholding an image is a special type of quantization
that separates the pixel values in two classes, depending
upon a given threshold value ath. The threshold function
ƒthreshold (a) maps all pixels to one of two fixed intensity
values a0 or a1; i.e.,
with 0 < ath < amax . A common application is binarizing
an intensity image with the values a0= 0 and a1 = 1 [10].
In this case we used 160 as a threshold value (ath),
2.4.4. Showing outlining haemozoin area in images
To show the outlining haemozoin area in original
image, we used Add Image function in ImageJ Overlay
menu.
2.4.5. Detecting total tissue area in green channel
This method is performed by detecting the total tissue
area by threshold the image and making the binary image
with ImageJ. Different with the haemozoin area, total
tissue area is strongly easy to determine in green channel.
We used 180 as a threshold value (ath), because with this
value total tissue area can be convert to black area in a
binary image. After that we selected the total black area
using CreateSelection command.
2.4.6. Measuring total tissue area
The purpose of this method is measuring the total
tissue area using Measure function in Analyze menu in
ImageJ. Detail script and flowchart of the plugin is show
in Fig 1 and 2.
Fig 1. Script Haemozoin Analysis Plugin
III. RESULTS
Average haemozoin area from 60 images defined with
our plugin are 3884,5 µm2, graphics of haemozoin area
in 60 images are show in Fig 3. Our plugin success
detected and measured the haemozoin area in a liver
histology in approximately 3, 91 seconds.
B2-2
IV. DISCUSSION
Images that comprise light objects on dark
background or dark object on a light background can be
In this plugin we choose to use splitting channels than
segmented by threshold operation. Based on that we
ColorDeconvolution plugin. We used splitting channels
segmented and detected the haemozoin area in blue
because ColorDeconvolution plugin failing to produce
channel by thresholding the image using ImageJ. ImageJ
an image that haemozoin area easy to identify. Color
automatically make binary image and then convert to
Deconvolution plugin commonly use for stain separation
mask after we apply thershold technique. The results is
in histological images. This plugin assumes images
image that divides into objects in black color and
generated by color subtraction (i.e. light-absorbing dyes
background in white color.
such as those used in bright field histology or ink on
To get measurement area result in µm not in pixel first
printed paper). Our experiment showed that in the blue
the scale of the image must be set using Set Scale
channel after we apply splitting channels process, a
command in Analyze menu. A known distance should be
haemozoin is easy to identify because only the
measured by fitting a line to the known distance using the
haemozoin area are coloring in dark and the other
straight line selection tool in the ImageJ toolbar. Then
component showed in light color.
open the Set Scale command, which will automatically
register the distance from the straight line selection.
Enter the Known Distance and the Unit of Length and
after selecting Global and then OK, the scale will
Original Image
automatically be calculated from the registered distance
[11]. A known distances we defined by capture a
micrometer slide in under 40× objective lens. With the
micrometer image then we define a scale using Set Scale
Substract Background,
command.
Rolling Ball Radius : 50 pixels
We also apply backround subtraction using rolling
ball algorithm before we splitting channels to do
background illumination correction in the images.
Background correction can be applied while acquiring
Split Channels
into Blue, Green images (a priori) or after acquisition (a posteriori). The
and Red Channels difference between these is that a priori correction uses
additional images obtained at the time of image capture
while in a posteriori correction, the additional images are
Threshold Green
not available and therefore an ideal illumination model
and Blue Channels
has to assumed. Substract background using rolling ball
algorithm is one of the a posteriori correction methods.
Substract background function is removes smooth
Create Selection to
continuous backgrounds from images. Based on the a
Measured the haemozoin
“rolling ball” algorithm described in Stanley Sternberg's
and total tissue area
article, “Biomedical Image Processing”, IEEE
Computer, January 1983.
Overall our plugin success measured the haemozoin
area in liver histology images, and the time need for
Show outline of
haemozoin area
analyze one images is approximately 3.91 seconds.
Other research that conductes by Silva et al [1] also
measured haemozoin area in liver histology images using
ImageJ, unfortunatelly the details process is not
Fig 2. Flowchart of haemozoin area plugin
explained so we can not compared with our methods.
12000
V. CONCLUSION
Haemozoin Area
10000
We have developed ImageJ plugin that can be used
measured the haemozoin area in liver histology images
of mice infected with Plasmodium berghei. Time need
for analyze one images is approximately 3. 91 seconds
using our plugin. Overall, our plugin worked very well to
measured the haemozoin area in liver histology images.
8000
6000
4000
2000
0
1
4
7
10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58
REFERENCES
Images Number
[1]
Fig 3. Chart of haemozoin area in 60 images
B2-3
Silva, A.P.C., Rodrigues, S.C.O., Merlo, F.A., Paixão, T.A., and
Santos, R.L. Acute and chronic histopathologic changes in wild
[2]
[3]
[4]
[5]
[6]
[7]
type or TLR-2-/-, TLR-4-/-, TLR-6-/-, TLR-9-/-, CD14-/-, and
MyD-88-/- mice experimentally infected with Plasmodium
chabaudi. Braz J Vet Pathol, 2011, 4(1), 5-12.
WMR UNICEF, World Malaria Report. Technical Report,
WMR and UNICEF, 2005
Baheti, R., Laddha, P., and Gehlot, R.S. Liver Involvement in
Falciparum Malaria – A Histo-pathological Analysis. JIACM
CM, 2003; 4(1): 34-8
ADACHI, K., TSUTSUI, H., KASHIWAMURA, S., SEKI, E.,
NAKANO, H., TAKEUCHI, O., TAKEDA, K., OKUMURA,
K., VAN KAER, L., OKAMURA, H., AKIRA, S., NAKANISHI,
K. Plasmodium berghei infection in mice induces liver injury by
an IL-12 and Toll-like receptor/myeloid differentiation factor
88-dependent mechanism. J. Immunol. Res., 2001,
167,5928–34.
ANDRADE JR HF., CORBETT CEP., LAURENTI MD.,
DUARTE MIS. Comparative and sequential histophatology of
Plasmodium chabaudi – infected BALB/C mice. Braz. J. Med.
Biol. Res, 1991, 24: 1209–18.
EGAN, T.J. Haemozoin (malaria pigment): a unique crystalline
drug target. Targets, 2003, 2(3).
[8]
SULLIVAN, A.D., and MESHNICK, S.R., Haemozoin:
Identification and Quantification. Parasitology Today, 1996,
12(4).
[9] Collins, T.J., ImageJ for microscopy. BioTechniques, 2007,
43:S25-S30.
[10] SYSKO, L.R., and DAVIS, M.A., From Image to Data Using
Common Image-Processing Techniques. Current Protocols in
Cytometry, 2010, 12.21.1-12.21.17.
[11] BURGER, W., and BURGE, M.J., Digital Image Processing An
Algorithmic Introduction using Java (1st Edition). 2008:
XX+565.
[12] Papadopulos, F., Spinelli, M., Valente, S., Foroni, L., Orrico, C.,
and Alviano, F., Common Tasks in Microscopic and
Ultrastructural Image Analysis Using ImageJ. Ultrastructural
Pathology, 2007, 31:401–407.
B2-4
Design of 3-phase Fully-controlled Rectifier
using ATmega8535
Mochammad Rif’an, ST.,MT., Ir. Hari Santoso, MS., Nanda Pratama Putra, ST.
Electrical Engineering of Brawijaya University
rifan@ub.ac.id
A fully-controlled three-phase full-wave rectifier is
shown in Fig. 2.
Abstract - Fully-controlled three-phase full-wave
rectifier typically uses SCR as an active component. This
paper will discuss the design and testing of rectifier by
using SCR and microcontroller ATMEGA8535[1]. The test
results show that a series of triggers can work well, which
is indicated from the observed values whose magnitudes
are all in accordance with theory. Under load resistive
output, the rectifier shows a zero value for the point of
ignition greater than 120 º , whereas when the load is
inductive, the mode of inverter can be applied following
the nature of the inductive load.
Figure 2.Fully-controlled three-phase full-wave rectifier[5]
I. INTRODUCTION
There are two main parts in the software designing
process. The first part is used for zero-crossing detector,
which will detect the zero condition of each phase using
the flowchart as shown in Figure 3.
Rectification is the process of converting current
from the alternating form (AC) into direct current
before being supplied to load[2]. Fully-controlled
three-phase full-wave rectifier can produce a directcurrent (DC) voltage which can be controlled. It is
done by adjusting the conduction intervals of each
SCR. Because the SCR can block voltage in both
directions, then it may be possible to reverse the
polarity of the output DC voltage and then give
power back to the AC supply from the DC side. In
some conditions, this converter operates in the
"inverter mode"[3]. SCR in the converter circuit is
commutated with the help of the supply voltage
during rectification mode of operation, and it is
known as "line commutated converter". Using the
same circuit, when it is operating in the inverter
mode, on the load side the commutation requires
emf, so that it is called "load commutated
inverters"[4].
Start
INT Pin low?
No
No
Yes
INT Pin
low checking > 5
Yes
Pin_output=high
II. METHODOLOGY
Pin_output=low
The outline of a rectifier system can be
described in a block diagram as shown in Figure 1.
Finish
Figure 3. Flowchart of zero-crossing detector
The second part is the main software programs used
on the microcontroller. The main program flow diagram
is shown in Figure 4.
Figure 1 System Block Diagram
B3-1
the phase sequence RST and RTS respectively shown in
Fig. 6 and Fig. 7.
START
ADC input signal (firing angle)
Detection of the condition of zero phase
R, S, and T
Show the point of
ignition through
the LCD
Timer OFF
- Define phase sequence
- Calculate the period of each wave phase
- Convert the ignition angle to be a constant time
- Perform calculations to determine the timing
sequence of the active 6 Ports as an SCR trigger
signal
Figure 6. The output signal S1 of microcontroller (input is
connected in the RST phase-sequence, with ignition angle 0º)
Timer ON
Activate the 6 Ports with the time sequence in
accordance with the conditions of phase sequence
Figure
4. Flowchart of main program
III. RESULTS AND DISCUSSION
3.1 Testing and Analysis of Zero-Cross Detector circuit
The purpose of testing the zero-cross detector circuit
is to determine whether the circuit is functioning well as
a series of 3-phase voltage sensor used to detect the zero
state of each phase that will be given to the
microcontroller . The results of the zero-cross detector
output circuit are shown in Figure 5.
Figure 7. The output signal S1 of microcontroller (input is
connected in the RTS phase-sequence, with ignition angle 0º)
Based on the test results, it can be seen that the signal
S2 on both RST and RTS phase-sequences shows the
most appropriate time, which is in accordance with
theory, i.e. starting from the angle of 90º to 210º for the
RST phase–sequence condition and from 330º to 90º for
the RTS phase-sequence condition.
3.3 Testing and Analysis of the Overall System
The purpose of testing the whole system is to
determine whether the Six-Pulse Control Unit system
can work well to trigger the gate of the 6 SCRs at the
time the three-phase full-wave controlled rectifier is
used to supply the load of resistive and inductive
characteristics with ignition at angles of 0 º to 150 º. The
block diagram of the entire system testing can be seen in
Fig. 8.
Figure 5. The output signal which shows the Interrupt Occurrence
on Edge Up with Reference to Phase R
The testing result shows that an interrupt has occurred
on the rising edge of the output signal, indicating that
the Zero Cross Detector[6] circuit can detect the state of
zero as expected during the design.
3.2 Testing and Analysis of Main Program on
Microcontroller
Testing of the main program was conducted to
determine whether the output signal is generated at the
phase sequence RST or RTS, and at different angles
according to the desired ignition time, so that it will
properly trigger the SCR. The output of the
microcontroller to signal S2 with ignition angle 45º with
B3-2
The test results after the addition of a low-pass filter
at the zero-cross detector is shown in Figs. 11, 12, 13
and 14.
Figure 11. Output voltage under resistive load at an ignition angle
of 30º
Figure 8. Block diagram of the entire system testing
The results of the tests with the ignition angle of 30º
can be seen in Fig. 9.
Figure 12. Output voltage under resistive load at an ignition angle of
90º
Figure 9. Resistive output voltage under load at 30º ignition angle
Based on the testing results, it can be seen from Fig.
9 that the output voltage of the rectifier is not in
accordance with the theory, because of the appearance
of noise at the output which was caused by SCR
switching. Consequently, the zero-cross detector cannot
detect the state of zero because of this noise. A
capacitor needs to be added in order to form a low pass
filter[7], as shown in Fig. 10.
Figure 13. Output voltage when loaded with inductive load with ωL/R
= 0.314 at an ignition angle of 30º
Figure 14. Output voltage when loaded with inductive load with ωL/R
= 0.314 at an ignition angle of 90º
The results of the measured quantities can be seen in
Table I and II.
Figure 10. Additional capacitors ( in the dashed box) to form a lowpass filter of order one on the zero-cross detector
B3-3
TABLE I. DATA MEASUREMENT RESULTS WITH R=200 Ω
LOAD AT DIFFERENT FIRING ANGLES α
Firing Angle
parameters
Measured
0º
30º
45º
60º
90º
120º
135º
150º
UdAV (V)
110
95
78
56
16
0,05
0,06
0,06
UdRMS (V)
110
95
80
62
25,5
0,4
0,55
0,50
IdAV(A)
0,55
0,45
0,34
0,27
0,067
0,001
0,001
0,001
IdRMS (A)
0,55
0,45
0,36
0,29
0,115
0,004
0,003
0,003
ITAV (A)
0,16
0,14
0,11
0,08
0,015
0,005
0,005
0,005
Is (A)
0,40
0,35
0,29
0,23
0,09
0,005
0,005
0,005
60º
90º
120º
135º
150º
UdAV (V)
110
94
79
57
11
0,1
0,1
110
96
81
63
27,5
1,6
2
1,7
IdAV(A)
0,50
0,44
0,36
0,255
0,045
0,001
0,001
0,001
IdRMS (A)
0,50
0,44
0,36
0,26
0,06
0,003
0,003
0,003
ITAV (A)
0,15
0,13
0,105 0,075
0,007
0,005
0,005
0,005
0,34
0,28
0,05
0,005
0,005
0,005
Is (A)
0,39
0,20
0,1
TABLE III. COMPARISON OF MEASUREMENT RESULTS TO
CALCULATION RESULTS UNDER LOAD R=200 Ω AT
VARIOUS IGNITION α
UdAV (V)
P
T
110 109,9
95
95,2
78
77,7
56
54,9
16
14,7
0,05
0
0,06
0
0,06
0
UdRMS (V)
P
T
110
110
95
96,8
80
81,4
62
62,3
25,5 23,9
0,4
0
0,55
0
0,50
0
Parameters measured
IdAV (A)
IdRMS (V)
P
T
P
T
0,55 0,55 0,55 0,55
0,45 0,47 0,45 0,48
0,34 0,39 0,36 0,41
0,27 0,27 0,29 0,31
0,067 0,07 0,115 0,12
0,001
0
0,004
0
0,001
0
0,003
0
0,001
0
0,003
0
ITAV (V)
Is (A)
P
T
P
T
0,16 0,183 0,40 0,45
0,14 0,159 0,35 0,39
0,11 0,129 0,29 0,33
0,08 0,092 0,23 0,25
0,015 0,025 0,09 0,09
0,005
0
0,005
0
0,005
0
0,005
0
0,005
0
0,005
0
Parameters measured
IdAV (A)
IdRMS (V)
ITAV (V)
P
T
P
T
P
T
0,50 0,55 0,50 0,55 0,15 0,183
0,44 0,47 0,44 0,48 0,13 0,159
0,36 0,39 0,36 0,40 0,105 0,129
0,25
0,12
0
0
0
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
TABLE IV. COMPARISON OF MEASUREMENT RESULTS TO
CALCULATION RESULTS AT R=200 Ω LOAD IN SERIES WITH
l=200 mH AT VARIOUS IGNITION α
UdRMS (V)
P
T
110
110
96
96,8
81
81,4
0,075 0,092 0,20
0,007 0,018 0,05
0,005
0
0,005
0,005
0
0,005
0,005
0
0,005
Based on the test results, can be summed up some
of the following:
In testing a gate trigger circuit blocks can generate
signals in accordance with that required to trigger the
SCR at a full three-phase controlled rectifier full wave.
But on testing the whole system when used directly to
trigger the SCR at rectifier under load resistive or
inductive, the circuit triggers may not work as well.
After the addition of a low pass filter components at the
zero cross detector, trigger circuit can work well so that
the output of the rectifier close to the value
corresponding to the theory.
Based on Table III, it can be seen that under resistive
load each parameter measured has almost the same
value as the calculation result. It also shows that an
increasing firing angle will be accompanied with
decreasing Is. The values of UdRMS, UdAV, IdRMS, and
IdAV will become smaller when the firing angle becomes
higher, and its value will be zero for the firing angle of
120º, 135º, and 150º. The existence of non-zero value
during the laboratory experiment using the firing angles
of 120º, 135º, and 150º, was caused by the fact that the
meter did not indicate the o value, even though it was
not used to measure anything. Figs. 11 and 12 shows
that each value of the output voltage ignition angle
tested was in accordance with the theory discussed
previously. The confomity to the theory will ensure that
the Six-pulse Control Unit can be used successfully to
provide triggers to the SCR gate at a three-phase fullwave controlled rectifier during resistive loaded.
UdAV (V)
P
T
110 109,9
94
95,2
79
77,7
0,30
0,16
0
0
0
IV. CONCLUSIONS
P=Experiment; T=Theory.
Angl
e
α
0º
30º
45º
63
62,3 0,25 0,27 0,26
27,5 25,18 0,045 0,055 0,06
1,6
0
0,001
0
0,003
2,0
0
0,001
0
0,003
1,7
0
0,001
0
0,003
Based on Table IV, it can be seen that under the
inductive load condition the results of experiment were
almost the same as the results of theoretical
calculations, at which time ωL/R = 0.314, both showed
a decreasing Is when firing angle getting bigger. The
values of UdRMS, UdAV, IdRMS, and IdAV will get smaller
when the firing angle getting bigger and its value will be
zero for the ignition angles of 120º, 135º, and 150º. The
existence of non-zero value during the laboratory
experiment using the firing angles of 120 º, 135º, and
150º was caused by the fact that the RMS-meter did not
indicate the 0 value, even when it was not used to
measure anything. When loaded inductively, inverter
mode can be applied, and the value of ωL/R is very
large. From Figure 13 and 14 it can be seen that each
value of the output voltage ignition angle tested was in
accordance with the theory discussed previously. The
confomity to the theory will ensure that the Six-pulse
Control Unit can be used successfully to provide
triggers to the SCR gate at a three-phase full-wave
controlled rectifier when loaded inductively.
The obtained measurment data were then compared
to the calculation of the theory, as shown in Table III
and IV.
Angl
e
α
0º
30º
45º
60º
90º
120º
135º
150º
55
11
0
0
0
P=Experiment; T=Theory
TABLE II. DATA MEASUREMENT RESULTS WITH R=200 Ω
AND L=200 mH LOAD AT DIFFERENT FIRING ANGLES α
Firing Angle
parameters
Measured
0º
30º
45º
60º
90º
120º
135º
150º
UdRMS (V)
57
11
0,1
0,1
0,1
[7]
Is (A)
P
T
0,39 0,45
0,34 0,39
0,28 0,33
B3-4
Atmel. 2006. 8-bit AVR with 8K Bytes In-System
Programmable Flash ATMega8535, ATMega8535. San Jose:
Atmel.
Mohan, Ned. 1989. Power Electronics Converters,
Applications, and Design Second Edition. John Willey & Sons,
Inc. New York.
Sen, P. C.1989. Principles Of Electric Machines and Power
Electronics. John Willey & Sons, Inc. New York.
Skvarenina, Timothy L. 2002. The Power Electronics
Handbook. Boca Raton, Florida : CRC Press LLC.
Shepherd, William. 2004. Power Converter Circuits. New York
: Marcel Dekker.
Atmel. 2003. AVR 182 : Zero Cross Detector. San Jose:
Atmel.
Niewiadomski, Stefan. 1989. Filter Handbook A Practical
Design Guide. Great Britain : Courier International Ltd,
Tiptree, Essex.
Design Of Boost Inverter for Setting Motor
Induction 3 Phase
1
1
Dedid Cahya Happyanto, 2Agus Indra Gunawan, 3Bregas Wiratsongko P
dedid@eepis-its.edu, 2agus_ig@eepis-its.edu,3bregasprasetyo@gmail.com
Abstract— At the end of this project created an electric
car that aims to reduce levels of air pollution, minimize
noise and saves fuel. This electric car uses 3-phase
induction motor movement is powered by Aki, Aki
connected to the output of the boost converter circuit. DC
output voltage is then converted into AC voltage using a
three-phase inverter. Inverter output voltage is used to
distribute three-phase induction motors. Three-phase
induction motor speed is regulated by the by applying the
method of V/f constant. Changes in induction motor speed
can be done in 2 ways, by setting the boost converter
output voltage and by adjusting the frequency of the
inverter 3 phase. Changes in each of the voltage and
frequency is set by the microcontroller. Based on the test
results obtained from the boost converter output voltage
can be set is of 48 Vdc and-220Vdc constant output current
of 2 amperes, the output frequency of the inverter can also
be set as the voltage changes from 0 - 50Hz. Thus
producing an electric motor which mengguanakan
induction motor drive with an adjustable speed. So the
electric motor can operate without causing any impact negative impact on the environment and society.
Index Terms— Boost Converter and Three Phase
Inverter.
I. INTRODUCTION
I
n the era of sophisticated globlalisasi, transportation
has become an indispensable need for humans to
perform daily activities - day. One of the commonly
used means of transportation is a motorcycle and a car.
With so much pollution and noise generated from flue
gas. On the other hand the use of the fuel used for
automobiles are relatively quickly exhausted in a
relatively short period of time anyway. One way that can
be used to overcome these problems is to replace the
combustion engine with an electric engine or in other
words turn your car into an electric car. Electrical
machines used are 3-phase induction motor because it
has several advantages such as:
a. The structure of 3-phase induction motors are
b. lighter (20% - 40%) than the DC current to power the
motors the same.
c. Unit price of 3 phase induction motor is relatively
cheaper.
d. 3-phase induction motor maintenance more efficient.
To support the operation of three phase induction motor
is required Aki (alteratif fuel source) and some power
electronic circuits such as boost converter and inverter 3
phase. With electric cars expected to reduce air pollution
levels, minimize noise, and saves fuel. So that the
electric motors used to operate without impact - negative
impact on the environment and society.
II. LITERATURE STUDY
A. Basic Theory
Boost Inverter is a tool to raise the low voltage DC to AC
high. Boost inverter consists of a combination of a boost
converter and inverter which can be used for 1 phase or 3
phase. at the end of this project boost inverter is used to
raise the voltage of 48 volts DC from the battery or the
battery to 220 Volt Inverter AC.Pada Boost has a system
of cooperation with the boost converter that raises the
voltage by using the inductor, but the difference in the
boost converter and boost inverter is the result output is
issued. In the boost converter outputs a DC voltage
output sinal is great but on the contrary, to boost the
output inverter of the voltage generated by the output
signal of the AC signal. At the end of this project the
boost inverter is made using a method of increasing the
voltage.
B. Method
The method that created the boost inverter is increased,
or in other words the voltage step-up. Intention to raise
the voltage here is to raise the voltage from 48 to 220
Volt 3 system must pass through the boost converter is
mounted, wherein the first boost designed a system of 48
Volts to 90 Volts DC, where the boost to the second
designed to enter the second boost the output of the boost
the first so when the boost output voltage of the first
issue was accepted by the second boost as well as enter
the boost of 3 to produce output 220 volts. When it
reaches the output 220 Volt DC keluaran dari boost
konverter ini masuk ke inverter untuk di rubah menjadi
keluaran sinyal AC.
III. SISTEM PLAN
In the design of these systems boost inverter
made consisting of:
B4-1
a.
b.
c.
d.
e.
f.
IV. SIMULATION RESULT
Boost konverter
Inverter
Optocoupler
Minimum System
Accu/power supply
Motor Induksi
This equipment will arise from a system design where
the system can be seen in schematic or block diagram
below:
Boost
1,2 dan
3
aki
Mikro 1
Fig2. Boost converter and the inverter circuit is simulated
inverter
motor
The following is a schematic in the simulation, where a
combination of boost konverter converterter.
Mikro 2
Fig1. Block diagram of boost inverter system
This block diagram describes a system where the tools
are made from 48 volts 48 volts DC into 220 volts AC
voltage with an increase in boost converter.
The working principle of this boost, when inserting the
battery voltage is 48 volts then the voltage will go to the
first boost, the first boost process that will increase the
voltage to 90 volts DC, when the boost voltage is the
voltage of the first issue will be used as the input voltage
boost 2 to issue 170 Volt DC, after the first boost and
boost both interact with each other then 3 will receive a
boost in output from the boost to the two to be used as
input and issued a 220 Volt DC. After the first boost to
boost to the three mutual interaction and generate a
voltage of 220 volts, the output of the Boost will be
converted into an AC signal by entering at the inverter
output. After all interact with each other then the output
will be in put in 3 phase induction motors.
Fig3. Results Boost Output Signal Converter 3 when given a 3
phase inverter.
The following is the output of the simulation result of the
last series between the boost converter and inverter.
Which of the insert 170 volts DC will generate a voltage
of about 300 volts AC.
Fig4. The Boost converter output series combination Boost converter
1,2 3 and inverter
The following is the output of the simulation results of
the boost converter circuit on the third step, which of the
insert 170 volts DC will generate a voltage of about 300
volts DC.
Search value L, R dan C :
Effisiensi =
a. Duty Cycle
Vout = (
)
b. Value Resistor
iL =
Fig5. Boost converter and the inverter circuit as a whole overall the
simulation results
The following is a series combination of a boost
converter 1,2 and 3, and join together in 3 phase inverter.
c. Value Capasitor
R=
d. Value Induktor
Lmin =
Fig6. overall circuit simulation results
B4-2
The following are the result of simulations in which the
output end 300 volts AC.
V. ANALYSIS
Of this system which when given a boost converter input
is expected to be approximately 170 tha give result 330
Volt DC output. From the above circuit can be explained
when the boost converter circuit in the given input it has
not resulted in increased tension, but after the entrance to
the second input is increased in response to a direct
increase the voltage boost converter which directly gives
the inductor. After that the output of the boost in tamping
by 3 phase inverter (AC). And the simulation results in
which the red color is green U blue V and W. system
where the signals U and V. W signal precedes This
method uses a series of stars.
[3] Aswadi. E Learning Document.BJJ FT UNP PADANG P4TK
MEDAN.2009
[4] Agus Cahya Setya Budi. Sistem Kontrol Kecepatan Motor
Induksi 3-Phase Penggerak Mobil Hybrid. Tugas Akhir : T.
Elektronika Politeknik Elektronika Negeri Surabaya Institut
Teknologi Sepuluh Nopember; 2011.
[5] Ainur Rofiq, Irianto, Cahyo Fahma S. Rancang Bangun AC – DC
Half Wafe Rectifier 3phasa dengan THD minimum dan Faktor
Daya Mendekati Satu Menggunakan Kontrol Switching PI
Fuzzy.Tugas Akhir : Teknik Elektro Industri Politeknik
Elektronika Negeri Surabaya Institut Teknologi Sepuluh
Nopember; 2006
VI. CONCLUSION
Be concluded that the results of the simulation signal
boost converter combined boost converter 1 through 3
produces a signal that SteadyState but the signal will be
steady state in the long term. This is due to ignoring the
value of efficiency so that no steady boost konverter state
then if given the inverter is in getting the results that there
is still a bit of signal noise on the signal.
REFERENCES
[1] Zhong Du, Burak Ozpineci, Leon M.Tolbert, John N.Chiasson.
DC-AC Cscacde H-Bridge Multilevel Boost InverterWith No
Inductors For Electric/Hybrid Electric Vehicle Application.
Boise State University.2009.
[2] Ramon O.Caceres, Ivo Barbi, IEEE. A Boost DC – AC Converter:
Analysis, Design, and Experimentation. IEEE Transactions On
Power Electronic.1999.
B4-3
AUTHOR
Dedid CH, born in Pasuruan, Indonesia,
December 27, 1962. Educational backgrounds:
Engineer in Electrical Engineering Institute of
Technology Sepuluh Nopember Surabaya,
Surabaya Indonesia (1986).
MT Electrical Engineering Institute of
Surabaya Indonesia (2002)
Post-graduate student in Electrical Engineering,
in Institute of Technology Sepuluh Nopember
Surabaya-Indonesia (2007- ow)
Agus Indra Gunawan, born in Nganjuk, Indonesia,
August 21, 1976.
Bregas W P, born in Surabaya, Indonesia, June 18,
1990. Educational backgrounds:
Engineer in Electrical Engineering Institute of
Surabaya
Indonesia
(2008-ow).
Android Smartphone Based for The Local
Directory Application of Public Utility
1
Arini, MT, 2Viva Arifin, MMSi, 3Chery Dia Putra, S.Kom
1,2,3Informatics Engineering Program
State Islamic University (UIN) Syarif Hidayatullah Jakarta
Ir. H. Juanda Street No. 95, Ciputat,South Tangerang– Banten
Email: arinizoel@yahoo.com, viva_mks@yahoo.co.id, cherydiaputra@gmail.com.com
Abstract—Mobile has become the one thing that
characterizes the lives of everyone present, so that
evolution is happening very quickly, not just a device used
to communicate, but the phone also has been deeply
involved in the life style, to multimedia. Smartphone is a
term of mobile (cellular phone) with multimedia and
computing capabilities are more advanced than the mobile
phones in general. Android is a smart phone that has a
complete platform starting from the operating system,
applications, developing tools, applications, market
applications, support for mobile industry vendors, and
even support from the community of Open Systems. Of
course this is an advantage not shared by other platforms.
This study examines the development of local application
directory that specifically discusses the Bintaro Sector1 to
Sector9.
Application
development
using
JAVA
programming language with tools ECLIPSE GALILEO
and other programming languages to access the server
using the Personal Home Page by using the MySQL
database server. For the method of data collection is done
by three stages, namely the field of research that includes
observations and interviews, library research, and similar
literature studies. For system development, researchers
used the method of Rapid Application Development (RAD)
which has 4 stages of the terms of the planning phase,
design phase, construction phase and implementation
phase. This application can facilitate users in finding
existing public facilities in the region Bintaro Sector 1 to
Sector 9. For the process of further development, these
applications can be expected to provide call features to be
able to contact the existing facilities.
Android is a complete platform starting from the
operating system, applications, developing tools,
applications, market applications, support of industry
vendors mobile, even the support of the community of
Open Systems. By looking at this development, android
has become extra ordinary powers. In 2009, estimates
reported by Canalys, the smartphone market to grow
android 1073.5% when there is no other platform that
reaches 100% growth [9].
Fig 1.The spread of the Smartphone Market
200000
100000
Index Terms—Local Directory, Public Utility, Android
Smartphone
0
Android iPhone
I. INTRODUCTION
Fig 2. Comparison Android Users in
Indonesia
source : Telkomsel, 2011
M
OBILE has become the one thing that
characterizes the lives of everyone present, so that
evolution is happening very quickly, not just a device
used to communicate, but the phone also has been deeply
involved in the life style, to multimedia [9].
Smartphone is a term of mobile (cellular phone)
with multimedia and computing capabilities are more
advanced than the mobile phones in general. This is due
to the combination of operating systems, hardware, and
applications that are much nicer on the smartphone.
As for the number of android smartphone users in
Indonesia, such as news quoted from Tempo Interactive
(www.tempo.co) based on data obtained from one of the
telecommunication operators in Indonesia, Telkomsel
which states that the number of Android users has
surpassed one of its competitors, the iPhone is based on
category of number of users of data services offered by
Telkomsel.
C1-1
Developed jointly between Google, HTC, Intel,
Motorola, Qualcomm, T-Mobile, Nvidia joined in the
OHA (Open Handset Alliance) with the purpose of
making an open standard for mobile devices (mobile
devices) [9].
Often times when in a new place and need
information about the place, for example where the
nearest restaurant or place of worship, many people like
the people outside the region or local community who
struggle to have to ask where or to whom. Usually people
use maps to determine the direction, but the map can also
be used to determine the location of public facilities such
as places of worship or a restaurant that difficult to find
and precisely determined, because the public facilities
are usually not included in the map.
In the case of the difficulty of finding the location of
public facilities that exist, it requires the device and
services that can assist in finding or determining one's
position. Bintaro is a strategic area visited by many
people, in addition to work, people visiting Bintaro also
for shopping, schools, etc. (Area Manager Directory
Bintaro, 2010:10).
From the results of the spread of the questionnaire,
many people do not know the exact area Bintaro location
of public facilities in the Area Bintaro. 76% of the Area
Bintaro not know the general location of the position of
the existing facilities in the Area Bintaro, and 24% of the
population Bintaro Regions Sector 1 to Sector 9 know
the location of existing public facilities in the Area
Bintaro. Therefore there is need for smartphone-based
application intended to determine the location of the
facilities visited by the public who wish around Bintaro
or
society
that
is
outside
Bintaro.
These system can assist communities in determining
public facility that can be viewed via mobile phones
using the Android operating system by using internet
access.
II. TEORITICAL BACKGROUND
The Smartphone become the next generation of
mobile computing (mobile) which will drive the
convergence between communications, computers, and
the use of electronic devices, three different
characteristics of traditional industries with low
interoperability.
PCMag Encyclopedia provides definitions
smartphone as a cellular phone with built-in applications
and internet access. Smartphones provide digital voice
services and text messaging, e-mail, Web browsing, and
video camera, MP3 player and video and even watch
TV. A smartphones can also run various applications,
change your phone to mobile computer (mobile
computer).
Additionally Pei Zheng and Lionel Ni defines a
smartphone as a new class in mobile phone technology
that is able to facilitate data access and processing
information with computing capability significantly.
Besides having the traditional functions contained on the
mobile phone such as call and sms, smartphones are
equipped with personal information management (PIM)
and communication to multiple media and wireless
access.
Basically, a smartphone is like a small computer
network in the form of mobile phones. It supports one or
more short-range wireless technologies like Bluetooth
and infrared, making it possible to transfer data via a
wireless connection in addition to cellular data
connection. Smartphones can provide mobility as a
computer, access to data everywhere, and
comprehensive intelligence to nearly every aspect of
business processes and everyday life. In addition, this
smart phone can be used as a terminal for e-commerce,
enterprise applications, and, location-based services
(Location Based Service). In short, be the future of
smartphones in mobile technology today, as it offers a
variety of features in improving wireless capabilities,
computing power and storage on-board. Today, people
are seeing as high-end smartphones, multifunctional,
business-oriented phones with high resolution color
displays and processors support the equivalent of
computer technology.In general, smartphone regarded as
one of the promising candidates to achieve that goal.
A. Location Based Service (LBS
Location-Based Services is an information service
that can be accessed through mobile devices over
cellular networks and has the ability to utilize the
location of the mobile device position. The same sense is
given by Open Geospatial Consortium (OGC, 2005)
regarding the LBS is a service IP - wireless that uses
geographic information to provide services to users of
mobile devices. Each service application that utilizes the
mobile terminal position (OGC, 2005). Kupper said :
Location Based Service (LBS) is a common name for a
new service where the location information into its main
parameters. Other terms are also given, that the LBS is
actually one of the added value of GSM cellular service.
LBS is not a system, but it is a service that uses
additional system support the GSM system. Basically,
some systems use the same basic principles, namely:
Triangulation. Thus, the principle is not much different
from the GPS system, it's just a satellite function is
replaced by BTS.
From some of the definitions above can illustrate
that the LBS as a combination of three technologies
(Figure 1).
Fig 3. LBS (Source : Shiode Et Al, 2004)
B. How LBS Work
The figure 4. bellow is the illustration of how the
LBS.
C1-2
Application Development (RAD) was chosen because
the applications will be built is a simple application.
Fig 4. How LBS Works (Source: Riyanto, 2011)
III. RESEARCH METHODOLOGY
A. Data Collection Methods
In the early stages of designing this application, first
conducted interviews with relevant parties in order to
obtain information about the problem and the application
needs to be designed. Party in question is Manager of the
Area Management Office Bintaro Ir. Riyaldi Lokaputra.
In addition to direct interviews, the next stage is
spread the questionnaire. To find out the problems and
the desire that is expected by prospective users of the
application, then made the spread of the questionnaire to
the respondent that will be used as sample data. Samples
are taken as many as 50, the researchers divided into two
parts, namely 15 respondents who already knew about it
about the location of existing public facilities in the Area
Bintaro, for this case is the community around Bintaro
Sektor1 to Sector 9. Furthermore, 35 respondents
addressed to the user community smarthphone android.
Sampling was done by purposive sampling technique.
Purposive sampling is a technique of determining the
sample with a certain consideration [11]. The reason the
use of this technique because we will develop
applications Local Directory associated with the existing
Public Facilities in the Area Bintaro, then the selected
respondents are people who are in the area Bintaro
Sector 1 to Sector 9.
Then we also use Library studies. At the stage of
collecting data by means of research literature, the
authors find references relevant to the object to be
examined. Other reference searches carried out in
libraries, bookstores, and online via the internet.
B. System Development Method
System development method that used is Rapid
Application Development (RAD). This method has four
stages of the development cycle, ie the terms of the
planning phase, design phase, construction phase and the
last is implementation phase. The selection method is
because the system is expected to have a design that can
be accepted by consumers and can be developed easily
because the design of the present system still needs
further development. Another reason this method is the
selection of system restrictions are needed in order that
the system has not changed. In addition, Rapid
1) Planning Phase
Combining the methods of field study reports the
results of user policies into a structured specification
using functional modeling to determine user needs. From
the analysis of such systems can be defined design
purposes, a viable proposal acceptable information.
Stage performed include :
1. Overview of Facility, which aims to find data about
the facilities that will be included in the application to
be made.
2. Problem identification or problem analysis. Identify
the problem or problem analysis aims to identify
existing problems, related to the application made.
3. Problem Solving. Is proposed settlement of the issues
in the search for the location of public facilities in the
form of restaurants (where food is unique), place of
worship, school, and banks.
2) Design Phase
Having drawn up the existing system including the
resolution of constraints or problems that exist, the next
stage is designing the proposed system in order to run
better and expected to overcome the problems that exist.
Applying the model of the desired user, the stages are
carried out are:
1. Designing processes that will occur in the system
using UML diagrams is by making the number 13
(thirteen) Activity Diagram, a Use Case and a Class
Diagram and 13 Sequence Diagrams Sequence
Diagram. In designing with UML, the author uses
Visual Paradigm software.
2. The design process required specification, by
translating the processes that occur in this system into
the form of a simple algorithm that will be
implemented in the form of the program.
3. The design of the interface, by making the screen
display design in the form of input-output which aims
to facilitate communication between user with the
system. After the design of the display screen is
formed then do the construction phase.
3) Construction Phase
At this stage a presentation of the design into the
program. In this stage the author uses the Java
programming language using the Eclipse platform and
the Android Emulator Galileo.
4) Implementation Phase
At this stage do application testing by performing
two stages of testing, the testing will be done
independent by the writer and testing to be performed by
the user of android smartphone users who will be using
this system. This stage focuses on the functional
requirements of a testing software, which ensures that the
input will be processed into output according to need.
Testing technique used is black box testing techniques
software testing is a method that tests the functionality of
the application as opposed to the internal structure or
C1-3
job-specific knowledge of the application code / internal
structure and knowledge of programming in general is
not required.
3. Designing Database
In this database, all types of data involved in the
process of occurring, and collected in the form of
presentation as follows:
IV. DISCUSSION
a. Place Table
A. Planning Phase
Bintaro Jaya is a residential area which is
increasingly
comprehensive,
integrated,
and
independent role in stepping on the age of 30 years.
Bintaro Jaya, which was developed in 1979 and
registered as a member of REI No. 1, managed by PT
Jaya Real Property, Tbk. Currently, it has built tens of
thousands of housing units and occupied by more than
22,000 kk. Vast stretches of sector 1 which still includes
the area south of Jakarta, to the Graha Raya, which is
included in the administrative area of Tangerang regency
administration. As a residential area that is well known,
Bintaro Jaya as the residence is equipped with various
facilities, such as commercial areas, offices, sports
facilities, education, health, shopping centers, places of
worship, transportation, and others. All this become
Bintaro as the residence choice of professionals, so that
called the Professional'sCity.
Design Phase
At this design stage, authors will design a system to
resolve the existing problems. The design of the system
that the author made the draft determination includes
actor, usecase design, drafting usecase scenario or
usecase narrative, activity diagrams, sequence diagrams,
class diagrams, and interface design.
Table 2. Place Table
No
1
2
3
4
5
6
7
8
Table 1. Actor
No.
Actor
1
User
2
Admi
n
Information
Users are actors who use android based mobile
smartphones. Users can only view the data
contained on this application
Admin has full rights over this data and
applications, including data editing, data updates,
and see all the contents of the data on the
application.
Type
INT(11)
VARCHAR (255)
VARCHAR (255)
VARCHAR (255)
VARCHAR (255)
VARCHAR (255)
INT (11)
VARCHAR (11)
Extra
Auto_Increment
b. Category Table
Table 3. Category Table
No
1
2
Field
Id_kategori
Kategori
Type
INT (11)
VARCHAR (255)
Extra
Auto_increment
c. Sector Table
Table 4. Sector Table
No
1
2
B.
1. Design Applications
a. Determination of Actor
Field
id_tempat
Nama
Alamat
Telp
Longitude
Latitude
Sector
id_kategori
Field
Id_sektor
Sector
Type
INT(11)
VARCHAR (255)
Extra
Auto_increment
a. Construction Phase
4. Coding Implementation
At this stage, carried out the implementation of
database designs, system designs, or design view. The
programming language used in the design of this system
is to use PHP and Java. To use the MySQL Server
database as data storage media. To run the server
application code is required, the design of these systems
use Apache. To the editor and unit tests are used Eclipse
Galileo. In the debugging phase of the eclipse plugin
authors using the Android Emulator using the SDK that
has been given. For the code can be found at the end of
the writing of this appendix. One example menu in this
application is the Main Menu which contains four
buttons, each button can provides the information
needed.
2. Designing Use Case
Fig 5. Screen Shot Main Menu
Fig 4. Use case
C1-4
Fig 11. Screen Shot of Searching
Fig 6. Screen Shot Nearby Place View
Fig 7. Screen Shot Sector View
Fig 12. Screen Shot of Searching Result
5. Software and Hardware
In order for these applications can be run properly and
correctly then it takes a device capable of supporting
these applications, both from the software and the
hardware. For that need to be considered a category of
devices that can run this application.
1. Mobile with OS Android
2. SmartphoneAndroidwith minimum API is 7
Fig 8. Screen Shot Place of Sector
Fig 9. Screen Shot of Category View
6. Implementation Phase
Before the program is applied, then the program
should be free from error. And to be free of errors it is
necessary to test to find errors that may occur as in the
language errors, logic errors and error analysis program.
This stage is done so that applications can continue to
use and runs well. As for the writer to do is application
maintenance, maintenance is done on the possibility of
error (errors) that occur in applications that are running,
so the need for periodic check or control.
Implementation of the applications implemented with
testing applications that have been built, if built is in
conformity with the expectations of the user, at this stage
if the system has not developed as expected the writer to
revise its application. Examination performed on all
matters relating to the application. Testing applications
with black box methods. Tests carried out in two phases,
namely an independent testing and testing by the user.
1. Self Testing
Independent testing done by running the application
bintaro this application directory and see if it matches
with the problem domain and the expected conclusion.
Independent test results can be seen in Table 5.
Fig 10. Screen Shot of View Restaurant
C1-5
Table 5. Application Test Table
No.
1.
2.
3.
4.
5.
6.
7.
Testing
Interface Page Splash Screen
Main Menu Interface
Interface tab page in the Sector
Button on the tab Sector sector
Category Tab son the application
Button of Category
Search button
2. In this application has not been a call feature that can
be contacted directly to the destination place. The
author hopes that the development in this issue.
Result
Good
Good
Good
Good
Good
Good
Good
VI. REFRENCES
[1]
[2]
[3]
2. Field Testing
Field testing conducted to determine the advantages
and disadvantages in this application. The author tested
the 50 people who becomes mapping in this study.
[5]
Table 6. Field Testing
No.
1.
2.
3.
4.
5.
6.
7.
8.
Testing
Over all of Application
Feature for user
Completely feature
Nearby feature
Driving direction feature
User Interface
Ease of access to applications menu
Compliance with Requirements
[4]
[6]
Result
Good
Good
Good
Good
Good
Good
Good
Good
[7]
[8]
[9]
[10]
[11]
After conducting an independent testing and field
testing, the authors concluded this application runs fine.
94% of people said that this application area Bintaro
good for use in a search of public facilities in the area
Bintaro Sector 1 to Sector 9.
[12]
[13]
[14]
V. CONCLUSION AND SUGGESTION
A. Conclusion
Form our research then it can be concluded as follows:
1. By analyzing and designing applications Bintaro
Directory is available applications that provide
information about the Bintaro, especially places to eat,
places of worship, banks, and schools.
2. This application can determine the distance from the
user's position to the position where that will be
addressed by utilizing the internet access.
3. Besides being able to determine the distance, this
application can also provide driving direction or a
direction from the position of the user to position the
destination.
4. Determination of the distance taken from the user's
longitude and latitude and the destination.
5. On the admin page, the admin can add new places that
have not been registered in the database server.
[15]
Anonim.2010. Direktori Pengelola Kawasan Bintaro
Hariyanto, Bambang. 2004. Sistem Manajemen Basis Data :
Pemodelan, Perancangan, dan Terapannya. Bandung:
Informatika.
Hariyanto, Bambang. 2004. RekayasaSistemBerorientasiObjek.
Bandung: Informatika.
Irfiyanda, Syukrina. 2009. Analisis dan Implementasi Informasi
Pembayaran Rekening Air Berbasis Mobile (Studi Kasus
Perusahaan Daerah Air Minum Kerta Raharja Kab. Tangerang).
Universitas Islam Negeri Syarif Hidayatullah Jakarta. Skripsi
tidak diterbitkan.
Kendall, K.E., dan Kendall, J.E. 2008. System Analysis
and Design 7th Edition.New Jersey: Prentice Hall.
Ladjamudin, Al Bahra Bin. 2005. Analisi Dan Desain Sistem
Informasi. Yogyakarta : GrahaIlmu
Maseleno, Andino. 2003. Kamus Istilah Komputer dan
Informatika. Dokumen tidak diterbitkan.
Misky, Dudi. 2005. Kamus Informasi & Teknologi. Jakarta :
EDSA Mahkota
Mulyadi, 2010.MembuatAplikasiUntuk Android
Nugroho, Bunafit. 2005. Database Relation Dengan MySQL.
Yogyakarta :Andi
Sugiyono, Prof. DR. 2009.Statistika Untuk Penelitian. Bandung
: Alvabeta
Suryadi, I Gede Iwan. Kepariwisataan. STMIK STIKOM bali.
Dokumen tidak diterbitkan.
Rahmawati, Yuli. 2008. Membangun Sistem Informasi Spasial
Fasilitas Umum Kesehatan (Studi Kasus : Puskesmas dan
Rumah Sakit Kota Administrasi Jakarta Selatan). Universitas
Islam Negeri Syarif Hidayatullah Jakarta. Skripsi tidak
diterbitkan.
Wulandari, Sri. 2010. Aplikasi Proses Hierarki Analitik (PHA)
Dalam Memilih Handphone. Universitas Pendidikan Indonesia.
Skripsi Tidak Diterbitkan.
Zahrudin, Muhammad. 2010. Impelementasi Simulator
Optimasi Rute Terpendek Berbasis Mobile menggunakan
metode Greedy dengan pendekatan Manhattan Distance (Studi
Kasus : Jalur Transportasi Darat Wilayah Administrasi Jakarta
Barat). Universitas Islam Negeri Syarif Hidayatullah Jakarta.
Skripsi tidak diterbitkan.
Arini, she acquire her Bachelor Degree (ST) from Brawijaya
University and Master Degree (MT) graduated from University of
Indonesia and cooperation with Uni Duisburg-Essen Germany. Now,
she work at the Faculty of Science & Technology, Informatics
Engineering, UIN Syarif Hidayatullah Jakarta
Viva Arifin, she acquire her Bachelor Degree (S.Kom) and Master
Degree (MMSI) from Gunadarma University. Now, she work at the
Faculty of Science & Technology, Informatics Engineering, UIN
Syarif Hidayatullah Jakarta
Cerry Dia Putra, he graduated from Informatics Engineering,
Faculty of Science & Technology, Informatics Engineering, UIN
Syarif Hidayatullah Jakarta
B. Suggestion
This application is of course still not perfect. There
are still many things to do to develop this application to
make it better again, among others:
1. The author expects to progress further, this application
can input data into the database through a gadget that
is owned by the user. So the user can add a few places
that have not been registered by the author.
C1-6
Tropical Rain Effects on Free-Space Optical and
30 Ghz Wireless Systems
1
M. Derainjafisoa, 2G. Hendrantoro
Department of Electrical Engineering, Institut Teknologi Sepuluh Nopember, Indonesia
Kampus ITS, Keputih-Sukolilo, Surabaya, 60111
1
mderainj@gmail.com, 2gamantyo@ee.its.ac.id
Abstract—The growth of technology in recent decades,
demonstrates an ongoing commitment to find new
elements for better improvement of quality of service in all
areas such as Free Space Optical (FSO) communication. In
any wireless communication system, propagating channel
is highly dependent on different weather conditions
particularly that reduce performances. For a system
requiring high link availability, the variable attenuation of
the atmospheric link is the main challenge in practice.
Ongoing research considers the performance of free-space
optical links over tropical rain which is a strong
turbulence that affects the channels. Millimeter Wave
communication is known as a rain dependence
communication. However, compare to FSO link, it
presents robustness to rain effect for propagation under a
kilometer link distances. The rain attenuation difference
between FSO and Millimeter wave was analyzed in order
to provide the optimal solution in terms of wireless
technology. The measurements and simulations described,
as a result, should achieve to a more suitable link distance
and a link availability prediction on FSO systems in
tropics.
Index Terms—Free-Space Optical, Link availability,
Millimeter wave, rain attenuation, Synthetic Storm
Technique.
I. INTRODUCTION
T
HE Free-Space Optical Communication technology
tries to carry out rising need for high bandwidth
transmission capability link along with safety and
simplicity in installation. Due to their high carrier
frequency in the range of 300 THz [1], it supplies
highest data rates. FSO link is license free, secure and
easily deployable. These features encourage the use
FSO as a solution to last mile access. In any wireless
communication system, transmission is influenced by
the propagating channel. The propagating channel for
FSO is atmosphere. FSO links are extremely weather
dependent that reduces the link availability and
reliability.
The increased signal losses and fades are the fallout
of the impairments due to the optical signal propagation
in free-space. The phenomena known as light absorption
at specific optical wavelengths comes from interaction
between photons and atoms or molecules that cause
extinction of the incident photon, rise of the temperature
and radioactive emission. Scintillation caused by
thermal turbulence within the propagation and
atmospheric scattering cause angular redistribution of
the radiation. Among atmospheric effects on optical
wireless communications link, rain is the most important
factor in tropics. The effort here is to focus on some of
the most important atmospheric attenuators and to
simulate their attenuation behavior using precise
mathematical relations, derived and improved over the
years. A conscious effort has been made to select the
well known and most suitable relations in modeling and
measurement the FSO and the Millimeter Wave
channel. Previously, a significant effort in this regard
has been shown. Effects of rain on FSO and GHz
frequency range links are studied and some
measurement results are presented. Specific attenuation
for FSO can reach up to 40dB/km for rain rate of 155
mm/hr. It further emphasizes the requirement of back up
link with least rain attenuation to achieve certain degree
of reliability of hybrid wireless network. [1]-[2].
Finding out a suitable modulation and coding scheme
for the FSO systems requests rigorous understanding of
the behavior of the FSO. The choice of the modulation
and coding format has to be based on measurements and
simulation of the performance of FSO systems under
unfavorable atmospheric conditions. These preliminary
investigations point out that the use of these schemes
will lead to better systems to combat fog and other
attenuators which limit the performance of the
technology [3].
The problems of the propagation of signals through
free space are generally difficult to solve due to the
impairments of those signals. It is quite impossible to
find exact transmission techniques. These conduct to
some problems that need throughout investigation as the
degradation of the signal imposed by rain attenuation.
Also find the most suitable technology for wireless
propagation in tropical areas between FSO and
Millimeter wave operating in 30 GHz.
Yet, the starting point of our thinking understood the
rain attenuation using SST as a measurement of the
channel. The objective of the research is that
understanding the impairment on the free space
propagation under tropical rain in order to provide the
appropriate technology need to solve that problem. To
do this task, the following step will be held, in first,
improve the quality of service in the field of free space
propagation in tropics. In the other hand, provide the
technology needed to reduce those impairments in order
C2-1
to obtain the best performances of the propagation. The
results of this research will benefit for the development
of research about the understanding both FSO and
millimeter wave operating in 30 GHz systems
performance under heavy tropical rain.
Much more investigation is desired as to the best of
our knowledge comprehensive channel models for
optical wireless communications which still have many
unanswered questions. The organization of the paper is
as follows: the first part presents the background of the
propagation in free space. The second part describes the
methodology adopted to predict the rain attenuation
using SST. Finally, a detailed analysis was done to
compare the performances for both Millimeter wave and
FSO systems considering operation and deployment.
generally, for FSO, given by the Carbonneau relation
[4]:
Att = 1.076 × R0.67 (dB)
(2.3)
Rain
The relationship between specific attenuation and
rain rate for millimeter wave communication is given by
[6]
γ R = kR α
Constant k and α, according to ITU-R model, is
given as:
(k H +k H +(k H -k H )cos 2 θcos2τ)
k=
2
(2.5)
(kH αH +kV αV +(kH αH -kV αV )cos2θcos2τ)
α=
(2.6)
2k
II. METODOLOGY
B. Synthetic Storm Technique (SST)
The synthetic storm Technique describes the value
of rain fall that moved on the line because of wind with
particular speed. The data range concern about some
interval of time which are January, February, November
and December 2008, January and February 2009, and
also January 2010 where the rain falls were heavy.
A rain event for a time minute sample was made.
The rain intensity records were done for each 10 seconds
and a sampling for each minute time was performed for
the calculation of the rain attenuation using the SST
method. The frequency operation is about the wireless
optical systems.
The attenuation using SST is calculated considering
the main wind orientation. Wind speed is used to
calculate the length of each segment link.
∆L = Vr × T ( Km ) ,
(2.1)
Where k H , kV , α H and α H for horizontal and vertical
polarization are given in [5].
C. Attenuation for FSO
The attenuation of the rain rate events from January
2008 to January 2010 was computed. Maximum rain
rate of 416.53 mm/hr and 351.66 mm/hr was recorded
respectively on January and February events 2009.
Fig.1 and Fig.2 below show the rain attenuation for
both North and East direction link. The attenuation
difference among the 0.5, 0.75, 1, 1.5 and 2 Km single
links increases linearly with the distance. The link
orientation shows a considerable difference. For the 2
km link length, the attenuation reaches up to 90.2 and
100.5 dB respectively for North and East direction link
at 0.001% outage probability. Rain attenuation
estimation results show that the North direction link has
the largest attenuation. This is caused by the main wind
directions in Indonesia which are from West and East.
CCDF Rain Attenuation Link North-South
1
10
0.5 km
0.75 km
1 km
1.5 km
2 km
0
10
Pb.[Attenuation > Absciss]%
A. Rain rate
Wind speed and orientation data range was provided
by the meteorology station at Juanda - Surabaya, a
division of the Department Of Climatology and
Geophysics Meteorology. The measurement of rain fall
was done in ITS campus of Surabaya. A disdrometer
optic combined with a rain gauge was used, and put on
roof of the mechanical engineering building. From this
calculation, it could found the rain fall value on the
ASDO software since January 01, 2008 until January 31,
2010.
(2.4)
-1
10
-2
10
-3
10
-4
10
Vr
where
is the wind speed on a line and T is a
sampling time of 10s.
A (k ) = ∑nN=0 aR (bk-n) × ∆L
i
where
∆L
Ai
0
N
(dB)
b
150
Fig. 1 CCDF rain attenuation for 90o direction link at 0.5, 0.75, 1, 1.5
and 2 km for FSO
(2.2)
is the rain attenuation for i=1, 2 ,…, n.
is the segment length, R is the rainfall intensity
(mm/h), aR
50
100
Rain Attenuation [dB]
is the rain attenuation (dB/Km) and is
A SST multilink is used to measure the rain
attenuation in Surabaya Indonesia. The rain attenuations
at 0.5 km link length, direction East (0o), for outage
probability 0.1%, 0.01%, 0.001% are respectively
2.913, 15.26 and 24.79 dB (fig. 1). It is shown clearly in
C2-2
fig.3, for the 2 km link orientation West (180o), a high
and considerable attenuation .The maximum attenuation
estimated for this link are 11.2, 60.3 and 90.2 dB
respectively on 0.1%, 0.01%, 0.001% outage
probability.
10
10
0.5 km
0.75 km
1 km
1.5 km
2 km
0
10
0.5 km
0.75 km
1 km
1.5 km
2 km
0
CCDF Rain Attenuation Link East
1
1
10
-1
10
-1
10
-2
10
-3
10
-4
10
-2
10
0
-3
10
50
100
150
Fig. 4 CCDF rain attenuation for 90o link from 0.5-2 Km for 30 GHz
frequency
1
-4
10
10
0
50
100
0
10
Fig. 2 CCDF rain attenuation for 0o direction link from 0.5-2 km for
FSO
1
10
link
link
link
link
link
0
10
0.5 km
0.75 km
1 km
1.5 km
2 km
150
0.5km
0.75km
1km
1.5km
2km
-1
10
-1
10
-2
10
-3
10
-2
10
-4
10
-3
10
0
-4
10
50
100
150
Fig. 5 CCDF rain attenuation for 45o direction link from 0.5-2 Km for
30 GHz frequency
0
50
100
o
o
o
10
o
Fig. 3 CCDF rain attenuation for 0.5 (0 ), 0.75 (45 ), 1 (90 ), 1.5 (135 )
and 2 km (180o) links for FSO
link
link
link
link
link
0
10
D. Attenuation for 30 GHz millimeter wave
The performance of the millimeter wave
communication system is corrupted by rain attenuation
which limits its exploitation for free space
communication link. Rain is one the major attenuating
factor at frequencies above 10 GHz. The theoretical
back ground between specific attenuation and rain rate is
given in [6]. The rain attenuation on the millimetre wave
operating in the frequency of 30 GHz was estimated
using the SST.
Estimation of the rain attenuation for both North and
East direction link are shown in Fig.4 and Fig.5 below.
Rain attenuation among the 0.5, 0.75, 1, 1.5 and 2 Km
point to point communication link do not diverge
considerably compared to the link direction but increase
linearly with distance. For the North orientation link,
attenuations are 13.51, 20.26, 27.02, 40.53 and 54.04 dB
at 0.01% outage probability; whereas at the same outage
probability, attenuations are 13.56, 20.23, 26.84, 40.47
and 54.42 dB for the North-East (45o) orientation link.
CCDF Rain Attenuation for 30GHz MM Wave
1
150
0.5km
0.75km
1km
1.5km
2km
-1
10
-2
10
-3
10
-4
10
0
50
100
150
Fig. 6 CCDF rain attenuation from 0.5-2 Km links for 30 GHz
frequency
For the multilink communication shown in Fig. 6
above, direction and length of links are 0.5 Km East (0o),
0.75 Km North-East (45o), 1 Km North (90o), 1.5 Km
North-West (135o), and 2 km West (180o). Rain
attenuations augment linearly with the distance and also
with the link orientation and are 27.39, 40.69, 55.79,
C2-3
CCDF Rain Attenuation Link North-East
1
10
80.34 and 96.5 dB respectively at 0.001 outage
probability.
0.5 km FSO
0.5 km 30GHz
0.75 km FSO
0.75 km 30GHz
1 km FSO
1 km 30GHz
1.5 km FSO
1.5 km 30GHz
2 km FSO
2 km 30GHz
0
10
III. SIMULATION AND RESULTS
A. Difference attenuation
From probability outage 0.1% to 0.004%, rain affects
the FSO link much more compared to the 30 GHz
Frequency link for both point-to-point and multilink
communication. However, under the sub mentioned
probability outage, the rain attenuation jeopardizes the
30 GHz Frequency badly independently of the link
length and orientation as shown in Fig.7, Fig.8 and
Fig.9. The difference attenuation (in dB) between FSO
link and 30 GHz Frequency according to the link
availability are shown in Table.1 and Table.2,
respectively for both point-to-point and multilink
communication. The negative values represent that the
rain attenuation on the 30 GHz Frequency link is greater
than its attenuation on the FSO link.
-2
10
-3
10
-4
0
50
100
150
Fig.9 Comparison of the rain attenuation for 45o direction link
from 0.5-2 Km for FSO and 30 GHz frequency
TABLE.1 DIFFERENCE ATTENUATION (IN DB) BETWEEN FSO AND 30
GHZ FREQUENCY SINGLE LINK
o
-3
0 Direction
Link
Link
Availability
0.5
0.75
1
1.5
2
99%
1.63
2.39
3.11
4.5
6.08
99.9%
1.58
2.34
2.86
4.68
5.47
99.99%
-2.95
-3.86
-5.38
-6.4
-6.3
Link Distance (Km)
o
10
-4
10
0
20
40
60
80
100
120
Fig. 7 Comparison of the rain attenuation for 0.5 (0o), 0.75 (45o), 1
(90o), 1.5 (135o) and 2 km (180o) links for FSO and 30 GHz frequency.
1
0.5 km FSO
0.5 km 30GHz
0.75 km FSO
0.75 km 30GHz
1 km FSO
1 km 30GHz
1.5 km FSO
1.5 km 30GHz
2 km FSO
2 km 30GHz
0
10
-1
10
-2
10
90 Direction
Link
Link
Availability
0.5
0.75
1
1.5
2
99%
1.66
2.5
3.33
4.99
6.65
99.9%
1.68
2.52
3.35
5.02
6.7
99.99%
-2.78
-4.17
-5.55
-8.33
-11.1
Link Distance (Km)
o
10
-1
10
-2
10
10
FSO 0.5km
30Ghz 0.5km
FSO 0.75km
30Ghz 0.75km
FSO 1km
30Ghz 1km
FSO 1.5km
30Ghz 1,5km
FSO 2km
30Ghz 2km
0
10
-1
10
45 Direction
Link
Link
Availability
0.5
0.75
1
1.5
2
99%
1.63
2.44
3.19
4.68
6.02
99.9%
1.62
2.52
3.31
5.03
6.08
99.99%
-2.44
-3.38
-4.9
-8.02
-7.84
Link Distance (Km)
TABLE.2 DIFFERENCE ATTENUATION (IN DB) BETWEEN FSO AND 30
GHZ FREQUENCY MULTILINK
Link Availability
-3
10
Link
Direction
-4
10
0
o
45
0
50
100
150
90
99.9%
99.99%
0.5
1.62
1.58
-2.61
0.75
2.42
2.52
-3.77
o
180
C2-4
99%
o
135
Fig.8 Comparison of the rain attenuation for 90o direction link from
0.5-2 km for FSO and 30 GHz frequency
Link
Length
1
3.32
3.36
-5.49
o
1.5
4.67
5.03
-7.63
o
2
6.01
5.47
-6.3
B. Analysis and discussion
In term of propagation distance, MMW is the best
issue to supply FSO system for propagation under a
kilometer range in a foggy region. However, in tropics,
rain affects strongly both MMW and FSO systems. The
simulations results shown that, for a link distances less
than 1 km, FSO link suffers more from rain attenuation
than the 30 GHz Millimeter Waves link for an outage
probability from 0.1% to 0.004%,. A real advantage of
MM-Wave compared to FSO is that they can work on
relatively short distances. Based on FSO technical specs
and installation statistics, most of the FSO links are
installed on distances no more than 1km, while
MM-Wave links are designed to work on distances up to
20km. This means that MM-Wave links have the very
significant gain margin which allows in penetrating 1km
distance even at very heavy rain [6] System designers
are free to choose the antenna size, which generally
dictates its gain. The size of the antenna determines the
amount of intercepted MMW energy and determines the
beam divergence, since the system is diffraction limited
[7]. Conversely, for long distance propagation,
millimeter wave transmission is affected more by rain,
as the simulation results for the 1.5 and 2km link length
because the carrier wavelength is closer to the size of a
rain drop. Rain drops can vary in size from 0.1mm to
10,0mm, and these will effectively disperse millimeter
waves, especially with carrier frequencies greater than
10 GHz.
Another important system performance is the data
rates. FSO can provides the highest data rates of 2.5
Gbps which can be increased to 10 Gbps using
Wavelength Division Multiplexing (WDM) due to their
high carrier frequency in the range of 300 THz. FSO
technologies offer optical capacity but are typically
deployed at lengths under a kilometer for reasonable
availability as discussed in the previous section. FSO
has a major time-to-market advantage over Millimeter
Wave. An FSO link can be operational in a few days.
IV. CONCLUSION
Tropical rain attenuation makes vulnerable
performances of FSO and MMW links based on the
simulation results. MMW links suffer less from rain
attenuation for short distance propagation but has
bandwidth limited and not secure. FSO offers solutions
for all of these problems but are typically deployed at
lengths under a kilometer for reasonable availability.
The benefits of FSO motivate a deep investigation on
understanding its behavior under heavy rain condition
and use it as last mile solution. The high and
considerable attenuation measured underlines that
reductions of these attenuations are needed which is the
more accurate approach for estimating performances of
the wireless optical link under such condition.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
C2-5
F. Nadeem, E. Leitgeb, O. Koudelka, T. Javornic, G. Kandus,
“Comparing the rain effects on hybrid network using optical
wireless and GHz links”, ICIET 2008, 17-18 October 2008,
Rawalpindi, Pakistan.
F. Nadeem, E. Leitgeb, M. S. Khan, M. S. Awan et al.
“Comparing the Fog Effects on Hybrid Network using Optical
Wireless and GHz Links”, CSNDSP 2008, pp. 278-282, Graz,
Austria, 23-25 July 2008.
Available:http://www.optikom.tugraz.at
Sheikh Muhammad S., Leitgeb E., Koudelka O. “Multilevel
Modulation and Channel Codes for Terrestrial FSO links”,
Beitrag und Präsentation zum International (IEEE) Workshop on
Satellite and Space Communications, September 2005 Siena,
Italy.
T.H. Carbonneau, David R. Wisely, "Opportunities and
challenges for optical wireless; the competitive advantage of free
space telecommunications links in today's crowded market
place", SPIE Conference on optical wireless communications,
Boston, Massachusetts, vol. 3232,119 (1998);
Specific attenuation model for rain for use in prediction methods,
ITU-R, P.838-1
Olsen, R.L., D. V. Rogers, D. B. Hodge, “The a.Rb relation in
the calculation of rain attenuation”, IEEE Trans. Antenna
Propag. AP-26(2), 318-329, (1978)
Gurdeep Singh, Tanvir Singh, Vinaykant, Vasishath Kaushal,
“Free Space Optics: Atmospheric Effects & Back Up”,
International Journal of Research in Computer Science eISSN
2249-8265 Volume 1 Issue 1 (2011) pp. 25-30 © White Globe
Publications www.ijorcs.org
Scott Bloom, “The last-mile solution: Hybrid FSO Radio”,
AirFiber, Inc. AirFiber, Inc.
First Aid Application
Based Android Smartphone
1
Qurrotul Aini, 2Husni Teja Sukmana, and Imamul Huda
Faculty of Science and Technology
Syarif Hidayatullah State Islamic University Jakarta
1
qurrotul.aini@uinjkt.ac.id, 2husniteja@uinjkt.ac.id
Abstract— Mobile computing technology advances rapidly
which has changed mobile phone into a smart phone device
with variety of applications on it. So that devices such as
smart phones have become the primary requirement for
users. Accidents can happen to anyone, anywhere and
anytime. First aid measures are very important to reduce the
impact of the accident. So researchers developed a First Aid
application on smart phone device which in this context is on
the Android platform with version 2.2 Froyo. First Aid
application that contains main features tutorials first aid
measures in accident, as well as some additional features of
drug information, info site nearest hospitals and dispensaries,
the types of rescue techniques, and calls the emergency
number. By using methodology development Rapid
Application Development system which consists of three
phases, namely planning, design and implementation
workshops, these applications are built using the Android
SDK Framework, Java Programming Language, Google
Maps as a spatial data service and system testing has been
done using the method of Black Box testing. The application
has capacity 5 MB and according to questionnaire result,
90% of respondents understood the application.
Index Terms—Android, first aid, froyo, smartphone.
I. INTRODUCTION
T
he development of telecommunications technology
evolution in communication gave birth to a
telecommunications device itself. The emergence of
smart phone technology which is capacity almost similar to
a personal computer, android as one subset of the software
for smart phones that include operating systems,
middleware and core applications released by Google.
Besides having a variety of features, android is also capable
to integrate with various Google services such as Google
Maps, in displaying visual map location information.
Activity resulted in a diverse society the number of
accidents that occurred in Indonesia either an accident at
work, traffic and natural disasters has increased, causing
the large number of casualties. The accident victim needs
help quickly and precisely. The aid can be made and given
before arrival of medical team; come to hospital or
someone will provide further assistance, known as the First
Aid. But the importance first aid was not accompanied by
sufficient knowledge in the community. Moreover, first aid
knowledge gained only from books, extracurricular school
and health education. Based on this background, the
research will be made on application of first aid android
smartphone.
II. LITERATURE REVIEW
Understanding of the applications came from English,
namely "to applicate" which means to apply or applied. But
in general understanding of the application program is a
package of ready-made and can be used. While the
meaning of the application is a computer program created
to help humans in performing certain tasks [1].
Computer it self relation with applications consisting
of multiple functional units to achieve the purpose of data
processing is:
1. The part that reads the data (input data or input units)
2. The part that manages the data (control processing unit)
3. Section who issued the results of data processing (Data
Output)
Besides understanding Application is a software unit
built to serve the need for some activities such as
commercial systems, games, community service,
advertising, or any process that is almost done man.
2.1 First Aid in Accident
First Aid in Accident is a temporary relief and care for
victims of accidents before getting help that is more perfect
than doctors or paramedics. This means that aid is not as
treatment or handling is perfect, but it was just a temporary
relief by first aid officer (medical officer or layman) who
first saw the victim. Ministration should be swiftly and
accurately by using the existing infrastructure at the scene.
First aid actions are done correctly will reduce the
disability or suffering and even save the victim from death,
but if action is not well done FIRST AID could even
worsen and even cause death due to accidents [2].
2.2 First Aid Purpose
The purposes of First Aid are as follows: [3]
a. Save lives or prevent death
1. Taking into account the conditions and
circumstances that threaten the victim
2. Implement Heart and Lung Resuscitation (CPR).
3. Find and fix the bleeding.
b. Prevent a more severe disability (prevent the condition
worsening)
1. Hold a diagnosis
2. Handle the victim with a logical priority.
3. Noting the conditions or circumstances (illness) is
hidden.
c. Supporting healing
1. Reduce pain and fear
2. Prevent infection
C3-1
3. Planning of medical rescue and right transportation
for victim.
2.3
Principle First Aid
Some principles to be implanted in First Aid officers
when facing incidents are as follows: [4]
a. Remain calm, do not panic. You expected to be a helper
rather than a murderer or a victim of the next (helped)
b. Use the eye with a sharp; brace your heart because you
have to have the heart to take action that makes the
victim screaming in pain for his safety, to do with agile
and precise movements without adding damage.
c. Consider the circumstances surrounding the accident,
the occurrence of accidents, weather etc.
d. Consider the person's condition is unconscious; there is
bleeding and wounds, broken bones, feeling much pain
etc.
e. Check the victim's breathing. If not breathing, check
and clean the airway and providerespiratory support (A,
B = Airway, Breathingmanagement).
f. Check the victim's pulse or heart rate. If the heart stops,
perform external cardiac massage.If there is severe
bleeding immediately stop (C= Circulatory
management)
g. Does the patient Shock? If you are looking for shock
and treat the cause.
h. After A, B, and C is stable, check the cause of or
concomitant injuries. If there do splintingbroken bones
to bones broken, do not rush tomove or brought to a
clinic or hospital before to bandage broken bone.
i. While providing assistance, you should also contact the
medical officer or the nearest hospital
2.4 Aid Priorities
There are several key priorities that must be
performed by a helper in helping the victims, namely: [5-7]
1. stop breathing
2. cardiac arrest
3. heavy bleeding
4. shock
5. unconsciousness
6. mild bleeding
7. fracture
2.5 Android
Android is a platform first truly open and
comprehensive approach to mobile devices, all software
that is enabled to run a mobile device without thinking of
ownership constraints that inhibit innovation in mobile
technology [8]. In another definition, android is a subset of
software for mobile devices that includes an operating
system, middleware and core applications released by
Google. While Android SDK (Software Development Kit)
provides tools and APIs needed to develop applications on
Android platform are using Java programming language.
Developed jointly between Google Android, HTC, Intel,
Motorola, Quallcomm, T- Mobile, NVIDIA joined in the
OHA (Open Handset Alliance) with the purpose of making
an open standard for mobile devices (Mobile Device) [9].
2.6 Architecture of Android
In architecture android, there is a Linux kernel and a
set of libraries for C / C ++ within a framework that provide
and manage the flow of the application process [10-12].
The following diagram shows the main components of the
Android operating system.
Fig. 1 Android Platform Architecture
2.7 JAVA
Java is a programming language that innovation can
be an option for a program that will run on various
operating systems. Java can be used for internet and
network-based applications. Java also allows the authors of
the program for use in big scale application that can run
without changes on the computer with the operating system
that supports Java. This is the most widely applied in
today's computers [13]. Java has some important
advantages, among others [14]:
1. Compatibility and stability
Code a Java program can run on an operating system
that has a runtime environment. And it has a lot of
mistakes that have been addressed, and the existence of
a virtual machine also supports the stability of java.
2. Monitoring and management
Java provides the functionality to monitor and manage
applications that typically have enterprise-scale
management using Java technology extension.
3. Enterprise desktop
Java provides integration with desktop facilities to
overcome the limitations of browser-based
applications.
4. XML
Java also provides support including the use of XML
digital signatures and streaming API for XML.
2.8 Eclipse IDE
Eclipse is an open source community that aims to
produce an open programming platform. Eclipse consists
of a framework that can be developed further, auxiliary
equipment to build and manage the software from the
beginning to launch. Eclipse Platform is supported by a
large ecosystem of technologies vendors, innovative
start-ups, universities, research institutions and individuals.
Many people are familiar with Eclipse as an IDE
(integrated development environment) for the Java
language, but the Eclipse is more than just a Java IDE.
Eclipse community has more than 60 open source projects.
These projects are conceptually divided into 7 categories:
1. Enterprise Development
2. Embedded and Device Development
C3-2
3.
4.
5.
6.
7.
Rich Client Platform
Rich Internet Applications
Application Frameworks
Application Lifecycle Management (ALM)
Service Oriented Architecture (SOA)
Generally, used Eclipse to build innovative,
industry-standard software, and tools along with its
framework help job is easier.
Fig. 2 Eclipse Galileo
III. RESEARCH METHOD
Developing multimedia application divided into two
stages, data collection and application multimedia
methods.
3.1 Method of Data Collection
1. Literature Studies
Researchers conducted a study of literature by reading
and studying books and e-book related to first aid,
android-based programming as well as books and articles
obtained from printed media and internet to support the
topics covered in preparation of this research.
2. Field Study
Observation data collection methods, researchers
differentiate into three parts, namely:
a. Observation
b. Interview
c. Questionnaire
3.2 Method of Application Development
A system development method that researchers use in
this study is method of Rapid Application Development
(RAD) which was introduced by James Martin in 1991.
RAD is a development cycle that is designed to provide
much faster development and higher quality results than
those achieved with traditional life cycle (SHPS). The
selection method is because the system is expected to have
a design that can be accepted by consumers and can be
developed easily because the design of the current system
still requires further development. Another reason this
method is the selection of system restrictions are needed in
order that system has not changed. In addition, Rapid
Application Development (RAD) was chosen because the
applications will be built an application that is built in a
fairly short period of time.
Rapid Application Development (RAD), which
researchers have used the following stages: [15]
1. Requirements
Activities charged with finding a general overview of
first aid, learning the culture or cultures android users,
analyzing some of first aid applications that have been
made previously and to identify features based on
application purposes.
2. Workshop Design
Activity in the content by designing proposed
application to be run better and expected to overcome
the problems which exist.
3. Implementation
Activities at the contents with a presentation of the
design into the program using the Java JDK (Java
Development Kit) as a programming language
integrated into Eclipse, and Android SDK (Software
Development Kit) and continued with the installation to
Android handsets.
Some of the reasons researchers use RAD in application
development for first aid on android smart phone:
1. First aid application is a simple application that was
developed by researchers and require a short time. This
is because all components are provided in the Android
application framework. RAD is very precise so that the
method is applied because this method emphasizes an
extremely short development cycle.
2. Existing needs can be well understood, the RAD
process enables to create a completefunctional system
in a short period of time.
3. RAD can develop applications quickly and sustainable
implement the design and specification of user
requirements using tools such as Java.
IV. RESULTS
Refer to RAD stages, the application built based on
following stages.
4.1 Requirement
As described in the previous chapter, in this stage,
researcher identified objectives and system requirements
information arising from those goals.
A.
System Prototype Development Goals
Development of prototype system aims to help user’s
android in providing first aid in an accident that both
happened to him or others.
B.
Finding Information Regarding First Aid
Information search purposes first aid aims to meet the
data completeness measures first aid handling. Here the
researchers conducted search first aid information through
books that discuss first aid, interviews to people who know
about first aid, and search for information on a certain
website about first aid (www.gotoaid.com).
C.
Learn About User Android
Purpose of studying culture is Android smart phone
users to know the habits of Android smart phone users
when interacting with applications and to maximize the
design of user interface or application interface will be
developed. Here researchers read a lot of good references
from printed books, e-book as well as supporting websites
C3-3
like
http://Admin.android.com
http://www.androidpatterns.com.
and
D.
Analyzing Some Applications FIRST AID
Researchers conducted an analysis on a number of
Android applications on FIRST AID with different media
platforms; the goal is to obtain a preliminary description of
the features, user interface and functionality making it
easier for researchers to do the innovations in the
application to be developed.
After doing the analysis, researchers can conclude
that in addition to feature a fairly complete presentation of
the material needed also an interesting application that
users can more easily understand the material interest and
which is on the application.
4.2 Workshop Design
In this stage, researcher design use case until interface
for the application.
Fig. 4 Use Case Diagram
D.
Activity Diagram
Activity diagrams describe the activities that occurred
in First Aid application begin until the activity stops.
A.
System Design
Here, researchers conducted a system design that will
be applied in application. The developed application is
called "First Aid on Mobile" for first aid in an accident with
combined of several technologies such as Google's APIs, as
well as the GPS is implemented into the smart phone
android.
First aid measures data that existed at first aid on
Mobile derived from several sources such as providers of
books that discuss about first aid [2-7], also a site that
displays first aid tutorial [16]. From these sources the
researchers collect material that will be displayed in the
First Aid application on Mobile.
User Interface Design
At this stage, the researchers designed user interface
or interface view of application.
Fig. 5 Activity Diagram FIRST AID
B.
E.
Sequence Diagram Design
Sequence diagram is interaction diagram which
expressed with time, or the other word said with the
diagram from top to bottom. Sequence diagrams express
every user of few streams that pass through a use case.
TindakanFIR
ST AID
: user
JenisTindakanFIR
ST AID
DetilTindakanFIR
ST AID
tindakan_FIRST
AID()
kategori_tindakan_FIRST
AID
jenis_tindakan_FIRST
AID()
show_list_jenis_tindakan_FIRS
T AID
detil_tindakan_FIRST
AID()
show_detil_tindakan_FIRS
T AID
Fig. 3 User Interface Design
Fig. 6 Sequence Diagram First Aid
Use Case Diagram
Use Case describes the interaction of actors in the
FIRST AID applications to be developed. In this context,
researchers chose Android Smartphone users as an actor.
Use Case Scenarios serve to explain more details
about modules in First Aid application based Android
Smartphone.
4.3 Implementation
In implementing this process there are several steps,
namely:
1. Write program code (coding), this stage is done using
application program android developers, Android
Developer Tools (ADT), Android SDK (Software
Development Kit).
C.
C3-4
2. Conducting the process of packaging by using facilities
provided by Android SDK.
3. Test the program by using the android smart phone
handsets, as well as perform debugging or repair the
program if necessary.
A.
Software Implementation
The software used in the building of this application is
as follows:
1. Windows 7 32 bit Operating System
2. Android SDK (Software Development Kit)
3. IDE (Integrated Development Environment) using
Eclipse Galileo
4. Android Developer Tools (ADT)
5. Android : Froyo 2.2 with Google API SDK Level 8
6. Java
7. XML
B.
Hardware Implementation
The hardware used in building this application is as
follows:
1. Intel Pentium Dual Core 2.6 GHz
2. Memori 3GB
3. VGA 1GB
4. Harddisk 250GB
5. Monitor
6. Mouse and Keyboard
7. Handset Smartphone Android (Samsung GT-S5660
Galaxy Gio)
C.
User Interface Implementation
Implementation is the stage where system is ready to
operate in actual stage, so it will be known whether the
system has been created completely according to plan or
not. In software implementation will be explained how this
system works, by providing system display.
researchers. Based on the results of testing, all functions
can be run well.
B. Beta Testing
Beta testing is testing conducted objectively where
testing is done directly to the field of public and not
restricted to certain circles. Testing is done by creating a
questionnaire to find out the opinions of respondents to
First Aid application on mobile, and then distributed to
some users to be filled which will serve as a sample and will
do the calculation to be taken its conclusion on the result of
making the application of this system. Based on testing
performed, 90% of respondents understood the steps to
help injured people in accident.
V. CONCLUSION
Based on testing of application and questionnaire, it
can be concluded that First Aid application based android
smartphone developed under the rules of medical action of
First Aid as its knowledge base which has capacity of 5 MB
memory. The application is able to provide information
about actions and supporting information which cover
location of hospitals and pharmacies, emergency numbers
and rescue techniques required in performing first aid in
accident. For the future research, the application could be
developed as a web-based application as continue to
function as mobile application. The web application is used
as a medium for knowledge base development system that
serves as a media updated. Moreover, it’s better to develop
for cross platform applications in other mobile platforms,
like Symbian or Blackberry.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
Fig. 7 Display of Main Menu
4.4 System Testing
The testing was used to examine new system by black
box method focuses on the functional requirements of
software.
A. Alpha Testing
Based on the testing plan has been prepared, it can be
done alpha testing, which is a test conducted by the
[13]
[14]
[15]
[16]
C3-5
A. Nugroho. Analisis dan Desain Sistem Informasi, Yogyakarta:
Andi, 2004.
N. Saubers, Semua yang Harus Anda Ketahui Tentang FIRST AID.,
Yogyakarta: PallMall, 2011.
M. Kartono, Pertolongan Pertama., Jakarta: PT Gramedia Pustaka
Utama, 2005.
S. Sudirman, Panduan FIRST AID, Jakarta: Restu Agung, 2008.
A. Thygerson, First Aid : Pertolongan Pertama Edisi Kelima.
Jakarta: Erlangga, 2011.
A. Yunisa, Pertolongan Pertama Pada Kecelakaan, Jakarta:
Victory Inti Cipta: Jakarta, 2010..
Peraturan
Menteri
Tenaga
Kerja
dan
Transmigrasi
(Permenakertrans) Nomor: PER.15/MEN/VIII/2008 tentang
Pertolongan Pertama pada Kecelakaan di Tempat Kerja.
R. Meier. Professional Android 2 Application Development.,
London: Willey Publishing, Inc, 2008.
A. Mulyadi, Membangun Aplikasi Android, Yogyakarta:
Multimedia Center Publishing, 2010.
M. Murphy, Beginning Android 2, Barkeley: APRESS, 2009.
M. Murphy, The Busy Coder’s Guide to Android Development,
United States of America: CommonsWare, 2008.
J. Steele, The Android Developer’s Cookbook: Building
Applications with the Android SDK, New York: Addison Wesley,
2010.
I.Horton, Beginning JavaTM 2: JDKTM 5 Edition, Indianapolis:
Wiley Publishing, Inc, 2005.
J. Friesen, Beginning JavaTM SE 6 Platform: From Novice to
Professional, USA: Apress, 2007.
Kendall & Kendall. 2008. System Analysis And Design. London:
Pearson International Edition 7th Edition.
First Aid. [Online]. Available: http://gotoaid.com
Singly-Fed Circularly Polarized Triangular
Microstrip Antenna With Truncated Tip Using
Annular Sector Slot For Mobile Satellite
Communications
1
Muhammad Fauzan Edy Purnomo and 2Sapriesty Nainy Sari
Department of Electrical Engineering, University of Brawijaya
1
mfauzane@yahoo.com and 2nainy_sari@yahoo.co.id
Abstract—In this research, a new model of the triangular
patch antenna with truncated tip using annular sector slot
embedded on the ground plane.The antenna instead of single
layer, singly-fed, small, wide bandwidth and analyzed by
using method of moments. The result for simulation among
of the new model antenna c3s, c3 and c1, in the case of
frequency characteristic, S-parameter, and input impedance
are good results. The bandwidth c3 is widest than the others.
The bandwidth c3sis almost the same with c1. It is caused by
using double truncated tip Is on the below side of the patch
antenna thus the total of vector current distributions become
increased are just around this area. In the case of antenna c3s,
the bandwidth decreased due to by using annular sector slot
embedded on the ground plane. Moreover, the bandwidth of
antenna c1 is also decreases, because of without using the
truncated tip Is, but it is slightly wider than antenna c3s.
Keywords-singly-fed, truncated tip, annular sector slot,
triangular-patch.
I. INTRODUCTION
To obtain circular polarization (CP)operation, some
designs by embedding a cross slot of unequal slot lengths
in the circular patch [1] or inserting slits of different
lengths at the edges of a square patch [2] or truncated tip
of equilateral triangular antenna [3] or using proximity
feed embedded on below radiating patch antenna [4] have
been proposed recently. In the case of the equilateral
triangular antenna with a truncated tip [3], for RHCP the
probe-fed is usually located in the right half of the
triangular patch. Conversely, LHCP radiation can be
obtained in the left half of the triangular patch. In this
research, the new phenomena happen if the patch antenna
changed become triple truncated tip with case Is> Ip(see
Fig.1), probe-fed RHCP and LHCP located on the left and
right half of the triangular patch, respectively. In the case
Is<Ip, the RHCP and LHCP can be obtained with the rule
above [3], but if case Is = Ip, both of RHCP and LHCP can
not happen, only linear polarization can be obtained. In
this case, the function of two truncated tip with length Is
are as switch to move variation of polarization, if the
probe-fed exist on the same place. In addition, the
function of Is or two of triple truncated tip canalso affect
the bandwidth. It makes the bandwidth become wider than
without it (see Fig.2). In the other hand, the slot antenna
both embedded on patch antenna or and on ground plane
cause the bandwidth antenna decreased, but its
advantaged isprobably to make small antenna. The
function of slot is decreasing the frequency
operationwherein the current path or guide wavelength λg
of the TM10 modewith slot is more length than the current
path without slot. The purpose of this research is to yield
the optimized result between the small antenna and the
slightly wide bandwidth.
II. METHODOLOGY
The methodology instead of literature study, qualitative
and quantitative analysis related microstrip antenna,
especially truncated tip using annular sector slot antenna,
feeding probe, and mobile satellite communications.
Ensemble version 8.0by moment method software is
used to design antennafor mobile satellite communication.
There are some advantages i.e. for industry and research
department to develop technology of telecommunication,
especially for design antenna with the new one of
technology.
The conclusion is enclosed of this paper that it
contributes the valuable analysis of singly-fedcircularly
polarized triangular microstrip antenna with truncated tip
using annular sector slot for mobile satellite
communications.
III. ANTENNA CONFIGURATION
Fig.1 shows the configuration of antenna design. The
triangularpatch has a side length of a = b and using a
conventional substrate (relative permittivity 2.17 and loss
tangent 0.0009). The antenna is fed by single probe which
located on right half for LHCP. Here, any three of small
triangular tip wherein two of all are the same side length
Is.It is affected to excite more magnitude current path
around this area which moving to y direction and x
direction, hence increasing the bandwidth. The others
triangular tip has side length Ipowing to the truncated-tip
C4-1
(top view)
substrate
LHCP
RHCP
patch slot
probe_fed
(side view)
substrate
ground
The bandwidth c3sis almost the same with c1. It is caused
by using double truncated tip Is on the below side of the
patch antenna thus the total of vector current distributions
become increased are just around this area. In the case of
antenna c3s, the bandwidth decreased due to by used
annular sector slot embedded on the ground plane.
Moreover, the bandwidth of antenna c1is also decreases,
because of without using the truncated tip Is, but it is
slightly wider than antenna c3s.
Fig.2 shows that the value of gain and axial ratio
(Ar)for simulation of new model antenna at the resonant
frequency. They are as followed that antenna c1 operates
at the frequency 2.76 GHz, gain RHCP= 6.66 dBic, Ar =
2.91 dB, antenna c3is frequency operation = 2.9 GHz, gain
LHCP = 6.98, Ar = 3.02 dB, antenna c3sis frequency
operation = 2.505 GHz, gain LHCP = 6.08, Ar = 1.75 dB.
In addition, eachantennais fed by probe-fed at the same
loci on the patch antenna. It is clear that antenna c1 and c3
did not satisfy the targets yet, especially the axial ratio. It
is due to by the loci of feeding are still not maximize yet
on the surrounding of patch antenna.Moreover, peak gain
antenna c3s at the frequency resonant is lowest than the
others. It is caused by annular sector slot embedded on the
ground plane can decreased of gain. In addition, the
bandwidth of gain c3 is the widest than the others. It is due
to used the truncated tip Is and without using annular
sector slot.
Aluminium plate
Fig.1. Configuration of simulated antenna, slot on the
ground plane
Moreover, the annular sector slot embedded on the
ground plane with wide of radial w = 1 mm, it is meant
for decreasing the frequency operation, therefore the
antenna can be design more small than the previous
antenna. In addition, the annular sector slot can appear the
others mode (TM20 and TM30) atthe higher frequency
operation. Placing of this slot on the around of ground
plane can affect the surface current path, and then effect
to the performance of antenna.
In this paper, the method of moment(Ensemble
version 8 software) is employed to simulate the model
with an infiniteground plane.Consideration of the efficient
thickness of the antenna(see Fig.1) allowed either the
substrate thickness for triangular patch to be defined with
the single substrate or single layer (h = 1.6 mm). The new
model antenna result (c3s) compared to the others that the
characteristic of the others antenna are the same design
but without annular sector slot (c3) and the same design
without truncated tip Is, without annular sector slot (c1) to
know the improvement of performance each other of
antenna.
IV. RESULTS
Fig.2 to Fig.4 shows the result for simulation among
of the new model antennac3s, c3 andc1, in the case of
frequency
characteristic,
S-parameter,and
input
impedance. The bandwidth c3is widest than the others.
Gain [dBic]
8
Gain-RHCPc1
Arc1
Gainc3
Arc3
Gainc3s
Arc3s
8
6
6
4
4
2
2
0
2.4
0
2.5
2.6 2.7 2.8
Frequency [GHz]
2.9
3
Fig.2. Gain and axial ratio vs frequency
Fig. 3 shows the relationship between the reflection
coefficient (S-parameter) and frequency for the simulation
Rx antenna. From this figure, it can be seen that the Sparameter of new model antennac3s at the resonant
frequency by comparison of the others (S11-c1 = -13.55 dB
and S11-c3 = -13.81 dB) is the best about by -21.07 dB. It
caused by used annular sector slot and thelocation of this
slot on the ground plane which likes the hyperbolic
position respect to null potential or central antenna. The
bandwidth of S-parameter of the new model antenna c3s is
the widest than the others. It is caused by loci of feeding
which optimized on that place of patch antenna compared
the others. In addition, it is also the effect of perturbation
C4-2
Axial ratio [dB]
perturbation, the effective excited patch surface current
path in the y direction is slightly more length than that in
the x direction, which gives the y-directed resonant mode
a resonant frequency slightly smaller than of the xdirected resonant mode. That is, the dominant mode
(TM10 mode) of the triangular patch can be split into two
near-degenerate orthogonal resonant modes of equal
amplitudes and 900 phase difference for LHCP operation.
area on the both of below patch antenna (Is) can enhance
the bandwidth of S-parameter.
S-parameter [dB]
0
-10
-20
-30
2.4
S-parameter c1
S-parameter c3
S-parameter c3s
2.5
2.6 2.7 2.8
Frequency [GHz]
2.9
3
Fig.3. S-parameter
150
100
100
50
50
0
0
Rin[Ω]
150
-50
-100
-150
2.4
Xin[Ω]
Fig.4 depicts the input impedance characteristic of Rx.
This figure shows that the real part of simulation is
difference of each others, but in the case antenna c3sreal
impedance by closed50 Ω at the frequency
operation.Moreover, the reactance part of new model
antennac3s is the best than the others by closed 0 Ω at the
resonant frequency.
-50
Re c1
Imc1
Re c3
Imc3
Re c3s
-100
Imc3s
the purpose design. If we want to design small antenna,
owning the enough bandwidth, antenna c3s will be better,
but if we want to design wideband antenna, so antenna c3
is should be chosen, especially for applicationmobile
satellite communication.
The future work will be done to design small antenna
that owning wide bandwidth,based on the new model
antenna by optimizing the perturbation area and slot.
REFERENCES
[1] H. Iwasaki, “A circularly polarized small-size
microstrip antenna with a cross slot,” IEEE Trans.
Antenna Propagat.,vol.44, pp1399-1401, Oct. 1996
[2] K.L. Wong and J.Y.Wu, “Single-feed small circularly
polarized square microstrip antenna,” Electron. Lett.,
vol. 33, pp.1833-1834, Oct.23,1997
[3] C.L. Tang, J.H. Lu, and K.L. Wong, “Circularly
polarized equilateral-triangular microstrip antenna
with truncated tip,” Electron. Lett., vol. 34, pp. 12271228, June, 1998.
[4] J. T. S. Sumantyo, K. Ito, D. Delaune, T. Tanaka, and
H.Yoshimura “Simple satellite-tracking dual-band
triangular
patch array antenna for ETS-VIII
applications,” Proc.IEEE Int. Symp. Antennas and
Propagation, pp. 2500–2503, 2004
Muhammad Fauzan Edy Purnomo was born in
Banjarmasin, Indonesia, in June 1971. He received the
B.E. and M.E. degrees in Electrical Engineering from
University of Indonesia, Jakarta, Indonesia in 1997 and
2000. He is presently with the Electrical Department
University of Brawijaya, Malang, Indonesia where he is
working toward as lecturer. His main interests are in
the areas of microstrip antennas, array antenna for
mobile satellite communications, and Synthetic
Aperture Radar (SAR).He has been ever be a student
member of the IEICE and IEEE.
-150
2.5
2.6 2.7 2.8 2.9
Frequency [GHz]
Fig.4. Input impedance
3
V. CONCLUSION
The new model antennawas studied in order to get a
compact, small, and simple configuration. The results of
characteristic performance among the new model antenna
are difference, but any two of them can be developed for
mobile satellite application. They are c3 and c3s depend on
Sapriesty Nainy Sari was born in Medan, Indonesia,
in April 1988. She received the B.E. and M.E. degrees in
Electrical Engineering from Institut Technology
Sepuluh Nopember, Surabaya, Indonesia in 2009 and
2011. She is presently with the Electrical Department
University of Brawijaya, Malang, Indonesia where she
is working toward as lecturer. Her main interests are
in the areas of Digital Signal Processing (DSP) and
Communication
Network.
C4-3
Improvement in Performance of WLAN
802.11e Using Genetic Fuzzy Admission
Control
Setiyo Budiyanto
Electrical Engineering Department, Faculty of Engineering, Mercu Buana University
JL. Raya Meruya Selatan, Kembangan, Jakarta, 11650
Phone: 021-5857722 (hunting), 5840816 ext. 2600 Fax: 021-5857733
budiys1@gmail.com
Abstract—The Development of WLAN is growing rapidly
along with the possibility of multimedia services
transmission through the wireless network. Until now,
WLAN has been recognised as a most reliable wireless data
communication and it has a transmission rate more speed
than WiMAX as well as celluler system. To ensure this
application and to maintain the quality of service, WLAN
needs an adaptive admission control system to guarantee
the channel availability for each connection request in high
speed transmission rate. In this paper, we proposed a
Genetic Fuzzy Algorithm as an admission control in WLAN
802.11e. Simulation will be done to investigated the ability
of Genetic Fuzzy Algorithm as an admission control in
WLAN 802.11e in relation with Quality of Service
Guarantee.
Keywords ; WLAN 802.11e, Admission Control, Genetic
Fuzzy System
I. INTRODUCTION
less understood and has a lot of interdependence whereas
fuzzy logic is able to work based on human intuition and
logic, the combination of the two algorithms is expected
to generate a mechanism that allows a better performance
improvement.
Studies that use fuzzy logic and genetic algorithms
on admission control as in [1] fuzzy logic is used as an
admission control in high speed networks such as ATM
(Asynchronous Transfer Mode), this scheme is an
improvement over conventional admission control
schemes. In [2], fuzzy logic tuned by genetic algorithm is
used for call admission control in ATM networks. ATM
technique that uses statical multiplexing complicate for
the application of mathematical models, hence the fuzzy
system is very suitable to be applied. Furthermore, [3]
Fuzzy Logic is used to control parameters such as cell
congestion status, load availability, and the total
interference. Fuzzy rules are used for admission criteria,
from the simulation results, obtained an improvement
when compared to classical admission control strategy.
Fuzzy logic is also used as an Adaptive Contention
Window (CW) based on the ambiguity of information
from the channel in [4],. This scheme then compared with
the scheme of differentiation, it was found that this
scheme has the capability of adaptation for streaming
applications. [5], the combination of fuzzy logic and
genetic algorithms used for optimal access network and
promising in the heterogeneous wireless networks.
In this paper, we propose the application of Genetic
Fuzzy System (GFS) as the admission control in WLAN
802.11e. This is a new proposal because no one has
proposed this scheme before. GFS will be applied as an
admission control by considering the parameters of
collition rate and network load so that decisions can be
taken with proper admission.
Since the release of the IEEE 802.11 standard in
1999, the applications running on the WLAN flatform
increasingly diverse, such as voice to video streaming.
But the Medium Access Control (MAC) on the IEEE
802.11 standard was originally designed for best effort
applications (such as e-mail, web browsing) so it can not
meet Quallity of Service requirements for many types of
new applications develop.
To support Quality of Service, Enhance Distributed
Channel Access (EDCA) was introduced in IEEE
802.11e WLAN standard, which was built as a derivative
of the Distributed coordination function (DCF), equipped
with a prioritization of the four access categories (AC).
This is achieved by variations in the size of contention
window (CW) in back-off mechanism in each category.
The continued development of multimedia services led
to the need in the admission control is increasing as well,
so it introduced an adaptive admission control to improve
II. IEEE 802.11E MAC LAYER [6]
performance of WLAN 802.11e. The use of fuzzy logic
and genetic algorithms as adaptive admission control has
The IEEE 802.11e refer to the specifications
been widely used in research compared to other artificial
developed by The IEEE for Wireless Local Area
intelligence algorithms, both algorithms are preferred to
Networks (WLAN). MAC Layer of The IEEE 802.11e is
apply, it is because they do not require complex
distinguished by the previous 802.11 standards by the
mathematical models and easy to implemented. Genetic
availability of the Access Category (AC) aimed at
algorithm allows to find solutions to problems that are
prioritization of data. Includes 802.11e Enhance
C5-1
Distributed Channel Access (EDCA). The IEEE 802.11e
has four transmission queues, each acting as a single
entity Enhance DCF, which is an Access Category (AC).
Therefore EDCA works with four AC, where each AC,
dealing with one different access channels
higher priority AC gets smaller AIFS values.
Formulation of AIFS as follows:
AIFS [AC] = SIFS + AIFS [AC] x Slot Time (2)
SIFS = Short Inter-Frame Space
b. CWmin, CWmax
This is the value of backoff counter which is uniformly
distributed random value between the contention
window CWmin and CWmax. The higher priority AC
getting smaller value of CWmin and CWmax.
c. TXOP (Transmission Opportunity) limit
This is the maximum duration of transmission after the
medium requested. TXOP obtained from the EDCA
mechanism called EDCA-TXOP. During the EDCATXOP, a station can transmit multiple data frames from
the same AC, where SIFS time period, split between the
ACK and data transmission sequence. The higher
priority AC get the larger of it's TXOP limit. TXOP
for each AC-I is defined as:
TXOP [i] = (MSDU [i] /R) + ACK +SIF +AIFS [i] (3)
Figure 1. Enhance Distributed Channel Access (EDCA) [7]
Each AC consists of a queue-free delivery and a
channel access function with its own parameters, namely
the minimum and maximum Contention Window
(CWmin, CWmax), Arbitration Interframe Space (AIFS)
and the duration of the Transmission Opportunity
(TXOP).
EDCA access mechanism can be described as
follows, when the medium is busy before the backoff
counter reaches zero, then the backoff should be stopped
temporarily, and the station must wait for a period of
AIFS. When the medium is idle again, during the AIFS,
backoff counter minus one.
After the transmission failed, a new CW value is
calculated by Presistence Factor (PF), which is also a
unique value according type of AC. CW value is
calculated based on the following provisions:
CWner [AC] ≥ ((CWold [AC] + 1) x PF) – 1
MSDU [i] is the packet length to the AC-I, ACK is the
time required to transmit an acknowledgment, R is the
physical transmission rate, SIF is the time period are
required by SIF, AISF [i] is the time of AIFS in AC-I.
III. IEEE 802.11E MAC LAYER [6]
This research consists of two major parts namely
designing a Genetic Fuzzy System (GFS) to be applied to
the 802.11e WLAN admission control and testing of the
scheme on a single WLAN system.
At this early stage, to design a model of genetic
algorithm scheme is applied to the WLAN 802.11e
EDCA admission control, as shown in Figure 2.
(1)
After entering the MAC layer, each data packet is
received from upper layers are assigned with a specific
user priority value between 0-7. Each packet of data that
has been given a priority value is mapped into the Access
Category. Applications that are background traffic such
as FTP, mapped into the Access Category AC_BK.
Applications that are best effort such as web browsing
mapped into Access Category AC_BE. Streaming video
applications such as video confference mapped into
Access Category AC_Vi, while the voice application is
mapped into the Access Category AC_VO.
EDCA parameters are periodically sent by Access
Point, parameters that are sent can be dynamically
changed depending on network conditions. EDCA
parameter values are different for each AC, the
parameters are:
a. AIFS (Arbitration Inter-Frame Space)
Each AC starting backoff procedure or begin
transmission after a period of AIFS replaces DIFS. The
C5-2
Figure 2. System Model of Genetic fuzzy Admission Control
In the figure 2 shown a modification of conventional
EDCA admission control system by adding the GFS as a
device that responsible for determining whether a request
will be served or not.
GFS output is obtained from the process of genetic
algorithms and fuzzy rules by observing the condition of
the network. The following figure illustrates this idea, as
a note that the genetic learning process aims to design or
optimize the KB (Knowledge Base). Consequently, a
GFS is a design method for fuzzy system of basic rules
that combine evolutionary techniques (genetic algorithms
to achieve automatic generation or modification to all or
part of the KB).
The model for the input membership functions used
is the triangular model, it is intended to avoid too much
overhead to the process of fuzzy on the MAC layer
Figure 4. Input Membership functions
Membership
Degre
0. 001
0. 005
0.02
0.03
Network load
Membership
Degre
0.3
0,4
0.5
0.6
Network load
Figure 3. Genetic Fuzzy Rule Base System [8]
A set of parameters describe the fuzzy rules,
fuzzy membership functions, and search a set of
parameter values that match based on the optimization
criteria. Knowledge Base Parameters, set the
optimization space, phenotype space that must be
changed into a single representation of genetics. For the
purpose of finding the genotype space, the GA requires
some mechanism to derive new variants of candidate
solutions. The purpose of search process is to maximize
or minimize a fitness function that describes the desired
behavior of a system.
A. Fuzzyfication Interface
Input fuzzy membership functions obtained from the
measurement network load and the collision rate is used
as fuzzy membership functions at the MAC layer [9].
Collision rate =
The number of packet collision
(4)
The number of packets transmitted
While the network load is used as the input fuzzy
membership functions were also measured at the MAC
layer. Network load is represented as a time when the
wireless medium busy represented by transmission
apportunity (TXOP), the time needed to transmit an
MSDU, TXOP obtained from equation (5)
TXOP[i] = (MSDU[i]/R) + ACK + SIF + AIFS[i] (5)
The total network load can be represented as the sum
of TXOP that existed at all AC I at all stations j, which is:
Net_load = ∑ queue_ength[j][i] xTXOP [i]
(6)
B. Genetic Process
The general form of Genetic Algorithm as described
by Goldberg [10]. Genetic Algorithm is a stochastic
search algorithm based on the mechanism of natural
selection and natural genetics. Genetic algorithms,
different from conventional algorithms, starting with the
initial set of random solutions called population. Each
individual in the population of individuals called single
(or chromosome), represents one potential solution to the
problem. Individuals evolve through successive
iterations, called generations. During each generation,
individuals were evaluated using a fitness measure. To
create the next generation, new chromosomes called
offspring (offspring) is formed by either merging two
individuals from current generation using crossover and /
or modify an individual using a mutation operator. A new
generation is formed with a good selection of individual
fitness according to their value. After several generations,
the algorithm converges on the best individual / superior,
which is expected to represent the optimal solution or
near optimal solution to one problem.
1) Generate An Initial Population Of Encodec Rules
Intial population can be generated from the output of
fuzification set. The code is obtained by concatenated
rule using AND Operator. Location on the chromosome
indicate the start and end of a particular rule. The total
number of fuzzy sets in the DB is L [11] :
L = La + Lc
where:
La = ∑Ni, where I = 1 to n
Lc = ∑Mj, where j = 1 to m
(7)
n and m is the number of input and output variables. Ni is
the number Ni represents the number of linguistic terms
C5-3
associated to input variable Xi and Mj the number of
linguistic terms associated to output variable Cj. The
general structure of a rule with AND operator is [16] :
if X1 is Y1 AND X2 is Y2 then Z1 is C1
(8)
X1, X2 are input variables, Y1, Y2 is Linguistic Value,
Z1 is output variables and C1 is value.
For implementing fuzzy rules, we uses fuzzy mamdani
membership functions used in Term set1: {High,
Medium, Low} while term set 2 consist of output label
set {Strong accept, Weak Accept, Strong Reject and
Weak reject} for output variables. Different combinations
will be utilized for chromosome representation scheme.
Using this methodology, a stronger rule can evolve with
every new generation.
3) Crossover and mutation
According to the theory of Genetic Algorithm, a
crossover operator selects substrings of genes of the same
length from parent individuals which are known as offsprings from the same point, replaces them and generates
a new individual. This point can be selected randomly
[13]. For designing chromosome, we used binary
encoding .Different rules can be represented in form of
chromosomes labeled as individuals. Here, single point
crossover operator has been implemented.
IV. IMPLEMENTATION
2) Fitness Function
The encoding scheme which is discussed as follows : The
rule be in the form : If A then C, whre A is Antecedent
and C is consequent: The predictive performance of a
rule can be summarized by a 2 x 2 matrix, which
called a confusion matrix, as illustrated in table. 1.
The labels in each quadrant of the matrix have the
following meaning: [12] :
Table 1. Confusion Matrix
Actual
Positive
Positive
Prediction
Negative
Prediction
Actual
Negative
TP
FP
FN
TN
V. CONCLUSION
Where ;
TP = True Positives = Number of examples satisfying A
and
FP = False Positives = Number of examples satisfying A
but not C
FN = False Negatives = Number of examples not
satisfying A but satisfying C
TN = True Negatives = Number of examples not
satisfying A nor
Therefore, CF(Prediction)=TP/(TP+FP)
(9)
Prediction accuracy is measured by (9) by looking for
proportion of the examples that have predicted class C,
that is actually covered by the rule antecedent. Rule
Completeness (true positive rate) can be measured by the
following equation.
Comp = TP / (TP + FN)
(10)
By combining (9) and (10) we can define a fitness
function such as:
Fitness = CP x Comp
This research will be implemented in Network
Simulator-2 or The NS-2. The NS-2 used is NS 2.28
because EDCA module is implemented in NS-2.28
version made by Sven Wietholter and Christian Hoene
from the Technical University Berlin Telecommunication
Networks Group, EDCA module was used because it has
been verified and used by many researchers by many
researchers as well as having good documentation
[14,15,16].
In the The NS-2.28, 802.11e WLAN is implemented
in the classes. Modifications done on a class-802_11e.h
mac, mac-802.11.cc, priq.h and priq.cc so it can measure
the collision rate and network load.
Fuzzy rule base is used for optimization or search
problems, and genetic algorithms are widely known and
used for global search techniques with the ability to
investigate a global search space for solutions that fit
with a single scalar performance measurement. In
addition to the ability to find the nearest optimal solution
in the a complex search space, the structure of the genetic
code and performance features of the genetic algorithm
makes it suitable as a candidate to be combined with
fuzzy system. The ability of the fuzzy system is expanded
with use of genetic algorithms to develop a broad range
of approaches to designing a fuzzy system fundamental
rules.
The use of soft computing especially fuzzy rule base
for admission control in 802.11e WLAN has been done,
however the results still can not meet the expected level
of QoS. Use of Genetic Fuzzy, is expected to optimize
the search for fuzzy membership functions so as to
improve QoS as expected..
Previous research was conducted on a fix WLAN
networks, in this study will be conducted on wireless
mesh networks with multimedia traffic so the proposed
algorithm can be tested more optimal.
(11)
REFERENCES
For each rule, degree of fitness is calculated according to
the above mentioned fitness function. According to a
defined termination criterion, new offspring is generated.
[1]
C5-4
Gao, Deyun and Cai, Jianfei, Admission Control in IEEE
802.11e Wireless LAN, Nanyang Technological
University, Singapore, University of Hongkong, 2006
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
Barolli, Leonard et.all, A CAC Scheme for Multimedia
Applications Based on Fuzzy Logic, Proceedings of the
19th International Conference on Advanced Information
Networking and Applications (AINA’05) 2005
Piyaratna, Sanka et.all., A Genetic Algorithm Tuned Fuzzy
Logic Based Call Admission Controller for ATM
networks, University of Adelaide, 1997
Dini, Paolo and Cusani,Roberto, A Fuzzy Logic Approach
to Solve Call Admission Control Issues in CDMA
Systems, EUSFLAT - LFA 2005
Naoum-Sawaya, Joe, Ghaddar, Bissan, A Fuzzy Logic
Approach for Adjusting The Contention Window Size in
IEEE 802.11e Wireless Ad Hoc Networks, University of
Waterlo, Canada, 2005
IEEE 802.11e, Wireless LAN Medium Access Control
(MAC) and Physical Layer Extension in the 2.4 GHz
Band, Supplement to IEEE 802.11 Standard, IEEE ,
September 1999
IEEE 802.11e, Wireless LAN Medium Access Control
(MAC) and Physical Layer Extension in the 2.4 GHz
Band, Supplement to IEEE 802.11 Standard, IEEE ,
September 1999
Oscar Cordon, et.al., Genetic Fuzzy Systems: Evolutionary
Tuning And Learning Of Fuzzy Knowledge Bases, World
Scientific Publishing Co. Pte. Ltd., Singapore, 2004.
Munadi, Rendy, R.Rumani, Layla, Performance Analyze
of IEEE 802.11e WLAN For Mixed Traffics TCP-UDP
Using Adaptif Admission Control Mechanism, IGCES
2008, Desember 2008, Malaysia
D. E. Goldberg, Genetic Algorithms in Search,
Optimization, and Machine Learning. Addison-Wesley,
1989
[11] Cordon, O., Herrera, F., Hoffmann, F., Magdalena, L.:
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
C5-5
Genetic Fuzzy Systems, Evolutionary Tuning and
Learning of Fuzzy Knowledge Bases, Advances in Fuzzy
system-Applications and Theory, Vol. 19, pp.8993,97,179-183,World Scientific, USA (2001)Electronic
Publication: Digital Object Identifiers (DOIs):
Freitas, A.: A Survey of Evolutionary Algorithms for Data
Mining and Knowledge Discovery, 31-36 , AAAI Press,
Brazil (2003)Article in a conference proceedings:
Akerakar, R., Sajja, P.S., Knowledge-Based Systems,
Jones and Bartlett, Massachusetts (2010)
Kevin Fall, Kannan Varadhan, “The ns Manual”, The
VINT Project, 2009
S.McCane, S.Floyd, NS Network Simulator ,available at :
www.isi. edu/nsnam/ns/
Wang, Yue, A Tutorial of 802.11 Implementation in ns-2 ,
MobiTab Lab.
Mankad, Kunjay, Sajja, Pritti Srinivas and Alkerkar
Rajendra,”Evolving Rules Genetic Fuzzy Approach-An
education Case study,” International Journal of Soft
Computing (IJSC), Vol. 2, No.1, February 2011
Kejik, Petr, Hanus, Stanislav, “Comparison of Fuzzy Logic
and Genetic Algorithm Based Admission Control
Strategies for UMTS System, available at :
http://www.radioeng.cz/fulltexts/2010/10_01_
006_010.pdf
Alkhawlani,Mohammed, Ayesh, Aladdin, Access Network
Selection Based on Fuzzy Logic and Genetic Algorithms,
Hindawi Publishing Corporation Advances in Artificial
Intelligence Volume 2008, Article ID 793058, 2008
Characterization of Tilted Fiber Bragg Grating as
a Sensor of Liquid Refractive Index
Eka Maulana1, Sholeh Hadi Pramono2, A. Yokotani3
Department of Electrical Engineering, Brawijaya University1,2
Department of Electrical Engineering, University of Miyazaki3
ekamaulana@ub.ac.id
Abstract— We have developed a fabrication technique of
tilted fiber Bragg gratings (TFBG) for measuring sensors
of refractive index of liquids. We demonstrated that a
simple technique using a combination of 266-nm laser and
a phase mask with a period of 1.065 µm was quite effective
for the fabrication of the TFBG. Using fabricated TFBGs
which had the tilted angles of 3.3˚, 6.7˚, 7.3˚, 8.0˚, and 9.9˚,
we tried to measure the refractive index of liquids which
have different indices. Water, ethanol, and glycerine
solutions (12%, 24%, 35%, 46%, 66%, and 84%) were
used as samples. For the measurement, a 10 mm long
TFBGs were covered with a sample liquid drops. The
transmission spectra in the cladding mode and core mode
were observed by an optical spectrum analyzer. We have
directed our attention to the fact that wavelength of
cladding mode shifts to be longer with the increase of
refractive index of sample liquids. Utilizing this
wavelength shift, we proposed a new measurement method.
As a result, we could successfully measured the refractive
index of liquids within a range from 1.00 to 1.41 with a
maximum sensitivity of 3.0x10-3. In addition, we have
found that a contact length of only 2.4 mm is necessary to
obtain 90% of signal intensity of 10 mm long TFBG..
Index Terms— Fiber Bragg Grating, refractive index,
phase mask, cladding mode.
are informations though the surface of the cladding can
be detected, while normal FBG can observe only
mechanical phenomena such as change in length since
the cladding prevents the optical information to the core
from the outside. Though the temperature indeed can be
defected using the normal FBG, this is also measured by
mechanical volume change due to the thermal expansion.
Fabrication of TFBG has been conventionally performed
using Lloyd mirror interferometer by 244-nm Ar+ ion
laser and the spectra observed in previous work4). In
addition, it has been reported that refractive index of
liquid is able to detect in principle by using the envelope
of the cladding mode of TFBG5). However, adjustment
of the Lloyd mirror causes a instability of the period of
the grating and furthermore, use of Ar+ ion laser results
in a large difficulty when this technique is considered to
apply on a commercial basis.
In this research, we have developed a simplified
fabrication technique of TFBG using a combination of
266-nm laser and conventional phase mask which is able
to use without a complicate optical adjustment. Besides,
we also tried to characterize the fabricated TFBG as a
sensor for refractive index of liquids. As a result, we
found a new method to estimate the index of liquid with a
wider measurement range compared to the conventional
method that has been reported in previous work5).
I. INTRODUCTION
T
he fiber sensors are widely used in physical sensing
such as temperature, strain, vibration, pressure,
liquid level and so on1). The increasing of fiber
sensor development for these purposes has many
advantages, ie, electro-megnetic immunity, small size,
stability,
harmless,
and
high
sensitivity2).
Photosensitivity of fiber core was reported by Hill et al.
in 19783). Since then, this invention has been a
significant background for the many kinds of fiber sensor
developments. Especially, the fiber Bragg gratings
(FBG) which consists of a periodic modulation of the
refractive index in the core of a single mode optical fiber
have become widely used for distortion sensors.
Recently, in addition to the FBG sensors, tilted fiber
Bragg grating (TFBG) have become to attract
considerable attention for sensing application. Because
in TFBG, not only the properties of the core but also
informations the cladding affects the reflection spectrum.
Therefore in TFBG, non-mechanical phenomena which
II. TFBG PROPERTIES
The FBGs are made using laser beam interference
technique. Fundamental structure of FBG and TFBG are
shown in Figure 1. A single core mode is produced in the
FBG transmission spectrum. Basically, the wavelength
shift in the core mode is used to detect mechanical
change in the fiber, therefore wavelength of core mode is
utilized to measure physical parameters6).
C6-1
Figure 1. Structure of (a) FBG (b) TFBG
The wavelength of core mode in the FBG (
can be calculated by the following equation,
,
)
period Ʌtfbg a long the longitudinal direction which
follows the equation 3.
,
(1)
where, neff.core is effective refractive index of the core,
and Ʌ is the grating period.
FBGs are intrinsically insensitive to the
environment, because the core modes are
well-screened from incident of the light from outside
due to the presence of the cladding7). On the other
hand, in the case of the TFBG which has a tilt angle θ,
not only the core mode but also a number of cladding
modes are observed in the reflection spectrum. The
reflection wavelengths of the cladding modes
(
) are calculated using the following
equation,
(3)
where, Ʌpm and θext are grating period of the phase
mask and the external tilt angle between the grating and
the sample fiber, respectively. A phase mask of 10 mm
long was used in this experiment.
, (2)
where, neff.cladding is the effective refractive index of
the cladding which is corresponding to each
reflecting order in the cladding mode. The
wavelength and amplitude of the cladding mode are
affected by optical properties of the surrounding
material. The spectral behavior in the core and the
cladding modes have beed reported in previous
work8).
III. EXPERIMENTAL METHOD
A. Fabrication of TFBG
The experimental setup for fabrication of TFBG is
shown in Figure 2. We used a 4ω Nd:YAG laser and a
wavelength converter to produce pulsed 266-nm UV
laser beam. The beam was reflected by five mirrors and
linierly focused on a sample fiber by three cylindrical
lenses. The laser pulses with an energy of 50 mJ/pulse
were produced by this laser. The beam was introduced to
an optical fiber core through a phase mask made of silica
glass. In this technique, periodically modulated UV
beam was produced by interference of diffracted two
laser beams due to the phase mask. The minimum
distance between the phase mask and the fiber sample
was approximately 1 mm. Tension of the fiber was kept
at 5.9x10-4 N during fabrication process with the UV
beam irradiation
A H2 loaded SBG-15 (Newport corp.) photosensitive
optical fiber was used for the sample fiber. This fiber is a
single mode and germanium- boron-codoped. We used
sample fibers of 250 mm long. The polymer jacket at the
center part of the fiber was removed by 50 mm long for
UV beam irradiation. The polymer jacket at both ends of
the fiber were also removed by 20 mm long in order to
connect an optical spectrum analyzer (OSA) and an ASE
light source. The fiber Bragg grating was made in the
center part of fiber using phase mask technique. A phase
mask with a grating period 1.065 µm was used. This
phase mask creates a grating in the fiber core with a
θext
Figure 2. Experimental setup
The fiber was fixed on a rotary stage to adjust the θext
easily. The θext were chosen at 5˚, 10˚, 11˚, 12˚ and 15˚,
which were corresponding to the incident angle of
modulated beam to the fiber surface. Since refraction
occurs between air and fiber material by the UV beam
during fabrication, the tilt angles in the fiber core
became 3.3˚, 6.7˚, 7.3˚, 8.0˚ and 9.9˚. The probe light
with a wavelength range from 1520 to 1610 nm was used
for the transmission spectral measurement. Two
mechanical splicers were used for connecting the fiber to
the light source and the OSA. The typical irradiation
period of the UV beam to obtain an enough intensity for
measurement was 20 minutes. We checked the spectral
change during TFBG fabrication.
B. Refractive Index Measurement
We put a droplet of sample liquid which covered
whole the TFBG whit a length of 10 mm for refractive
index measurement. Experimental setup to measure the
refractive index of liquid is shown in Figure 3. Two
holders were used to keep the TFBG stable on the glass
plate without applying tension. The liquids used were
include water, ethanol, and glycerine solutions.
Optical
Spectrum
Analyzer
Figure 3. Refractive index measurement procedure
The reason why we have chosen the glycerine solution
is that the refractive index of the solution can be easily
C6-2
adjusted in wide index range by simple mixing of the
glycerine and water9). The concentration of glycerine
solution used in the experiment was from 12% to 84%.
The detailed information for index of sample was
summarized in Table 1.
Table 1. Refractive index of sample
Sample
Ref.Index
Air
1.0003
Water
1.3255
Ethanol
1.3539
Glycerine 12%
1.3417
Glycerine 24%
1.3579
Glycerine 35%
1.3727
Glycerine 46%
1.3876
Glycerine 66%
1.4146
the core mode compared to 7.3˚-TFBG was observed.
The peak wavelength and the intensity were 1580 nm
and 2 dB, respectively. The cladding mode were
observed in a range form 1520-1578 nm. The maximum
intensity of the cladding mode was 11.5 dB at 1548 nm.
We observed two coupling modes in the cladding mode
as well.
LP11
LP1n
LP2n
Glycerine 84%
1.4389
We observed the core and the cladding modes in the
transmission spectra and investigated the relationship
between the spectral change and to the change in
refractive index.
Figure 5(a). transmittance spectra 7.3˚-TFBG
IV. RESULT
LP11
A. Characteristic of Fabricated TFBG
Transmittance spectra of 0˚-TFBG (namely normal
FBG) after two minutes irradiation time is shown in
Figure 4. Only the core mode LP11 was observed in this
spectrum. The intensity of the core mode LP11 was 9.5
dB at 1544 nm.
LP1n
LP2n
Figure 5(b). Transmittance spectra 8˚-TFBG
LP11
Figure 4. Transmittance spectra 0˚-TFBG
Figure 5(a) shows the transmittance spectra of the
7.3˚-TFBG after 20 minutes irradiation time. We
investigated not only core mode LP11, but also the
cladding modes in a spectral range approximately from
1520 to 1563 nm. The core mode shifted to the longer
wavelength and reached 1565 nm, and its intensity
decreased to 2.5 dB. The maximum intensity of 9 dB in
the cladding mode was observed at a wavelength of 1546
nm. In the cladding mode, two coupling modes (LP1n and
LP2n) were observed.
Figure 5(b) shows the transmittance spectra of the
8˚-TFBG after 20 minutes irradiation. In the case of
8˚-TFBG, the longer wavelength and smaller intensity of
Figure 5(c). Transmittance spectra 9.9˚-TFBG
Figure 5(c) shows the transmittance spectra of the
9.9˚-TFBG after 20 minutes irradiation. In this case, the
core mode was disappeared. The cladding modes were
observed in a range from 1520 to 1580 nm. Although
superposition of the two coupling modes were not
apparently observed, we could confirm that both of them
were present because the wavelength difference between
two adjacent peaks were almost same as the those in the
case of 8˚-TFBG.
C6-3
B. Characteristic of Liquid Mesurements
Using the 8˚-TFBG which exhibits the strongest
coupling in cladding mode, the experiments for
measuring the refractive index of liquids were
performed. The result are shown in Figure 6. In Figure 6,
spectral change of a selected one peak in the cladding
modes is indicated. We selected a peak at 1543.76 nm
which was the strongest peak in 8˚-TFBG. As shown in
the figure, the peak wavelength shifted to longer as the
refractive index of liquid become larger. As a result,
wavelength shift from 1543.76 to 1544.40 nm
corresponding to the index from 1.00 to 1.41. In addition
to the wavelength change, the smooth change in the peak
intensity was also observed. Laffont et al. have used this
change in intensity to estimate the index of liquid and
demonstrated that the refractive index of liquids from
1.35 to 1.44 was able to estimated by calculating area of
the envelope of cladding modes. In this work, we
concentrated to investigate the change in wavelength
intending the wore accurate measurement.
The resolution of refractive index is shown in Figure
8. We estimated the spacial resolution of detection using
8˚-TFBG, The resolution of the optical analyzer is 20
pm, then the sensitivity of refractive index change by this
method is estimated to be 9.2x10-2 to 3.0x10-3 depending
on the index.
Figure 8. Resolution of refractive index
The spacial resolution of detection is shown in
Figure 9. Using 8˚-TFBG, it was found that only 2.4 mm
is necessary to contact with the sample liquid in order to
get 90% of signal change at 10 mm droplet.
Figure 6. Refractive index measurement of liquids in 8˚-TFBG
Figure 9. Spacial resolution of detection in 8˚-TFBG
V. DISCUSSION
Figure 7. Relative wavelength shift of TFBG with different tilt angle
Similar experiments have been done for other TFBG
with different tilt angles. Figure 7 shows the relative
wavelength shift to the refractive index of the liquids and
air. The maximum correlation of relative index is
achieved at 8˚-TFBG. As the tilted angle become larger,
the correlation between the index and obtained data
become stronger.
Two modes have been investigated in the TFBGs
transmittance spectra. These modes are the core mode
and the cladding mode. The amplitude of transmittance
spectra become larger by irradiation time. When the
tilted angle of TFBG became larger, the amplitude of
core mode getting smaller and its wavelength shift to the
longer. In cladding mode, transmittance spectra shows
two coupling modes. At 9.9˚-TFBG, the core mode
almost disappeared, it has a cladding mode only. The
amplitude of 9.9˚-TFBG transmittance spectra after 20
minutes different with others. It has small amplitude.
We investigated one peak in cladding mode for liquid
measurement. The results show that wavelength shifts to
the longer when the refractive index of liquid was
C6-4
increase. Relative wavelength shift to be increasing
exponentially against the refractive index of liquid
sample. The maximum transmittance intensity and
sensitivity were achieved at 8˚-TFBG. Wider dynamic
range was achived at 3.3˚ and 6.7˚-TFBG. Figure 10
shows the peak wavelength as a function of refractive
index of liquids. The square dot shows the measured
peak wavelength with 8-TFBG. By interpolating the
measured data, a smooth correlation curve has been
obtained in the index range from 1 to 1.41. The circle dot
indicates the data from previous work presented by
Laffont et al.5). Comparing these two curves in Figure 10,
we can say that our estimation method indicates the
wider measurement range. Our refractive index range
was 0.41 while the Laffont’s method range was 0.09.
technique and 266-nm laser.
The fabricated TFBG could be used to measure
refractive index of liquid from 1 to 1.41 at
8˚-TFBG
The sensitivity was 9.2x10-2 to 3.0x10-3 for
8˚-TFBG depending on the index.
It was found that only 2.4 mm was enough to
contact for measurement refractive index of
liquid.
ACKNOWLEDGMENT
We would like to acknowledge and extend our
gratitude to Allah SWT, the greatest creator who makes
everything possible and to the following person who
have made the completion of this paper among those: Dr.
Agung Darmawansyah, our research team, our advisor,
and all of the member of Photonic Applications
Laboratory in UoM.
REFERENCES
[1]
[2]
[3]
[4]
Figure 10. Peak wavelength compared by Laffont method in 8˚-TFBG
Resolution of refractive index measurement depend
on optical spectrum analyzer resolution, it was 20 pm.
The resolution was measured by this method estimated to
-2
-3
be 9.2x10 to 3.0x10 .
[5]
[6]
[7]
VI. CONCLUSION
Several TFBGs have been fabricated and its
transmittance spectra during fabrication has been
investigated. We have also measured the refractive index
of liquid using wavelength shift monitoring in a cladding
mode and transmittance response of TFBGs by liquid
droplet. Based on our experiment, we conclude that:
TFBG could be fabricated using phase mask
[8]
[9]
C6-5
X Dong, et.al.: Tilted Fiber Bragg Grating; Principle and
Sensing Applications, Photonic Sensor, 1,6-30, 2011.
Yin S et.al.: Fiber Optic Sensors. Second Edition, CRC
Press, New York, 2008.
Hill K.O and Meltz G: Fiber Bragg Grating Technology
Fundamentals and Overview, Journal of Lightvawe
Technology, 15(8), 1263-1276, 1997.
Othonos A and Kalli A: Fiber Bragg Gratings, Artech
House, Boston, 1999.
Laffont G and Ferdinand G: Tilted Short-period Fibre
Bragg Grating Induced Coupling to Cladding Modes for
Accurate Refractometry, Meas.Sci. Technol. 12,
765-770, 2001.
R Kasyhap: Fiber Bragg Gratings, Academic Press, San
Diego, 1999.
Erdogan E: Cladding-mode Resonance in Short- and
Long-period Fiber Grating Filters, J. Opt. Soc. Am. A, 14
(8), 1760-1773, 1997.
A Cusano, et.al.: Single and Multiple Phase Shifts Tilted
Fiber Bragg Grating, Research Letters in Optic,1-4, 2009.
Rheims J: Refractive Index Measurement in the near-IR
using Abbe Refractometer, Meas. Sci. Technol, 8,
601-605, 1997.
Video Streaming Analysis on Worldwide
Network Interoperability for Microwave Access
(WiMAX) 802.16d
Dwi Fadila Kurniawan, Muhammad Fauzan E.P. dan Widya Rahma M.
Department of Electrical Engineering Faculty of Engineering UB
df_kurniawan@ub.ac.id, mfazanep@ub.ac.id
Abstract-In the recent years, more and more
information services require high-speed data access. Video
streaming is a real-time service with high-speed data
access which conveys information such as audio and video
networks using Internet Protocol (IP). Using the streaming
technology clients can play the video in real time
condition. However, it’s strongly influenced by the
bandwidth. Insufficient bandwidth for the streaming
process will cause losses and greater delay[1]. Therefore,
in order the video streaming service to approach its ideal
conditions it is necesary to be applied on a network which
has a high speed data access and large bandwidth. Such
conditions can be fulfilled by the WiMAX network
802.16d, because it is the network technology based on
international standard IEEE 802.16 which enable to
transfer
data to wireless broadband access as an
alternative to cable or DSL. WiMAX can provide the
folllowing types of access : fixed, nomadic, portable and
mobile wireless broadband on the line of sight (LOS) and
non line of sight (NLOS) conditions[2]. Based on
calculations, by varying the distance 1 km - 15 km between
transmitter and receiver for LOS and 1 km - 5 km for
NLOS, the value of the propagation losses on NLOS is
found to be much larger than on LOS. In LOS conditions,
the value of bit error probability is smaller than the NLOS
conditions for all types of modulation. The best conditions
occur in LOS using QPSK modulation with 2.6 Mbps data
rate with bit error probability 2.6184x10-45 and packet
loss probability of video streaming is 9.1200x10-4.
I.
INTRODUCTION
The current telecommunications technology has
evolved to the needs of high-speed data access. Video
streaming is a real-time service with high-speed data
access which conveys information such as audio and
video networks using Internet Protocol (IP). Using the
streaming technology, in ideal conditions clients can
play the video in real time. Ideal conditions of the video
streaming is strongly influenced by the bandwidth.
Inadequate bandwidth in the process stream will cause
the loss and greater delay. Therefore, in order that
service streaming video applications approach ideal
conditions it needs to be applied on a network that has a
high data access speed and wide bandwidth. Terms -
conditions can be met by the WiMAX.
WiMAX is a basic standardized IEEE 802.16
technology that allows transfer of data to access
wireless broadband access as an alternative to cable or
DSL (Digital Subscriber Line). WiMAX can provide
access to the type of fixed, nomadic, portable and
mobile wireless broadband to the condition of LOS and
NLOS. Just with one Base Station, the theoretical
coverage of the cell radius could reach 50 km. WiMAX
also includes QoS features that enable services such as
voice and video with low delay. According to the
WiMAX Forum, the system can transmit data at speeds
up to 75 Mbps per carrier for the type of fixed and
portable access. In a network with mobile access types,
based on its specifications, it can generate speeds of
more than 15 Mbps with a radius up to 3 km. This
indicates that WiMAX technology can be used through
the notebooks and PDAs which can be implemented on
mobile phones.
In this paper, the calculation of video streaming
parameters on the network of WiMAX 802.16 analyzed
were pathloss, bit energy to noise ratio (Eb / No), bit
error rate (BER), packet loss probability of streaming
video, delay end-to-end throughput as well.
II. METHODOLOGY
The first step of methodology used in this paper is
modeling the system in order to simplified architecture
of end to end WiMAX network created to facilitate the
calculation and analysis of data end-to-end delay. Fig. I.
below shows the picture of end-to-end delay of video
streaming in IEEE 802.16d WiMAX network.
tpacketizatio
tenc tprop ttrans
tprop ttrans
tdepacketization
tprop ttrans tdec
tw
Fig. I. Modelling end-to-end Delay on the 802.16d
WiMAX network
C7-1
With the same calculations for different distances
between the BS to the SS, the obtained results as shown
in Fig.2.
Based on that model Performance parameters of video
streaming on the 802.16d WiMAX network being
analyzed include the end to end delay, propagation
losses, energy bit per noise, bandwidth, probability of
bit error, packet loss, and throughput. Performance is
reviewed from several conditions, namely LOS, NLOS
outdoor and indoor NLOS.
The calculation of performance parameters of video
streaming on the 802.16d WiMAX network, consist of
the value of pathloss, RSL, the probability of bit errors,
packet loss probability, end to end delay and
throughput. All performance parameters are computed
on LOS and NLOS conditions. To simplify the process
of analysis and calculation, some secondary data used is
as shown in Tables I and II, it shows the specifications
of the base station and CPE (Customer Premises
Equipment) on the WiMAX IEEE 802.16d.
Fig. 2. Graph of the distance values at RSL LOS
Fig. 2. shows that the greater the distance between BS
and SS, the smaller RSL (received power level of the
receiver).
Non Line of Sight (NLOS) Conditions
In this condition, the value loss of NLOS
propagation will be calculated with the distance
between the transmitter and the receiver changes from a
distance of 1 km - 5 km. By using equation (3), the
value of path loss in NLOS conditions can be calculated
as follows:
TABLE I
BASE STATION DEVICE SPECIFICATION[3]
Parameter
Transmitter Power
Maximum EIRP
Value
27 dBm
44 dBm
TABLE II
SUBSCRIBER STATION DEVICE SPESIFICATION [4]
Parameter
Profile
Receiver Power
Antenna Gain
Value
Outdoor
NLOS
24 dBm
17 dBi
 d 
PL = A + 10γ log   + ∆ PL f + ∆ PL h + s
 d0 
…..(3)
For the distance between transmitter and receiver as far
as 1 km, the value of path loss can be calculated by
using
the
steps
as
follows:
- Calculation of reference path loss value (A)
 4π d 0 
A = 20 log 

 λ 
…………..…(4)
With λ= c/f = 0,086 m,
d0 = 100 m. Than,
 4 × 3,14 × 100 
A = 20 log 

0 ,086


= 83,3231 dB
Indoor NLOS
24 dBm
13 Bi
A. Calculation of Propagation loss (Pathloss)
LOS Condition
The calculation of LOS propagation loss is often
called the Free Space Loss (FSL). The calculation of
this attenuation will be used to calculate the amount of
power received by the Receiver Signal Level (RSL).
In this condition the value of free space loss will be
calculated if the distance between the transmitter and
the receiver changes from a distance of 1 km - 15 km
and if the system works at a frequency of 3.5 GHz. By
using equation (1), loss propagation at a distance of 1
km in LOS conditions can be calculated as follows[1]:
- Calculation of path loss (γ)
c
γ = a − b .h t +
ht
………………(5)
If the area observed is assumed in urban areas, the
value of a, b and c using the data in Table III for the
terrain type B.
FSL = 32.45 + 20 log d + 20 log f………...(1)
= 32.45 + 20 log 1 + 20 log 3500
=195.6604
TABEL III
PARAMETER UNTUK TIPE TERRAIN YANG BERBEDA [3]
By using equation (2), the calculation of the signal level
at the receiver can be calculated as follows:
RSL ( dB ) = EIRP − FSL + G r …….(2)
C7-2
Parameter
Tipe A
Tipe B
Tipe C
A
4.6
4
3.6
B
0.0075
0.0065
0.005
C
12.6
17.1
20
Efek shadow (s)
10.6
9.6
8.2
- Calculation of the frequency correction factor (∆PLf)
If f = 3.5 GHz = 3500 MHz, then
17 ,1
γ = 4 − 0 , 0065 × 32 +
32
= 4,3264
the frequency correction factor (∆PLf) value is
3500
∆PLf = 6 log
= 1.4582
2000
Calculation of antenna height correction factor
user (∆PLh)
h 
∆PLh = −10.8 log r 
2
Fig. II. Graph of RSL value and distance on NLOS
Based on Fig. II, it can be seen that the greater the
distance between BS and SS, give greater the value of
path loss. Fig. II also shows a comparison of the RSL to
the distance in NLOS conditions. The farther the
distance between BS and SS, RSL value will lower it
means that power received of receiver getting weaker.
......................(6)
If r = 2 m, then:
2
∆PL = −10.8 log  = 0
2
h
- Shadow fading variation (s)
S value can be seen in Table III. For the terrain type B,
the value of s = 9.6 dB. For different types of terrain,
the magnitude of the constants a, b, c, and the shadow
effect (s) which depends on the type of terrain it can be
seen in Table III. Having obtained the required values,
the calculated value of the path loss for NLOS
conditions with a distance of transmitter and receiver as
far as 1 km as follows:
B. Calculation of Energy-bit per Noise (Eb / No)
The calculation of Eb / No value will be used for the
measurement of bit error probability. In the calculation
of Eb / No below, use the under conditions LOS lowest
value of RSL, ie -188.8214 dBm and -106.8852 dBm in
outdoor NLOS and -110.8852 dBm in indoor NLOS.
Condition of Line of Sight (LOS)
The calculation of Eb / No with the bandwidth (B) =
3.5 MHz, using QPSK modulation technique with a
data rate (R) used = 2.6 Mbps is as follows:
 d
PL = A + 10γ log 
 d0

 + ∆ PL f + ∆ PL h + s

 1000 
PL = 83.2942 + 10 × 4.3264 log 
 + 1.4582 + 0 + 9.6
 100 
= 137.6451 dB
In NLOS condition, two types of CPE were used, ie
outdoor and indoor NLOS NLOS with the antenna gain
respectively - each of 17 dBi and 13 dBi. By using
equation (3.7), the calculation of the signal level at the
receiver side with NLOS outdoor CPE and the distance
between BS and SS as far as 1 km can be calculated as
follows:
= -188.8214 – 10 log(2.6x106) + 228.6 dBW –
10 log(273 + 37)
= -165.2973 dB
In the same way it’s possible to obtain the value of Eb /
No using QPSK modulation technique 4 Mbps, 16QAM with data rate 5.3 Mbps and 7.9 Mbps, and 64QAM with data rate 11.9 Mbps and 13.2 Mbps.
…..(7)
On the other hand, the calculation of the signal level
at the receiver side with indoor NLOS CPE and the
distance between BS and SS as far as 1 km can be
calculated as follows:
Conditions Non Line of Sight (NLOS)
In NLOS conditions two types of CPE,outdoor and
indoor were used. Value of Eb / No for outdoor NLOS
using QPSK modulation techniques with a data rate (R)
used = 2.6 Mbps is:
= -106.8852 – 10 log(2.6x106) + 228.6 dBW
– 10 log(273 + 37)
= -83.3611 dB
C7-3
….(11)
Pvs-size= headerUDP/RTP/IPv6+(PLa + PLv)
= 480 + (1920 + 6720)
= 9120 bit
= 1140 byte
Big each packet of audio and video as well as the
number of audio packets and video packets streaming
video applications using IPv6 generated per second can
be calculated as follows:
Pa-size = headerUDP/RTP/IPv6 + Pla
…….….(12)
= 480 + 1920 = 2400 bit
Pv-size = headerUDP/RTP/IPv6 + PLv
…….…(13)
= 480 + 6720 = 7200 bit
Pa = BCODECA/PLa
……………………….(14)
= (64.103)bps / 1920 bit = 33,333 paket/s
Pv = BCODECA/PLv
………………………(15)
= (224.103)bps / 6720 bit = 33,333 paket/s
Ba = Pa-size x Pa
………………………...(16)
= 2400 bit x 33,333 paket/s
= 79999,2 bps = 80 kbps
Bv = Pv-size x Pv
……….………………..(17)
= 7200 bit x 33,333 paket/s
= 239997,6 bps = 240 kbps
Value of Eb / No for indoor NLOS using QPSK
modulation techniques with a data rate (R) used = 2.6
Mbps can be calculated:
= -110.8852 – 10 log(2.6x106) + 228.6 dBW – 10 log(273 + 37)
= -87.3611 4dB
In the same way to obtain the value of Eb / No for
outdoor and indoor NLOS CPE using QPSK
modulation technique 4 Mbps data rate, 16-QAM with
data rate 5.3 Mbps and 7.9 Mbps, and 64-QAM with
data rate 11.9 Mbps and 13.2 Mbps (Table IV).
TABLE IV
VALUE EB / NO OF MODULATION TECHNIQUE IN LOS AND
NLOS CONDITIONS
Eb/No
Mod
QPSK
2.6 Mbps
QPSK
4 Mbps
16-QAM
5.3 Mbps
16-QAM
7.9 Mbps
64-QAM
11.9 Mbps
64-QAM
13.2 Mbps
LOS
Outdoor
NLOS
Indoor
NLOS
2.9530e-017
4.6120e-009
1.8361e-009
1.0952e-017
1.7104e-009
6.8093e-010
5.7289e-018
8.9473e-010
3.5620e-010
2.2851e-018
3.5689e-010
1.4208e-010
8.8969e-019
1.3895e-010
5.5317e-011
7.0075e-019
1.0944e-010
4.3569e-011
So the actual bandwidth of video streaming that is
expressed by the equation 18:
Bvs= Bv + Ba + bandwidth overhead
…….(18)
= 240000 bps + 80000 bps +
{5% x (240000 + 80000)bps}
= 336000 bps
D. Calculation of Loss Packet Video Streaming
The probability of packet loss in streaming video with
headerUDP / RTP / IP is 60 bytes (8 byte UDP header,
RTP header 12 bytes, and 40 byte IP header) and
payload of 840 bytes of video and audio payload of 240
bytes[6], ie:
Table IV shows that the greater the data rate used, the
1.1
smaller the value of bit energy per noise. Energy value
of the smallest bit per noise present in LOS conditions,
while in NLOS conditions, CPE outdoor NLOS noise
energy per bit larger than the indoor NLOS CPE.
= (60 + 840 + 240) x 8 x 10-7
= 9,120 . 10-4
C. Calculation of Bandwidth Video Streaming
Video streaming will be analyzed using the
H.264/AVC video codec with codec bandwidth between
64 kbps - 240 Mbps and AAC-LC audio codec for the
codec bandwidth of 16-576 kbps. The format used is a
CIF image with a frame rate of 30ms. By using
equations 9 and 10 it will get the value of streaming
video data packets on IEEE 802.16d WiMAX network
using IPv6, namely:
Condition of Line of Sight (LOS)
The value of the BER, or often called the probability
of bit error (PBE), using QPSK modulation technique
with 2.6 Mbps data rate and Eb / No = 2.9530e-017 can
be calculated using equation (19).
Pbe . QPSK
PLa = BCODECA x frame rate
…………….(9)
= (64.103)bps x (30.10-3) s = 1920 bit
PLv = BCODECv x frame rate
…………….(10)
= (224.103)bps x (30.10-3) s = 6720 bit
= Q
(

= Q 

2
Eb
No




2 × 2.9530 × 10
-17
= Q (7.6851 x 10-9)
With x = 1.0425 x 10-4, then :
So that large data packets streaming video on IEEE
802.16d WiMAX network using IPv6 by equations 11
are as follows:
C7-4
)
……………(19)
Conditions Non Line of Sight (NLOS)
The calculation of the value of bit error probability
and packet loss for NLOS conditions equal to the
calculations in LOS conditions, where the difference is
only on the value of Eb / No only.
with :
There for,
Then the probability of packet loss in video
streaming WiMAX 802.16d networks with QPSK
modulation is calculated using equation (20), namely:
Fig. III. GraphBit Error Rate of WiMAX 802.16d network
Based on the Fig III., it can be seen that the value of bit
error probability is very small in LOS conditions. While
in NLOS conditions, the probability of bit error in the
NLOS outdoor larger than the indoor NLOS. This can
occur because of differences in CPE specifications are
used in both circumstances.
On the other hand, according to equation (21), the
probability of bit error on QAM modulation can be
calculated as:
Table V.
Probability Packet Loss Video Streaming
Probability of Packet Loss
Type
of
Outdoor
Indoor
Modulation LOS
NLOS
NLOS
QPSK 2,6
9.1200e-004 9.1200e-004
9.1200e-004
Mbps
QPSK
4
9.1200e-004 9.1200e-004
9.1200e-004
Mbps
16-QAM
9.1200e-004 9.2000e-004
9.1705e-004
5,3 Mbps
16-QAM
9.1200e-004 9.1705e-004
9.1519e-004
7,9 Mbps
64-QAM
9.1200e-004 9.1346e-004
9.1292e-004
11,9 Mbps
64-QAM
9.1200e-004 9.1330e-004
9.1282e-004
13,2 Mbps
For 16-QAM,
 M 1 / 2 − 1 
3 log 2 M  Eb 



1 − erfc
log 2 M  M 1 / 2 
2( M − 1)  No 
1/ 2
Pb16 −QAM ==
Pb16 −QAM ==
2
3 
3 log 2 16
× × 1 − erfc
5.7289 × 10−18
log 2 16 4 
30
)
2
(

...(21)

1/ 2



Then the probability of packet loss in streaming
video 802.16d WiMAX network with 16-QAM
modulation is calculated using equation (22) namely:
For 64-QAM,
Pb 64 − QAM ==
Pb16 − QAM ==
 M 1 / 2 − 1

 1 − erfc
log 2 M  M 1 / 2  
2
2
log 2 64
×
7 
× 1 − erfc
8 
3 log 2 M  Eb 


2 ( M − 1)  No 
(
1/ 2
3 log 2 64
8 . 8969 × 10 −19
126



)
1/2



Then the probability of packet loss in streaming
video 802.16d WiMAX network with 64-QAM
modulation is calculated using equation (3.22) namely:
Fig. IV. Graph of Packet Loss probability of Video Streaming on
LOS and NLOS condition
C7-5
[4] Datasheet CPE Easy ST dan Pro ST Airspan
Network Inc.
[5]Kwang-Cheng Chen and J. Roberto B. de Marca,
2008 Mobile WiMAX. London : John Wiley &
Sons
[6]Forouzan, Behrouz. 2000. Data Communication and
Networking. United States : McGraw-Hill
[7]Freeman, Roger L. 1994. Reference Manual for
Telecommunications Engineering 2nd Edition.
Toronto : John Wiley & Sons
Fig. IV shows the relationship between the packet loss
probability of video streaming with data on LOS and
NLOS conditions. In QPSK modulation type, ie the data
rate of 2.6 Mbps and 4 Mbps, the packet loss
probability is very small streaming video since the
value of bit error is also small. While on the QAM
modulation type, the higher the data rate, the lower the
probability of packet loss video streaming.
IV. CONCLUSIONS AND RECOMMENDATIONS
Based on the calculation and analysis of the video
streaming performance on the 802.16d WiMAX
network, the conclusion is obtained as follows: The
packet loss probability of video streaming in LOS
conditions have the same value on all modulation
techniques, ie 9.1200 x 10-4. This can occur because
the value of bit error probability is very small. On the
other hand, in NLOS conditions by using QPSK
modulation technique the value is 9.1200 x 10-4. Using
QAM techniques, the value of packet loss probability is
inversely proportional to data rate.
First author, Dwi Fadila Kurniawan received the Master Degree
in CDMA (Code Division Multiple Access) Multimedia from the
Institute of 10 November, Surabaya, in 2001. He worked as a lecturer
in electrical engineering departement the University of Brawijaya,
Malang, Indonesia. His research has been in the areas of microwave,
antenna propagation, and mobile communication.
The second author, Muhammad Fauzan Edy Purnomo was born
in Banjarmasin, Indonesia, in June 1971. He received the B.E. and
M.E. degrees in Electrical Engineering from University of Indonesia,
Jakarta, Indonesia in 1997 and 2000. He is presently with the
Electrical Department University of Brawijaya, Malang, Indonesia
where he is working toward as lecturer. His main interests are in the
areas of microwave, mobile communication, microstrip antennas,
array antenna for mobile satellite communications, and Synthetic
Aperture Radar (SAR). He has been ever be a student member of the
IEICE and IEEE.
References
[1]Schwartz, Mischa. 1987. Telecommunication
Network. Addison-Wesley.
[2]Andrews, Jeffrey G., Arunabha Gosh, Rias
Muhamed. 2007. Fundamental of WiMAX :
Understanding Broadband Wireless Networking.
Massachusetts : Pearson Education Inc.
[3]Datasheet Base Station MicroMaxd Airspan Network
Inc.
The third author, Widya Rahma received the engineering degree
in telecommunication from Electrical Department University of
Brawijaya in 2010. His research has been in the areas of the mobile
communication and microwave.
C7-6
Statistical Beam Propagation of Terrestrial FreeSpace Optical Communication Using
Gamma-Gamma
Ucuk Darusalam, 1Purnomo Sidi Priambodo, Harry Sudibyo and Eko Tjipto Rahardjo
Study Program of Opto-Electrotechnique & Laser Application
Department of Electrical Engineering, Faculty of Engineering
Universitas Indonesia, Depok, Indonesia
E-mail: 1pspriambodo@ieee.orgpspriambodo@ee.ui.ac.id
Abstract—We present an experimental of Free-Space
Optical Communication (FSOC) system characteristical
performance in a turbulence medium. The FSOC system is
fiber detection via TPS (Tube Propagation Simulator)and
using 1550 nm optical modem as the main source of
communication and EDFA with output of +23 dBm).The
index structure of 10-15 - 10-13as representation atmosphere
index turbulences areused for calculateintensity distribution
model (scintillation) using gamma-gamma. The results of
experiment shows that in the weak to moderate scale of
turbulence, highest value of mean SNR and high quality
mean BER are achieved for spherical waves. While from
measurements the <BER> is in the range of 10-6 – 10-11.
Keywords-component; free-space optical communication
(FSOC), turbulence media, Scintillation,< Prfade>,< BER>.
I. INTRODUCTION
Free-Space Optical Communication (FSOC) system
has been implemented widely in many contries so rapidly
because its provide high link of capacity, free-license,
low cost of deployment, easy of maintenance, and could
be integrated with existing communication system [1][3]. The latest development of FSOC is used as an
integrated space-terestrial network e.g. to enhance
communication links for satelite to satelite crosslinks, upand-down between space platforms and aircraft, ships,
and other ground platforms, and among mobiles and
stationary terminals terestrial [4]. The potential of link
capacity of FSOC have been achieved at the scale of 4.10
Gb/s with the length of transmission 2.4-Km amplified by
Erbium Doped Fiber Amplifier (EDFA) [5]. Enormous
bandwidth also have been investigated by modulated
32x40 Gbit/s of WDM system in FSOC over 1,2-Km
using laser diode 1550-nm [6]. FSOC also has been
integrated with the broadband network in Japan with the
length transmission of 2-Km by implementing 800-nm
laser diode and tested on natural environment such as rain
and fog [7]. Another work also showed that FSOC has
many advantages of free of EMI, inexpensive
deployment and more faster, while RF signal was
transported through the link [8]. Evenmore FSOC have
been used as wireless broadband in order to support the
optical fiber system in metropolitan area by using LED
transmitter [9].
The major problem of FSOC system is the media of
propagation is atmosphere, which its natural
charactheristics of light for example light attenuation
caused by the O-H absorption such as rain, fog, and
snow. Other degradation are caused by Rayleigh and Mie
scattering and difraction as well. The physical properties
of atmosphere are random fluctuation in temperature,
pressure and wind speed. These charactheristics cause the
medium of the atmosphere behave random fluctuation
index of refraction. The random fluctuation of the
propagation medium is called turbulence, which the size,
dimension and density of the air change randomly in
space and time. Due to index of refraction fluctuates
randomly, the intensity of light that propagate along the
medium suffer attenuation and scintillation. Those all
phenomena finally degrade the strength of signal
performance of FSOC in the receiver system.
Some research works have been devoted to study
these phenomena intensively in order to enhance the
FSOC performances and mitigate the effect of
turbulence. The signal strength suffers from degradation
and also has fluctuation of intensity at the receiver then
the FSOC system is designed to be amplified by EDFA.
The EDFA is configured in a saturated regime condition
to boost the signal strength at the receiver, in order to
mitigate signal successfully[10] [11]. The effect of
turbulence also cause the beam wandering in receiver
side. To overcome this wandering problem, an array
detector system has been implemented at receiver and
have been reported succesfully [12]. Another work also
studied intensively in transmitter system in order to
mitigate the effect of turbulence. The multi input multi
output (MIMO) method is implemented, where the
multiple laser diodes are applied at the transmitter and
multiple photodetectors are applied at the receiver
system. The MIMO system is reported succesfully
overcoming the turbulence effect and enhance the FSOC
performance [13].
In this work we use the FSOC of fiber detection
method in order to analize its performance in turbulence
medium. The turbulence media is using Tube Propagation
Simulator (TPS) that designed to capable simulate the
turbulence as well as at atmosphere. We use optical
modem of 1550 nm as the main source of optical
communication equiped with EDFA with output of +23
dBm. The output beam of EDFA is collimated and
transmitted through TPS, passing the turbulence media.
C8-1
TPS is modelled in laboratory by a simulation of using of
plastic tube filled with the flow of hot water vapour
(steam) and mixed with the cold air at temperature of
160C. This turbulence media is called random media of
tube propagation simulator. The motivation of this work is
to investigate the characteristical performance influenced
by scale of turbulence that is designed from weak to
strong turbulence.
II. BASICS THEORY
Free-Space optical communication system is
lightwave communication that using free-space for
propagation medium rather than optical fiber. The
fundamental difference of FSOC and optical fiber
communication systemis that FSOC is not guided at freespace, whereas the optical fiber is guided and immune
from the surrounding noise. The FSOC are consists of
three main parts TX, Free-space terestrial medium, and
RX [14]. FSOC consist of laser diode as an optical
transmitter (TX) and photodiode as the optical detector
(RX) gathered with the optical system to direct the
optical beam [14]. The laser beam is collimated via
optical lens configuration system, which called telescope
transmitter to direct the optical beam through air to reach
the receiver lens and focused into the photodetector as as
shown in Fig. 1.
=
Where η is the quantum eficiency of photodetector, Ps is
the received signal intensity, ν is the light frequency, h is
planck constant, and B is the bandwidth. Eq. 1 is the
value of signal to noise ratio (SNR) by ignoring the
background illumination, circuit and thermal noise which
called limited shot-noise. The parameter of BER (bit
error rate) in the form basic modulation OOK (on-off
keying) is represented as [15]:
=
(2)
√
Eq. 2 represents the value of BER by considering the
random noise in the photodetector that lead to mistaken
bit from 0 to be 1 or vice versa [15].
In the case of turbulence in weak scale that lead to
irradiance fluctuation or scintillation the governing
equation PDF (probability of density function) on the
photodetector is the lognormal and modelled in Eq. 3:
Pr
〈
Direct Detection Method
Beam
Expander
〈
~
>= $ %& (
=< "
√
〉+
,((3)
Where-& .0, 1 + 13 4 = -& 56 is the flux variance and
dependent upon the diameter of aperture. Due to the
nature of turbulence is random fluctuation, the value of
SNR is no longer of deterministic but rather than mean
value and can be expressed as [15]:
Atmosphere
T
X
(1)
R
X
〉 =
8 9: < =
; >
(4)
<
In the presence of optical turbulence the PDF (Eq.3) is
considered as conditional probability that must be
averaged over the PDF of the signal in order to determine
the unconditional mean of BER [15]:
Receiver Lens
Figure 1. The FSOC system.
There are several advantages of FSOC system which
are listed as implementing smaller antenna (telescope),
smaller size and weight of the components, power
concentration in a very narrow beam, and enormous
bandwitdh. FSOC, furthermore is considered to be more
compact, simple configuration device, and inexpensive
compared to its technological competitor such as optical
fiber and microwave communication. For that reason, it
is now being developed so vastly for many areas of
communication. The simple configuration of FSOC
system is the direct detection method, which means at the
receiver side the optical beam is directly collimated by a
lens onto photodetector, as shown in Fig. 1. The benefit
of Direct Detection method are simple and unnecesary to
use the optical fiber as the point of focus spot from the
receiver lens. This method also reduce the effect of beam
wandering. However there is a disadvantage of direct
detection method due to shot noise caused by influenced
of environment temperature outdoor. Also a mandatory
requirement to locate the detector outside the door.
Moreover it requires the optical filter to reduce the
background noise of another optical sources such as
comes from the sun or another that may be detected by
the photodetector.
When the turbulence is assumed absence on the
medium of propagation or in the atmosphere the SNR is
represented as [15]:
Pr
〈
~
〉+
>= $ %& (
,(
(5)
√
%& ( is the gamma-gamma distribution of unit mean as
the representation of PDF:
&
=< "
( =
?@ ABC
D E D F
(
?9@ / H
I?H@ .2KLM(4,( > 0 (6)
And for the case of spherical wave the parameter of α and
β as the representation of atmospheric trubulences, are:
L=
(7)
M=
S
b
R
a
.UVC<
NOP R
_] aH
\
X<]
RWXB .XYZ< B .[\C [ ^ à
Q
c[]
\b
X<]
S
<
[^
R .[XC WXB .\VC
a
NOPR
aH
X<]
RWXB .VZ< B .\<Z< C [ ^ à
Q
(7)
III. METHOD OF EXPERIMENT
The FSOC system on the experiment is shown in
Fig.2. And the TPS (Tube Propagation Simulator) is also
shown in Fig. 3. While the the scheme of turbulence scale
is shown in Table 1. Optical modem of 1550 nm is used
in FSOC system, equiped with EDFA with gain of +23
C8-2
a. <Prfade> of strong turbulence
dB. Beam collimator is collimating the beam output of
EDFA transmitted through turbulence media in TPS and
reach the lens focuser at Unit RX. Focused beam from
lens focuser directed into fiber (SMF/MMF) to measure
the scintillation by Power Meter.
○Plane Wave
*Spherical Wave
▼Gaussian Wave
b. <BER> of strong turbulence
○Plane Wave
*Spherical Wave
▼Gaussian Wave
c. <Prfade> of moderate turbulence
Figure 2. Experiment diagram of FSOC fiberdetection method.
○Plane Wave
*Spherical Wave
▼Gaussian Wave
d. <BER> of moderate turbulence
Figure 3.Set-up of turbulence medium as the optical beam
propagation (random media of tube propagation simulator).
○Plane Wave
*Spherical Wave
▼Gaussian Wave
Table. 1. The scale of turbulence designed in TPS.
Turbulence
Exhaust
Intake Cold
Steam
Fan
Air
Generator
Off
Off
Off
No
Turbulence
On
On
Off
Weak
(TAC = 16 0C)
Turbulence
On
On
On
Moderate
(TAC = 16 0C)
(TSG = 30-50
Turbulence
0
C)
On
On
On
Strong
(TAC = 16 0C)
(TSG> 510C)
Turbulence
e. <Prfade> of weak turbulence
IV. RESULTS AND DISCUSSIONS
○Plane Wave
*Spherical Wave
▼Gaussian Wave
The results of simulation with gamma-gamma
distribution is compute the means Probability of fade
(<Prfade>) and BER (<BER>) for each scale turbulence
and model of beam waves (plane wave, spherical wave,
dan gaussian wave). The results are shown in Fig 4.a-f.
f. <BER> of weak turbulence
○Plane Wave
*Spherical Wave
▼Gaussian Wave
Figure 4.The characteristical performances of FSOC system with
gamma-gamma simulation for wach beam waves model at three scale of
turbulences.
While the results of measurement of FSOC system at
TPS for the various scale of turbulence are shown in Fig.
5.a – f:
C8-3
Table 2. The characteristics of means Probability of fade<Prfade> and
<BER> from simulation of beam waves.
Strong Turbulence
Turbulence
Conditions
Srong
Turbulence
Moderate
Turbulence
Weak
Turbulence
a. <Prfade> of strong turbulence
Strong Turbulence
Charactheristics of <Prfade>&<BER>
<Prfade>&<BER>achieved at lowest order at
spherical wave
<Prfade>&<BER>achieved at lowest order at
gaussian wave.
<Prfade>achieved at the same order for all
beam waves
<BER> achieved at lowest order at spherical
wave.
The comparison of results from simulations and
measurements can bes summarized as shown on Table 3
as follows:
Table 3. The characteristics of <Prfade> and <BER> from
measurements.
Turbulences
Simulations
Measurements
Orde
Orde
Orde
Orde
<Prfade> <BER> <Prfade> <BER>
-2
-4
-4
10
10
10
10-6
Srong
Turbulence
10-2
10-4
10-4
10-6
Moderate
Turbulence
10-1
10-8
10-10
10-11
Weak
Turbulence
b. <BER> of strong turbulence
Moderate Turbulence
From the the results of simulation and measurements
can be well understood that:
• The order <Prfade> and <BER> is getting lower
as the rise of <SNR>
• The order <Prfade> and <BER> is getting lower
as the rise of scale turbulences.
• The difference results of simulation and
measurements is caused by the index
scintillation in gamma-gamma model is the
means from three beam waves model.
• The results from measurements is more exact
value from the scintillation distribution of the
received power.
• The gamma-gamma simulation by all means is
approaching the results of measurements.
• The highest characteristical performances of
FSOC system is well achieved at strong
turbulences with the order of <Prfade> and
<BER> are 10-1, 10-8, 10-10, and 10-10,
respectively.
c. <Prfade> of moderate turbulence
ModerateTurbulence
d. <BER> of moderate turbulence
Weak Turbulence
e. <Prfade> of weak turbulence
Weak Turbulence
f. <BER> of weak turbulence
Figure 5. <Prfade> and <BER>measurement in three scales of
turbulences.
From the results of simulation using gamma-gamma
which the distance of propagation is L = 1000 m can be
summarized as the Table 2 as follows:
The effect of intensity deterioration caused by the
random absorption, diffraction, and scattering of beam
wavesthat cause a random mean SNR. The higher
intensity fluctuation on the photodetector causes the
probability of fade become more higher also. The fade
probability means that the profile intensity is fluctuated
by the beam wandering at the receiver as well. The beam
spot also moves randomly, hence the photodetector
receive a random wandering at the same time. This can
degrade the performance of the FSOC system in fiber
detection method. The higher of scale of turbulence will
rise up the fade probability, hence the mean BER is going
to lower quality as the low of mean SNR. The high value
C8-4
of mean SNR will cause the mean BER also goes to a
high quality.
The beam wandering occurs along the random
medium (tube propagation simulator) can be explained by
considering that the profile of Gaussian beam from
transmitter is spherical wave. The divergence of the
optical beam will occur when the random medium exhibit
random index refraction structure. Due to this divergence,
the mean SNR also decreases signifficantly, espescially
in the case of strong turbulence. Beam wandering also
exhibits the strong scintillation in the photodetector hence
this lead to degrade the intensity of received power,
hence it causes the mean SNR decrease.
In order to enhance the performance of FSOC system
in the scheme of fiber detection method, a new technique
is required to elevate the value of mean SNR. The mean
SNR value can be elevated by rising the received power
in the photodetector and minimize the effect of
scintillation or fluctuation signal power. Rising the power
received by photodetector could be achieve using large
aperture lens at the receiver, to minimized effect of the
beam divergence, due to turbulence. The large aperture
of receiving lens could anticipate the spot movement of
Gaussian beam hence still focused to sensing area of
photodetector as well. While to minimize the strong
fluctuation of the received power or scintillation could be
achieved by using spatial diversity. By using the spatial
diversity system, scintillation in the photodetector can be
minimized [16]. Another technique to reduce the effect of
scintillation due to turbulence is by using photodetector
with a large sensing area. This technique is called array
photodetector, which sensing large optical beam after
propagate through the random media and mixed the
signal output from each photodetector into the signal
processing method in order to maintain the BER quality
[12].
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
CONCLUSION
We report the characteristical performance of FSOC
system in the fiber detection method by using random
media of tube propagation simulator. TPS is employing
steam of hot water vapour that mixed with cool air from
Air Conditioner.By those could be obtained conditions of
Rytov variance factors at various scale turbulence (weak
to strong scale). From the simulation and experiments the
lowest order of Probability of fade <Prfade> and <BER>
is achieved at weak turbulence which the characteristical
of <BER> are 10-8 by simulation and 10-11 by
measurement. Hence the performance of system degrade
by the presence of turbulence. On the other hand, for
weak, moderate and strong scale of turbulencesstill
contribute higher mean of SNR and BER, it means that
the characteristical performances of FSOC system is high
(lowest order of <Prfade> and <BER>). From the
experiments the quality of BER is still in the limit of
system performance, i.e. in the range of 10-4 – 10-8 and
10-6 – 10-11.
[11]
[12]
[13]
[14]
[15]
[16]
C8-5
H. E. Nistazakis, T.A. Tsifsis, and G.S. Tombras, “Peformance
analysis of Free-space optical communication systems over
atmospheric turbulence channel, IET Communication,” vol. 3 iss.
8, pp. 1402 – 1409, 2009.
E. Ciaramella, Y. Arimoto, G. Contestabile, M. Presi, A. D’Errico,
V. Guarino, and M. Matsumoto, “1.28 Terabit/s (32x40 Gbit/s)
WDM ransmission System for Free Space Optical
Communications,” IEEE Journal on Selected Areas in
Communications, vol. 27, no. 9, pp.1639 - 1645, Dec. 2009.
Kazuhiko Wakamori, Kamugisha Kazaura, Member, IEEE, and
Ikuo Oka, Experiment on Regional Broadband Network Using
Free-Space-Optical Communication Systems, Journal of
Lightwave Technology, vol. 25, no. 11, pp. 3265 - 3273, Nov.
2007.
Vincent W.S. Chan, “Free-Space Optical Communications,”
Journal of Lightwave Technology, vol.24 no.12, pp. 4750-4762,
Dec. 2006.
Dong-Yiel Song et al., “4 × 10 Gb/s terrestrial optical free space
transmission over 1.2 km using an EDFA preamplifier with 100
GHz channel spacing,” Optic Express, Vol. 7, No. 8, pp. 1634 –
1645, Oct. 2000.
E. Ciaramella, "1.28 Terabit/s (32x40 Gbit/s) WDM Transmission
System for Free Space Optical Communications," IEEE Journal
on selected areas in communications, Vol. 27, No. 9, pp. 1639 1645, Dec. 2009.
Kazuhiko Wakamori, "Experiment on Regional Broadband
Network Using Free-Space-Optical Communication Systems,"
Journal of Lightwave Technology, Vol. 25, No. 11, pp. 3265 3273, Nov., 2007.
Hakki H. Refai, Transporting RF Signals over Free-Space Optical
Links, Free-Space Laser Communication Technologies XVII,
edited by G. Stephen Mecherle, Proceedings of SPIE Vol. 5712,
pp. 46 - 54, 2005.
E. Leitgeb, Free Space Optics – Broadband Wireless Supplement
to Fiber-Networks," Free-Space Laser Communication
Technologies XV, G. Stephen Mecherle, Editor, Proceedings of
SPIE Vol. 4975, pp. 57 - 68, 2003.
Mohammad Abtahi et al., "Suppression of Turbulence-Induced
Scintillation in Free-Space Optical Communication Systems
Using Saturated Optical Amplifiers," Journal of Lightwave
Technology, Vol. 24, No. 12, pp. 4966 - 4973, Dec., 2006.
Yoon-Suk Hurh et al., Weather-Insensitive Optical Free-Space
Communication Using Gain-Saturated Optical Fiber Amplifiers,“
Journal of Lightwave Technology, VoL. 23, No. 12, pp. 4022 4025, Dec., 2005.
Shiomi Arnon & Norman S. Kopeika,"Free-space optical
communication: detector array aperture for optical communication
through thin clouds," Optical Engineering 34(2), pp. 518-22,
February 1995.
Stephen G. Wilson et al.,"Free-Space Optical MIMO
Transmission With Q-ary PPM," IEEE Transactions on
Communications, Vol. 53, No. 8, pp. 1402-1411, Aug., 2005.
Olivier Bouchet et. al, Free-Space Optics, London W1T 5DX
Newport Beach, CA 92663 UK, 2004.
Larry C. Andrews and Ronald L. Philips, Laser Beam Propagation
through Random Media, 2nd Ed., SPIE Press, Washington USA,
2005.
W.O. Popoola et al., “Free-space optical communication
employing subcarrier modulation and spatial diversity in
atmospheric turbulence channel," IET Optoelectron., Vol.2 No.1),
pp. 16–23, 2008.
Maximum Power Point Tracking
Using Fuzzy Logic Control for Buck Converter
in Photovoltaic System
Mahendra Widyartono1), Sholeh Hadi Pramono2), and M. Aziz Muslim3
1)
Student of Master Degree Program, 2)3)Lecturers
Electrical Engineering Department, Engineering Faculty, Brawijaya University, Malang, Indonesia
mahe.lucretia@gmail.com, sholeh_hp@ub.ac.id, muh_aziz@ub.ac.id
Abstract—Photovoltaic (PV) systems are power
source systems that have non-linear current – voltage
characteristics (I-V) under different environments
condition. The system consists of PV generator (cells,
modules, PV array), energy storage (batteries), buck
converter, and resistive load. The proposed maximum
power point control is based on fuzzy logic to control the
switch of the buck converter. Buck converter is used to
convert DC input voltage that varies into controlled DC
output voltage at a desired output voltage. Voltage and
current output from PV module were used as input
parameters of fuzzy control to generate optimum duty
cycle so that maximum power can be generated in
varying operating condition. With fuzzy MPPT, the
current and voltage through the load is drop from 1,10
A to 1,06 A and from 16,60 V to 15,95 V. Using proposed
maximum power point tracking (MPPT) method, the
system have better stability even in dynamic operating
conditions.
Index Terms—Photovoltaic system, Maximum Power
Point Tracking, Fuzzy Logic Controller, Buck Converter.
I.
INTRODUCTION
Photovoltaic energy applications have been
increasing along with the rapid depletion of
conventional energy sources such as petroleum,
natural gas and coal [1]. These applications include
water pumps, refrigerators, air conditioners, vaccine
storage, electric vehicles, military and aerospace
application. PV energy considered to be the primary
energy in many countries that have a large solar
iradiation. PV system technology developed rapidly
along with the development of technology in power
systems to provide safe and pollution-free energy
sources.
PV system is power source system with non-linear
I-V characteristic under different environments
conditions (temperature and solar iradiation) [1]. The
system consists of PV generator (cells, modules, PV
array), buck converter, energy storage (batteries) and
resistive load. The simplest PV system has no
electronic control [2]. This simple system can not
control the PV system to generate maximum power.
To overcome this limitation, electronic circuits are
introduced to control the battery charging, conversion
DC to DC voltage and convert DC voltage to AC
(inversion).
D1-1
DC-DC converter is used to convert the DC input
voltage that varies into controlled DC output voltage at
the desired voltage level. The basic form of DC-DC
converter is buck converter, are also called step-down
converter. As the name implies, step-down converters
produce a dc output voltage of the average lower than
the input dc voltage. In DC-DC converter, average
output voltage is controlled by the duration of the
switch on and off (ton and toff). This method is known
as pulse-width modulation (PWM) switching [3].
Maximum Power Point Tracking (MPPT) is a subsystem designed to extract maximum power from
power source [4]. In the case of solar power source,
the maximum point varies due to the influence of
changes in electrical characteristics as function of
temperature, solar iradiation, heating and others. With
the change of temperature and solar iradiation, the
voltage and current output of the PV modules are also
changing and reducing efficiency of PV systems.
MPPT maximizes power output of the panels in
different conditions to detect the best working point
of the power characteristics and then controls the
current or voltage on the panel [4].
General requirement for MPPT is simple and low
cost, fast tracking the changing conditions, and
fluctuations of small output [5]. More efficient
methods for solving this problem becomes very
important. Fuzzy Logic Controller (FLC) is suitable
to control a non-linear systems through manipulating
its membership function and rule base. Choosing the
parameters to obtain the maximum operating point
and a good control system depends on the designer
experience [5]. Due to the nature of PV system is
non-linear, i.e. current and voltage that varies
depending on environmental conditions, it is very
important to operate the PV system at the condition of
maximum power point. This will improve the
efficiency of PV systems.
II.
PHOTOVOLTAIC MODELING
A. Photovoltaic cell model
The simplest equivalent circuit of PV cell is
current source paralel with a diode (Fig. 1). The
output of current source proportional to the amount of
solar iradiation on the PV cell (IPV). In this model, the
May 30-31,
open circuit voltage andd short circuit current is a key
parameter [6].
This point represents
epresents the maximum efficiency in
convertering
ing sunlight into electricity [7].
[7
Figure 1. Equivalent circuit of one diode PV model.
model
Short circuit current depends on the intensity of
sunlight, while the open circuit voltage is affected by
the material and temperature. Equations of this model
are :
1 "1$
/
"2$
ln
1
1*"3$
I-V
V characteristic of PV cells can be defined as
follows :
exp
!"
#
%&
$
'
1( #
#
"4$
Where IPH is the photovoltaic current, ID is the diode
current, RS is the series resistance and RP is paralel
resistance of PV cells.
B. I-V curve of PV cells
The relationship of current-voltage
voltage is used to
measure the electrical characteristics of PV devices.
devices
The I-V curve describes the flow of voltage through
the imposition of a short-circuit
circuit current ISC to open
circuit voltage VOC. This curve is used to obtain the
level performance of PV systems (cells,
cells, modules, PV
array). I-V
V curve is obtained by performing
experiment with exposing PV cells or modules at the
level of constant iradiation, maintaning the cell
temperature, varying the load resistance, and then
measuring the resulting current and voltage
volt
[7]. I-V
curve for Wuhan Rixin MBF75 PV modules can be
seen in Figure 2. Horizontal axis is for voltage and
vartical axis is for current.
PV cells can operate on a wide range of areas of
current and voltage. Simply by varying the load
resistance from zero
ero (short circuit) to infinity (open
circuit), it is possible to determine the highest
efficiency PV cells deliver maximum power. Because
power is the result of voltage multiplied by current,
then the point of maximum power (Pm) appears in the
I-V curve where the outcome of the current (Imp)
multiplied by voltage (Vmp) is maximum. No power is
generated on the condition of short circuit or open
circuit conditions, so that maximum power is
generated only at one point on the curve called “knee”.
Figure 2. I-V characteristic for Wuhan Rixin MBF75 PV module
III.
FUZZY MPPT
The purpose of fuzzy control is to extract
maximum power from PV modules at a certain
certa level
of solar iradiation [8]. Fuzzy logic control has several
advantages such as suitable for use on systems that are
not linear and cann work with imprecise inputs.
inputs Fuzzy
control (Fig. 3) using input voltage (V) and current (I)
and generates a duty cycle (D) output
outp that used as
buck converter input. The input voltage and current for
fuzzy control is from the output voltage and power
from PV module. Fuzzy logic control consist of three
stages : fuzzification, inference method and
defuzzification.
Figure 3. Fuzzy logic controller.
controller
A. Fuzzification
Fuzzification stage is a stage where the input
variable was changed into the language (linguistic)
based on the membership
mbership function. Triangular and
trapesium membership function with seven fuzzy
subsets VVS (very veryy small), VS (very small), S
(small), M (medium), B (big), VB (very big), VVB
(very very big) is used (Fig. 4). Variable V and I are
used as input, and variable D as output.
B. Inference method
At this stage, Mamdani method is used to control
the generated output.
utput. The design of basic rules (rule
base) consists of 49 fuzzy control rules. This rule
implemented by computer and used to control the duty
cycle of buck converter in order to obtain maximum
power from PV modules in different conditions. These
rule were expressed as IF-THEN
THEN statements as follows
R1 : IF V is VVS and I is VVS THEN D is VVB.
D1-2
R2 : IF V is VVS and I is VS THEN D is VVB.
Fuzzy rule base used for fuzzy MPPT can be seen
in Table 1.
TABLE I.
FUZZY RULE BASE
I
VVS
VS
S
M
B
VB
VVB
VVS
VVB
VVB
VVB
VVB
VB
VVB
VVB
VS
VVB
VVB
VVB
VB
B
M
S
S
VVB
VVB
VB
B
M
S
VS
M
VVB
VB
B
M
S
VS
VVS
B
VB
VVS
VVS
S
VS
VVS
VVS
VB
B
M
S
VS
VVS
VVS
VVS
VVB
M
S
VS
VVS
VVS
VVS
VVS
(a)
C. Defuzzification
At this stage, the outputs of fuzzy logic control is
changed from linguistic variables into numeric
variables using membership function. With
defuzzification, fuzzy logic control can generates
analog output signal that can be converted into digital
signals and control the power converter of MPPT
system. Centroid type of defuzzification is used for
this research.
IV.
(b)
SIMULATION OF FUZZY MPPT
The MATLAB/Simulink software is used for the
simulate fuzzy MPPT with PV module and resistive
loads. The system consist of :
A. PV Module Block
Simulate the non-linear I-V characteristic of
Wuhan Rixin MBF75 PV module. Table 2
summarized specifications of the PV module.
TABLE II.
PV MODULE SPECIFICATION
Brand
Model
Material
Power output (max)
Voltage output (max)
Current output (max)
Working temperature
Wuhan Rixin
MBF75
Polycrystalline Silicon
75 W
17,5 V
4,29 A
21,5 V
4,72 A
-0,35% / °C
+0,036% / °C
- 40 ~ 90°C
(c)
Figure 4. Membership function : (a) first input V, (b) second input
I, (c) output D.
B. Fuzzy Controller Block
Simulates the fuzzy MPPT process and computes
the desired duty cycle of the buck converter using
solar panel voltage and current. The fuzzy controller
block performs the fuzzification, inference method,
and defuzzification process.
C. Buck Converter Block
DC – DC converter is used to convert the DC input
voltage that varies into controlled DC output voltage at
the desired voltage level. Buck converter (Fig. 6)
produces a dc output voltage of the average lower than
the input DC voltage. A capacitor (C) with a value of
2200 µF is used to reduced module ripple voltage.
The equation of the buck converter circuit is as
follows :
-./
-0
<
D1-3
1
2 3 . - − ./ #/ − 5 6(5)
1
1
= (./ − .59: )(6)
-0
8
= + #=>? (./ − .59: )(7)
Figure 5. Simulation of fuzzy MPPT system using MATLAB/Simulink.
Cu
rre
nt
(A)
V
1000 W/m2
200 W/m2
Figure 6. Buck converter circuit (L=1mH, C=2200µF,
RL=80mΩ, RC=5mΩ).
t (s)
D. PWM Block
Generates the pulse signals for the buck converter
based on the desired duty cycle.
Figure 7. Load current without fuzzy MPPT and with fuzzy MPPT
E. Load
A 15 ohm resistive load is connected to PV
module via buck converter.
V.
RESULT AND ANALYSIS
This chapter contains the results and analysis of
fuzzy MPPT simulation system (Fig. 5). Fuzzy MPPT
simulation uses two operating conditions. Case 1
without the fuzzy MPPT (e.g., direct connection of
module PV and the load). Case 2 with fuzzy MPPT.
The solar iradiation used for simulation is vary
between 200 ~ 1000 W/m2. Figure 7 and 8 show the
load current and voltage characteristic of the two
conditions.
With the change of solar iradiation, the current
through the load R = 15 ohms also changed. Figure 7
shows that in case 1 (without fuzzy MPPT) by
changing of solar iradiation from 1000 W/m2 to 200
W/m2 resulted the load current drop from 1,38 A to
0,81 A. While case 2 (using fuzzy MPPT), the load
current drop from 1,10 A to 1,06 A.
This condition is also hold for the load voltage
(Fig. 8). In the condition without fuzzy MPPT (case
1), the load voltage drop from 20,74 V to 12,18 V.
While using the fuzzy MPPT (case 2) the load voltage
down from 16,60 to 15,95 V. These results indicates
that PV system using fuzzy MPPT has better
performance than PV system without using fuzzy
MPPT.
Vo
lta
ge
(V)
1000 W/m2
200 W/m2
t (s)
Figure 8. Load voltage without fuzzy MPPT and with fuzzy MPPT.
VI.
CONCLUSION
From the simulation results and analysis, it can be
concluded that PV system using fuzzy MPPT has
better performance than the system without MPPT.
This can be seen on the load current and voltage
curves which are more stable than the system without
MPPT. System with fuzzy MPPT have better stability
even in dynamic operating conditions.
D1-4
Mahendra Widyartono received Bachelor
Degree from Institut Teknologi Sepuluh
Nopember, Surabaya, Indonesia, in 2006, in
electrical engineering. Currently, he is
working toward Master Degree in Electrical
Engineering Department at Brawijaya
University, Malang Indonesia. His current
research interest is solar power system and
renewable energy.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
Masoum, M, A, S an Sarvi, M. 2005. A new fuzzy-based
maximum power point tracker for photovoltaic applications.
Iranian Journal of Electrical & Electronic Engineering, Vol. 1.
Iran.
Markvart, T and Castaner, L. 2003. Practical Handbook of
Photovoltaics Fundamentals and Applications. Elsevier
Advanced Technology. Oxford, Inggris.
Mohan N, Undeland T, M, and Robbins, W, P. 1995. Power
Electronics. Converters, Applications, and Design. (2nd
Edition). John Wiley & Sons, Inc.
Daoud, A, Midoun, A. A Fuzzy Logic Based Photovoltaic
Maximum Power Tracker Controller. Department of
Electronics, Faculty of Engineering, University od Sciences
and Technology of Oran. Algeria.
Patcharaprakiti N., et al. 2005. Maximum power point
tracking using adaptive fuzzy logic control for grid-connected
photovoltaic system. Elsevier Ltd.
Khaligh, A and Onar, A, C. 2010. Energy Harvesting : Solar,
Wind, and Ocean Energy Conversion Systems. CRC Press
Taylor & Francis Group. Boca Raton, Florida.
Foster, R., et al, A. 2010. Solar Energy Renewable Energy
and the Environtment. CRC Press Taylor & Francis Group.
Boca Raton, Florida.
Simoes M,G, Franceschetti N, N, Friedhofer, M. 2008. A
Fuzzy Logic based Photovoltaic Peak Power Tracking
Controller, IEEE-ISIE International Symposium on Industrial
Electronics, Vol. 1,pp. 300-305.
Sholeh Hadi Pramono received Bachelor
Degree from Electritrical Engineering
Department, Brawijaya University in 1986.
He received his Master Degree and
Doctoral Degree both from University of
Indonesia, in 1995 and 2010, respectively.
Since 1987 he is with Electrical
Engineering
Department,
Brawijaya
University. His current research interest
includes fiber optics, telecommunication
and renewable energy.
M. Aziz Muslim received Bachelor Degree
and Master Degree from Electritrical
Engineering
Department
of
Institut
Teknologi Sepuluh Nopember, Surabaya,
Indonesia, in 1998 and 2001, respectively.
In 2008 he received Ph.D degree from
Kyushu Institute of Technology, Japan.
Since 2000 he is with Electrical Engineering
Department, Brawijaya University. His
current research interest is computational intelligence and its wide
applications in electronics, power systems (including renewable
energy), telecommunications, control systems and informatics.
D1-5
A Computational Fluid Dynamics Study of 6.5
Micron AA 1235 Annealing Treatment in Sided
Blow Inlet – Outlet Furnace
Ruri A. Wahyuonoa), Wiratno A. Asmoro b), Edy Sugiantoro c), and Muhamad Faisal d)
a,b,d)
Department of Engineering Physics, Institut Teknologi Sepuluh Nopember Surabaya
c)
PT. Supra Aluminium Industri (SAI), Jalan raya Kasrie 146 Pandaan - Pasuruan
E-mail address: r_agung_w@ep.its.ac.id, wiratno@ep.its.ac.id, eds@supra-aluminium.co.id,
muhamad.faisal11@mhs.ep.its.ac.id
Abstract— Annealing is the last stage of aluminum foil
production process which often causes undesired condition
of foil. It is mostly caused by improper treatment of
annealing. In this paper, annealing treatment for
aluminum alloy AA 1235 in foil annealing furnace (FAF)
has been analyzed. A combined study was conducted by
means of Computational Fluid Dynamic (CFD) to evaluate
thermal distribution inside the two FAF A and FAF B
during heating. Furnace’s performance from the
temperature control response and conduction time for
heating of aluminum is also analyzed. The FAF A has a
better temperature distribution than FAF B, but there is
saturated airflow between the aluminum roll in second
stage. Based on temperature control response, settling
time of evaporation temperature is achieved about 4 hours
for FAF B and can’t be reached in FAF A whereas it is
desired to be reached in 1 hour. It is suggested to change
the proportional mode control to higher value in order to
get fast settling time since the furnace employs PID
controller. There is big different between theoretical and
actual conduction time of aluminum foil that indicates
improper work of insulating material of furnace so that
there is much heat losses.
Keywords—CFD, Annealing treatment, FAF, AA1235.
I. INTRODUCTION
E
very rolling mill company especially aluminum foil
production, there are many kind of defect and
undesired quality of aluminum foil. They are caused by
improper conditions and treatments of two main
production processes which consist of rolling and
finishing [1] – [2]. It employs a set of cold works in
rolling process that include some passes through rolling
to obtain the desired thickness of aluminum foil. At the
next process, finishing, aluminum foil is also passed
through some steps. They are separating, slitting,
rewinding, annealing and packaging [3].
Rolling process that usually employed in aluminum
company is categorized to cold work. It reduces
thickness of aluminum coil to particular thickness of
aluminum foil by external force from work rolls. It is
operated below the re-crystallization temperature of
aluminum alloy. There is also coolant oil that sprayed
along the surfaces of aluminum coil and work rolls
during rolling process. This coolant oil is added to avoid
direct surface friction between work rolls and coil of
aluminum which caused many defects [3] – [4].
The consequent of rolling process is carrying coolant
oil which embedded inside the rolled aluminum. The
carrying oil needs to be removed from the rolled
aluminum. Therefore, annealing is employed as the
purpose. Basically, annealing is a heat treatment given
to soften the metal due to cold work [1] – [5]. It removes
physical stress of the metal so that some of the
mechanical properties are back to normal. Annealing in
aluminum foil production is a part of finishing process.
It is employed to remove both physical stress and
carrying coolant oil of aluminum foil [2], [3].
It is often found that the aluminum foils still have
worse wet-ability (usually called as weta) and some of
them are too sticky. It is induced by non-evaporated
coolant oil trapped inside the roll of aluminum.
Improper heat treatment of annealing process may be
caused by insufficient heat in the furnace or temperature
control in the furnace. In this study we collaborate with
PT. Supra Aluminium Industri (SAI), one of growing
aluminum industries in Indonesia, which mostly work
with AA 1235 (non heat treatable alloy). As the problem
remains, data of quality control of SAI show that there is
still undesired quality of final product after pass through
annealing process.
Foil annealing furnaces (FAF) employed in SAI are
two kind of sided blow inlet-outlet furnace. They have
specific structure and dimension, thermal distribution
characteristic, and also response to the temperature
control. However, the treatment of annealing process
that given to roll of aluminum foil is same to the all of
furnace [2]. The aim of this study is to evaluate the
overall annealing treatment including temperature
control, thermal distribution, and furnace’s performance
since AA 1235 must be treated in proper heat treatment
to get high quality product.
II. THEORETICAL APPROACH AND REAL PROCESS
Annealing is an additional heat treatment to soften the
metal. It removes physical stress resulting from cold
working (cold rolling) given during overall production
process of rolled aluminum foil [1] – [3]. Annealing
D2-1
applies heat by convection through the atmosphere
inside an annealing furnace. To avoid oxidizing any
un-evaporated lubricant residues or forming magnesium
oxide on magnesium-bearing alloys, annealing may be
carried out in a dry, inert (low O2) atmosphere such as
nitrogen gas. A large integrated aluminum rolling plant
may have its own nitrogen generating plant for this
purpose [2], [6].
Non-heat-treatable aluminum alloy, commonly
heated for 1.5 – 2 hours in the range of operating
temperature 635 – 765oF or equivalent to 335 – 445oC
[6], [7]. The heat released to the rolled aluminum has
another objective. It also evaporates the carrying coolant
oil inside rolled aluminum. So that’s why, annealing
chamber must be dry (very low oxygen intensity) to
avoid oxidizing of coolant oil in the surfaces.
Annealing process in SAI is batch annealing. It
means loading a furnace with a batch of metal, roll of
aluminum foil, and holding it there until the annealing
process is complete. Rolls of aluminum foil are annealed
as a single batch, depending on the size of the foils and
the size and shape of the furnace [2]. In batch annealing,
heat conveyed by the furnace atmosphere to the outside
surfaces of the foils must be conducted through the
metal to the innermost layers, and sufficient time must
be allowed for all parts of each foil to absorb enough
heat to achieve the planned anneal [1], [3]. Batch
annealing is an efficient approach and is the most
commonly used method in high-production aluminum
foil mills.
240
220
Evaporating (225 oC ~ 15 hr)
200
Temperatur (oC)
180
Drying (180 oC ~ 60 hr)
160
140
Pre-heating 2 (160oC ~ 8 hr)
120
Pre-heating 1 (130 oC ~ 8 hr)
100
60
0
10
20
30
40
50
60
Time (hours)
III. METHOD
First analysis of the annealing problem is temperature
and airflow distribution since decrease in temperature in
some volume of chamber can induce incomplete
evaporation. This has caused to some rolls of aluminum
foil is still in worse wet-ability and/or sticky. This
analysis is conducted based on the simulation results of
Computational Fluid Dynamics (CFD) simulation
intended to analyze temperature and air flow
distribution in the empty and filled chamber of Foil
Annealing Furnace (FAF). The furnace is distinguished
as FAF A and FAF B which the aluminum is in specific
orientation inside the chamber.
A. Computational Fluid Dynamics
The computational fluid dynamics, usually
abbreviated as CFD, is a branch of fluid mechanics
using numerical methods to analyze and solve problems
that involve flows of fluid. Numerical method is built by
employs the governing equations such as conservation
of energy, momentum and continuity. Energy
conservation is determined as equation shown below
[8], [9].
∂
(ρE ) + ∇.(υr (ρE + p )) = ∇.keff ∇T
(1)
∂t
r
+ ∇.(τ eff .υ ) + S h
where keff is effective conductivity which the value is
equal to sum of k and kt (thermal conductivity for the
presence of turbulence). The two terms on the right side
represent the energy transfer by conduction and
viscosity dissipation.
For the solid region (i.e. newborn body), energy
transfer is calculated by employing equation as follow
[8], [9]:
r
∂
(2)
( ρh) + ∇.(v ρh) = ∇.( k∇T ) + S h
∂t
where ρ is solid density, h is sensible enthalpy, k is
80
40
shock. The thermal shock effect can be reduced by
applying graded pre-heating.
70
80
90
Fig. 1. Heat treatment scheme of batch annealing in SAI
The annealing scheme is describe as three stages of
thermal treatments (See Fig. 1). They are heating,
soaking and cooling. In heating stage, aluminum foil is
heated to particular temperature up to 5 hours.
Temperature setting depends on the thickness of
aluminum foil. Soaking stage holds annealing chamber
temperature to particular value for 15 – 20 hours. This
step also includes evaporating and drying. The last
stage, cooling, chills amount of rolled aluminum for two
hours inside the annealing chamber. It’ll be pulled out
later if the temperature reaches 70oC [2].
Evaporation of carrying coolant oil of 6.5 micron
aluminum foil is estimated to occupy 15 hours of
heating with temperature setting 225oC. Since the
presence of receding fold (RF) on aluminum foil after
annealing, it is identified that the mechanism of
preheating (big temperature different) induces thermal
conductivity constant of newborn, T is newborn skin
temperature, and Sh is volumetric heat source.
The equation (1) and (2) are complemented by
continuity and conservation of momentum defined
below:
∇. u = 0
ρ
du
= F − ∇ p + µ∇ 2 u
dt
(3)
(4)
where p is normal pressure (N/m2), F is body force on
solid region.
For natural-convection flows, faster convergence of
numerical calculation can be retained with the
Boussinesq model. It sets the fluid density as a function
of temperature. The Boussinesq model is represented by
equation below:
(5)
ρ = ρ 0 (1 − β (T − T0 ))
D2-2
where β is thermal expansion coefficient (1/K), T0 dan
ρ 0 represent the operational parameter. This model is
accurate as long as the density changes are small or it is
valid if satisfying for β (T − T0 ) << 1.
2
qn


Veq
&

Qcv = ∑ m& e q heq +
+ gz eq 


2
q =1


2


Vip
− ∑ m& ip  hip +
+ gz ip 


2
p =1


& = ρAV , (10) can be rewritten as,
Since m
pn
B. Thermodynamic Analysis
The analysis of conduction rate has been developed.
The heat transfer and thermodynamic (control volume)
approach is used to determine how long the aluminum
foil steadily reaches the setting temperature furnace. The
first assumption is the type of material must be solid or
rigid body so that the roll of aluminum foil is same as
rigid cylinder. The Fourier equation that represent
conduction rate is given below [10].
q
dT
= −k
A
dx
(6)
Since the aluminum foil is assumed to be cylindrical, the
cross-section area become a circle. The equation (6) can
be written:
qr = − kA
dT
dT
= − k (2πrL )
dr
dr
(7)
2
qn


Veq
&

Qcv = ∑ ρAeqVe q heq +
+ gz eq 


2
q =1


2
pn


Vip
− ∑ ρAipVip  hip +
+ gz ip 


2
p =1


(8)
In evaporating phase, the temperature different is 65oC.
The aluminum foil on evaporation temperature (225oC)
has conductivity coefficient 222 W/m K. In this study,
specification of aluminum roll is 82 cm width, 34 cm
OD (Outer Diameter) and ID (Inner Diameter) 8 cm. By
using (8), the heat needed for aluminum foil roll is 5.413
kW. As the time setting for transient response during
pre-heating to evaporating is 1 hour, the released heat
which needed is approximately 51,356 kWh .
The energy balance on a control volume is given as
equation below.
2
pn


dE cv
V
= Q& cv − W& cv + ∑ m& i  hi + i + gz i 
dt
2
p =1


2


V
− ∑ m& e  he + e + gz e 
2
q =1


(11)
The calculation of energy in control volume (furnace)
provides the data of heat accumulated inside the
chamber. From this value, it can be determined the
theoretical settling time of air chamber and annealed
aluminum in the FAF.
Integrating (7) for r1 to r2 in left side and T1 to T2 in right
side, so that we obtain:
2πkL(T1 − T2 )
qr =
r 
ln 2 
 r1 
(10)
IV. RESULT AND DISCUSSION
A. Thermal Distribution
FAF A has the setting basket to hang the roll of
aluminum foil in front-rear direction. The heat is blown
to spherical surface of aluminum roll. The orientation of
aluminum rolls inside the chamber makes the capacity
of FAF A is only 32 rolls. The CFD simulation result of
temperature and airflow distribution is given as follow.
(9)
Fig. 2. Temperature distribution of FAF A in (a) rightleft view and (b) front-rear view
qn
There are no work applied in the control volume so that
the value of W& cv equals to zero. Potential energy
difference of inlet and outlet can be neglected since the
value is too small. Steady state analysis results as
follow:
This Fig. 2 expresses temperature distribution of FAF
A chamber. It shows that the chamber has better
temperature distribution in bottom to half of chamber
height. The upper stage reaches temperature setting in
the center.
D2-3
Fig. 3. Airflow distribution of FAF A in (a) rightleft
view and (b) front-rear view
Fig. 6. Airflow distribution of FAF B in (a) rightleft
view and (b) front-rear view
Based on Fig. 3, lower velocity magnitude of air is
distributed in upper side. However, it won’t affect much
to annealing in FAF A since the upper side of chamber
isn’t fully filled by rolls of aluminum foil. It shows that
the airflow distribution is almost well (average airflow
magnitude is about 2.09 ms-1). The higher airflow
magnitude (3.35 – 4.19 ms-1) is only distributed in
bottom of chamber.
Fig. 4. Temperature distribution on roll of aluminum foil
inside FAF A
The distributed heat in the roll of aluminum annealed
in FAF A is shown as Fig. 4. In the figure shows that
almost all of aluminum is well treated by the proper
heat, especially in the side closed to inlet flow. This
condition can minimize the weta and/or sticky of
aluminum foil.
Based on the result on Fig. 5, temperature of FAF B
chamber can’t reach the set point in almost all of area.
Only several rolls of aluminum in the upper side get air
temperature 1 – 2 oC lower than the set point
temperature. The highest airflow magnitude (2.66 – 3.2
ms-1) in FAF B is achieved in bottom to half of chamber
then it drops until 0.38 ms-1 on the upper side. Fig. 6
clearly shows that the fourth stage of aluminum rolls
isn’t supplied adequate airflow to blow up the vaporized
carrying oil.
Fig. 7. Temperature distribution on roll of aluminum foil
inside FAF B
Comparing to FAF A, the distributed heat in the roll
of aluminum annealed in FAF B isn’t better as shown as
Fig. 7. In the figure above shows that the second row of
aluminum foil inside the chamber isn’t get the proper
heat. Only half of aluminum foil rolls in first row is
heated in the desired temperature. It probably induces
some weta condition of aluminum rolls.
B. Evaluation of Temperature Control and Furnace’s
Performance
The temperature control on both FAF A and FAF B is
recorded in overall annealing time which consumes 93 –
95 hours. Notice that the FAF A and FAF B employ PID
controller to control the temperature of annealing.
Transient and steady response of temperature control in
FAF A is given in the figure below.
Fig. 5. Temperature distribution of FAF B in (a) rightleft view and (b) front-rear view
D2-4
Table 2. Settling time for evaporation temperature
set point of FAF B
250
Temperature (K)
200
150
100
Air Temperature at Zone 1
Metal Temperature
50
0
0
10
20
30
40
50
60
70
80
90
Time (hours)
Fig. 8. Response of temperature control of annealiang
process in FAF A
The steady conduction time of aluminum foil rolls in
FAF A is theoretically obtain by dividing conductive
heat transfer by accumulative evaporation heat. Based
on thermodynamic calculation, the conduction time of
aluminum foil in FAF A is 0.152 hour. This value is
lower than the actual conduction time which needs 1 – 4
hours to settle. The detail data of settling/conduction
time of aluminum foil in FAF A is as in Table 1.
Table 1. Settling time for evaporation temperature
set point of FAF A
Zone
T Set ± 0.5oC
Tinit evap.
1
1
0.96
2
1
0.96
3
1
0.96
Metal Temp.
~
4
The recorded temperature response of annealing in
FAF B is follow.
250
Temperature (K)
200
150
100
Metal Temperature at Zone 1
Metal Temperature at Zone 2
50
0
0
10
20
30
40
50
60
70
80
90
Time (hours)
Fig. 9. Response of temperature control of annealiang
process in FAF B
The evaporation settling time based on control
response in FAF B (see Table 2) is about 1 hour for
aluminum in zone 1 and can’t be reached for aluminum
in zone 2 (see Fig. 9). The conduction time for annealing
aluminum foil in FAF B is 0,042 hour theoretically.
Zone
T Set ± 0.5oC
Tinit evap.
1
1
0.97
2
1
0.93
Metal Z1
1
0.97
Metal Z2
~
~
Both of FAF A and FAF B have a quite big different
of conduction time theoretically and actual, taken from
response of temperature control. As usual, this data
analysis indicates that improper heat treatment is
occurred while annealing. Two conditions that might
become the cause of this condition are undesired heat
process/heat transfer and inappropriate control mode for
temperature annealing.
C. Discussion
Considering the result of temperature and airflow
distribution from CFD simulation, FAF A has good
thermal distribution than FAF B. However, the settling
time to evaporating phase in FAF A is about 1 – 4 hours.
The FAF B has average settling time to evaporation
phase about 1 hour. It fit to transition setting time of
pre-heating and evaporating. The temperature and
airflow distribution for FAF B is worse than FAF A.
That is caused by the profile of airflow inside the
chamber is different. Comparing to furnace that has
inlet-outlet in the side of chamber, the airflow of FAF B
is worse than the FAF A. This condition is caused by the
geometry of blade sticked in the inlet and outlet zone is
different to FAF A. The orientation of aluminum foil roll
is also affect to airflow distribution. It is recommended
to change the blade of FAF B as the FAF A has or
change the orientation of aluminum roll annealed in the
chamber to get better het treatment.
Comparing settling time of metal/aluminum in to
conduction time of aluminum oil rolls in each FAF,
there is big different value. Theoretically in FAF A, it
only need about 0,016 – 0,152 hour to reach the set point
temperature. However in fact, aluminum foils need 4 to
reach the set point temperature. Moreover aluminum foil
in FAF B can’t achieve the set point temperature. It is
probably caused by two problems explained in the
previous point. They are improper tuning controller and
the condition of insulating material. Due the FAF A and
B still using PID controller, it should be check that the
value of proportional constant, time derivative and time
integral setting is based on transient response. Since the
settling time is too much longer then the proportional
constant should be substituted to higher value to get fast
response. The other problem may be caused by the
insulating material (e.g. glass wool, grafite, gypsum,
etc) inside the annealing chamber doesn’t work properly
so that there is so much heat losses during the annealing
process. In order to reduce the heat losses, it is
recommended to check the condition or thermal
conductivity of the insulating material.
D2-5
V. CONCLUSIONS
REFERENCES
Annealing treatment of AA 1235 in foil annealing
furnace has been analyzed. FAF A has quite better
temperature and airflow distribution which is set point
temperature and higher airflow magnitude is distributed
in upper side of chamber. The worst temperature and
airflow distribution is possessed by FAF B. The
upper-right side of chamber gets lower temperature so
that it may induce weta. There is heating problem due
the big difference between conduction time of real
process and theoretical calculation. It is probably caused
by improper control tuning of PID controller in the
furnace and heat losses by under works of insulation
material.
VI. ACKNOWLEDGMENT
Thanks to Dr.-Ing. Doty D. Risanti and Dyah Sawitri,
M.T. for the helpful comments on the analysis of
annealing treatment. This study was supported by PT.
Supra Aluminium Industri Pasuruan for giving
measurement data of FAF and Indonesia-Germany Fast
Track Scholarship from Directorate of Higher Education
for the grant.
[1]
The Aluminum Association. 2007. Rolling Aluminum: From the
Mine Through the Mill. The Aluminum Association, Inc.
[2] Visual Quality Characteristic of Aluminum Sheet and Plate, the
Aluminum Association Inc., 4th Edition February 2002
[3] Annisa Kesy Garside, “Penentuan Setting Parameter Proses
Finishing Rolling untuk Aluminium Foil dengan Thickness Exit
7 Mikron di PT. Supra Aluminium Industri.” Laporan Magang
Dosen, Program Hibah A1, Jurusan Teknik Industri – FT,
Universitas Muhammadiyah Malang, 2005.
[4] Smith, W. F. 1990. Principle of Materials Science and
Engineering 2nd Edition. New York: McGraw-Hill Publishing
Company.
[5] Jing Zhang, Fusheng Pan, Rulin Zuo, Chenguang Bai. The low
temperature precipitation in commercial-purity aluminium
sheets for foils. Journal of Materials Processing Technology.
2008; 206: 382 – 387.
[6] Ozgul Keles, Murat Dundar. Aluminum foil: Its typical quality
problems and their causes. Journal of Materials Processing
Technology. 2007; 186: 125 – 137.
[7] R. J. Vidmar. (1992, August). On the use of atmospheric plasmas
as electromagnetic reflectors. IEEE Trans. Plasma Sci. [Online].
21(3).
pp.
876—880.
Available:
http://www.halcyon.com/pub/journals/21ps03-vidmar
[8] J. Blazek. Computational Fluid Dynamics: Principle and
Applications. ELSEVIER SCIENCE ltd. 2001.
[9] Fluent manual. Modeling Heat Transfer. Fluent Inc.. September
29. 2006.
[10] Incropera, F. P. and D. P. DeWitt. 1996. Fundamentals of Heat
and Mass Transfer 4th Edition. U.S.A.: John Wiley & Sons, Inc.
D2-6
The Height Control Systems of
Hydraulic Jack Using Takagi Sugeno
Fuzzy Logic Controller
1
Fitriana Suhartati1, Ahmad Fahmi2
Electrical Engineering Department of Brawijaya University,
2
Electrical Engineering of State University of Malang
1
fitrianas@ub.ac.id
Abstract— More diverse types of cars with different
heights create difficulties when technician make
improvements on the part located under the car, this
condition will affect the time taken and the results of
repairs. Therefore, it is necessary to drive an automatic
height adjustable hydraulic jack according to the type and
height of the car to be repaired. This research designed a
hydraulic jack height control system based fuzzy algorithm
using Takagi Sugeno method. Takagi Sugeno fuzzy
controller was programmed into the microcontroller
AT89S51. The level sensor consisted of a series of
potentiometers and the disk can change the level of the jack
to a voltage shift with an average error of 2.63%. Fuzzy
inputs are error and ∆error position, each using three
membership functions. While the fuzzy output is the
magnitude of the voltage that goes to the hydraulic pump
that consists of 1.21 V, 1.35 V, 1.55 V, and rule base
consists of 9 rules. Based on the test results, Fuzzy
Inference System can work as expected and the system
generates an error of 0.05% to 1573% for no-load test, and
generates an error 1.81% to 2.67% for the test with a load
of 26 kg.
transfer function or the dynamic equations of the
systems.
In this research, hydraulic jack model used pump
actuator as a DC motor, height sensor used
potentiometer, and height controlling used pump
controlling actuated by DC motor.
II. HYDRAULIC SYSTEMS
The hydraulic systems is a combination of various
mechanical components such as oil reservoirs, pumps
and accessories, actuators, valves, connections and
conductor, hydraulic oil contained many non-linearity
such as pressure-flow characteristics of the regulator
valve, due to the frictional forces on the actuator and the
drying-moving parts of valve-valve, wear between the
valve and its seat.
As a result, a wide range of phenomena arise due to
non-linearity of this nonlinearity. (Hayashi, S., 2002).
Keywords—hydraulic jack, Takagi Sugeno fuzzy, error,
∆error
I. INTRODUCTION
The development of automotive rapidly that increase
the number and variety of vehicles make maintenance
and repairs are done differently for each type of car. One
variation of the car is different height of each type. Such
as jeep would be higher than a sedan and car
modification cars that usually lower.
The various height of the car raises a matter of
convenience for people who will do the repair parts are
located at the bottom (under the car) and ultimately will
affect the time taken and the results of repair. This
problem is necessary to drive automatic hydraulic jack
can be adjusted to the person who will fix the car.
This research designed a height control systems of
hydraulic jack model based on Takagi Sugeno fuzzy
algorithms which is simple and reliable to be applied as a
control on a wide range of systems using only input
variables and output without having to know the system
Figure 1 Bottle Type Hydraulic Jack Schematic
Courtessy: www.hyjack.net
Where:
D3-1
a.
b.
c.
d.
e.
f.
g.
Reservoir
Main Cylinder
Piston
Pump
Screws relief
Oil channel to the piston
Check Valves
below:
III. DESIGN AND IMPLEMENTATION
The height control of this hydraulic jack model as
shown in figure 2.
Figure 4 Block Diagram of Fuzzy Logic Controller
Courtessy: Kuswadi, S., 2000
Figure 2 Block Diagram of Control Systems of Hydraulic Jack
Based on the design of the hardware block diagram in
fig. 2, there are two ways of working tools, which
controls the initial speed and control of the opening and
closing the valve. Height that was desired and that has
been achieved perform through the LCD monitor.
Variable used in the design can be seen in Fig. 3
below.
a.
Determine the input variables and output variables.
Input variables for fuzzy controller are error and
∆error, while the output variable is deltaOutput,
with
Err(n)=SP(n)–PV(n)
(1)
deltaErr(n)=Err(n)–Err(n-1)
(2)
Output(n)=Output(n-1)+deltaoutput(3)
b. Fuzzification is a process to convert crisp input
become fuzzy input.
Figure 3 Variable Used in Jack
Figure 5 Diagrammatic of Membership Function of Error
where:
Hreq = height total desired
H2 = Height of objects that affect Hreq
Hreff = Set point as a result of the difference between
H2 and Hreq
Hcurr = output of jack, from condition has not lifted up
to Hreff
DC motors drive a prototype lever hydraulic jack
rated voltage source of +12 volt dry batteries as a power
supply.
The desired height is determined by the A (65mm), B
(75mm) and C (85mm), height of the car with a sliding
potentiometer. Microcontroller as a fuzzy logic
controller receives input from the difference in height of
one of the buttons with sliding potentiometer and
calculate the height of the jack of potentiometer that has
been converted by the ADC.
After the weight and speed are involved in each range,
, so that fuzzification process result membership degree
of each input value.
Further evaluation rule, where the entries have been
involved in the rule base, defuzzification using Takagi
Sugeno method generate the output (control signal).
IV. FUZZY LOGIC CONTROLLER
Fuzzy logic controller algorithm as shown in fig. 4
D3-2
Figure 6 Diagrammatic of Membership Function of ∆Error
c.
Fuzzy logic controller rules is based on experience
and in the form of If-Then. After the crisp input is
converted into fuzzy input, according to Takagi
Sugeno method it is processed based on 9 standard
rules below:
1. If error = PS and ∆error = PS, then output =
1.21V.
2. If error = PS and ∆error = M, then output =
1.21V.
3. If error = PS and ∆error = PB, then output =
1.21V.
4. If error = M and ∆error = PS, then output =
1.35 V.
5. If error = M and ∆error = M, then output =
1.35 V.
6. If error = M and ∆error = PB, then output =
1.21 V.
Table 5 Testing Result of Hreq 65 mm and Href 53
7. If error = PB and ∆error = PS, then output =
1.55 V.
8. If error = PB and ∆error = M, then output =
1.55 V.
9. If error = PB and ∆error = PB, then output =
1.55 V.
Href (mm)
53
V. TEST AND ANALYSIS
A. Height Sensor Testing
The test of existing potentiometers results errors
ranged from 1.54% to 4:16% with an average error
2.63%.
63
Table 1 Testing Result of Hreq 65mm and Href 65mm
65
Actual Height
(mm)
65.5
65.6
66.2
65.5
66.3
LCD Display
(mm)
62
63
62
61
62
Actual Height
(mm)
62.4
62.7
61.8
61.4
62.3
Href (mm)
1. No Load Testing
LCD Display
(mm)
65
66
66
66
66
Actual Height
(mm)
52.4
51.5
51.1
51.5
51.4
Href (mm)
B. Control Systems Testing
This test use three references as the desired height
(Hreq) are 85 mm, 75 mm, and height of 65mm and as
reference (href) are 75 mm, 65 mm, 59 mm and 53 mm.
Measurement using the actual height of the jack-term
slide.
Href (mm)
LCD Display
(mm)
52
52
51
52
51
73
LCD Display
(mm)
71
72
71
72
72
Actual Height
(mm)
71.3
72.5
70.5
71.9
72.2
From the test results shows that the average error for
testing without a load range from 0.05% to 1.573% and
for testing with a load of 26 kg of 1.81% to 2.67%.
Href (mm)
59
LCD Display
(mm)
60
60
60
60
60
Actual Height
(mm)
60.2
59.6
60
60
59.7
Href (mm)
53
LCD Display
(mm)
54
54
54
54
54
Actual
Height (mm)
53.7
53.6
53.7
53.5
53.5
VI. CONCLUSIONS
From the test results can be drawn the following
conclusions:
1. Potentiometer circuit and the disk is used as a
height sensor is able to change the height of the jack
to shift electrical quantities of voltage with an
average error of 2.63%.
2. ADC 0804 series that is used to convert analog data
into digital data as input to the microcontroller with
an average error 0.52%.
3. DAC 0808 series that is used to convert digital data
into analog data with an average error of 1.495%.
4. Fuzzy Logic Controller to work in accordance with
a system that is expected and the system generates
an error of 0.05% up to 1573%. For testing without
load and 1.81% to 2.67%. For testing with a load of
26 kg.
Href (mm)
75
2.
LCD Display
(mm)
75
76
76
76
76
Actual Height
(mm)
75.5
76.4
76.3
76.4
76.3
REFERENCES
[1]
[2]
[3]
[4]
[5]
Testing with Load 26 kg
D3-3
ATMEL Corp. 8-Bit Microcontroller with 4 Kbytes Flash
AT89C51,ATMEL,(www.atmel.com), 1996.
Elektro Indonesia.Teknologi Sistem Fuzzy, ElektroIndonesia.
(http://www.elektroindonesia.com), 1995.
Hayashi,S.,
Nonlinear
Phenomena
in
Hydraulics
System,Tohoku University, 2002.
Kuswadi, Son, Kendali Cerdas, EEPIS Press, 2000.
Malvino, Albert Paul. Prinsip-Prinsip Elektronika, Edisi ketiga,
Alih bahasa: Hanapi Gunawan, Penerbit Erlangga, Jakarta,
1996.
[6]
Mihajlov,Miroslav,Vlastimir Nicolic, Dragan Antic, Position
Control of electro-hydraulic Servo System Using Sliding Mode
Control
Enhanced
by
Fuzzy
Controller,
Facta
UniversitatisSeries:Mechanical Engineering Vol I No. 9, Serbia
Montenegro, 2002.
[7] Reznik,Leonid, Fuzzy Controllers,Victoria University of
Technology, Newnes, Melbourne, 1997.
[8] Ross, Timothy J. Fuzzy Logic With Engineering Aplication.
McGraw-Hill Inc., 1995.
[9] Sullivan, James A, Fluid Power:Theory and Applications.
Prentice-Hall,Virginia, 1975.
[10] Zuhal, Dasar Teknik Tenaga Listrik dan Elektronika Daya, PT.
Gramedia Pustaka Umum, Jakarta, 1993.
[11] www.hyjack.net/animation
Fitriana Suhartati was born in Sidoarjo on 17th October 1974. She
hasgraduated from electrical engineering magister program at Sepuluh
Nopember Institute of Technology, Surabaya, Indonesia, in 2003. Her
major field of study is control systems.
Since 1998, she has teached in electrical engineering department of
Brawijaya University, Malang, Indonesia. She has many publications,
and she written a book Adaptive Control Systems (TEUB, Malang,
2009). Her research interest is in the field of intelligent control systems
and adaptive control systems with application to motor and solar
energy systems.
Ahmad Fahmi was born in Malang on 31st July 1973. He
hasgraduated from electrical engineering magister program at Sepuluh
Nopember Institute of Technology, Surabaya, Indonesia, in 2005. His
major field of study is control systems.
Since 1998, he has teached in electrical engineering department of
State University of Malang, Indonesia. He has many publications, and
he written a book Robotics (UM, Malang, 2009). His research interest
is in the field of control systems and fuzzy logic with application to
motor and renewable energy.
D3-4
An Application of Adaptive Neuro Fuzzy
Inference System (ANFIS) with Subtractive
Clustering for Lung Cancer Early Detection
System
Mochamad Yusuf Santoso1) Syamsul Arifin2)
1) Department of Engineering Physics, Faculty of Industrial Technology ITS Surabaya Indonesia
1)
uchups55@gmail.com, 2)syamsul@ep.its.ac.id
Abstract
Cancer is a disease that related with
uncontrolled cell growth. To date, lung cancer is one of the
most deadly disease. An application of ANFIS with
subtractive clustering for lung cancer early detection system
was developed in this study. Characteristic and chest x-ray
datas were used in this study. The data used to build best
ANFIS model, that will be applied in the software. Results
from the software was validated with doctor’s decission.
Parameters that used to determine system performance are
RMSE, VAF, and the succes rate. The best ANFIS model
for characteristic data was obtained in ra = 0,4; RMSE for
training = 0,1193, RMSE for testing = 0,2030, VAF for
training = 93,34%, VAF for testing = 82,28%, the success
rate of software for training data = 96 % and for testing
data = 96%. While for chest x-ray data, the best model was
obtained in ra = 0,4; RMSE training = 0,0185, RMSE testing
= 0,1063, VAF training = 99,85%, VAF testing = 94,84%, the
success rate of software for training data = 95,56 % and for
testing data = 88,46%.
Index Terms: ANFIS, characteristic, chest x-ray, lung
cancer, subtractive clustering
I. INTRODUCTION
Cancer is a disease associated with the uncontrolled
cell growth. Today, lung cancer is one of the most deadly
desease. According to World Health Organization (WHO),
every year there are more than 1.3 million cases new of
lung cancer and bronchitis in the world, and the mortality
rate approximately 1.1 million [ HYPERLINK \l "rhd11"
1 ]. In Indonesia, 1 of 1000 persons is a new sufferer of
lung cancer, it means that more than 170.000 new
sufferers annually2]. Both in Indonesia and other
developed countries, reported that most of cases diagnosed
when it’s were in advance stage (stage III and IV).
An artificial intellegent system for lung cancer early
detection based on characteristic and chest x-ray was
designed by [ HYPERLINK \l "Ari11" 3 ]. But, this
software has the best performance around 66,6%. Then,
the aim of this study is to develop these artificial
intellegent with deifferent clustering method, subtractive
clustering. It be expected to improving system
performance. The used subtractive clustering, for making
decicion based doctor’s experties consistently. This system
work for helping doctor to make decission.
1.1 Lung Cancer Identification
Lung cancer is a disease characterized by
uncontrolled cell growth in tissues of the lung. It is also
the most preventable cancer. Cure rate and prognosis
depend on the early detection and diagnosis of the
disease. Lung cancer symptoms usually do not appear
until the disease has progressed. Thus, early detection is
not easy. Many early lung cancers were diagnosed
incidentally, after doctor found symtomps as a results of
test performed for an unrelated medical condition4].
There are two major types of lung cancer: non-small
cell and small cell. Non-small cell lung cancer (NCLC)
comes from epithelial cells and is the most common type.
Small cell lung cancer begins in the nerve cells or
hormone-producing cells of the lung. The term “small
cell” refers to the size and shape of the cancer cells as
seen under a microscope. It is important for doctors to
distinguish NSCLC from small cell lung cancer because
the two types of cancer are usually treated in different
ways. Lung cancer begins when cells in the lung change
and grow uncontrollably to form a mass called a tumor (or
a lesion or nodule). A tumor can be benign
(noncancerous) or malignant (cancerous). A cancerous
tumor is a collection of a large number of cancer cells that
have the ability to spread to other parts of the body. A
lung tumor can begin anywhere in the lung [
HYPERLINK \l "Per03" 5 ].
1.2 Adaptive Neuro Fuzzy Inference System (ANFIS)
Adaptive Neuro Fuzzy Inference System (ANFIS)
is combination of fuzzy inference system (FIS) which
illustrated in neural network architecture. Fuzzy
inference system which used is first order TakagiSugeno-Kang (TSK) model, for computaion simplicity
and convenience6].
ANFIS structure consists of five layers represent
neural network architecture is shown in fig1. The square
node is an adaptive node, it means parameter’s value can
change in the midst of training process. The circle node is
non adaptive node with fixed value. There are different
equation for each layer.
D4-1
1.3 Subtractive Clustering
Subtractive clustering method was proposed by [7].
The method make each data points are considered as the
candidates for cluster center. In subtractive clustering, a
data point with the highest potential, which is a function
of the distance measure, is considered as a cluster center.
The potential of each data point is estimated by the
following equation:
Fig. 1. ANFIS structure [
HYPERLINK \l "Jan93" 6 ]
Layer 1
Mathematical equation for this layer dependent on
type of membership function. For example, if gaussian
membership function:
,
=
=
= 1,2
(1)
= 1,2
(2)
=
, =
Fig 1 illustrated an ANFIS with two inputs (x and
y). The output ( , is input’s membership degree.
Membership function that used is gaussian with parameter
σ and c, called premis parameter. It’s value can be
determined from ANFIS training in MATLAB software.
Layer 2
Layer 2’s fuction is to multiply every input signal
which comes from layer 1’s output. The equation is:
=µ
.µ
, = 1,2
(3)
, =
The node in layer 2 is non adaptive node (fixed
parameter). Node number in this layer show the created
rule number.
Layer 3
Every node in layer 3 is non adaptive node that
show normalized firing strength, the i-th node output ratio
with all node output. The equation is:
!"
=
,
= 1,2
(4)
, =
!# $!
If there are more than two rules, then the fuction can be
extend, devided wi with all w for entire rules.
Layer 4
Every node in this layer is adaptive node with
equation:
=
& + ( +
(5)
%, =
In layer 4, there are normalized firing strength from layer
3 and parameter p, q, r, called consequent parameter. As
wll as premis parameter, it’s value resulted from ANFIS
training in MATLAB software.
Layer 5
There is a single node for summing all outputs form
layer 4. The equation is:
∑ !+
= " " "
(6)
), = ∑
!"
Layer 5’s output will be used for making decicion from
the created system [6].
, = ∑234 -./0"-01 /
6
where 5 =
(7)
78
Pi is the potential of i’th data point, n is the total number
of data points, xi and xj are data vectors in data space
including both input and output dimensions, γ is a
positive constant and is selected as 4, and ra is a positive
constant defining the neighborhood range of the cluster or
simply the radius of hypersphere cluster in data space.
Each time a cluster center is obtained, the data
points that are close to new cluster center are penalized in
order to facilitate the emergence of new cluster centers.
The revising of the potential is done by subtraction as
shown in the following equation:
<-
/=" = # /
A
>
?@
, ∗ = , − ,; .
(8)
where B = C. D
Pi* is the i-th’s new potential value, η is squash factor, a
positive constant greater than 1. The positive constant rb
is somewhat greater than ra and it helps avoiding closely
spaced cluster centers.
To accept or reject new cluster center, the following
criteria was suggested by [7]:
If
E"∗
E F
> H̅
E"∗
<H
accept
else if
E F
as a cluster center and continue
reject and end the clustering process
else
Let KL 2 = shoertest of the distances between
all previously found cluster centers
if
MN"O
78
+
E"∗
E F
accept
and
≥1
as a cluster center and continue
else
reject
set the potential at
to 0. Select the
data point with the next highest potential as
the new and re-test
end if
end if
Subtractive clustering has four parameters, namely,
accept ratio H̅, reject ratio ε, cluster radius ra and squash
factor η (or rb). These parameters have influence on the
number of rules and error performance measures. Large
values of H̅ and ε will result in small number of rules.
Conversely, small values of H̅ and ε will increase the
number of rules. A large value of ra generally results in
fewer clusters that lead to a coarse model. A small value
of ra can produce excessive number of rules that may
D4-2
result in an over-defined system. The suggested values for
η and ra are 1.25≤ η ≤ 1.5and 0.15 ≤ ra ≤0.30 [7].
compared with doctor’s decicion. The validation’s result
show the success rate of software for making decision.
II. METHOD
output
1
0.5
0
-0.5
0
5
10
15
20
25
30
data kera = 0,4 testing
35
40
45
50
1.5
1
output
2.1 ANFIS Modeling
In this early system development, characteristic and
chest x-ray datas were used in training for ANFIS
modeling. Characteristic data provides information about
normal and suspected patient. There four kinds of
characteristic data for identification: amount of cigarette
consumed per day, duration of smoking, occupation, and
cough. Chest x-ray data that used for training was
obtained from [8].
Subtractive clustering was employed in training
process. This method will generate data to make it’s
natural membership function. One of the subtractive
clustering’s parameters is ra, becomes varibael for
obtaining some ANFIS models. Fig 2 shows the
subtractive clustering’s parameters in MATLAB and fig 3
shows the ANFIS training process for ra = 0,4.
ra = 0,4 training
1.5
0.5
0
-0.5
0
5
10
15
20
25
data ke-
Fig 4: Validation graph
III. RESULTS AND DISCUSSIONS
3.1 ANFIS Model
3.1.1 Characteristics Data
The result of ra variation in design for ANFIS model
for characteristics data shown in table 1. It’s also show
the result from ANFIS model validation for training and
testing datas, represented in RMSE and VAF.
Table 1 Characteristics Data Validation Result For ra Variation
Fig. 2. Subtractive clustering’s parameters
Fig. 3. ANFIS training process
2.2 ANFIS Model Validation
The best ANFIS model, choosed upon it’s validated
result, Root Mean Square Error (RMSE) and Variance
Accounted For (VAF). The model validated with the real
results, doctor’s decicion. The best ANFIS model has
minimum RMSE value and maximum VAF value. Fig 4
shows the characteristics validation graphics for ra = 0,4.
The blue point is the real value and the red one is the
estimated value. From fig 4, it was obtained that mainly
the red point coincide with the blue one. It means that the
ANFIS model can estimates the output almost perfect.
2.3 Software Design and Validation
The best ANFIS model used as basis for designing
early detection software. The software’s result will be
ra
MF
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
47
42
28
19
15
9
8
6
5
4
RMSE
Training Testing
0,1235
0,2033
0,1231
0,2030
0,1241
0,2056
0,1193
0,2030
0,5389
2,2692
0,5374
1,1564
0,3627
1,0305
0,1856
0,2893
0,1908
0,2641
82,39
44,92
VAF (%)
Training Testing
93,99
82,46
94,01
82,49
93,30
81,76
94,34
82,28
-15,37
-2107,9
-32,31
-473,01
47,82
-360,36
86,23
64,27
85,43
70,23
82,39
44,92
From table 1, it’s obtained that greater ra value, the
number of membership function will be smaller. The
smallest RMSE and the highest VAF for training were
obtained fot ra = 0,4. For testing datas, the smallest
RMSE was obtained for ra = 0,2 and 0,4; while the
highest VAF was obtained for ra = 0,2. The best ANFIS
model is the model which use ra = 0,4 because it has the
smallest RMSE for training. Moreover, ra = 0,4 has
smaller number of membership function than ra = 0,2.
Acoording to [7], excessive number of rules that may
result in an over-defined system.
Table 2 VAF and RMSE ANFIS Model Comparison for Characteristics
Data
Data
Training
Testing
D4-3
Cluster
2 MF
3 MF
SUBTRACTIVE
2 MF
3 MF
SUBTRACTIVE
RMSE
0,49921
0,60747
0,1193
0,415454
0,34802
0,2030
VAF (%)
52,3337
57,3042
94,34
66,60485
54,15432
82,28
With ra = 0,4, it was resulted 19 clusters. It means
for each inputs, the are 19 membership functions in
gaussian form. Then, this model used to design early
detection software for characteristics data.
The comparison for RMSE and VAF result for
characteristics data between this study and [3] shown in
table 2. From table 2, it was obtained that the result of this
study better than [3]. It was caused by subtractive
clustering method can produces membership function
naturally, so that it’s more suittable with the system.
3.1.2 Chest X-ray Data
The result of ra variation in design for ANFIS model
for characteristics data shown in table 1. It’s also show
the result from ANFIS model validation for training and
testing datas, represented in RMSE and VAF.
ra
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
Table 3 Chest X-rays Data Validation Result For ra
Variation
RMSE
VAF (%)
MF
Training Testing Training Testing
2
0,1550
0,3229
89,20
52,74
5
0,1131
0,0646
89,20
98,05
5
0,0210
0,0702
99,80
97,75
5
0,0185
0,1063
99,85
94,84
4
0,0316
0,1210
99,55
93,13
4
0,0572
0,1963
98,53
81,95
3
0,1689
0,2233
87,17
77,76
3
0,1584
0,2760
88,71
65,03
2
0,1546
0,1837
89,25
84,73
2
0,1754
0,2012
86,16
81,74
Table 4 VAF and RMSE ANFIS Model Comparison for Chest X-ray
Data
Data
Training
Testing
Cluster
2 MF
3 MF
SUBTRACTIVE
2 MF
3 MF
SUBTRACTIVE
RMSE
0,364194
0,29113
0,0185
0,25199
0,275235
0,1063
VAF (%)
48,8421
56,5538
99,85
65,45689
62,49139
94,84
3.2 Early Detection Software
The software have been created shown in fig 5 and
fig 6. Tabel 5 and table 6 give the comparison of
software’s success rate. Form those tables, it was obtained
that either this study or [3]’s study resulting sotfware with
success rate more than 90%. It was caused by success rate
calculation based linguistic variable only. In creating
software code, there is a value which used for making
decicion. For both studies, it’s value are not fixed.
Fig. 5. Early detection characteristics data software
Table 4 shows that ra variation not always resulting
different number of membership function. Althougt there
are any same number for several ra value, all of it RMSE
and VAF value are different. It shows that change of ra
will resulting the change of ANFIS parameters, premis
and consequent.
From the training result, minimum error and best
VAF were obtained for ra = 0,4. While for testing,
minimum error and best VAF were obtained for ra = 0,4.
So, the model with ra = 0,4 was choosed to the best
ANFIS model. This model then used to design early
detection software for chest x-rays data. With ra = 0,4, it
was resulted 5 clusters. It means for each inputs, the are 5
membership functions in gaussian form.
The comparison for RMSE and VAF result for chest
x-rays data between this study and [3] shown in table 4.
From table 2, it was obtained that the result of this study
better than [3]. It was caused by subtractive clustering
method can produces membership function naturally, so
that it’s more suittable with the system. Beside that,
image processing in [3] was done for all chest x-ray
region. In this development, chest x-ray was processed
with GLCM, take only the region that contain chest
image.
Table 5 Characteristics data software’s success rate comparison
Data
Training
Testing
Cluster
2 MF
3 MF
SUBTRACTIVE
2 MF
3 MF
SUBTRACTIVE
Success rate
96%
98%
96%
91,43%
97,15%
95,56%
Fig. 6. Early detection chest x-ray software
D4-4
Table 6 Characteristics data software’s success rate comparison
Data
Training
Testing
Cluster
2 MF
3 MF
SUBTRACTIVE
2 MF
3 MF
SUBTRACTIVE
[5]
Success rate
91,43%
97,15%
95,56%
90%
85%
88,46%
IV. CONCLUSIONS AND FUTURE WORKS
A new method of lung cancer detection system has
been developed by using ANFIS subtractive clustering.
It’s can improve system’s performance. The best ANFIS
model for characteristics data was obtained in ra = 0,4;
RMSE for training = 0,1193, RMSE for testing = 0,2030,
VAF for training = 93,34%, VAF for testing = 82,28%,
the success rate of software for training data = 96 % and
for testing data = 96%. While for chest x-ray data, the
best model was obtained in ra = 0,4; RMSE training =
0,0185, RMSE testing = 0,1063, VAF training = 99,85%,
VAF testing = 94,84%, the success rate of software for
training data = 95,56 % and for testing data = 88,46%.
For further study, this system can be improved not
only for detection, but alsocan determine lung’s stage.
Moreover, this system could be applied in a website that
could be accessed by everyone.
ACKNOWLEDGEMENT
The study described in this paper has been
developed in Instrumentation and Control Laboratory,
Engineering Physics Department, Faculty of Industrial
Technology, Institut Teknologi Sepuluh Nopember. The
authors acknowledge the Fast Track DDIP (Double
Degree Indonesia Prancis) Grants from DIKTI
(Directorate General of Higher Education Indonesia) for
having supported.
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
V. REFERENCES
[1]
[2]
[3]
[4]
rhd. (2011, Maret) FAJAR Online. [Online].
http://www.fajar.co.id/read-20110302235529merokok-dan-kanker-paruparu
Tjandra Yoga Aditama, "Situasi Beberapa
Penyakit Paru di Masyarakat," Cermin Dunia
Kedokteran, pp. 28-30, 1993. [Online].
http://www.kalbe.co.id/files/cdk/files/10SituasiPe
nyakitParu084.pdf/10SituasiPenyakitParu084.htm
l
Syamsul Arifin, "Design of Artificial Intellegence
Software for Lung Cancer Diagnosis using
Adaptive Neuro Fuzzy Inference System," in 3rd
International Conferences and Workshops on
Basic and Applied Sciences, Surabaya, 2011, p.
T002.
American Society of Clinical Oncology, "Guide
to Lung Cancer," Alexandria, 2011.
[15]
[16]
[17]
[18]
D4-5
Perhimpunan Dokter Paru Indonesia, "KANKER
PARU
:
PEDOMAN
DIAGNOSIS
&
PENATALAKSANAAN DI INDONESIA,"
2003.
Jyh-Shing Roger Jang, "ANFIS: AdaptiveNetwork-Based Fuzzy Inference System," IEEE
TRANSACTIONS ON SYSTEMS, MAN, AND
CYBERNETICS, VOL. 23, NO. 3, pp. 665-685,
1993.
S. L. Chiu, "A Cluster Estimation Method with
Extension to Fuzzy Model Identification," in
IEEE Internat. Conf. on Fuzzy Systems, 1994, pp.
1240-1245.
Sungging Haryo Wicaksono, "Design of Lung
Cancer Prediction System Based On Image
Pattern Recognition," Surabaya, 2011.
Sylvia Ayu Pradanawati, "Pengembangan Sistem
Kecerdasan Buatan Berbasis Adaptive Neuro
Fuzzy Inference System Untuk Diagnosa
Penyakit Kanker Paru-paru," ITS Surabaya, 2011.
Badan Pusat Statistik, "Perkembangan Beberapa
Indikator Utama Sosial-Ekonomi Indonesia,"
Jakarta, 2010.
Sri Kusumadewi, Analisis dan Desain Sistem
Fuzzy Menggunakan Tool Box MATLAB.
Yogyakarta: Graha Ilmu, 2002.
Ratih Setyaningrum, "Kemampuan Expert System
- ANFIS untuk Diagnosa Kesehatan Pekerja
Industri dan Mencari Solusinya," in Seminar
Nasional
Aplikasi
Teknologi
Informasi,
Yogyakarta, 2007, pp. L15-L20.
Agus Priyono et al., "Generation of Fuzzy Rules
with Subtractive Clustering," Jurnal Teknologi,
pp. 143-153, 2005.
Tarig Faisal, Mohd Nasir Taib, and Fatimah
Ibrahim, "Adaptive Neuro-Fuzzy Inference
System for diagnosis risk in dengue patient,"
Elsevier: Expert System with Application, pp.
4483-4495, 2011.
Adi Purnomo and Sulistyo Puspitodjati, "Aplikasi
Pemrograman C# untuk Analisis Tekstur Kayu
Parquet dengan Menggunakan Metode Grey
Level Co-occurrence Matrix (GLCM)," Fakultas
Teknik Industri Universitas Gunadarma, 2009.
Arna Fariza, Afrida Helen, and Annisa Rasyid,
"PERFORMANSI NEURO FUZZY UNTUK
PERAMALAN DATA TIME SERIES," pp. D77D82, Juni 2007.
Mesothelioma Health Information. [Online].
http://www.mesotheliomahealth.org/images/normlungX.jpg
Elisna Syahruddin. (2006, Juni) kankerparu.org.
PID Design for 3-Phase Induction Motor Speed
Control Based on Neural Network
Levenberg Marquardt
Dedid Cahya H1, Agus Indra G2 , Ali Husein A3, Ahmad Arif A4
1
2
dedid@eepis-its.edu, agus_ig@eepis-its.edu,3ali husein@eepis-its.edu,4arive_76ers@rocketmail.com
Abstract— PID control is a control that is often used in
the industrial world because the PID control can overcome
the existing problems. But it still has a PID control
weaknesses in terms of tuning which is done by trial and
error. Many adaptation methods are used for PID
parameter tuning to, but this time it will be used Neural
Network (NN) training method using the Levenberg
Marquardt (LevMar). With the use of NN training
LevMar would eventually get an optimal PID parameters
are used to drive a three phase induction motors with fast
to match what is desired. The result of using NN LevMar
without PID is enough good where the rise time is almost 1
second, osilation about 20 RPM, but the overshoot is too
big almost half of the set point given.
Index Terms—PID, Neural Network, Levenberg
Marquardt, tuning, trial and error.
Induction motor rotation settings can be done with a
variety ways, by changing the number of pairs of poles,
set the grid voltage, or by adjusting the size of the
frequency. For the regulation of induction motor
rotation by changing the grid voltage, will produce a
limited rotation arrangement (the narrow setting). While
the settings using the frequency changes, the change was
done in more rounds can be smooth or linear according
to the change in frequency.
b. Rotary Encoder
This sensor is used to convert rotation into linear
motion or digital signals. This sensor monitors the
rotation of a rotary movement of the tool, which in this
case is the wheel that is connected to the 3-phase
induction motor to determine the rotary motion of the
motor.
I. INTRODUCTION
C
urrent technological developments have created
a variety of technological advances, particularly in
the field of control technology. One of the controllers
are still widely used in industrial process control
systems is the PID controller.
PID controller parameter adjustment requires the
strengthening of the proportional gain (Kp), integral
gain (Ki), derivative gain (Kd) in an induction motor
parameter changes, such as changes in load torque. To
obtain the desired performance, PID controllers with
fixed gain can be used to plant induction motor with
parameter changes in a particular range. As for the
conditions outside the range, strengthening the PID
controller parameters need to be adjusted again.
Neural Network with Levenberg Marquardt training
method will be used for the tuning process in order to
gain Kp, Ki, and Kd is appropriate if the plant works
outside of a predetermined range, so the plant will
continue to work optimally.
II. LITERATURE STUDY
a. Induction Motor
Opto
Piringan
Fig. 1. Rotary Encoder
c. Inverter
Inverters used to convert the DC voltage source into
the AC source, where the resulting stress can be a
constant or variable value. A voltage source inverter is
called an inverter (voltage source inverter) when the
output voltage constant while the current source inverter
(current source inverter) if the output current constant
and the variable DC link inverter (variable DC linked
inverter) when the output voltage can be controlled or
controlled larger and smaller than the input voltage.
d. PID control
Characteristics of the PID controller is strongly
influenced by the large contribution of the three
parameters P, I and D. Tuning constants Kp, Ti, and Td
will lead to protrusion of the nature of each element.
D5-1
One or two of the three constants can be tuned more
prominent than others. Constant that stands out that will
contribute to the effect on the response of the system.
Fig. 2. Block diagram PID controller
e. Neural Network
Neural network or artificial neural networks (ANN) is
a distributed information processing structure in the
form of directed graph. The advantages of this neural
network is a network can learn where there are two
stages in the operation of the ANN. At this stage of
learning to adjust to any provision of input connections
to the network produces the desired output with the
structure and parameters of the optimal ANN. There are
two stages of learning, namely supervised learning (with
supervision) and unsupervised learning (without
supervision). While the initial testing phase input in the
form of the unknown information is given as input the
network. Each cell will perform computation by
activation function connection with the influence of
weight gained during the learning process.
f. Levenberg Marquardt
Levenberg Marquardt algorithm can be performed
using the second derivative approach without having to
calculate the hessian matrix. If the feed forward neural
network using the work function of the sum of square
hessian matrix can be approximated by:
(1)
And the gradient can be calculated by:
(2)
With J is the Jacobian matrix containing first derivatives
of the weighting network error and bias network.
Levenberg Marquardt algorithm can be calculated with
the approach to compute the Hessian matrix in which to:
(3)
Weighted so that repairs can be calculated:
(4)
Where I is the identity matrix and e is a vector of size pno
that can be determined by the equation JTJ. With the
input dimensions are ni, nh dimension is hidden and the
output dimension is no. So the total weight can be
calculated by
(5)
So the dimension of the Jacobian matrix is pno x w
while the Hessian matrix dimensions w x w.
x = the weights and biases in the network
Jacobi matrix is a matrix of first derivatives of the
weights and the bias error in the network. Jacobi matrix
between input layer and hidden layer is a matrix that
contains the error derivative of the weights between
input layer and hidden layer along with the bias. While
the Jacobi matrix between the hidden layer and output
layer contains the error derivative of the weights
between the hidden layer and output layer along with the
bias.
● Jacobi matrix element between the input layer and
hidden layer
(6)
● Jacobi matrix element in bias hidden layer
(7)
● Jacobi matrix element between the hidden layer and
output layer
(8)
● Jacobi matrix element in bias output layer
(9)
III. SISTEM PLAN
At this stage the hardware design and make the
Levenberg Marquardt algorithm for neural network PID
tuning that aims to control the plant to fit the desired.
Here is a block diagram of control system Neural
Network with Levenberg Marquardt training method.
NNOut
Output
PLANT
Input
NNErr
Input
NN
Hidden Layer
dan
Bias Hidden
Bobot NN
I-H
(V)
Output NN
dan
Bias Output
Bobot NN
H-O
(W)
Fig. 3. Block diagram NN LevMar Without PID
Where have 1 piece of data input to the NN NN where
it will split itself into two pieces of input nodes, 3 hidden
nodes and one output node. So the dimension of the
matrix jacobiannya to 2 x 13 where the "2" is derived
from the data input into the NN and "13" comes from the
number of weights and biases that are connected from
the input layer to hidden layer. So the dimension of the
hessian matrix is 13 x 13 which is a product of the JT
with J.
Here are the stages of the control system are assigned
the value of NN LevMar for weights and biases are
connected:
a. Step 1
The first step taken to set the initial values of the
weights connected between the input and hidden layer
(v) and the hidden and output layer (w) are random.
Determine the value of µ and β where there are no
D5-2
provisions on how much the value of µ and β, but many
studies using µ = 0.1 and β = 10.
b. Step 2
The second step is performed by calculating the
function forward (feed forward)
~ Each hidden node input signal summing weights as
follows:
be carried out continuously until the same error with the
error limit.
While the overall LevMar NN algorithm is the
following can be summarized as follows:
START
Inisialisasi Jaringan
Random bobot V dan W
Set µ dan β
Where:
Zinh = signal input for hidden node h
Xi = value input for node i
Vih = value weight between input node i and hidden
node h
V0h = value bias at hidden nodeh
~ Calculate the value of hidden nodes based on the
sigmoid activation function.
Hitung maju pada simpul
Hidden dan Ouput
Zh = f(v,in), Yo = f(w,zh)
Hitung sse
=
Where:
Zh = value hidden node h
~ Each hidden node output signal summing weights as
follows:
1
2
=1
(
−
)2
Hitung matriks Jacobian
J(x)
Hitung selisih bobot
∆ = (
+ )−1
Dimana, g = JTe
Where:
Youto= signal input to output node o
Zh = value hidden to node h
Who = value weight between hidden node h and output
node o
W0o = value bias at output node o
~ Calculate the value of the output node based on the
sigmoid activation function
Koreksi pembobot
=
+ ∆
Ebaru < SSE
TIDAK
YA
Reduce (µ/β)
Increase (µxβ)
YA
Where:
Yo = value output node o
TIDAK
c. Step 3
The third step is done by calculating the value of
error where the error function approximated by Sum
Square Error (SSE). Where to compare the value of each
output (Yo) with a target (tk) with the following
equation
d. Step 4
The fourth step is done by calculating the Jacobian
matrix (J) which contains the first derivative of the
weights and bias error.
error < error limit
END
Fig 4. Flowchart NN LevMar
IV. SIMULATION RESULT
In the Sub is the result of simulation by the method of
training neural networ LevMar NN to be compared with
simulations by using this NN BackPro where the
comparison of the results can be seen which one is better
for NN training.
A. NN simulation LevMar
e. Step 5
The fifth step is done by calculating the difference in
weight
f. Step 6
The sixth step is done by calculating a new weight
Having obtained the new weights are calculated as the
error back to the step 3. If the error is reduced, the new
do (µ / β) and return to step 2 to step 7. If a new error is
not reduced then do (µxβ) and return to step 5. This will
D5-3
change set point that is not large overshoot of about 200
RPM.
Fig 5. Sheet NN settings LevMar
Set Point = 300 RPM
Set Point = 1000 RPM and load 50%
Fig 6. Graph speed of 300 RPM for NN LevMar
Fig 9 .Graph of speed with load of 50% for the NN LevMar
From the graph it was found that rise time of 0.5
seconds. For the overshoot of about 150 RPM.
Stabilized within 20 seconds and the steady state error of
about 15 RPM.
From the graph the response to load change 50% found
that the system is able to restore to its original set point
while the open loop system can not return to its original
state.
Set Point = 1500 RPM
B. NN simulation BackPro
Fig 7. Graph-speed 1500 RPM for NN LevMar
Fig 10. Sheet NN settings BackPro
From the graph it was found that rise time of about 1
second. For the overshoot of about 600 RPM. Stabilized
within 20 seconds and the steady state error of about 20
RPM
Set Point = 300 RPM
Set Point altered by the period 2000
Fig 11. Graph speed of 300 RPM for NN BackPro
From the graph it was found that rise time of 0.5
seconds. For the overshoot of about 150 RPM. and
steady state error of about 100 RPM.
Set Point = 1500 RPM
Fig 8. Graph of velocity with the period 2000 to NN LevMar
From the graph the response to changes in set point
period of 2000 found that the change in set point showed
that when the transformation of a large enough set point
overshoot produced large about 500 RPM, but when the
D5-4
V. ANALYSIS
From the simulation results performed breaking can be
done to analyze the results of these simulations which
are categorized into 3 pieces, namely the fixed speed,
the speed varies according to period, and administration
expenses
Figure 12 Graph of the speed of 1500 RPM for NN BackPro
From the graph it was found that rise time of about 1
second. For the overshoot of about 600 RPM. Stabilized
within 25 seconds and the steady state error of about 20
RPM.
Set Point altered by the period 2000
Figure 13 Graph of velocity with the period 2000 to NN BackPro
From the graph the response to changes in set point
period of 2000 found that the change in set point showed
that when the transformation of a large enough set point
overshoot produced enough large about 200 RPM, but
when the change set point that is not too large overshoot
of about 50 RPM.
Set Point = 1000 RPM and load 50%
Fixed speed
On testing the speed of response is obtained that
LevMar NN NN LevMar takes about 1 second rise time
and began to stabilize an average of 20 seconds. Its
steady state error of not more than 20 RPM. However
Overshoot produced quite large for large yag RPM
about 500 RPM
While the testing of the response rate was obtained
that BackPro NN NN BackPro takes about 1 second rise
time and began to stabilize an average of 30 seconds. Its
steady state error of not more than 50 RPM. However
Overshoot produced quite large for a large RPM about
500 RPM
Speeds vary according to the period
Of testing the speed of response to changes in NN
LevMar set point was found that the changes will affect
a very large overshoot in terms of its becoming quite
large about 500 RPM with a rise time of about 1 second.
Changes are not too big to be a more rapid rise time of
0.5 seconds.
While testing the response speed of NN backPro to
changes in set point was found that the changes will
affect a very large overshoot in terms of its becoming
quite large about 200 RPM with a rise time of about 1
second. Changes are not too big to be a more rapid rise
time of 0.5 seconds.
Load given
To test NN LevMar to changes in load was found that
when the load under 54% of the system is able to return
to the set point, but typing a given load of more than
55% rate can not be returned to its set point
To test NN BackPro to changes in load was found that
when the load under 54% of the system is able to return
to the set point, but typing a given load of more than
55% rate can not be returned to its set point.
VI. CONCLUSION
Fig 14. Image speed with load of 50% for the NN BackPro
From the graph the response to load change 50% found
that the system is able to restore to its original set point
while the open loop system can not return to its original
state.
Once the testing is done by performing a simulation on
visual basic 6.0 obtained the following conclusion:
~ Overshoot LevMar NN generated by nearly half from
a set point given
~ Time to reach set point with a range of 1 second with
the assumption that the provisions of the motor is used
in accordance with the motor simulation
~ NN LevMar achieve faster convergence which is
about 20 seconds while the NN BackPro takes about 30
seconds
~ The process of adaptation is performed for the NN
LevMar tend to be slow compared to NN where NN
BackPro BackPro able to respond to a change of pace
well with the relatively short time is about 60 seconds.
~ The oscillations that occur in less than LevMar NN
NN BackPro for high RPM, while about 10 RPM to low
D5-5
RPM, while about 20 RPM to high RPM BackPro to
about 20 RPM and low RPM to about 100 RPM.
REFERENCES
[1] Ratna Ika Putri, Mila Fauziyah, Agus Setiawan,
Penerapan Kontroller Neural Fuzzy Untuk
Pengendalian Motor Induksi 3 Fase Pada Mesin
Sentrifugal, INKOM, Vol. III, No. 1-2, Nopember
2009.
[2] N. N. R. Ranga Suri and Dipti Deodhare, Parallel
Levenberg-Marquardt-based Neural Network
Training on Linux Clusters - A Case Study, AI &
Neural Networks Group Centre for Artificial
Intelligence & Robotics Bangalore
[3] Cia Ju Wu, A Neural Networks Based Method For
Fuzzy Paramater Tuning Of PID Controllers,
Journal of the Chinese Institute of Engineers, Vol.
25, No. 3, pp. 265-276; 2002
[4] Rahmat, Perbandingan Algoritma Levenberg
Marquardt Dengan Metode Backpropagation
Pada Proses Learning Jaringan Syaraf Tiruan
Untuk
Pengenalan
Pola
Sinyal
Elektrokardiograph, Seminar Nasional Aplikasi
Teknologi Informasi 2006 (SNATI 2006) ,
Yogyakarta; Juni 2006
[5] Tianur, Kontrol Kecepatan Motor Induksi
Menggunakan PID-Fuzzy, Jurusan Teknik
Elektronika, Politeknik Elektronika Negeri
Surabaya;2010
[6] Huailin Shu, dkk, Decoupled Temperature Control
System Based on PID Neural Network, ACSE 05
Conference, Cairo Egypt; Desember 2005
[7] Cesar Souza, Neural Network Learning by the
Levenberg-Marquardt Algorithm with Bayesian
Regularization (part 1) ,
cesarsouza.
Blogspot.com;Nopember 2009
[8] Cesar Souza, Neural Network Learning by the
Levenberg-Marquardt Algorithm with Bayesian
Regularization (part 2) ,
cesarsouza.
Blogspot.com; Nopember 2009
AUTHOR
Dedid CH, born in Pasuruan, Indonesia,
December 27, 1962. Educational
backgrounds:
Engineer in Electrical Engineering
Institute of Technology Sepuluh
Nopember
Surabaya,
Surabaya
Indonesia (1986).
MT Electrical Engineering Institute of
Technology
Sepuluh
Nopember
Surabaya, Surabaya Indonesia (2002)
Post-graduate student in Electrical Engineering, in
Institute
of Technology
Sepuluh
Nopember
Surabaya-Indonesia (2007- ow)
D5-6
Agus Indra Gunawan, Born Agustus 21,
1976. Educational backgrounds:
Engineer in Electrical Engineering
Institute of Technology Sepuluh
Nopember
Surabaya,
Surabaya
Indonesia
Ali Husein Alasiry, born in Maluku,
Indonesia,
Oktober
27,
1973.
Educational backgrounds:
Engineer in Electronics Engineering
from Institute Technology Sepuluh
Nopember (ITS) Surabaya(1998).
M.Eng degree in Mechanical and
Control Engineering from Tokyo
Institute of Technology (2004)
Ahmad Arif Asror, born in Pasuruan,
Indonesia,
Januari
15,
1990.
Educational backgrounds:
(2008-ow).
Zelio PLC-based Automation of Coffee
Roasting Process
1
M. Aziz Muslim, 2Goegoes Dwi N, 3Ali Mahkrus
Electrical Engineering Department, Faculty of Engineering, Brawijaya University
1
muh_aziz@ub.ac.id, 2nisways@gmail.com, 3aly_alfath@yahoo.com
Abstract— “Pusat Penelitian Kopi dan Kakao
Indonesia” located in Jember, East Java, has developed
coffee roaster machine, with the capacity of 10 kg as well as
50 kg. For economic and reversible reasons, it uses wood
for combustion. In fact, starting from roasting process up
to tempering process the machine must be operated by field
operator step by step.
To overcome these difficulties, we developed an efficient
centered automation systems for overall processes. This
system can be operated in automatic mode as well as in
manual mode. To implement this system, we employed
Programable Logic Controller (PLC) type Zelio
SR3-B261BD (as the main PLC) and SR3-XT43BD (as an
extended PLC). PLC programming is done using Function
Block Diagram (FBD) model. Pt-100 and Thermostat are
used to sense temperature at roaster cylinder and
temperer. Experiments showing satisfactory results
compared to previous system. For example, with automatic
mode, using 10 kg coffee with water content of 12% and
fuel wood with water content of 15,33% resulting medium
class roasted coffee in 25 minutes, whis is 15 minutes faster
than using previous system.
Index Terms— Coffee Roaster Machine, Temperature
Sensor Pt-100, Thermostat, Zelio PLC SR3-B261BD,
SR3-XT43BD, Function Block Diagram.
I. INTRODUCTION
I
ndonesia is one of the largest coffee-producing
countries in the world. Statistic shows that 95,9% of
1,30 million Ha plantation area is owned by the personal
and the rest is owned by PTPN and private. Hence, this
sector plays an important role as income generator for
Indonesian farmer [1][2].
Coffee been processing has big impact in the quality
of coffee. One of important steps in the processing is
roasting process. “Pusat Penelitian Kopi dan Kakao
Indonesia” located in Jember, East Java, is a central for
Coffee and Cacao research in Indonesia. They have
developed a coffee roaster machine, with the capacity of
10 kg as well as 50 kg. For combustion, wood fuel is
used with the reasons of economic, reversible, and
according to market needs. In fact, starting from roasting
process up to tempering process the machine must be
operated by field operator step by step, which is of
course not efficient [1][3].
To overcome these difficulties, we developed an
efficient centered automation systems for overall
processes. This system can be operated in automatic
mode as well as in manual mode. We proposed to use
Zelio PLC as the bases of automation system. Using a
PLC system means flexibility, since we can modify the
system (such in case of relay aging or system
modification) easily, without having to replace all
existing instruments, by modifying the program inside
PLC.
II. COFFE PROCESSING
A key part of coffee production is roasting process. In
this process, aroma and distinctive taste of coffee from
the coffee beans are formed through heat treatment.
Coffee beans naturally contain quite a lot of potential
compounds forming distinctive taste and aroma of
coffee. During roasting process, there are three stages of
physical and chemical reactions run in sequence, i.e,
evaporation of water from the beans, the evaporation of
volatile compounds (such as aldehydes, furfural,
ketones, alcohols and esters), and the process pyrolysis
(browning beans). Pyrolysis itself is basically a
decomposition reaction of hydrocarbon compounds such
as carbohydrates, hemicellulose, and cellulose which
generally occurs after the roaster temperature above
180oC. Chemically, the process is characterized by the
release of CO2 gas in large amounts of space roaster.
Being physically, characterized by changes in the
original color of coffee beans a yellow-green to brown.
Roasting time varied from 7 to 40 minutes depending on
the type of equipment, quality of coffee, as well as fuel.
Roasting process terminated when the aroma and taste of
coffee has reached the desired one and the color is
changed from the original seeds of yellow-green to dark
brown,
blackish-brown,
and
black
[1].
There are two processes conduct by two machines
developed at the “Pusat Penelitian Kopi dan Kakao
Indonesia”. The first process is the roaster process,
conduct by two kind of machines, first is with capacity of
10 kg (depicted in Figure 1, while the inner side of the
roaster is presented in Figure 2) and the other is with 50
kg of capacity. After completion of roasting process, the
temperature of the coffee bean outing from the roaster
cylinder must be reduced quickly below 30oC so that the
taste and aroma of coffee do not lost due to the hot
temperature. The process is called tempering process.
D6-1
This process is conducted by means of tempering
machine, such as depicted in Figure 3.
including a monitor (display) small, integrated on the
module and extension modules are either direct
conversion of the temperature sensor to the voltage or
the addition of discrete I/O, as well as direct monitoring
capabilities to facilitate the Human Machine Interface
(HMI). In general, Zelio also has analog and discrete I
/O. We used Zelio PLC with the type of SR3-B261BD. It
has 16 input ports, which are can be set as 16 discrete
input or 10 discrete input with 6 analog inputs. This PLC
has 10 ports of discrete output. Figure 4 shown the Zelio
SR3-B261BD.
Figure 1. Roaster of 10 kg Capacity
Figure 4. The Zelio SR3-B261BD
The extension analog module used in this study is the
Zelio SR3-XT43BD (depicted in Figure 5.). Using this
extension module we can control directly the analog
input and analog output. The analog input of the PLC are
voltage, current, and an output of temperature sensor
pt-100. While the output is the form of voltage.
Figure 2. Schematic Diagram of Inner Side of the Roaster
Figure 5. The Zelio SR3-XT43BD
As an advantageous of using Zelio PLC is it supports
Function Block Diagram (FBD) as the programming
languange which is relatively simpler than using ladder
diagram. Further description of FBD follows.
Figure 3. Tempering Machine
III. PROGRAMMABLE LOGIC CONTROLLER (PLC)
SYSTEM
A. Zelio PLC
Zelio is generally referred to mini PLC, applied in
small industries. Zelio can also be used for simple
Automation Lighting Control with free customized
program. Even it can also functioned as the master
control security system. Equipped with several features,
B. Function Block Diagram (FBD)
FBD is a graphical data flow-based programming,
widely used for control process purposes involving
complex calculations and analog data acquisition.
Visually, this kind of programming can be seen in Figure
6. In Figure 6, the left side is the system input, while the
right side is the system output. In between is the space to
place the desired functions. The functions is in the form
of an image block. Of which block is the function of
logic, as in Figure 7 and the non-logic, as in Figure 8.
D6-2
IV. AUTOMATION OF ROASTING PROCESS
A. General Description
Figure 11 shows the overall process controlled by our
proposed automation system. A block diagram of our
proposed process control schema is depicted in Figure
12.
Figure 6. Function Block Diagram
Figure 7. Logic Function Block Diagram
Figure 11. The Overall Coffee Bean Process
Figure 8. Non-logic Function Block Diagram
Figure 12. Block Diagram of The Proposed Process Control
Schema
Output latching using ladder diagram is shown in
Figure 9, while the same function using FBD is shown in
Figure 10. In both figures, x1 is input 1 (a normally open
push botton), and x2 is input 2 (a normally closed push
button), while Y1 is motor driving output. When x1 is
pressed, the output (motor) will remain in ON condition
unless x2 is pressed.
Figure 9. Output Latching using ladder diagram
Figure 10. Output Latching using FBD
In accordance with the block in Figure 12, the parts of
the system include:
1. Zelio (mini PLC) functioned as the main
controller of the entire process.
2. Fan/blower system that serves as a controller in
roaster burner.
3. AC motors, we used 4 AC motors placed in
space heater motor, vacuum coffee beans, coffee
stirrer, and vacuum blower heat.
4. DC motors, we used DC motors for
closing/opening
doors
of roaster
and
closing/opening doors of temperer.
5. Sensors, consist of temperature sensors Pt-100,
thermostat, and limit switches.
6. Alarms are used when there is a shortage of wood
in burner or system failure, feeding raw material
to be roasted and when overheating occurs.
B. Process Control Strategy
In this section, control strategy of overall process
including the task of controller is described. All of this
explanations are based on the processes in Figure 11 and
control schema Figure 12. The explanation is as follows:
1. Roasting process. This process begins by manually
heating the burning area under cylinder roaster.
D6-3
Starting from the initial heating, cylindrical roaster
was switched on. This is done to avoid expansion
of part of the cylinder. When the temperature of the
hot cylinder has reached the desired temperature
(150o-170oC), the coffee beans are automatically
inserted into the cylinder by using a vacuum
suction machine with a capacity of 36.85 kg of
coffee beans. However, in practice, the engine
vacuum is not constantly suck all the coffee beans,
but with PLC control, a vacuum is sucking an
average of 10 Kg in 45 seconds, then release the
suction that eventually entered into a cylindrical
roaster in 20 seconds. The process was repeated 5
times so it can suck an estimated 50 kg in 5 minutes
and at the same time put the coffee beans into rotary
cylinder. The cylinder is then rotated constantly by
the motor, to make roasting process evenly occurs
on the coffee beans. A temperature sensor is placed
to measure real time in-cylinder temperature and
sending
this
data
to
the
PLC.
The heating process of the cylinder is continued to
reach the temperature of 180oC to 220oC (± 10% of
the 200o C) according to the cylinder heat rise and
their normal temperature in roasting. The
difference in reference temperature is caused by the
difference of water content of wood used in
combustion and water content of roasting coffee.
The smaller the water content of coffee and wood,
the heat will rise faster than the high water content.
To supply oxygen into the heating system we used a
DC motor as fan / blower. Motor speed can be set
according to the analog signal sent to the fan. If the
combustion temperature exceeds 200oC, then the
fan will be turned off. When the temperature
decreases from the desired value (from 170oC to
130oC), then the fan will rotate with a maximum
speed. When the fan is spinning at a maximum, but
still cannot raise the combustion temperature, and
even drops below 115 oC, then the PLC will send a
signal to the siren failure as a sign that there has
been a fuel shortage, this is followed the turning
off of a fan if the temperature is below the
minimum fan system. Once the wood fuel is
inserted, the temperature will rise to the desired
point.
After the roasting process goes according to plan,
and the set temperature and time is reached, the
cylinder remain rotates while the door opening.
Roasted coffee will go out to the tempering
machine. Before the doors of roaster open, the
temperer mixer and cooling blower are activated
first.
1. The Tempering Process. This process is the final
step. Within 15 seconds before coffee bean set to
tempering, a mixer has been activated, followed
by the fall of all coffee bean in the temperer. This
is done to prevent the buildup of the hot coffee
beans into one side of the temperer. When the
temperature of the coffee bean is already under
30oC (read by temperature sensor of thermostat
type), the beans will be issued through the holes
in the tempering and go into the container of
roasted coffee beans.
Program of proposed control strategy in FBD is given
in Figure 13. The program is then loaded into the PLC’s
memory to implement the control strategy.
Figure 13. Function Block Diagram of The Proposed Control Strategy
V. EXPERIMENTAL RESULTS
In automatic mode control, duration of roasting
process varies according to the water content of fuel
wood, water content of roasting coffee and weight of
coffee. The higher the water content of wood the longer
roasting duration. Table 1 summarized the results of
three experiments.
Table 1. Experimental Results
Trial
Roaster
Type
1
2
3
10
50
50
Coffee
Weight
(Kg)
10
30
30
Water
Content of
Coffee (%)
12
11,8
12
Water
Content of
Wood (%)
15,33
12,69
13,82
Roating
Duration
(minutes)
25
34
35
Table 1 shows satisfactory results since it still less
than the maximum roasting time (40 minutes). The most
important point actually is not in the reducing of roasting
time, but in the ease work of the operator. Using the
proposed control system, in automatic mode, all
processes can be controlled easily from control panel .
VI. CONCLUSSIONS
In this study, we developed an efficient centered
automation systems for overall processes of producing
roasted coffee. This system can be operated in automatic
mode as well as in manual mode. A Zelio PLC of type
SR3-B261BD (as the main PLC) and SR3-XT43BD (as
an extended PLC) were employed as process controller.
D6-4
PLC programming is done using Function Block
Diagram (FBD) model. Pt-100 and Thermostat are used
to sense temperature at roaster cylinder and temperer.
Experiments showing satisfactory results compared to
previous system. In the term of roasting time, maximum
roasting time is reduced from 40 minutes 35 minutes.
And using the proposed control system, in automatic
mode, all processes can be controlled easily from control
panel .
M. Aziz Muslim received Bachelor Degree and
Master Degree from Electrical Engineering
Department of Institut Teknologi Sepuluh
Nopember, Surabaya, Indonesia, in 1998 and
2001, respectively. In 2008 he received Ph.D
degree from Kyushu Institute of Technology,
Japan. Since 2000 he is with Electrical Engineering
Department, Brawijaya University. His current
research interest are control systems and computational intelligence
including its wide applications in electronics, power systems,
telecommunications, control systems and informatics.
ACKNOWLEDGMENT
Authors thank to “Pusat Penelitian Kopi dan Kakao
Indonesia” at Jember, for giving us facility and
opportunity to conduct this research.
REFERENCES
[1]
[2]
[3]
Suharyanto, Edi, Sri Mulato, Pengolahan Primer dan Sekunder
Kopi, Pusat Penelitian Kopi dan Kakao Indonesi, Jember, 2006
Mulato, Sri, Development and Evaluation of a Solar Cocoa
Processing Center for Cooperative Use in Indonesia,
Dissertation.Hohenheim University. Stuttgart-Hohenheim,
2001
Ali M, M. Aziz M, Goegoes D.N, “Otomatisasi Mesin Sangrai
Kopi Berbasis PLC Zelio Berbahan Bakar Kayu”, Bachelor
Degree Thesis, Malang, 2011
D6-5
Goegoes Dwi Nusantoro received Bachelor
Degree from Brawijaya University and Master
Degree from Gajah Mada University, Indonesia,
in 1999 and 2005, respectively. Since 2006 he is
with Electrical Engineering Department,
Brawijaya University. His current research
interest are electronics control systems (embedded
control system) and robotics.
Ali Mahkrus joined Electrical Enegineering
Department of Brawijaya University as an
undergraduate student in 2005. This study is a
part of his endeavour for fulfilment of Bachelor
Degree. Finally he was awarded the Degree in
2011.
Prediction of CO and HC on Multiple Injection
Diesel Engine Using Multiple Linear Regression
Bambang Wahono, Harutoshi Ogai
Graduate School of Information, Production and Systems, Waseda University
bambangwahono80@fuji.waseda.jp
Abstract—In recent years, diesel engine has been
equipped some control devices such as multiple injection
equipment with common rail system and turbocharger. In
order to control the large number of control parameter
appropriately in consideration of CO and HC as the
engine output objectives, the model construction which
reproduces the characteristic value of CO and HC from
control parameter is needed. In this research, the multiple
linear regressions were applied to construct the engine
model. Using the experimental data of a single cylinder
diesel engine, the prediction model of CO and HC on
multiple injection diesel engines was built and compared
with the conventional method on estimation accuracy.
Index Terms—Engine,
injection
modeling,
MLR,
multiple
I. INTRODUCTION
T
HE big problem in the diesel engine is the exhaust
gas emission such as CO and HC. CO and HC are
harmful emission not only for human health but also
for the environment. Many approaches have been
proposed to reduce these emissions [1]. In recent years,
diesel engine has been equipped some control devices
such as multiple injection equipment with common rail
system and turbocharger. In order to control the large
number of control parameter appropriately in
consideration of CO and HC as the engine output
objectives, the model construction which reproduces the
characteristic value of CO and HC from control
parameter is needed. In this research, the multiple linear
regressions were applied to construct the engine model.
Using the experimental data of a single cylinder diesel
engine, the prediction model of CO and HC in multiple
injection diesel engines was built.
II. CONSTRUCTION OF MODELING
In this research, we used the multiple linear
regressions to construct the mathematical modelling of
the multiple injection diesel engine.
A. Multiple Linear Regression
Multiple linear regression (MLR) refers to the
establishment of multiple linear regression model
utilizing historical sample data. Multiple linear
regression is the theory and method, a mature and
quantificational analysis method [2]. Multiple linear
regression is a method used to model the linear
relationship between a dependent variable and one or
more independent variables. The dependent variable is
sometimes also called the predictand, and the
independent variables are called the predictors. MLR is
based on least squares. The model is fit such that the
sum-of-squares of differences of observed and predicted
values is minimized.
B. The Mathematical Model Equation
The model expresses the value of a predictand variable
as a linear function of one or more predictor variables
and an error term:
(1)
y i = β 0 + β 1 xi1 + β 2 xi 2 + ..... + β p xip + ei
Where, yi is the predictand variable in observation i.
xip is the value of pth predictor variable in observation i.
β0 is the coefficient constant. βp is the coefficient on the
pth predictor. p is the total number of predictors. ei is the
error term.
The model (1) is estimated by least squares, which
yields parameter estimates such that the sum of squares
of errors is minimized. The resulting prediction equation
is:
(2)
yˆ i = βˆ 0 + βˆ1 x i1 + βˆ 2 x i 2 + ..... + βˆ p x ip
Where, the variables are defined as in (1) except that
“^” denotes estimated values. The error term in equation
(1) is unknown because the true model is unknown. Once
the model has been estimated, the regression residuals
are defined as:
(3)
eˆi = y i − yˆ i
Where, yi is the observed value of predictand in
observations i and ŷ i is the predicted value of predictand
in observations i.
The difference between the observed value yi and the
predicted value ŷ i would on average, tend toward 0. For
this reason, it can be assumed that the error term in
equation (1) has an average or expected value of 0 if the
probability distributions for the dependent variable at the
various level of the independent variable are normally
distributed. The error term can therefore be omitted in
calculating parameters [3].
Then, the sum of squared residuals (SSE) equation is:
SSE =
n
∑ eˆ
i =1
2
i
(4)
Where, n is the number of observation. The sum of
squared regression equation is:
D7-1
SSR =
n
∑ ( yˆ
i =1
− yi ) 2
i
(5)
Where, y i is the mean of the y values. In the case of
simple regression, the formulas for the least squares
estimates are:
β=[ β0 β1 β2 ….. βp]T =(XT X)-1 XY
(6)
Where, X and Y are the following column vector and
matrix:
1 x11 x12 .. x1 p 
 y1 


y 
1 x21 x22 .. x2k 
X =
and Y =  2 
.. .. .. .. .. 
 .. 


 
 yn 
1 xn1 xn2 .. xnp 
R = 1−
SSR
F =
tj =
SSE
SSE
SST
(7)
p
(8)
( n − p − 1)
βˆ
(9)
j
c ij [ SSE
cij=(XT X)-1
Fig. 1. Single cylinder diesel engine experimental device
( n − p − 1)
i,j=0,1,2…,p
]
(10)
Where, F is the statistic value, tj is relationship
parameter R and t statistic value. The correlation
coefficient R indicated that the matching level of the
calculation datum by the regression equation and the
original datum, the result is the better when R is more
close to 1. Statistic values indicate the significance of the
multiple linear regression equation, whose values obey F
distribution. On the condition of less effective regression
analysis result, the statistics values of t correspond to non
significant variables should be rejected in turn according
to the value of tj. Then the regression analysis will be
carried out again with the remaining significant factors.
Finally, the prediction model of output is identified.
Fig. 2. Single cylinder diesel engine schematic view
The experimental device of this research is Yanmar
TF70 V-E diesel engine with 4 cycle horizontal type
water-cooling and equipped with a turbocharger.
Table I Specification of a diesel engine
Engine type
4-cycle, 1cylinder, DI
78 mm × 80 mm
Bore × Stroke
Top clearance
0.98 mm
Con-rod length
115mm
Compression ratio
21.4
Cylinder capacity
0.382L
Maximum output
5.5/2600 KW/min-1
Full-length
640 mm
Full-height
474 mm
Full-width
330.5 mm
III. EXPERIMENT
In this research, we used the single cylinder diesel
engine experimental device to get the experiment data.
We used experiment data to build the model.
A. Experimental Device
Combustion model could be applied to calculating
many engine indices, BSFC, gas emission, pressure, etc.
In the current research, multiple linear regression model
is constructed, and only CO and HC are taken as
optimization objectives. These objectives are formulated
from groups of experiment data. The experiments with
multiple injections are performed on a diesel engine
experimental device (in Fig. 1) included the diesel
engine schematic view (in Fig. 2) whose specifications
are listed in Table I. The multiple injections include two
pilot injections and main injection, as shown in Fig.3.
Fig. 3. Multiple injection pattern
B. Experimental Condition and Result
In this research, the diesel engine is set with three
stage injection pilot1 injection, pilot2 injection and main
injection. Then, the rotation speed 1500 rpm, EGR rate
0% and the engine temperature is set about 90 degrees.
The engine control parameters are set as Table II and the
D7-2
engine optimization objectives are listed as table III.
Table IV shows the data was obtained from Diesel
engine. In Table IV, x1, x2, x3 and x4 as control
parameters (input). Then, y1 and y2 represents the
characteristic value of the optimization objective
(output).
Control
Parameter
x1
x2
x3
x4
Table II Diesel engine control parameters
Variation
Meaning
Unit
Range
Pilot 1 injection
deg.
-70,-50,-40,-3
timing
ATDC
0
Pilot 2 injection
deg.
-40,-20,-15
timing
ATDC
Main injection
deg.
-8~0
timing
ATDC
Engine Speed
rpm
1500
Then, the regression equation for CO and HC can be
expressed as:
y1= –24.6057+0.0060x1–0.0041x2–0.0030x3+0.0174x4
y2= –60.3584–0.1760x1–0.1421x2+0.0543x3+0.1951x4
The predicted results of CO and HC are evaluated
using correlation coefficient. The predicted result,
experiment data and absolute error of CO and HC is
showed in Table VI, Fig. 4 and Fig. 5
Table VI The predicted result, actual data, absolute error of CO and HC
CO (%)
actual
Table III Optimization objectives
Optimization
Meanin
Uni
Objective
g
t
y1
CO
%
pp
y2
HC
m
piot1(T)
deg
x1
-70
-70
-70
-50
-50
-50
-50
-50
-50
-50
-50
-40
-40
-40
-30
-30
-30
-30
-30
-30
-30
-30
-30
-30
-30
-30
Table IV Data from input and output
engine
piot2(T) main(T)
speed
deg
deg
r.p.m.
x2
x3
x4
-40
-3
1523
-40
-8
1524
-40
-2
1524
-40
-4
1522
-40
-3
1525
-40
-2
1527
-40
-6
1522
-40
-5
1522
-40
-4
1523
-40
-3
1521
-40
-2
1520
-20
-5
1522
-20
-4
1522
-20
-3
1523
-20
-2
1521
-20
-6
1524
-20
-5
1521
-20
-4
1522
-15
-6
1525
-15
-5
1525
-15
-4
1522
-15
-3
1522
-15
0
1522
-15
-7
1524
-15
-6
1523
-15
-5
1523
CO
%
y1
1.64
1.6
1.62
1.72
1.75
1.75
1.71
1.73
1.72
1.73
1.64
1.65
1.65
1.69
1.7
1.77
1.71
1.76
1.82
1.78
1.73
1.68
1.75
1.8
1.78
1.73
HC (ppm)
No
HC
ppm
y2
257
253
253
253
252
252
250
251
249
252
249
248
246
247
245
245
244
243
245
244
244
243
243
244
243
244
predict
error
actual
predict
1
1.64
1.599228
2.49%
257
254.5442
0.96%
2
1.6
1.631564
1.97%
253
254.4677
0.58%
3
1.62
1.613605
0.39%
253
254.7935
0.71%
4
1.72
1.705824
0.82%
253
250.7749
0.88%
5
1.75
1.75494
0.28%
252
251.4144
0.23%
6
1.75
1.786686
2.10%
252
251.8588
0.06%
7
1.71
1.71181
0.11%
250
250.6663
0.27%
8
1.73
1.708817
1.22%
251
250.7206
0.11%
9
1.72
1.723193
0.19%
249
250.97
0.79%
10
1.73
1.68546
2.57%
252
250.6342
0.54%
11
1.64
1.665097
1.53%
249
250.4934
0.60%
12
1.65
1.686789
2.23%
248
246.1185
0.76%
13
1.65
1.683796
2.05%
246
246.1728
0.07%
14
1.69
1.698172
0.48%
247
246.4222
0.23%
15
1.7
1.720926
1.23%
245
244.3264
0.27%
16
1.77
1.785008
0.85%
245
244.6943
0.12%
17
1.71
1.729905
1.16%
244
244.1635
0.07%
18
1.76
1.744282
0.89%
243
244.4128
0.58%
19
1.82
1.781749
2.10%
245
244.1788
0.34%
20
1.78
1.778756
0.07%
244
244.2331
0.10%
21
1.73
1.723653
0.37%
244
243.7023
0.12%
22
1.68
1.72066
2.42%
243
243.7566
0.31%
23
1.75
1.71168
2.19%
243
243.9195
0.38%
24
1.8
1.767373
1.81%
244
243.9295
0.03%
25
1.78
1.74701
1.85%
243
243.7887
0.32%
1.73
1.744016
0.81%
244
243.843
0.06%
26
average absolute
error
1.32%
average absolute
error
IV. RESULT
Based on the multiple linear regression equation, the
regression coefficients β for each objective are listed as
Table V:
j
1
2
3
4
Table V regression coefficients
y1-CO(%)
y2-HC(ppm)
β
βo
β
βo
-0.176
0.0060
0
-0.004
-24.605
-0.142
-60.358
1
7
1
4
-0.003
0.0543
0
0.0174
0.1951
Fig. 4 The prediction and experiment of CO
D7-3
error
0.37%
minimum absolute error of CO is 0.03% and the average
absolute error of CO is 0.37%. From Fig. 4 and Fig. 5,
the coefficient correlation of CO is 0.87968 and the
coefficient correlation of HC is 0.958748. It can be
regarded that the multiple linear regression method can
effectively estimate the objectives.
V. CONCLUSIONS
Fig. 5 The prediction and experiment of HC
Differences of the predicted values with the
experiment data are shown in Fig. 6 and Fig. 7.
Fig. 6 The difference of predicted value and experiment data of CO
In order to control the large number of control parameter
appropriately in consideration of CO and HC as the
engine output objectives, the model construction which
reproduces the characteristic value of CO and HC from
control parameter is needed. In this research, the
multiple linear regressions were applied to construct the
engine model. The accuracy of predictions made using
multiple linear regression models depends on how well
the regression function fits the data, there should be
regular checks to see how well a regression function fits
a given data set. This can be done through regular
updates or monitoring to ensure that the error values are
always below a pre-specified error threshold. A pre
analysis of the control parameter is necessary for
successful MLR predicting as the result of the analysis
showed. In this paper, we have reported our predicting
model of CO and HC in multiple injection diesel engines
by multiple linear regression. It can be regarded that the
multiple linear regression method can effectively
estimate the objectives.
ACKNOWLEDGMENT
The author would like to thanks INPEX Scholarship
Foundation for financing the study in Waseda
University, Japan.
REFERENCES
[1]
[2]
Fig. 7 The difference of predicted value and experiment data of HC
From table VI, Fig. 6 and Fig. 7, the maximum absolute
error of CO is 2.57%, the minimum absolute error of CO
is 0.07% and the average absolute error of CO is 1.32%.
The maximum absolute error of HC is 0.96%, the
[3]
[4]
D7-4
P.K. Karra, S.C. Kong, “Diesel Engine Emissions Reduction
Using Particle Swarm Optimization”, Combustion Science and
Technology, 182:7,Taylor and Francis, pp. 879–903.
J. Zhang, Y. Li, J. Cao,” Sensor situation based on the multiple
linear regression forecast”, 2011 IEEE International Conference
on Computer Science and Automation Engineering (CSAE),
pp.47-50, China, 10-12 June 2011.
N. Amral, C.S.Ozveren, D. King,” Short term load forecasting
using Multiple Linear Regression”, 42nd International
Universities Power Engineering Conference, pp.1192 – 1198,
Brighton,
4-6
Sept.
2007.
Acceptance of Mobile Payment Application in
Indonesia
Hendra Pradibta
Informatics Management, State Polytechnics of Malang
ndropradibta@yahoo.com
Abstract - The Unified Theory of Acceptance and Use
of Technology (UTAUT) model with two additional
variables (perceived of Risk and perceived of trust) is used
to assess the users’ perception towards the adoption of
mobile payment application. The results of this research
show that from the correlation analysis performance
expectancy, effort expectancy, social influence and
perceived of trust have positive relationship towards
behavioural intention to use mobile payment application.
Whereas, perceived of risk hasnegative relationship on
behavioural intention. From the results, it can be
concluded that although Indonesian consumers have
willingness to use mobile payment application, they are
still considering trust and risk as main determinant to use
mobile payment application.
Keywords: mobile payment, UTAUT, risk and trust
I. INTRODUCTION
Over the past several years the number of mobile
telephone users has increased significantly. Driven by
the sharply increasing of mobile usage and the
evolvement of its functionality (Viehland and Leong,
2007), mobile telephone has become common devices
in our daily activities(Hwang et al., 2007). More and
more people rely on mobile devices to conduct their
business activities. Furthermore, this condition has
underpinned the shift of traditional commerce to
electronic commerce, which utilizes technology to
support the performances. From several forms of
electronic commerce, mobile payment has turned into
one of the promising business model in the near future.
Mobile payment enables consumers to performs their
activities such as paying goods and services using
mobile devices (Kim et al., 2009a). Mobility and wide
reach characteristics which are addressed with mobile
telephone has supported the penetration of mobile
telephone adoption for commerce activities.It can be
seen from several countries that has been utilized
mobile payment as alternative payment methods for
commerce transactions, such as MobilPay (Germany),
Paybox (United Kingdom), PayDirect (United States)
and many more (Norman, 2002). Concerning
Indonesia, within five years time, the number of mobile
telephone subscriber has increased dramatically from 30
million in 2004 to 180 million users by the end of 2009
(Asia, 2010). This number shows that Indonesia has big
potential opportunity to get exploit in mobile telephone
services such as mobile payment. Moreover,
development towards Cash Less Society also supported
the growth of this mobile service application. Yet, the
adoption of mobile payment has not been as good as
expected. User’s perceptions towards mobile payment
still need to be observed further to tackle this issues.
The greatest problems in the adoption of technology
such as mobile payment is lower respond from
users(Schierz et al., 2009). Therefore, this research
attempts to describe user’s perceptions towards the
adoption of mobile payment.
This research propose Unified Theory of Acceptance
and Use of Technology (UTAUT) model combine with
trust and risk to investigate user’s behavioural intention
towards technology adoption. So far, there is little
research regarding the acceptance of mobile payment in
Indonesia. Thus, this research will provide a
comprehensive study about factors that affecting
consumer to adopt the technology. The findings of this
research would be beneficial for both academics and
business practitioners (vendors). From academic point
of view, this research offers a framework to identify the
determinants of user’s perception on mobile payment
application. On the other hand, from business
perspective, this research would give guidance in order
improve the mobile payment usage in Indonesia.
II. LTERATURE REVIEW
A. Mobile Commerce
The widespread usage of mobile telephone and the
emerging of wireless technology have shifted the
business activities in general. Many businesses have
utilized mobile telephone as an alternative instrument
for delivering services (Medhi et al., 2009), such as
advertising, commerce and financial.
Many types of mobile financial application emerge
and are available in several countries such as MobilPay
(Germany), Paybox (UK), PayDirect (US) and many
more (Norman, 2002).In line with the widespread of
mobile commerce, mobile payment has appeared as an
interesting subject and become promising mobile
services in thefuture(Karnouskos and Vilmos, 2004).
Several industries have been influenced by the emerging
of mobile payment applications, including financial
services, retail, telecommunication, information
services, entertainment, and technology (Smith, 2006).
E1-1
It seems that business practitioners have assured that
mobile payment may generate competitive advantages
for their business.
Theoretically, mobile payment can be defined as a
payment for services or goods which is conducted
through mobile devices instead of paying with cash
(Karnouskos, 2004). In this business model, mobile
devices are employed for initiation, authorization and
completing the payment process.
III. RESEARCH FRAMEWORK
Performa
Perceived
Effort
Behaviour
Perceived
Social
B. Mobile Payment Application in Indonesia
Fig.1 Research Framework
Over the past several years there has been dramatic
increase of mobile telephone market in Indonesia. In
2004, the number of mobile telephone subscribers in
Indonesia reached 30 million users and dramatically
increased to 180 million only in five years time
(Euromonitor, 2010). The mobile subscriber penetration
in 2009 demonstrates that Indonesia is potential market
for mobile telephone and mobile services. It is
understandable if Indonesia has predicted to be the third
largest mobile telephone market after China and India
(Asia, 2010).
This research adopts the Unified Theory of
Acceptance and Use of Technology (UTAUT) model
(Venkatesh et al., 2003) as the basic framework to
investigate the user’s perception of mobile payment
application. Unified Theory of Acceptance and Use of
Technology model has been introduced to explain how
users’ differences influence the adoption of technology (
Park et al., 2007). Basically, UTAUT model is develop
from eight user acceptance models: Theory of Reasoned
Action, Technology Acceptance Model, Motivational
Model, and Theory Planned Behaviour, a combined
Theory of Planned Behaviour/Technology Acceptance
Model, model of PC Utilization, innovation diffusion
theory and social cognitive theory (Venkatesh et al.,
2003). From the research, Venkantesh et al. (2003) has
formulated four key constructs (performance
expectancy, effort expectancy, social influence, and
facilitating condition) as direct determinants of
behaviour intention and behaviour, while attitude
toward using technology, self efficacy and anxiety are
theorized not to be direct determinants of intention. In
addition gender, age, experience, and voluntariness of
use are considered to moderate the impact of the direct
determinants on behaviour intention and behaviour. The
most important finding in this study is that UTAUT
model can perform 70% of variance in usage intention
better that TAM studies alone. However, UTAUT
model is not perfect in describing user’s behaviour. In
specific technology adoption such as mobile commerce,
revision and modification may be needed (Venkatesh et
al., 2003). Hence, this research framework has added
perceived of risk and perceived of trust as new core
constructs to investigate user’s perception towards
technology usage.
Innovation in mobile applications seems become an
attractive and promising market, thus, it encourages
telecommunication companies to develop various
mobile applications such as mobile news, horoscope,
logos, ringtones, and mobile payment. Apparently,
among these applications mobile payment has risen into
promises mobile application. According to Central Bank
of Indonesia (Bank.Indonesia, 2010) total of electronic
money transactions including mobile payment
transactions has reached 60 billion rupiah by the end of
June 2010. This number has increased 50% from the
same month in 2009. Yet, only few providers are
offering this form of services to their consumers. It is
obvious that mobile payment market still have
opportunity to be expanded further. There are several
mobile payment applications that are familiar for
Indonesian consumer and are used as research objects in
this research, i.e. Nada SambungPribadi application,
download content application, mobile banking
application, and e-wallet application.
Alongside with these ICTs’ developments, it seems
that users’ perceptions about e-payment have been
increasing. According to Central Bank of Indonesia
(Bank.Indonesia, 2010), in May 2010 total of electronic
money circulation in public increased four times
compared with that in 2009. It is obvious that market
has accepted the adoption of electronic money as the
complement of cash money. However, not all of
consumers are familiar with the use of electronic money
in e-payment model such as mobile payment. Therefore,
it is still needed to investigate why consumers still take
into consideration to use e-payment and what factors
which influence them to use it.
A. UTAUT
Unified Theory Acceptance and Use of Technology
consist of four key constructs as direct determinants of
behavioural intention and behaviour (Venkatesh et al.,
2003). Performance expectancy, effort expectancy, and
social influence are as direct determinants of
behavioural intention, whereas, facilitates condition as
direct determinant for behaviour. This research has
proposed key constructs from UTAUT model, which
hold direct relationship with behavioural intention.
Thus, performance expectancy, effort expectancy and
social influence are selected as the key determinants that
E1-2
respondents, 66 peoples (66%) are male and 34 of them
are female which represents 34% of total respondents.
In addition, a reliability analysis was used for
measured the questionnaire items. Validity is defined as
the extent to which the indicator really measures the
model (Bryman and Bell, 2007). Because this research
has adopted concept from previous research and
literature reviews as constructive model (Kim et al.,
2009b; Wu and Wang, 2005; Venkatesh et al., 2003),
therefore, it can be considered as valid concept.As
expected the results of the measurement is shows good
degree of reliability since each variables obtain
coefficient alpha greater than 0.5(Hair et al., 1995;
Cronbach, 1951).
Table 1. Cronbach’s Alpha
influence users’ perception towards mobile payment
application.
Performance expectancy is used to measures how
much people perceive that using technology is useful to
improve job performance. This determinant has three
items used to investigate user’s perception regarding the
technology adoption. Effort expectancy is defined as
how ease of use is associated with the user’s experience
and their acceptance to adopt the system. Four items
will be performed in this determinant to examine user’s
perception regarding the technology adoption. Social
influence refers to how people feel that others’ advices
and recommendation to employ a certain systems is
important. Three questions have been proposed in the
online survey to examine user’s perception regarding
the technology adoption. These questions have been
developed based on items used in estimating UTAUT
model.
Constructs/ Variables
Performance
Expectancy (PE)
Effort Expectancy
(EF)
Social Influence (SI)
Perceived of Risk
(PoR)
Perceived of Trust
(PoT)
Behavioural Intention
(BI)
Fig. 2 UTAUT Model
Cronbach’s
Alpha
0.870
N of
items
3
0.905
4
0.611
0.859
3
4
0.847
4
0.855
3
Furthermore, the correlation analysis technique was
used to measures the relationship between variables.
This analysis is commonly used in quantitative analysis,
which is appropriate for this research. Pearson’s Product
moment is tends to be more accurate to measure
relationship between numeric data (Oakshoott: 2001),
hence, this research employed Pearson’s to analyse the
relationship between behavioural intention variables
with five other variables (performance expectancy,
effort expectancy, social influence, perceived of risk
and perceived of trust).
B. Perceived of Risk
Perceived of risk is defined as the extent to which
the prospective user expects the technology (mobile
payment application) to be risky. This determinant using
four items, which investigate the users’ perception
regarding losses or harm that will be occurred within the
adoption of technology. These four questions have been
used as determinant items in previous research in which
investigate the usage of mobile commerce technology
(Wu and Wang, 2005).
V. RESULT AND ANALYSIS
C. Perceived of Trust
It can be seen from the data collection that the
majority of the respondent’s age is between 20 to 30
years old that represents 84% from the total
respondents. Respondents with age above 30 are 13
peoples (13%) and 3 peoples are under 20 years old
with 3% of total.
Perceived of trust is defined, as consumers’ believe
that particular system will be conducted based on the
procedures and their expectations. This construct
proposes four questions, which focus on the users’ trust
regarding the process of mobile payment application.
These four questions are developed based on the
previous research in mobile payment nature (Kim et al.,
2009b).
From the results about respondent’s mobile payment
applications it can shows that Nada SambungPribadi
(NSP) is the most popular mobile payment application
with 44 peoples from the total 100 respondents. Mobile
banking application gains 36% (36 peoples) from the
total of 100 respondents, and place it in the second
position for the mobile payment application usage.
IV. METHODOLOGY
The online survey was launched in May 2010 until
July 2010 by placing the address (URL) in social
networking site (Facebook) and group associated with
mobile application services. The survey received 100
responses from the consumers. From the total of 100
Correlations analysis is performed to measure
the relation between two variables. The Correlations
E1-3
combination of two additional variables (perceived of
risk and perceived of trust) will provide in depth insight
regarding the successful and the failure of mobile
payment adoption.
table below shows that all variables are significantly
correlated with Behavioural Intention variables, which
is shown by the significance value (p<0.05 and p<0.01).
From all five variables in the model research, four
variables have positive correlation with behavioural
intention (i.e. PE= 0.453, EF= 0.259, SI= 0.350, and
PoT= 0.521). Whereas, perceived of risk has negative
correlation (PoR= -0.274).
REFERENCES
[1]
Table 2. Correlations
Pearson
Correlation
(BI)
[2]
BI
PE
EF
SI
PoR
PoT
1
.453
.259
.350
-.274
.521
[3]
Nonetheless, the correlation table indicates weak
association towards behavioural intention because it
shows value <0.8. Relation between perceived of trust
variable with behavioural intention is more likely to be
the highest value (0.521). In conclusion it can be
validated that each variables (partially) have significant
correlation with behavioural intention, but, it is
considered as weak relationship.
[4]
[5]
[6]
[7]
VI. CONCLUSION
[8]
As mention previously, this study attempts to describe
user perception towards mobile payment in terms of
applying UTAUT model combined with risk and trust.
1) The results show positive relationship for
performance expectancy, effort expectancy, social
influence, and perceived of trust towards
behavioural intention to use mobile payment. It
indicates that if performance expectancy value
increase, the value for behavioural intention
variable also increases. This condition also applies
to others variable.
2) Perceived of risk has negative relationship on
behavioural intention. It shows that if user
interpretation concerning the risk towards mobile
payment application is low, user intention to use
mobile payment application is also low.
3) Fivevariables (performance expectancy, effort
expectancy, social influence, perceived of trust and
perceived of risk) have significant relationship on
the users’ acceptance of mobile payment
application, thus, it can be used to predict the result
of behavioural intention variable
From the descriptive analysis, it appears that
Indonesia consumers are having willingness to use
mobile payment as alternative payment for their
transactions. Yet, they still consider trust and risk as
main determinant on their perception towards mobile
payment. From managerial implication, the research
confirms the important of these five variables as the
main determinants in order to increase the adoption of
mobile payment application. Therefore, mobile payment
providers have to concerning their application to meet
the user needs and wants, particularly on five variables
above. This research has validated the use of UTAUT
model to investigate user’s acceptance regarding the
adoption of technology. Moreover, with the
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
E1-4
Asia, D. M. A. (2010) Indonesia:Mobile Penetration
Available
at:
http://comm215.wetpaint.com/page/Indonesia%3A+Mobil
e+Penetration (Accessed: 29 June 2010).
Bank.Indonesia. (2010) 'Jumlah Uang Elektronik ',
[Online].
Available
at:
http://www.bi.go.id/web/id/Statistik/Statistik+Sistem+Pem
bayaran/Uang+Elektronik/JmlUang.htm (Accessed: 30
July 2010).
Bryman, A. and Bell, E. (2007) Business research
methods. Oxford Oxford Univ. Press.
Christer, C. (2006) Proceedings of the 39th Annual Hawaii
International Conference on System Sciences (HICSS'06)
Track 6.
Cronbach, L. (1951) 'Coefficient alpha and the internal
structure of tests', Psychometrika, 16, (3), pp. 297-334.
Euromonitor (2010) Mobile Telephone Subscribtions
Available
at:
http://portal.euromonitor.com/Portal/Statistics.aspx
(Accessed: 15 March 2010).
Hair, J. F., Anderson, R. E., Tatham, R. L. and Black, W.
C. (1995) Multivariate Data Analysis
Hwang, R. J., Shiau, S. H. and Jan, D. F. (2007) 'A new
mobile payment scheme for roaming services', Electronic
Commerce Research and Applications, 6, (2), pp. 184-191.
Karnouskos, S. (2004) 'Mobile payment: A journey
through existing procedures and standardization initiatives',
IEEE Communications Surveys and Tutorials, 6, (4), pp.
44-66.
Kim, C., Mirusmonov, M. and Lee, I. (2009a) 'An
empirical examination of factors influencing the intention
to use mobile payment', Computers in Human Behavior,
26, (3), pp. 310-322.
Kim, C., Tao, W., Shin, N. and Kim, K. S. (2009b) 'An
empirical study of customers' perceptions of security and
trust in e-payment systems', Electronic Commerce
Research and Applications, 9, (1), pp. 84-95.
Medhi, I., Gautama, S. N. N. and Toyama, K. (2009) 'A
Comparison of Mobile Money-Transfer UIs for NonLiterate and Semi-Literate Users', in Greenberg, S.,
Hudson, S. E., Hinkley, K., RingelMorris, M. and Olsen,
D. R.(eds) Chi2009: Proceedings of the 27th Annual Chi
Conference on Human Factors in Computing Systems,
Vols 1-4. pp. 1741-1750.
Norman,
M.
S.
(2002)
M-Commerce:
Technologies,Services,and Business Models. John Wiley
&Sons, Inc.
Oakshott, L. (2001) Essential Quantitative Methods for
Business Management and Finance.
Park, J. K., Yang, S. J. and Lehto, X. R. (2007) 'Adoption
of mobile technologies for Chinese consumers', Journal of
Electronic Commerce Research, 8, (3), pp. 196-206.
Schierz, P. G., Schilke, O. and Wirtz, B. W. (2009)
'Understanding consumer acceptance of mobile payment
services: An empirical analysis', Electronic Commerce
Research and Applications.
Smith, A. D. (2006) 'Exploring m-commerce in terms of
viability, growth and challenges', International Journal of
Mobile Communications, 4, (6), pp. 682-703.
Venkatesh, V., Morris, M. G., Davis, G. B. and Davis, F.
D. (2003) 'User acceptance of information technology:
Toward a unified view', MIS Quarterly: Management
Information Systems, 27, (3), pp. 425-478.
Viehland, D. and Leong, R. (2007) 'Acceptance and Use of
Mobile Payments', ACIS 2007 Proceedings. pp. Wu, J.-H.
and Wang, S.-C. (2005) 'What drives mobile commerce?:
An empirical evaluation of the revised technology
acceptance model', Information & Management, 42, (5),
pp. 719-729
Attitude Consensus of Multiple Spacecraft with
Three-Axis Reaction Wheels
Harry Septanto1, Bambang Riyanto Trilaksono2, Arief Syaichu-Rohman2 and Ridanto Eko
Poetro4
1
Center of Satellite Technology, Indonesian Institute of Aeronautics and Space
School of Electrical Engineering and Informatics, Insitut Teknologi Bandung
4
Faculty of Mechanical and Aerospace Engineering, Insitut Teknologi Bandung
1
hseptanto@gmail.com
2,3
a result in [1].
Abstract— This paper deals with attitude consensus of
multiple spacecraft in a team where each spacecraft
applies three-axis reaction wheels. Two control laws for
two different cases are presented. The attitude consensus
of multiple spacecraft is achieved under a connected
information exchange topology. Simulations run to verify
the effectiveness of the control laws in reaching the
attitude consensus.
Index Terms—Attitude consensus, connected graph,
quaternion feedback, three-axis reaction wheels.
I. INTRODUCTION
Recently, many research efforts have been dedicated
to analysis and design spacecraft formation flying via
consensus approach. W. Ren in [1] proposed
quaternion-based control laws for three different cases,
including attitude alignment under undirected and
directed information-exchange graph. Y. Igarasi et al in
[2] addressed passivity-based attitude synchronization
on
3
under
strongly
connected
information-exchange graph. W. Ren in [3] proposed
Modified Rodriques Parameters-based control laws,
including passivity-approach attitude synchronization
under undirected connected information-exchange
graph. H. Du et al in [4] proposed Modified Rodriques
Parameters-based continuous finite-time attitude
controllers for single spacecraft case and attitude
synchronization
case
under
directed
information-exchange graph. C. G. Mayhew et al [5]
proposed a quaternion-based hybrid feedback controller
for attitude synchronization under connected and acyclic
information-exchange graph. All of these researches
used dynamic model of spacecraft rigid body with
external control torque.
In practice, an actuator that may generate above type
of control torque is, for example, thruster. Nevertheless,
there is another actuator that also usually taken into
account on the research field about spacecraft control
system, i.e. reaction wheel. This actuator is not an
external control torque. Principally, it absorbs
(distributes) angular momentum from (to) the
spacecraft.
This paper addresses to attitude consensus of multiple
spacecraft in a team where each spacecraft applies
three-axis reaction wheels. This research is motivated by
II. KINEMATIC AND DYNAMIC OF RIGID BODY
SPACECRAFT
A. Vector Notation
Consider a vector as follows: =
ℱ . Here, ℱ is
a vetrix associated to the inertial reference frame, i.e. a
column matrix whose three unit vectors , and ;
ℱ =
.
∈ ℝ is a column matrix
whose three components of expressed or decomposed
into the inertial reference frame.
Note that subscript “ ” in ℱ and superscript “ ” in
denote the frame of interest, i.e. inertial reference frame.
The other two frames of interest would be denoted by
subscript/ superscript “ ” and “ ” for the spacecraft’s
fixed body frame and the spacecraft’s desired frame,
respectively. For brevity, the inertial reference frame,
the spacecraft’s fixed body frame and the spacecraft’s
desired frame may write the inertial frame, the body
frame and the desired frame, respectively.
B. Rotation Matrix
Rotation matrix is a matrix to transform a vector
expressed in one frame to be expressed in another frame.
The set of all rotation matrices—for simplicity, a
rotation matrix is denoted by —is the special
orthogonal group in ℝ , i.e.
∈ SO 3 whose
1 0 0
properties:
=
,
=
= 0 1 0! = "
0 0 1
and det
= 1.
In a rigid body general case, a rotation matrix for
transforming a vector expressed in the inertial frame to
be expressed in the body frame, & , is defined as
follows:
&
= ℱ& ∙ ℱ
(1)
Let ℱ& and ℱ are rotating w.r.t. (with respect to) each
other. Then relationship between ℱ& and ℱ is time
dependent. It implies the rotation matrix of these frame
is also time dependent, & = & ( . Take time
derivative of ℱ& w.r.t. ℱ —note that it could be
interpreted as how observers in ℱ see the motions of
ℱ& —then (2) is satisfied.
E2-1
)
*ℱ+
*,
- = .)*&0/
S78&& 9ℱ& ⇔ )
*,
*ℱ+ ;
*<
)
*&/1
)
-
*&/2
5 & × ℱ& =
- 3 =4
- = ℱ& S78&& 9
*,
*,
D. Kinematic Equation
(2)
where 4
5 & = 8&& ℱ& = =4&&
4&&
4&& >ℱ& is the
angular velocity vector of ℱ& w.r.t. ℱ which is
decomposed in ℱ& and
0
−4&&
4&&
?78&& 9 = @ 4&&
0
−4&& B is denoting the
&
&
−4&
4&
0
&
skew-symmetric matrix of 8& . Note that 4
5 & is also
called absolute angular velocity of ℱ& since it rotates
w.r.t. the inertial frame, ℱ .
Consider & = ℱ& ∙ ℱ ⇔ ℱ = ℱ& & , where
& =
& , and take time derivative of ℱ w.r.t. ℱ ,
then (3) is satisfied.
)
*ℱC ;
*<
- =0=)
*ℱ+ ;
*<
⇔ 0 = ℱ& )?78&& 9
−?78&&
9
&
-
&
&
+ ℱ& )
+)
*
+C
*<
*
+C
*<
-
--⇒)
*
+C
*<
- =
(3)
C. Attitude Representation Using Unit Quaternion
From the Euler’s theorem, an angular displacement of
a frame w.r.t. another frame can be obtained by a single
rotation about an axis over an angle. The former axis and
angle correspond to column matrix F and angle G ,
respectively, whose properties: F = F and F F = 1.
The Euler parameter composed of F and G is defined
as follows:
J > ∈ K (unit 3-sphere)
I +J J=1
)
−J >
H ⊗ H∗ = H∗ ⊗ H = =1
0
0 0> = U
*<
)
*J+C
*<
- =X
*Y+C
*<
)
*J+C ;
*<
-Z
(8)
*Y+C
*<
- =
= − J& 8&&
(9)
? J& + I& " 8&&
(10)
In the unit quaternion, the successive rotation can be
represented by the quaternion multiplication as follows,
[9]:
(5)
(6)
(7)
Note that the rotation matrix can be related to H
through Rodriques formula whose the map : K →
SO 3 , where H = −H .
(11)
where H&* = H *
H& . Note that map is a
group homomorphism, [10]. For convenience in
notation, the rotation matrices is written as follows:
H&* , * = * = H* ∗ = H * and
&* =
H& .
& =
Consider &* = ℱ& ∙ ℱ* ⇔ ℱ* = ℱ& &* , where
= *& . Take time derivative of ℱ* w.r.t. ℱ* , then
&*
0 = ℱ& )?78&&* 9
&*
−?78&&* 9 &* .
+)
* +[
- -.It
*< *
implies )
)
that ℱ& 8&&*
⇔ 8&&* = 8&&
* +[
*< *
=
Noting
=4
5 &* = 4
5& −
&
&
4
5* =
− 8* 9
− 8* , then the
unit quaternion representation of attitude kinematic for
&* is as follows:
ℱ& 78&&
*H+[
*<
- =X
*
*Y+[
*<
)
*J+[ ;
*<
- Z
*
(12)
where
(4)
The Euler parameter (or also called the unit
quaternion)
has
properties:
the
quaternion
multiplication (5) and conjugate of a unit quaternion (6)
that satisfies (7)
H∗ = =I
*H+C
where
O
I I −J J
H ⊗H =.
3
I J +I J +J ×J
declared in (3); the time derivative
of rotation matrix & can be represented in a unit
quaternion as follows:
where I = cos ∈ ℝ , J = F sin ∈ ℝ and ‖H‖ =
O
* +C
*<
H&* = H* ∗ ⊗ H& = H * ⊗ H&
Rotation matrix & represents the attitude history of
ℱ& w.r.t. ℱ . Practically, it may be obtained—even
indirectly—from reference sensor(s), e.g. star tracker,
sun sensor and geomagnetic sensor. Meanwhile, 8&&
may be measured by means of rate gyro sensor or
computed from the attitude data.
H = =I
Consider )
)
*J+[
*<
*Y+[
*<
- =
*
= − J&* 8&&*
? J&* + I&* " 8&&*
(13)
(14)
Through fact (7) and (11), then (12)-(14) is defined as
kinematic equation of attitude error.
E. Dynamic Equation of Spacecraft with Three-Axis
Reaction Wheels
The following assumptions are used in deriving the
equation of the rigid spacecraft dynamic with 3-axis
reaction wheels:
• The location of the reaction wheels’ mass center
is in the origin of ℱ& .
E2-2
•
The rotation axis of each reaction wheels is
coincidence with ] -axis of ℱ& , ^ -axis of ℱ&
and z-axis of ℱ& , respectively.
•
The transversal components of inertia moment
of each reaction wheels are not considered
effective in operation.
•
There is no external torques applied to the
system.
Hence, the equations of the spacecraft dynamic are
given as follows, [6]:
`)
_
*8++C
*<
- +` )
*8+aC
` ))
*<
- = −?78&& 97`8&& + ` 8&b 9
*8++C
*<
- +)
*8+aC
*<
--=
c
`
⇔.
`
`
3
`
&
.−?78&
97`8&&
+`
8&b
8&
lm & n q
(
p
` k
3
` k 8&b p
km
np
( o
j
93 = r
)
*8++C
)
*<
*8+aC
*<
-
-
s
(16)
Using a fact 2.17.3 in [7], then (17) is equivalent to
(15) and (16).
_
`−`
`−`
)
)
*8++C
*<
*8+aC
*<
- = −?78&& 9`8&& − ?78&& 9` 8&b −
- = ?78&& 9`8&& + ?78&& 9` 8&b + ``
c(17)
III. SPACECRAFT FORMATION REPRESENTATION
A. Dynamic and Kinematic Equation
Index of a spacecraft in a formation composed by
t -spacecrafts would be denoted by subscript (and
sometimes both sub- and superscript) “ u ” and its local
neighbor would be denoted by subscript (and sometimes
both sub- and superscript) “ v ”, where 1 ≥ u, v ≥ t.
Then, the first equation of (17) becomes as follows:
7` x − `
_
x
*8+
9m
+ y
y C
*<
−? )8&
& x
x
n = −? )8&
-`
x
8b
& x
x
& x
x
−
- ` x 8&
x
& x
x
c
)
(18)
Kinematic equation that represents attitude of u
spacecraft body frame, ℱ& x , w.r.t. desired frame, ℱ* , is
*H+
y [
*<
- =X
*
*Y+
y [
*<
)
*J+
y [
*<
;
- Z
*
(19)
Meanwhile, kinematic equation that represents
attitude of u spacecraft body frame, ℱ& x , w.r.t. its
neighbor body frame, ℱ& z , is as follows:
)
(15)
where a symmetric positive definite matrix ` ∈
ℝ × kgm is the moment inertia of the spacecraft; a
diagonal positive definite matrix ` ∈ ℝ × kgm is
the inertia moment of the reaction wheels;
∈
ℝ Nm is the control torque produced by the reaction
wheels; and 8&b ∈ ℝ rads
is the column matrix of
reaction wheels absolute angular velocity decomposed
in ℱ& . In practice, a reaction wheel has a limited angular
velocity in operation. However, in this paper, the
reaction wheels are assumed to operate within the
saturation limit.
One may rewrite (15) as follows:
&
&
&
`
.−?78& 97`8& + ` 8b 93 = .
`
the following:
*H+
y + {
*<
-
& z
*Y+
=.
y + {
*<
)
*J+
y + {
*<
;
-
& z
3 (20)
B. Information-Exchange between Spacecraft in
Formation
Information-exchange between spacecraft is modeled
by a directed graph as follows:
| } ≜ •} , € } , • }
(21)
where •} = ‚ƒ x |u = 1,2, … , t‡ is the node set;
€} ⊆ •} × •} is the edges set; and •} = ‰ x z ∈
ℝ}×} is the adjacency matrix of the graph |} , where
1 ≤ u, v ≤ t ∈ ℤ and u ≠ v.
Note that undirected graph is a special case of
directed graph. In a directed graph, the edge 7ƒ x ƒ z 9 ∈
€} denotes that spacecraft v can receive information
from spacecraft u ; ƒ x is the parent node and ƒ z is
child node; and ‰ x z = 1 . Meanwhile, the edge
denoted by 7ƒ x ƒ z 9 in the undirected graph
corresponds to edges 7ƒ x ƒ z 9 and 7ƒ z ƒ x 9 ; and
‰ x z = ‰ z x = 1 . If there is no edge between
spacecraft u and spacecraftv, then ‰ x z = ‰ z x = 0.
A balanced graph is a graph whose ∑}zŽ ‰ x z =
}
∑zŽ ‰ z x , where 1 ≤ i, j ≤ n ∈ ℤ. Every undirected
graph has symmetric•} . Thus every undirected graph
is a balanced graph. An undirected graph is connected if
there is an undirected path between every pair of
different spacecraft.
IV. PROBLEM STATEMENT
As noted above, there is a redundancy in representing
the same physical orientation using unit quaternion. The
relationship between quaternion and physical
orientation is the following: H = −H ∈ • 3 ,
where H ∈ K . Through this facts, agreement value of
attitude in • 3 would be represented using two values
of unit quaternion in K .
Nevertheless, when the attitude consensus is
achieved, the agreement value on • 3 × ℝ is the
following:
)
& x
7H&
x
( = ±HF 9, 8&
& x
x
( = 8F -as ( → ∞
(22)
where & x ±HF ∈ • 3 and 8F ∈ ℝ3 , for 1 ≤ u ≤
t. Regarding to (7), the corresponding agreement value
could be written on • 3 × ℝ as (23) or on K × ℝ
as (24).
E2-3
7
& x & z
7H&
x & z
where “
7H&
x & z
( = ±U9,_
_8&
( − 8&
( =“
_8&
( − 8&
( =“
x
& x
& z
z
( = ±U,_
×
x
& x
= =0
& z
z
×
-as ( → ∞
×
-as( → ∞
0> , 1 ≤ u, v ≤ t and u ≠ v.
0
(23)
V. MAIN RESULT
(24)
Definition IV-1 Let t − spacecrafts in a team (18),
where 1 ≤ u, v ≤ t and u ≠ v, exchange the information
each other such that there is a corresponding graph of
the information state transmission direction model, |} .
The attitude consensus is achieved, if each spacecraft
has a control torque depending on information states of
its local neighbor, ” x , such that the spacecrafts in a
team satisfy the following conditions:
•
•
•H& x & z 0 − U• > 0 ⇒
u— •H& x & z ( − U• = 0
<→˜
™8&&
x
x
0 − 8&
u— ™8&&
<→˜
x
x
& z
z
& z
z
( ™ = 0,
where
)H& x & z 0 , 8&& xx 0 − 8& z 0 - ≠
−U, “ × . In addition, −U, “ × is an isolated point,
where if the spacecrafts in a team is there at ( = 0, then
they will agree to stay there for all the time, i.e. attitude
consensus is achieved since( = 0. Note that this first
case corresponds to the leaderless consensus case.
& z
Now, consider the second case corresponding to the
leader-following consensus case. Let there is an addition
spacecraft—a virtual leader spacecraft—having index
“ ” as a root of the corresponding graph |}š that
may “transmits” (and not “receives”) the information
states to one, several or all spacecrafts in a team (18),
where 1 ≤ u ≤ t, 1 ≤ v ≤ t + 1 = and u ≠ v. In this
case, the attitude consensus is achieved, if each
spacecraft has a control torque depending on
information states of its local neighbor, ” x , such that
the spacecrafts in a team satisfy the following
conditions:
•
•
•H& x & z 0 − U• > 0 ⇒
u— •H& x & z ( − U• = u— •H&
<→˜
<→˜
U‖ = 0
™8&& xx
0 −
u— ™8&&
<→˜
u— •8&&
<→˜
x
x
x
x
& z
8& z
0 ™>0⇒
( − 8&
( −
& z
z
8&& **
( ™=
( • = 0,
First, it is worth to consider some following lemmas.
Lemma V-1 If for 1 ≤ u, v ≤ t and u ≠ v , ‰ x z =
‰ z x , then
∑}xŽ ∑}zŽ ‰ u
= ∑}xŽ 8&&
Proof:
See [1].
x & *
( −
where
)H& x & * 0 , 8&& xx 0 − 8&& ** 0 - ≠
−U, “ × . In addition, −U, “ × is an isolated point,
where if the spacecrafts in a team is there at ( = 0, then
they will agree to stay there for all the time, i.e. attitude
consensus is achieved since( = 0.
)8&&
v
− 8&
x
x
x ›
7∑}zŽ
x
‰
œ = •€ ∑}xŽ ∑}š
zŽ ‰ x
}
œž = •€ Ÿ 8&&
Proof:
xŽ
x ›
x
z
}š
_+€&
x & z
9
•H&
x & z
− U• ,
zŽ
z
*J+ y + { ¡
¢
*<
& z
›
} }š
= •€ Ÿ Ÿ ‰ x
xŽ zŽ
z
8&&
x & z
x & z
Ÿ‰ u
œž = 2•€ ∑}xŽ ∑}š
zŽ ‰ x
›
- €&
& z
z
€&
u v
Lemma V-2 Let
then
0 ™>0⇒
( − 8&
Remark IV-2 Regarding the argument of globally
asymptotically stable for quaternion-based attitude
control in [8], the attitude consensus designed to follow
Definition IV-1 is necessary for globally asymptotically
stable guarantee.
v
n
€&
x & z
7I&
x & z
− 19
*Y+ y + {
*<
_
›
x
x & z €& x & z
Since spacecraft u uses information states of
spacecraft v that measured at its own frame, i.e. ℱ& z ,
hence the information states is supposed that measured
at spacecraft u frame, i.e. ℱ& x . Therefore, 4
5 u v =
4
5
u
−4
5
v
= ℱ&
x
œž = •€ ∑}xŽ ∑}š
zŽ ‰
›
)8&&
x
x
& x
x z )8& x
− 8&
−
& z
z
-.
& z
8& z
›
- €&
Then,
x & z
.
Suppose ‰ x z = ‰ z x and 8&& ** = “ × , where
= t + 1. Then, following Lemma V-1, œž becomes as
follows:
}
œž = •€ Ÿ 8&&
xŽ
x ›
x
}š
Ÿ‰ u
zŽ
v
€&
x & z
Theorem V-3 Let t −spacecrafts in a team exchange
the information state (i.e. attitude H& x ) each other such
that the corresponding graph of the information state
transmission direction model, |} , is connected. Let there
is an addition spacecraft—a virtual spacecraft—having
index “ ” as a root of |}š that may “transmits”
(and not “receives”) the information states to one,
several or all spacecrafts in a team (18). Let, for
1 ≤ u ≤ t, 1 ≤ v ≤ t + 1 =
and u ≠ v, the control
torque of each spacecraft in a team is as follows:
E2-4
€&
( + £ 8 x 8&
(
& x
x
x & z
(25)
where positive scalar •€ ∈ ℝ¤¥ , symmetric positive
definite matrix £8 x ∈ ℝ × > 0 , 7` x − ` x 9 ∈
ℝ × > 0 and ‰ x z is the element of a corresponding
adjacency matrix of |}š . If H& * ∈ K is a constant
and 8&& ** = =0 0 0>› = “ × then consensus is
achieved with the agreement value 7 & x ±HF , 8F 9 ∈
• 3 × ℝ , where:
HF = H&
and
•
= H&
*
x
8F = 8
= 8&
∞ = “3×1
•
& x
x
( → ∞ = H&
z
( → ∞ = 8&
& z
z
(→∞ ;
(→
§
1
t
œž = œž + Ÿ 8
u=1
u §
u 7` x
−`
x
x
98
u
u
. Taking time
8&&
9¨
(
x
x
©
Considering Lemma V-2, (18) and (25), one has
}
œž = Ÿ 8&&
xŽ
}
= − Ÿ 8&&
x
x ›
7−?78&& xx
x
_−?78&
x
& x
x ›
£8 x
x
9`
8&&
x
x
9` x 8&&
x
x
x
_
8&b xx − £8 x 8&&
≤0
Noting that œž = 0 implies 8&&
x
x
=“
×
x
x
9
(18) and (25), €& x & z = “ × , if 8&& xx = “
Clearly,€& x & z = “ × implies H& x & z = ±U .
. Considering
×
= •€ ∑}zŽ ‰
.
Note that −U, “ × is an isolated point, then the
spacecrafts in a team will agree to stay there for all the
time, if 7H& x & z , 8&& xx 9 correspond to this point at
( = 0. Otherwise, they will asymptotically approach to
the agreement attitude such that H& x & z ( = +U at
( → ∞. However, the situation of H& x & z = ±U has
the same agreement value in the physical space • 3 ,
i.e. & x & z U = & x & z −U . The situation also
implies
−HF .
Therefore,
& x +HF = & x
regarding LaSalle’s theorem, they have globally
asymptotically stable guarantee, i.e. attitude consensus
is achieved.
•
•
HF = H&
€&
x
8F = 8&
& x
x
x & z
( + £ 8 x 8&
& x
x
(
(26)
( → ∞ = H&
z
( → ∞ = 8&
& z
z
( → ∞ ; and
( → ∞ = “3×1
The proof may be obtained through the same way as the
proof of Theorem V-3 above.
VI. NUMERICAL EXAMPLES
Simulations of some cases are presented here to verify
the main results. All simulations are run use following
constraints:
• H * = −0.7071 ∗ =1 0 1 0>T
1 0.1 0.1
• `
= 0.1 0.1 0.1!
0.1 0.1 0.9
0.8 0.1 0.2
• `
= 0.1 0.7 0.3!
0.2 0.3 1.1
0.9 0.15 0.3
• `
= 0.15 1.2 0.4!
0.3
0.4 1.2
• `
= 0.01"
• `
= 0.015"
• `
= 0.02"
0 1 0 0
1 0 1 1
• • š = ‰x z =r
s
0 1 0 0
0 0 0 0
• •ℰ = 0.05
• £8 = ` 1 ; £8 = ` 2 ; £8
=`3
To simplify the notation, unit quaternion of
spacecraft
u
is
denoted
by
§
1
2
3
I
J
J
J
H& x =
.
u
u
u
u
1
0
-1
0
100
200
300
400
500
600
Time(s)
700
800
900
1000
0
100
200
300
400
500
600
Time(s)
700
800
900
1000
0
100
200
300
400
500
600
Time(s)
700
800
900
1000
0
100
200
300
400
500
600
Time(s)
700
800
1
0
-1
Remark V-4 The addition spacecraft having index
“ ” in Theorem V-3 has the same property as “virtual
leader” in [1], i.e. transmits (and not receives) the
desired agreement value.
2
ε(i)
1
0
-1
1
3
ε(i)
Corollary V-5 Consider Theorem V-3 and suppose there
is no a virtual spacecraft. Hence, there is only |} and
the information exchange topology is a connected. Let,
for 1 ≤ u, v ≤ t and u ≠ v, the control torque of each
spacecraft in a team is as follows:
x z
where positive scalar •€ x z ∈ ℝ¤¥ , symmetric positive
definite matrix £8 x ∈ ℝ × > 0, 7` x − ` x 9 ∈
ℝ × > 0 and ‰ x z is the element of a corresponding
adjacency matrix of |} . Therefore, consensus is
achieved with the agreement value 7 & x ±HF , 8F 9 ∈
• 3 × ℝ , where:
Proof:
Proof:
Consider a corresponding Lyapunov candidate function
as follows:
œ = œ + ∑tu=1 8 uu 7` x − `
2
derivative of it, one has
x
η(i)
z
1
= k € ∑}š
zŽ ‰ x
ε(i)
x
0
-1
i=1
i=2
900
1000
i=3
desired
Fig 1. Attitude of spacecrafts in a team in according to Theorem V-3
E2-5
ACKNOWLEDGMENT
η(i)
1
0.5
0
0
50
100
150
Time(s)
200
250
300
0
50
100
150
Time(s)
200
250
300
The authors would like to gratefully acknowledge
suggestions and comments of one of our draft paper on
the same subject from Prof. Wei Ren and his former
student. This research is partly supported by Beasiswa
Unggulan, BPKLN, Kemendikbud, Indonesia.
ε(i)
0.5
1
0
-0.5
2
ε(i)
1
REFERENCES
0.5
0
0
50
100
150
Time(s)
200
250
300
0
50
100
150
Time(s)
200
250
i=1
i=2
i=3300
[1]
3
ε(i)
1
0
-1
Fig 2. Attitude of spacecrafts in a team in according to Corollary V-5
Two figures above show that the spacecrafts in a team
converge to the same attitude. For a case with desired
attitude “information” from the virtual spacecraft, the
multiple spacecraft converge to the desired attitude (Fig.
1). For a case without any desired attitude, the multiple
spacecraft converge to the common attitude (Fig. 2).
VII. CONCLUDING REMARK
This paper presented two control laws for two
different cases. The attitude consensus of multiple
spacecraft is achieved under a connected information
exchange topology. Simulations run to verify the
effectiveness of the control laws in reaching the attitude
consensus.
W. Ren, “Distributed Attitude Alignment in Spacecraft
Formation Flying,” In. J. Adapt. Control Signal Process, 2007,
21, pp. 95-113.
[2] Y. Igarashi, T. Hatanaka, M. Fujia and M. W. Spong,
“Passivity-Based Attitude Synchronization in SE(3),” IEEE
Transaction on Control System Technology, 2009, Vol. 17, No.
5.
[3] W. Ren, “Distributed Cooperative Attitude Synchronization and
Tracking for Multiple Rigid Bodies”, IEEE Transactions on
Control Systems Technology, vo. 18, No. 2 March 2010.
[4] H. Du, S. Li and C. Qian, “Finite-Time Attitude Tracking
Control of Spacecraft With Application to Attitude
Synchronization”, IEEE Transactions On Automatic Control,
Vol. 56, No. 11, November 2011.
[5] C. G. Mayhew, R. G. Sanfelice, J. Sheng, M. Arcak and A. R.
Teel, “Quaternion-based hybrid feedback for robust global
attitude synchronization”, Submitted to IEEE Transactions on
Automatic Control (Received: December 12, 2011) .
[6] P. C. Hughes, “Spacecraft Attitude Dynamics”, Dover
Publication, Inc., 2004.
[7] D. S. Bernstein, “Matrix Mathematics: Theory, Facts and
Formula”, Princeton Universiy Press, 2009.
[8] S. M. Joshi, A. G. Kelkar and F. T.-Y. Wen, “Robust Attitude
Stabillization of Spacecraft using Nonlinear Quaternion
Feedback,” IEEE Transactions on Automatic Control, 1995, Vol.
40, No. 10.
[9] M. D. Shuster, “A survey of attitude representations”, The
Journal of the Astronautical Sciences, Vol. 41, No. 4,
October-December 1993, pp. 439-517.
[10] C. G. Mayhew, R. G. Sanfelice and A. R. Teel, “On the
non-robustness of inconsisten quaternion-based attitude control
system using memoryless path-lifting schemes”, 2011 American
Control Conference on O’Farrell Street, San Francisco, CA,
USA, June 29-July 01, 2011.
E2-6
Implementing Naïve Bayes Classifier and Chi
Square on the Abstract to Classify Research
Publication Topics
1
Imam Fahrur Rozi 1, Rudy Ariyanto 2
Magister Teknik Elektro Universitas Brawijaya, 2 Politeknik Negeri Malang
1
imam.rozi@gmail.com, 2 rudy.ariyanto@poltek-malang.ac.id
Abstract—Abstract based-classification of research
publication topics, in the past has been very much done in
the manual way. Sometime, it could be done difficultly by
operator who don’t have any good knowledge on all of
research topics including its coverage. In this research, we
develop an automatic Indonesian publication document
classifier using Naïve Bayes Classifier (NBC). The first
step of classification process is preprocessing (including
stemming process, removing stopword and feature
selection). That pre-processing processes will be applied
both in training process of training document collections
or testing process using testing documents. Training result
will be matched with testing document. The experiment
result shows that NBC could be used to classify abstract
document based on its topics effectively. NBC also has a
simple algorithm, so it makes simple to be
implemented.Feature selection by using Chi Square also
could improve the accuracy of this classification system,
even in this system evaluation the effect of Chi Square is
not too significant because of the limited dataset used in
training or learning process.
Index Terms—Text Classification, Naïve Bayes, Chi
Square, Feature Selection
I. INTRODUCTION
N
AIVE Bayes Classifier has been one of the core
frameworks in the information retrieval research
for many years. Recently Naïve Bayes is emerged as
research topic itself, because it sometimes achieves
good performance on various tasks, compared to more
complex learning algorithms, inspite of wrong
independence assumptions on Naïve Bayes.
Naïve Bayes is also an attractive approach in the text
classification task because it is simple enough to be
practically implemented even with a great number of
features. This simplicity enables us to integrate the text
classification and filtering modules with the existing
information retrieval system easily. It is because that the
frequency related information stored in the general text
retrieval systems is all the required information in naïve
Bayes learning. No further complex generalization
processes are required unlike the other machine learning
methods such as SVM or boosting. Moreover,
incremental adaptation using a small number of new
training documents can be performed by just adding or
updating frequencies.
Several earlier works have extensively studied the
naive Bayes text classification. They used English text
to be classified. NBC also could be implemented to
classifiy text using Bahasa Indonesia, for example:
Indonesian News Document Classification (Wibisono,
2005).
In this research, we implement NBC to classify
research publication abstract document based on its
topic11 .This application was developed as web based
application, that we could access it online using internet
connection. We also combine NBC with Chi Square
stochastic test as feature selection method.
II. UNITS
Some equations used in this research, are shown in
this section.
A. Chi Square Test (χ2)
Chi-square testing (χ2) is a well-known discrete data
hypothesis
testing
method
from
statistics,
whichevaluates the correlation between two variables
and determines whether they are independent or
correlated [2]. The test for independence, when applied
to a population of subjects, determines whether they are
positively correlated or not. χ2 value for each term t in a
category c can be defined by equation (1) [1].
χ2 t, c = (1)
Where N: is the total number of training documents, A is
the number of documents in c containing t, B is the
number of documents not in c containing t, C is the
number of documents in c not containing t, D is the
number of documents not in c not containing t. χ2 was
used in TC problem [3] and showed promising results
[1].
In text classification or categorization, the
implementation of Chi Square is used to measure the
lack of independence between a term or word or feature
(t) and a category or class (c) and could be compared to
the Chi-Square distribution with one degree of freedom
to judge the extremness [9].
1
Web application resulted from this research could be accessed at
http:// dropbox.psmi.polinema.ac.id/tc
E3-1
B. Naïve Bayes Classifier
Naïve Bayes classifier is a simple probabilistic
classifier based on applying Bayes Theorm (from
Bayesian statistics) with strong (naive) independence
assumptions which assumes all of the features are
mutually independent. It uses Bayessian algorithm for
the total probability formula, the principle is according
to the probability that the text belongs to a category
(prior probability), the next would be assigned to the
category of maximum probability (posterior
probability). In simple terms, a naïve Bayes classifier
assumes that the presence (or absence) of a particular
feature of a class is unrelated to the presence (or
absence) of another feature [4].
Suppose the training sample set is devided into k
categories, denoted as
= { , , , … , } , the
prior probability of each category is denoted as
,
where j = 1,2,3,…,k.
=
For an arbitrary document denoted as
, … , , … , ! , whose feature words are denoted as
, where j = 1,2,3,…m, belongs to a specific category
. To classify the document , is to calculate the
probability of all documents in this case of a given ,
i.e. the posterior probability of category , calculated
by the formula as follows [1] :
"
=
# $% "&' # &'
# $%
"
=
()*
∈
"
(3)
According to Bayesian hypothesis, the feature words w1,
… wj …, wmof di = w1, … wj …, wmareindependent, the
joint probability distribution is equal to the product of
the probability distribution of the various feature words,
i.e [5,6]
"
=
,…,
,…,
!"
(
= Π
,=1
"
therefore formula (3) becomes as follows [5,6]:
()*
∈
"
=
(
Π
,=1
()*
∈
"
(4)
this formula called as classification formula. Where the
value of
is the sample size of category devided
by the total number of training set samples, denoted as .
. There are many ways to calculate
"
, the
simplest way is
"
C. Text Classification Evaluation (Precision and
Recall)
Two evaluation methods frequently used to
evaluate the performace of information retrieval system
areprecision dan recall. These methods are also
commonly used as evaluation methods in text
classification system. There are four terms used to
calculate precision and recall (true positives, true
negatives, false positives, and false negatives). For
classification task, all of that terms compare the results
of the classifier under test with trusted external
judgments. The terms positive and negative refer to the
classifier's prediction (sometimes known as the
observation), and the terms true and false refer to
whether that prediction corresponds to the external
judgment (sometimes known as the expectation). This is
illustrated by the table below:
Table 2. Precission and Recall variable
Predicted
Class
(Observation)
(2)
Bayesian text classification is to maximize the value of
equation (2). Obviously, for all the categories given, the
denominator p(di) is a constant. Therefore, solving the
maximum value of equation (2) is converted into solving
the formula followed [1] :
()*
∈
number of the category , V is the total number of
categories. M is used to avoid the problems caused by to
small 3 4 [7].
=
/%0 /0 1 2
True
False
Actual Class
(Expectation)
True
False
a
b
c
d
By referring Table 2, then precision (P) and Recall (R)
could be define as follows[1,8]:
6 = )/ ) + 9
(5)
: = )/ ) + ;
(6)
Note that a+bis the total number of docs that the
classifier reported as positiveor true. For example, if a
classifier classifies 25 docs to belong to a certain
category, out of which 20 truly belong to the category
(are true positives) and a total of 30 docs belong to the
category, then the precision is 20/25=0.8 and recall is
20/30=0.67.
III. METHODS
Text classification process is used to classify several
text documents into some classes or some categories.
For example email classification that classify emails
into some classification based on its content, maybe
entertainment email, job purposed email, advertisement
email or spam.
In this research we use one of supervised learning
method to classify the text document. Supervised
learning means that it uses the data collection (called as
training or learning document or dataset) which have
been manually classified into some categories
before.This learning result is used to classify the testing
documents (testing datase) then. Table 1 ilustrates text
classification process.
, where 3 4 is the
number of training document with feature attribute 5
among the category , 34 is the training document
E3-2
Table 1. Text classification illustration [1]
Category
C1
...
Cm
Training Documents
d1
C11
...
Cm1
...
...
...
dj
C1j
...
Cmj
used Chi Square Test as feature selection method. This
method was briefly explained at Units section.
Testing Documents
dj+1
C1(j+1)
...
Cm(j+1)
...
...
..
dn
C1n
...
Cmn
Generally, text classification goes through three main
steps: pre-processing, text classification and evaluation.
Pre-processing phase aims to make the text documents
suitable to train the classifier. Then, the text classifier is
constructed and tuned using a text learning approach
against from the training data set. Finally, the text
classifier gets evaluated by some evaluation measures
i.e. recall, precision, etc. The following sections are
devoted to these three phases.
A. Dataset Gathering
Data used in this research are collections of paper
abstract of SENTIA. SENTIA is national seminar held
by Politeknik Negeri Malang on information technology
including its application.
We used 276 training data. These data was collected
from abstract collection of SENTIA 2011 and SENTIA
2010. SENTIA is an annual seminar organized by
Politeknik Negeri Malang. For testig purpose we used
148 testing data that were collected from SENTIA 2009.
B. Pre-processing
Preprocessing is used to make the documents that will
be processed are ready and suitable to be processed by
NBC. There are 3 types of preprocessing method used in
this research.
1. Stopword Removing
Stopword is words that are not relevant to be
processed. It means that stopwords are not relevant or
unnecessary in classifying document to a category. For
example, yang, di, kepada, untuk etc.By removing
stopwords, we could reduce the dimension of words that
will be processed by classifier. It means, we could also
reduce the time spent to do the classification or learning
process.
2. Stemming
Stemming is process to which attempt to reduce a
word to its stem or root(basic) form. For example:
Menghitung
Perhitungan
hitung
hitung
In several natural language processing, stemming is not
relevant or is not necessary to be done. But in this text
classification research, stemming is relevant to be
implemented, because this teks classification process
just aware in basic form of the word, not into the word
form effected by its affix. For example, we just need to
process the basic form of hitung instead the verb
resulted by adding prefix me- to hitung (menghitung).
3. Chi Square Test as Feature Selection Method (χ2)
In order to reduce the word dimension, we applied
feature selection method. For this current research we
C. Learning and Classifying
Because we used Naïve Bayes Classifier (NBC) as
classifier method and NBC is one of machine learning
algorithm, so we need some dataset. First, NBC would
learn from dataset to produce some probabilistic models
of each words in each category. By using the
probabilistic models returned from learning process,
NBC will justify or classify some sentences for the most
proper category.NBC equations was briefly given at
Units section above.
D. Evaluation
To evaluate the classifier system performance, we
used Precision and Recall that were explained also at
Units section above.
IV. RESULT AND DISCUSSION
Dataset that was used for learning and testing process,
we got from collection of SENTIA’s publication
abstract.
These abstract (training data) classified into 11
categories (Electronics and Control System, Informatics
and Computer, Electricity, Telecommunication,
Bioengineering, Economic and Bussiness, Government,
Education, Chemical, Machine Engineering and Civil
Engineering). These training data will be used to train
and make the classifier probabilistic model.
Table 3. Comparison between Classification without Preprocessing
Result and Classification with Preprocessing
Time
(second)
Classification without
Preprocessing
Classification with
Preprocessing
207.8
116.6
Table 3 shows the comparison of time spent to do
both of classification with preprocessing and
classification without preprocessing. Figure 1 gives the
graphical illustration of data shown in Table 3. It shows
that preprocessing implemented in text classification
could reduce time spent to do classification process. By
implementing the preprocessing before classification
process, we could reduce the number of word or feature
to be processed, so that time spent to classify text could
be reduced too.
Fig. 1. Comparison between Time resulted by Classification without
Preprocessing and Classification with Preprocessing
E3-3
To evaluate the system performance, we calculate
precision and recall. In this paper we just show precision
and recall calculation result of Electronics (ELE) and
Informatics (IT) category. We choose both of ELE and
IT because they have the a lot of dataset used in training
process.Table 4 shows the value of precision and recall
resulted in Electronics (ELE) category and Informatics
(IT) category.
make text classification system could perform
optimally. By using more big number of training
documents, probability model resulted by Naïve Bayes
could be better than using training documents in small
number. By using the better probability model, the
system could optimally classify the documents .
Table 4. Comparison of precision and recall value between ELE’s and
IT’s
ELE
Classification
without
Preprocessing
Classification
with
Preprocessing
IT
Precision
Recall
Precision
Recall
0.4
0.08
0.33
0.41
0.4
0.12
0.4
0.51
Fig. 3. Comparison of precision and recall value effected by number of
learning/training documents (ELE and IT)
Figure 2 ilustrates the effect of preprocessing
implementation in precision and recall value. This recall
and precision value taken from precision and recall
calculation in IT category. It shows that implementation
of preprocessing also takes effect in precision and recall
value. Preprocessing could improve the preformace of
text classification system, it shown by precision and
recall value resulted by classification with preprocessing
that is bigger than precision and recall resulted by
classification
without
preprocessing.By
using
preprocessing, the words which are not relevant to be
processed could be reduced.
.
In this research, we used stemming and stopword
removing
as
preprocessing
method.
That
implementation of preprocessing was aimed to reduce
the text dimension to be processed, by removing
irrelevant words.
We also try to combine the preprocessing method
with Chi Square Test. Chi Square is also aimed to reduce
the number of words that would be processed.We
calculatedX2 for each term, and then we selectedsome
highest ranked terms.The only selected terms will be
processed using Naïve Bayes.
V. CONCLUSION
Fig. 2. Comparison of precision and recall value effected by
preprocessing implementation
Number of training documents also take effect in
system performance. For example, we show in Table 5,
number of dataset that were classified in ELE and IT.
Table 5. Comparison of number learning documents in ELEand IT
category.
Number of training
document in a certain
category
ELE
IT
45
85
Figure 3 shows that recall value resulted by IT is
bigger that ELE’s recall value. Even precision value of
IT is same as ELE, we still could see that number of
dataset used in training/learning process is important to
There are some items we could conclude from this
research.
a. There are some kinds of method that could be used
to classify text document. One of them is Naïve
Bayes Classifier (NBC).
b. Preprocessing aims to reduce the dimension of the
text document that would be processed. By using
preprocessing, not all of the words inside the
document will be processed, but only the selected
word that will be processed. By using
preprocessing, the classification accuracy could be
increased. Stemming and removing stopword that
was used in this research colud be used as
preprocessing method. Also Chi Square could be
used to reduce the dimension or feature to be
processed.
c. From the experiment result, it proved that NBC
could be used to automatically classify the text
document. The simplicity of NBC makes this
algorithm simple and fast in processing (training
and classifying).
d. From the experiment result, the value of Precision
and Recall are relatively small. It shows that
performance of this system is not optimal. It was
caused by data training provided or used were not
large enough. The larger dataset provided and used
on training or learning process will take a good
effect in classifier system performace.
E3-4
[4]
ACKNOWLEDGMENT
We would like to thank to DR. M. Sarosa Dipl. Ing.,
M.T. for allowing us to copy the collection of SENTIA’s
publication abstract.
[5]
[6]
REFERENCES
[1]
[2]
[3]
[7]
Thabtah, Fadi. Ali, Mohammad. Zamzeer Mannam and Hadi,
M.W. 2009. Navie Bayessian Based on Chi Square to
Categorize Arabic Data. Communication of IBIMA Vol. 10,
ISSN: 1943-7765.
Snedecor, W., and Cochran, W. 1989. Statistical Methods,
Eighth Edition. Iowa State University Press.
Yang, Y., and Pedersen, J.O. 1997. A comparative study on
feature selection in text categorization. In Proc. of Int'l
Conference on Machine Learning (ICML), , pp. 412-420.
[8]
[9]
E3-5
Kim.S., Han.K., Rim.H., Myaeng.S., 2006. Some Effective
Techniques for Naive Bayes Text Classification, IEEE
Transactions
on
Knowledge
and
Data
Engineering,18(11),1457-1466.
WANG Jun-ying, GUO Jing-feng, HUO Zheng, 2006. Design
and Implementation of Chinese Text Categorization System,
Microelectronics & Computer.23.
Gao Yuan, Liu Da-zhong,2008.A Comparison Study of Chinese
Text Categorization, Science& Technology Information,.2.
Yang Ye, Peng Hong, Lin Jia-yi, Chen Shao-jian, 2004. The
Bayesian Text Categorization Based on Extraction of Effectual
Features, Systems Engineering.9(22).
Manning, D. Cristopher, Prabakhar Raghavan dan Hinrich
Schutze. 2009. An Introduction to Information Retrieval.
CambridgeUniversity Press.
Zheng, Zhaohui. Wu, Xiaoyun. Srihari, Rohini. 2004.Feature
Selection for Text Categorization on Imbalanced
Data.SIGKDD Exploration, Vol. 6, Issue 1.
Analysis and Implementation of Combined
Triple Vigenere Cipher and ElGamal
Cryptography Using Digital Image As a
Cryptographic Key
KomangRinartha1), AgungDarmawansyah2), Rudy Yuwono3)
STMIK Stikom Bali1)
Electrical Engineering Department University of Brawijaya2,3)
1)
rinartha@stikom-bali.ac.id, katak_negara@yahoo.com
2)
agungdarmawansyah@yahoo.com, 3)rudyyuwono@yahoo.com
Abstract—Cryptography commonly use in our life,
especially on internet communication. On internet
communication there are many possibilities of secret
information which is not allowed to the public, therefore it
needs an information security for the information that is
confidential and important to its safe and intact accepted
by the users. One way that can be used in securing the
delivery of information, is encryption of codes that are not
easily to be understood.
The research proposes a cryptographic algorithm which
is a combination of cryptographic Triple Vigenere Cipher
with ElGamal cryptography. In general, data security
using key that contain only a number and letter, but in this
research also developed a key in the form of digital images.
From the results, Triple Vigenere Cipher cryptography
and ElGamal cryptography can be combined in order to
form a new cryptographic algorithm, which is done by
changing the mathematical model in each of these
cryptographic algorithms. In this research, the message is
secured in the form of a picture message with the type of
bitmap and jpeg. The result of decryption of secured
messages using combined cryptographic algorithms have
100% similarity level to the original bitmap picture
messages and the level of similarity varies for jpeg picture
messages.
Keywords: Cryptography,
ElGamal, Digital Image
Triple
Vigenere
II. THEORY OF CRYPTOGRAPHY
There are several definitions of cryptography that
has been presented in the literature. Definitions used in
the old books (before the 1980's), states that
cryptography is the science and art to maintain
confidentiality by encrypting messages into a form that
is difficult to understand. This definition may be
appropriate in the past because cryptography is used for
safety critical communications such as communications
in the military, diplomats, and spies. But this time is
more than just privacy cryptography, but also for the
purpose of data integrity, authentication, and
non-repudiation.
2.1. Vigenere Cipher Cryptography
Vigenere Cipher (1523-1596) was one of the
classic cryptographic techniques named after Blaise de
Vigenère. Vigenere Cipher is a polyalphabetic
cryptography that using keywords to perform the
encryption process. Each letter of the message is
encrypted with a key letter is associated, as well as the
decryption process.
Cipher,
I. INTRODUCTION
C
ryptography is a method used to encode the
information needed while maintaining the
confidentiality of public communications or to prove the
authenticity of a message.
Vigenere Cipher and ElGamal cryptography,
generally using the numbers of each digit only limited
numbers with the numbers 0 to 9, so that when the attack
occurred, each digit has a substantial opportunity to be
known. To anticipate the possibility of tracking the
private and public keys, digital images are used which is
a collection of numbers 0 to 255.
Figure 1.Vigenere table
E4-1
In Vigenere Cipher cryptography, there are several
processes performed on a message. An original message
secured using encryption process. Decryption is
performed to read the messages that have been
encrypted. Keys used in encryption and decryption
process is the same key, this is because the Vigenere
Cipher is a symmetric cryptography.
Encryption is a process that converts plaintext to
ciphertext.Vigenere Cipher encryption process is done
using a mathematical model or by using the Vigenere
table.
C = P + K mod26 ..................................... (1)
Decryption is the reverse with encryption, a
process that converts cipher text into plain text. As well
as encryption, decryption can be performed using a
mathematical model or by using the Vigenere table.
produceoutputplain textmessage.
On several cryptographic systems like RSA,
ElGamal and Diffie-Hellman requires fast powering
theorem and Fermat's Little Theorem.
2.3. Entropy
In an efficient cryptography, a key must be used to
secure data. Perfect secrecy is difficult to achieve. But
the best thing to do is build a computationally secure
cryptography.
In
the
secrecy
that
is
notcompletelyperfect there is a possibility some of the
cipher text shows key information. So that Shannon
introduced a concept called entropy to calculate the
uncertainty of a result.
EntropyH(x)
ofxis
avalue
thatdependson
theprobabilityp …p ofthe possibility ofx.
H p …p
P = C − K mod26 ..................................... (2)
With H p … p is theentropy, p is theprobability
ofappearance ofthe symbolto thei.
Where C is a cipher text, P is the plain text, K is the
key and the mod 26 is the remainder (modulus) with 26.
2.2. ElGamalCryptography
In the ElGamal cryptography there are three
processes used, such as create public key by the
recipient, the process of encryption by the sender, and
the decryption by the receiver.
Create public key which is performedby the
userwhowillreceive themessage, requiringthreeinputs
areprocessed
toproduceanoutputandis
writtenin
mathematical form:
A = g modp ................................................... (3)
With the first input is p, the second input is g and
the third input is a which is processed to produce output
in the form of numbers A.
For the encryption process is done by those who
would do the sending messages with the public key
provided by the parties who would receive the message,
requiring three inputs are processed to produce the two
outputs are written in mathematical form:
III. ANALYSIS
3.1. Triple Vigenere Cipher
Initial
analysis
ofthis
study
wasto
analyzeandimprove thestrength ofVigenereCipher
Cryptography.
VigenereCipherCryptographywillbe
strengthenedtoTripleVigenereCipher.
Picture
messagesonthis
algorithmwillbe
processedoneach
pixelcomponenttoeachpixelcomponentofthe
key
imageon the correspondingcoordinates.
Vigenere Cipher cryptography in securing digital
images useVigenere table as a reference that has been
modified so that all the color intensity from 0 to 255 can
be processed.
0
1
2
3
4
5
.
.
.
.
252
253
254
255
C = g modp ................................................... (4)
C = m. A modp .............................................. (5)
With the first input is a plain text message m, the
second input is a number k which is a random number,
the third input is the numbers (A, p, g) which is a public
key that is processed to produce output in the form
of C and C . Generally, the resulting cipher text is
(C ,C ).
For the decryption process to be performed by the
receiver, it requires three inputs which are then
processed into an output and is written in mathematical
form:
message = C
= −∑ ! p log p .................. (7)
1
2
3
4
5
6
.
.
.
.
253
254
255
0
2
3
4
5
6
7
.
.
.
.
254
255
0
1
3
4
5
6
7
8
.
.
.
.
255
0
1
2
4
5
6
7
8
9
.
.
.
.
0
1
2
3
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
252
253
254
255
0
1
.
.
.
.
248
249
250
251
253
254
255
0
1
2
.
.
.
.
249
250
251
252
254
255
0
1
2
3
.
.
.
.
250
251
252
253
255
0
1
2
3
4
.
.
.
.
251
252
253
254
Figure 2.Modification of the Vigenere Table
The mathematical model of the Vigenere Cipher
encryption process which has been modified will bea
form of:
C = P + K mod256 ............................... (8)
The mathematical model Vigenere Cipher decryption
process which has been modified will be a form of:
P = C − K mod256 ............................... (9)
C modp ........................ (6)
Withthe first inputis C , the second inputisC and
thethirdinputisthe numbers (a, p) which is processedto
E4-2
Where C is a ciphertext, P is the plaintext, K is the key
and mod 256 states remainder (modulus) with 256.
Triple Vigenere Cipher has three processes in the
process of encryption and decryption. Because it has
three processes in each process, the possible modes are:
a. EEE
(encryption(encryption(encryption(message))))
TVig(message,K1,K2,K3) = (P + K1 + K2 +
K3)(mod 256)
b. EED
(decryption(encryption(encryption(message))))
TVig(message,K1,K2,K3) = (P + K1 + K2 K3)(mod 256)
c. EDD
(decryption(decryption(encryption(message))))
TVig(message,K1,K2,K3) = (P + K1 - K2 K3)(mod 256)
d. EDE
(encryption(decryption(encryption (message))))
TVig(message,K1,K2,K3) = (P + K1 - K2 +
K3)(mod 256)
e. DDD
(decryption(decryption(decryption(message))))
TVig(message,K1,K2,K3) = (P - K1 - K2 - K3)(mod
256)
f. DDE
(encryption(decryption(decryption(message))))
TVig(message,K1,K2,K3) = (P - K1 - K2 +
K3)(mod 256)
g. DEE
(encryption(encryption(decryption(message))))
TVig(message,K1,K2,K3) = (P - K1 + K2 +
K3)(mod 256)
h. DED
(decryption(encryption(decryptio(message))))
TVig(message,K1,K2,K3) = (P - K1 + K2 K3)(mod 256)
Vigenere Cipher and ElGamal, this algorithm also
processes each pixel component of the input image. The
selection of a prime number p is the number 257 with the
same process on the ElGamal algorithm. Processes in
the combined algorithm can be described as follows:
Mathematical form of create public key :
A=((g + 1)(TVig(a1,a2,a3,a4)+ 1) (mod 257) – 1) ...... (14)
Mathematical form of encryption :
C1 = ((g + 1)k (mod 257) – 1) ............................ (15)
C2 = ((m + 1)(A + 1)k (mod 257) – 1) .............. (16)
Mathematical form of decryption:
Message= (((C1 + 1)(TVig(a1,a2,a3,a4) + 1))−1 (C2 + 1)
(mod 257) – 1) ..................................................... (17)
IV. DISCUSSION
4.1. ETV Create Public Key
ETV create public key process in the application
program is shown in Figure 3 and Figure 4.
Figure 3. ETV Create public key
3.2. ElGamal Cryptography
In the ElGamal cryptography there are three
processes used, such as create public key by the
recipient, the process of encryption by the sender, and
the decryption by the receiver.The processes inthe
ElGamalcryptographycan be explained asfollows:
Mathematical form of create public key :
A = ((g + 1)(a+1) (mod 257) - 1) ......................... (10)
Mathematical form of encryption :
C1 = ((g + 1)k (mod 257) - 1) ............................. (11)
C2 = ((m + 1)(A + 1)k (mod 257) - 1) ............... (12)
Figure 4.Result of ETV create public key
Mathematical form of decryption :
Message= (((C1 + 1)(a + 1))−1 (C2 + 1)(mod 257) – 1)
..................................................................................................... (13)
4.2. ETV Encryption
ETV encryption process in the application program
is shown in Figure 5 and Figure 6.
3.3. Combine Algorithm (ETV)
Cryptography to be built is a public-key
cryptography is the process includes create public key,
encryption and decryption. The entire process will use
the digital image as input variables. Similar to Triple
E4-3
V. CONCLUSION
Figure 5. ETV encryption
Figure6.Result of ETV encryption
4.3. ETV Decryption
ETV decryption process in the application program
is shown in Figure 7 and Figure 8.
This research proposed a new cryptographic
algorithm design is a combination of Triple Vigenere
CipherCryptographyand ElGamalCryptography. From
the design, implementation and testing of cryptographic
software obtained the following conclusion:
1. Triple Vigenere Cipher cryptography and
ElGamalcryptography can be combined using a
mathematical model of the Triple Vigenere Cipher
and ElGamalthat have beenmodified, which have
similar characteristics with ElGamal cryptography.
Characteristics of the algorithm are a public key
cryptography, message expansion in cipherimage,
time of encryption process is not much different
from the ElGamal encryption, decryption processing
time and create public key processing time longer
than the ElGamalcryptography and Triple Vigenere
Cipher, combined cryptography has several modes
as well as the Triple Vigenere Cipher cryptography
is used in create public key and decryption, as well
the combined algorithm has four private key used in
the decryption process, thus increasing the security
of a digital image of the message can be obtained.
2. In the combined algorithm, the similarity of
messages and decrypt the result would be obtained if
the message in the form of bitmap digital images,
with 100% similarity in pixels and visually. When
using jpeg digital image, the result will be the same
visually, but the pixels are not 100% the same, due to
the compression techniques used in the jpeg image.
3. In the combined cryptography, entropy of the public
key encryption will affect the results of the message.
The greater the entropy of the public key, then the
encrypting result has a high randomness.
REFERENCES
Figure 7. ETV decryption
[1] Adha, R.“Vigènere Cipher RotasiBerlapis”.
http://www.informatika.org, October 2010.
[2] Afif,
S.“Kriptografi”.
http://javanusco.files.
wordpress.com, October 2010.
[3] Anonymous. “PSEUDOCODE STANDARD”.
http://users.csc.calpoly.edu, October 2010.
[4] Anonymous. “Citra Digital”. http://www.ittelkom
.ac.id, October 2010.
[5] Anonymous.
“PengantarKriptografi”.
http://
www.informatika.org, October 2010.
[6] Anonymous.
“LandasanMatematikaUntukKriptografi”.
http://haryanto.staff.gunadarma.ac.id, January 2011.
[7] Defls, H danKnebl, H. “Introduction to
Cryptography - Principles and Applications”.
Springer-Verlag Berlin Heidelberg.2007.
[8] Gonzalez, C dan Woods, E. “Digital Image
Processing Second Edition”. Addison-Wesley
Publishing Company, Inc.1993.
[9] Guiliang,
Weiping,
XiaoqiangdanMengmeng.
“Digital Image Encryption Algorithm Based On
Pixels”. Intelligent Computing and Intelligent
Systems (ICIS), IEEE International Conference on.
ISBN: 978-1-4244-6582-8. Page 769 – 772.
Xiamen, China.2010.
Figure 8.Result of ETV decryption
E4-4
[10] Hassan, M danHamdan, T. “Alternatives To Visual
Cryptography For Colored Images”. Electronics,
Circuits and Systems. ICECS. 12th IEEE
International
Conference
on.
ISBN: 978-9972-61-100-1.
Page
1
–
4.
Gammarth.2005.
[11] Hoffstein, J., Pipher, J dan Silverman. “An
Introduction to Mathematical Cryptography”. 233
Spring
Street,
New
York:
Springer
Science+Business Media, LLC.2008.
[12] Jeyamala, C., GopiGanesh, S dan Raman, G.S. “An
Image Encryption Scheme Based On One Time Pads
—
A
Chaotic
Approach”.
Computing
Communication and Networking Technologies
(ICCCNT), International Conference on, ISBN:
978-1-4244-6591-0. Page 1 – 6. Karur.2010.
[13] Joux, A. “Algorithmic Cryptanalysis”. 6000 Broken
Sound Parkway NW, Suite 300: Taylor and Francis
Group, LLC.2009.
[14] Kreherdan Stinson. “A LATEX Style File for
Displaying Algorithms”. http://kambing.ui.edu,
October 2010.
[15] Massandy,
T.
“AlgoritmaElgamalDalamPengamananPesanRahasi
a”. http://webmail. informatika.org, October 2010.
[16] Mollin, R. “An Introduction to Cryptography
Second Edition”. 6000 Broken Sound Parkway NW,
Suite 300: Taylor & Francis Group, LLC.2007.
[17] Oppliger, R. “Contemporary Cryptography”. 685
Canton Street, Norwood: Artech House, Inc.2005.
[18] Rinartha,
K.
“Pengamanan
Citra
Digital
DenganMenggunakanPengembanganKriptografiKu
nci Public Elgamal”.
Prosiding Seminar
NasionalTeknologiInformasidanAplikasinya
Volume 2, Malang: PoliteknikNegeri Malang.2010.
[19] Rukmono, A. “Triple Vigenère Cipher”. http://www.
informatika.org, October 2010.
[20] Suhartana, G. “Pengamanan Image True Color 24
Bit MenggunakanAlgoritma Vigenere Cipher
DenganPenggunaanKunciBersama”.
http://ejournal.unud .ac.id, October 2010.
[21] WahanaKomputer,
Tim
PenelitiandanPengembangan.
“KonsepJaringanKomputerdanPengembangannya”,
Jakarta: SalembaInfotek.2003
[22] Yogaswara, R. “Implementasi Public Key
Cryptography pada Multimedia Messaging Service
MenggunakanEnkripsiElGamal”.
http://www.ittelkom .ac.id, October 2010.
E4-5
Heart Rate Variability Analysis on Sudden
Cardiac Death Risk RR Interval by Using
Poincaré Plot Method
Ponco Siwindarto1, I.N.G. Wardana1, M. Aris Widodo2, M. Rasjad Indra2
1
Faculty of Engineering, Universitas Brawijaya
2
Medical Faculty, Universitas Brawijaya
e-mail: ponco@ub.ac.id
Abstract— Death is the end of the biological functions
that sustain life. Death can occur gradually or suddenly.
When a person’s heart suddenly stops beating effectively
and breathing ceases, the person is said to have
experienced sudden cardiac death (SCD). SCD is not the
same as actual death. The important difference between
the two dies is that the SCD patients is potentially
reversible. If it is reverse quickly, the brain will not die.
SCD mortality rate in patients with heart disease who
carry the risk of SCD can be reduced by using the
implantable cardioverter defibrillator (ICD). ICD
treatment reduced mortality by about 30%, however, it
still requires a relatively large cost for the ICD
implantation. By the large cost of ICD implantation, it
should be installed only in patients who will experience
SCD. This study extracted the characteristics of people
who would experience SCD by using the Poincaré plot
method, which is one of the methods of heart rate
variability analysis. The study resulted that the Poincaré
Plots of RR intervals from a person who would experience
SCD were characterized by relatively large values of SD1
and SD2, and small value of SD21.
Index Terms — Heart rate variability, Poincaré plot,
sudden cardiac death, SD1, SD2, SD21.
I. INTRODUCTION
D
EATH is the end of the biological functions that
sustain life. Death can occur gradually or can also
suddenly. Sudden death can be caused by many things.
It could be due to sudden blockage of the airway as in
the case of strangled, or could be due to the sudden
cessation of cardiac function.
When a person’s heart suddenly stops beating
effectively and breathing ceases, the person is said to
have experienced sudden cardiac death (SCD). SCD is
defined as an unexpected death due to heart problems,
which occurs within a short interval of time (generally
one hour of symptoms onset) in a person with known or
unknown heart disease[1].
SCD is not the same as actual death. In actual death,
the brain also dies while on the SCD, the brain is still
alive. The important difference between the two dies is
that the SCD patients is potentially recoverable. If it is
recovered quickly, the brain will not die.
In the United States, cases of SCD are about 400,000
deaths per year, mainly in men 20 to 64 years age [2],
and cases of cardiac arrests are about 300,000 per year
[3].
SCD mortality rate in patients with heart disease
who carry the risk of SCD can be reduced by preventive
treatment, by using the implantable cardioverter
defibrillator - ICD [4]. The preventive treatment with
ICD reduced mortality by about 30% [5].
ICD therapy was initially given to patients who
survived
from cardiac arrest or who failed in
farmokologis therapy [6]. But in its development,
several studies have shown that ICD is also effective for
patients who had never suffered cardiac arrest or
sustained ventricular tachicardia (VT) [7] - [9].
The ICD therapies have successfully reduced the
mortality rate, however, the ICD implantation still
requires a relatively large cost. The cost of ICD
implantation; including devices, leads, and the cost of
the hospital; is about $30,000 to $40,000 [10]. For the
large cost of ICD implantation, it should be installed
only in patients who will experience SCD. The problem
is, how to know whether a person will experience SCD
or not. This study extracted the characteristics of people
who experience SCD by using the Poincare plot method,
which is one of the methods of heart rate variability
analysis.
II. HEART RATE VARIABILITY (HRV)
Heart rate is the number of heart beat for one minute
periode of count, and the unit is beat per minute (bpm).
Heart beats are caused by electrical depolarization of the
heart muscle. The depolarization of the upper cardiac
chambers, called atria is visualized by the P-wave. The
Q, R and S waves, which create the QRS complex,
represent the depolarization of cardiac lower chambers
known as ventricles. The interval between successive
heart beats is called RR interval (RRI) and it is the
distance between the consecutive QRS complexes,
usually measured as the distance between the RR waves
as shown in Fig. 1.
Variation of instantaneous heart rate or RR interval is
a consequence of constant interaction between the
intrinsic activity of the sinus node and the influence of
E5-1
x = ( x1 , x 2 ,...x N )
and two auxiliary vectors:
x + = (x1 , x 2 ,...x N −1 )
x − = (x 2 , x3 ,...x N )
Fig. 1. RR interval (RRI) duration derived from an ECG
(1)
so that each point:
the autonomic nervous system, various substances
circulating in the blood and present in the heart tissues
[11]. The control of heart rate is modulated by both
sympathetic and parasympathetic branches of
autonomic nervous system as well as many other
autonomic reflexes.
Breathing is one of the important factor modulating
heart rate [11]. It causes heart rate acceleration during
inspiration and its deceleration during expiration.
Another example of a separate system regulating the
heart rate are the changes in blood pressure modulated
by baroreflex.
All these systems and reflexes are responsible for
changing of the duration of RR interval from one beat to
another and this phenomenon is called heart rate
variability (HRV) [12]. HRV is a strong and
independent predictor of mortality following an acute
myocardial infarction [13]. The higher HRV the better
prognosis in survivors of myocardial infarction or
patients with heart failure.
A number of parameters are used in HRV analysis.
The Poincaré plot of RR intervals is one of the recent
methods. The analysis of Poincaré plot is an emerging
method of nonlinear dynamics applied in HRV analysis.
(x
+
i
)
, xi− ,
i = 1....N − 1
in the plot corresponds to two successive heart beats.
Poincaré plot provides sumary information as well as
detailed beat-to-beat information on the behavior of the
heart [16] . Poincaré plot shown in Fig. 2 can be divided
into three regions. All points described by consecutive
cardiac beats of equal duration (RRn = RRn+1) are
located on the identity line. The points above the identity
line correspond to all prolongations (RRn < RRn+1), and
the points below this line represent all shortenings of the
interval between two consecutive beats (RRn >RRn+1).
There are several descriptors in the Poincaré plot. The
following 3 descriptors of the Poincaré plot were used in
the study [14], as shown in Fig. 3.
SD1 is the standard deviation calculated from
Fig 3. Descriptors of Poincaré plot, SD1 and SD2 as parameters that
measure short- and long- term HRV
resulting distribution if all points of the Poincaré plot are
projected on a line perpendicular to the line of identity
(known as the “width”). SD1 measure the dispersion of
points in the plot across the identity line. This parameter
is usually interpreted as a measure of short-term HRV
presented by [15]:
SD1 = Var ( x1 )
Fig. 2. Poincaré plot of RR interval
(2)
where Var(x) is the variance of x,
and:
III. POINCARÉ PLOT
Poincaré plot is a graphical representation of temporal
correlations within the RR intervals shown in Fig. 1,
where each RR interval is plot as a function of the
preceding RR interval [14]. The Poincaré plot is shown
in Fig 2, where the duration of the current cardiac beat
(RRIn) is represented on the x axis, and the duration of
the following beat (RRn+1) on the y axis.
Suppose an RR interval data vector of length N [15]:
x1 =
x+ − x−
(3)
2
SD2 is the standard deviation calculated from
resulting distribution if all points of the Poincaré plot are
projected on the identity line (known as the “length”).
This standard deviation measure the dispersion of points
along the identity line. This parameter is usually
E5-2
interpreted as a measure of long-term HRV presented by
[15]:
SD2 = Var( x2 )
(4)
where Var(x) is the variance of x, and
x1 =
x+ − x−
(5)
2
SD21 is the ratio of SD2 to SD1.
SD21 = SD2 / SD1
Fig. 4. RR interval time series (a) RR Interval Time Series for record
30 and (b) the 9000 s to 9300 s selected segment
(6)
by using Eq. 2, 4, and 6, and the result were SD1 = 51.2
ms, SD2 = 58.9 ms, and SD21 = 1.15 ms.
IV. METHODS
This study used time series of RR intervals (RRI)
extracted from the data from two Physionet ECG
database [17]. The first was Sudden Cardiac Death
Holter Database. We used 17 ECG recordings (each of
about 24h in duration), which is ended with ventricular
fibrillation. The second was The MIT-BIH Normal
Sinus Rhythm Database. We used 17 long-term ECG
recordings of subjects referred to the Arrhythmia
Laboratory at Boston's Beth Israel Hospital (now the
Beth Israel Deaconess Medical Center). Subjects
included in this database were found to have had no
significant arrhythmias.
The first step of the study was determining the
positions of each R waves in records of all the ECG
samples. Second, computed the RR intervals (RRI),
and then ploted them as RR Interval Time Series. The
Poincaré plots were made in all 5-minute segments
randomly selected from the RR Interval Time Series.
For the sample of sudden Cardiac Death Database, the
segments were selected randomly from the beginning of
the RRI Time Series to shortly before the onset of
ventricular fibrillation. Poincaré plot descriptors SD1,
SD2, and SD21 were computed based on Eq. 2,4, and 6.
Outliers were removed before the average of SD21 were
computed. The HRV characteristic related to sudden
cardiac death was obtained by comparing the average
value of SD21 between The Sudden Cardiac Death
Holter Database and The MIT-BIH Normal Sinus
Rhythm Database.
V. RESULT
The RR Interval Time Series.
Fig. 4 shows an example of RR Interval Time Series
for Sudden Cardiac Death Holter Database, taken from
record 30 (a). This RR Interval Time Series then was
randomly selected for 10 segments, each 5-minutes in
duration. Fig 4(b) shows one of the segments, which is
9000 s to 9300 s of time interval.
The Poincaré Plot
Each selected segment of RR interval time series was
used for generate the Poincaré plot. Fig. 5 is the
Poincaré plot of the segment selected in Fig 4. The
descriptor values for this Poincaré plot were computed
Fig. 6. Poincare plot for segment 9000 s to 9300 s of RR Interval Time
Series of record 30 Sudden Cardiac Death Holter Database
The same procedure then was applied to all records of
the database. The result is shown in Table I. There are
three descriptors that will be observed, that is SD1,
SD2, and SD21.
It can be seen from the table that the average value of
SD1 and SD2 for Sudden Cardiac Death Holter
Database are larger than for MIT-BIH Normal Sinus
Rhythm Database. For Sudden Cardiac Death Holter
Database, the average values are SD1=163.3ms and
SD2=155.2ms instead of SD1=28.92ms
and
SD2=87.96ms for MIT-BIH Normal Sinus Rhythm
Database. In other words, RR interval of Sudden
Cardiac Death Holter Database has more variability on
both long-term and short-term variability than of
MIT-BIH Normal Sinus Rhythm Database.
SD21, the ratio of SD2 to SD1 has a smaller value on
the Sudden Cardiac Death Holter Database than on the
MIT-BIH Normal Sinus Rhythm Database. The average
values are SD21=0.989 for Sudden Cardiac Death
Holter Database and SD21=3.474 for MIT-BIH Normal
Sinus Rhythm Database. This descriptor shows a
relative variability between the long-term and the
short-term variability.
E5-3
[2]
Table I. Comparison of Poincaré plot between Sudden Cardiac
Death Holter Database and MIT-BIH Normal Sinus Rhythm
Database
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
Rec: record number
VII. CONCLUSION
[15]
Poincaré Plots of RR intervals from a person who
would have sudden cardiac death are characterized by
the relatively large values of SD1 and SD2, and small
value of SD21.
[16]
[17]
REFERENCES
[1]
A. A. Sovari. (2011, Dec 5). Sudden cardiac death. Medscape
Reference: Drugs, Diseases, and Procedures. Available:
http://emedicine.medscape.com/article/151907-overview#show
all.
[18]
E5-4
Concise Dictionary of Modern Medicine. McGraw-Hill
Companies, 2002.
Lloyd-Jones D, Adams RJ, and Brown TM. Heart disease and
stroke statistics---2010 update. Circulation 2010;121:e46--215.
Al-Khatib, SM, Sanders GD, Bigger JT, Buxton AE, Califf RM,
and Carlson M. Expert panel participating in a Duke's Center for
the Prevention of Sudden Cardiac Death conference. Preventing
tomorrow's sudden cardiac death today: part I: Current data on
risk stratification for sudden cardiac death. Am Heart
Journa,2007; 153: 941-950.
Kuck KH, Cappato R, Siebels J, and Ruppel R. Randomized
comparison of antiarrhythmic drug therapy with implantable
defibrillators in patients resuscitated from cardiac arrest: the
Cardiac Arrest Study Hamburg (CASH). Circulation 2000; 102:
748–754.
Mirowski M, Reid PR, and Mower MM. Termination of
malignant ventricular arrhythmias with an implanted automatic
defibrillator in human beings. N Engl Journal Med 1980;
303:322-324.
Zipes DP, Camm AJ, and Borggrefe M. ACC/AHA/ESC 2006
guidelines for management of patients with ventricular
arrhythmias and the prevention of sudden cardiac death: a report
of the American College of Cardiology/American Heart
Association Task Force and the European Society of Cardiology
Committee for Practice Guidelines (Writing Committee to
Develop Guidelines for Management of Patients With
Ventricular Arrhythmias and the Prevention of Sudden Cardiac
Death). Journal Am Coll Cardiol 2006;48:e247-e346.
Zwanziger J, Hall WJ, Dick AW. The cost effectiveness of
implantable cardioverter-defibrillators: results from the
Multicenter Automatic Defibrillator Implantation Trial
(MADIT)-II Journal Am Coll Cardiol 2006;47:2310-2318.
Mark DB, Nelson CL, dan Anstrom KJ. Cost-effectiveness of
defibrillator therapy or amiodarone in chronic stable heart
failure: results from the Sudden Cardiac Death in Heart Failure
Trial (SCD-HeFT) Circulation 2006;114:135-142.
Kadish A & Mandeep Mehra. Heart Failure Devices.
Implantable Cardioverter-Defibrillators and Biventricular
Pacing Therapy. Circulation. 2005; 111: 3327-3335.
R. Hainsworth, Physiology of the cardiac autonomic system in:
Clinical guide to cardiac autonomic tests edited by M. Malik,
Kluwer Academic Publishers, London, 1998, pp. 3-28.
Heart rate variability. Standards o measurement, physiological
interpretation, and clinical use. Task Force of the Working
Groups on Arrhythmias and Computers in Cardiology of the
ESC and the North American Society of Pacing and
Electrophysiology (NASPE), European Heart Journal 93, 1996,
pp. 1043-1065.
Bigger JT, Fleiss JL, Steinman RC, Rolnitzky LM, Kleiger RE,
Rottman JN. Frequency domain measures of heart Standards of
heart rate variability. European Heart Journal, Vol. 17, March
1996.
P Guzik, Jaroslaw P, Tomasz K, Raphael S, Karel HW, Andrzej
WT, and Henryk W. Correlations between the Poincaré Plot and
Conventional Heart Rate Variability Parameters Assessed during
Paced Breathing. Journal Physiol. Sci. Vol. 57, No. 1, Feb.
2007. pp. 63–71.
J. Piskorski, P. Guzik. Filtering Poincaré plots. Computational
Methods in Science and Technology 11(1), 2005. pp. 39-48.
Toichi M, Sugiura T, Murai T, Sengoku A. A new method of
assessing cardiac autonomic function and its comparison with
spectral analysis and coefficient of variation of R-R interval.
Journal Auton Nerv Syst. 1997;62:79-84.
Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov
PCh, Mark RG, Mietus JE, Moody GB, Peng C-K, Stanley HE.
PhysioBank, PhysioToolkit, and PhysioNet: Components of a
New Research Resource for Complex Physiologic Signals.
Circulation 101(23):e215-e220 [Circulation Electronic Pages;
http://circ.ahajournals.org/cgi/content/full/101/23/e215]; 2000
(June
13).
Lung Cancer Prediction In Imaging Test Based
On Gray Level Co-Occurrence Matrix
Sungging Haryo W.1). Agus M. Hatta2), Syamsul Arifin3)
Department of Engineering Physics, Faculty of Industrial Technology
ITS Surabaya Indonesia 60111,
email: mr.pascal@yahoo.com1), agus.hatta@yahoo.com2), syamp3ai@ep.its.ac.id3)
Abstract— Lung cancer is the most causing death
among all types of cancer for both men and women, based
on World Health Organization (WHO) with 19.7%
precentage from all cancer. This report presented results
of a research of lung cancer prediction in imaging test
based on image processing and extracted using Haralick
Features in Gray Level Cooccurence Matrix. 9 Extracted
Haralick Features used as input data in 2 Generalized Bell
membership functions of ANFIS. The results are validated
by comparing training and testing results with the analysis
of radiologist. Moreover the results indicate that this
method can predict the present of malignant cell of 92%
accuracy in imaging test.
Index Terms— ANFIS, Chest X-Ray, Gray Level
Cooccurence Matrix, Haralick
I. INTRODUCTION
Lung cancer is the most causing death for both men
and women based on World Health Organization
(WHO) data with 19.7% precentage from all cancer [3].
Every year, more than 1.2 million lung cancer case have
been diagnosed. People who inhale cigarrete smoke
from other smokers (also called as secondhand smokers
or passive smokers) also increase lung cancer risk,
although they are not smokers.
The lung cancer can be cured easily in initial stage
but may be impossible in the advanced stage. In the
other hand early detection of lung cancer patient is
difficult because it’s prognosis would appear when it
comes to advanced stadium. Many of early lung cancers
were diagnosed incidentally, after the radiologist found
symstomps as a result of test performed for an unrelated
medical condition [1]. Imaging test is needed in lung
cancer diagnosis process to discover the present of
malignant cell in the lung after a patient is suspected
from his/her medical history. Many researches in
medical image processing field has been proposed for
this purpose. A key function in different image
applications is feature extraction. The feature is a
characteristic that can capture a certain visual property
of an image either globally
For the whole image, or locally for objects or
regions. Different features such as color, shape, and
texture can be extricated from an image. Texture is the
variation of data at different scales. A number of
methods have been developed for texture feature
extraction. They can be extracted from co-occurrence
matrices and wavelet transform coefficients. Then, they
are stored as feature vectors. In this paper, chest X-Ray
image is processed and extracted by using Gray Level
Co-occurrence Matrix, then the extracted features stored
as inputs in Adaptive Neuro Fuzzy Inference Systems
(ANFIS) for lung cancer prediction in imaging test.
II. LITERATURE REVIEWS
A. Lung Cancer
Lung cancer is a disease characterized by
uncontrolled cell growth in tissues of the lung. It is also
the most preventable cancer. Cure rate and prognosis
depend on the early detection and diagnosis of the
disease. Lung cancer symptoms usually do not appear
until the disease has progressed. Thus, early detection is
not easy. Many early lung cancers were diagnosed
incidentally, after doctor found symtomps as a results of
test performed for an unrelated medical condition [2].
There are two major types of lung cancer:
non-small cell and small cell. Non-small cell lung
cancer (NCLC) comes from epithelial cells and is the
most common type. Small cell lung cancer begins in the
nerve cells or hormone-producing cells of the lung. The
term “small cell” refers to the size and shape of the
cancer cells as seen under a microscope. It is important
for doctors to distinguish NSCLC from small cell lung
cancer because the two types of cancer are usually
treated in different ways. Lung cancer begins when cells
in the lung change and grow uncontrollably to form a
mass called a tumor (or a lesion or nodule). A tumor can
be benign (noncancerous) or malignant (cancerous). A
cancerous tumor is a collection of a large number of
cancer cells that have the ability to spread to other parts
of the body. A lung tumor can begin anywhere in the
lung [3].
E6-1
Figure 1 X-Ray image of (a) normal lungs and (b) lung cancer
May 30-31,
Once a cancerous lung tumor grows, it may or may
not shed cancer cells. These cells can be carried away in
blood or float away in the natural fluid, called lymph,
that surrounds lung tissue. Lymph flows through tubes
called lymphatic vessels that drain into collecting
stations called lymph nodes, the tiny, bean-shaped
bean
organs that help fight infection. Lymph nodes are
located in the lungs, the center of the chest, and
elsewhere in the body. The natural flow of lymph out of
the lungs is toward the center of the chest, which
explains why lung cancer often spreads there. When a
cancer cell leaves its site of origin and moves into a
lymph node or to a faraway part of the body through the
bloodstream,
am, it is called metastasis [4].
The stage of lung cancer is determined by the
location and size of the initial lung tumor and whether it
has spread to lymph nodes or more distant sites. The
typee of lung cancer (NSCLC versus small cell) and stage
of the disease determine what type of treatment is
needed.
B. Gray Level Co-occurrence
occurrence Matrix
Texture analysis is an essential issue in computer
vision and image processing, such as in remote sensing,
content based image retrieval.
retrieval The Gray-Level
Co-occurrence
occurrence Matrix (GLCM) is one of the statistic
method that can be used as texture analysis based on the
extraction of a gray-scale
scale image. Coocurance Matrix is
extracted by considering the relationship between
betw
two
neighborhood pixels. Based on the hypothesis that in a
texture configuration reccurence is occurs, the first pixel
is known as a reference and the second is known as a
neighbor pixel [5].
features which are extracted from texture analysis.
These features contain the information about the image
such as homogeneity,
ty, contrast, the complexity of the
image, and etc. They are used in many applications such
as biological applications and image retrieval.
Figure 2 Direction of calculation in Haralick texture features
Based on figure 2.3, Haralick texture features can be
extracted from eight directions. Four equations are used
:
∈
,
, 0, )|
,
,
, ∠ , ,
!"
,
,
,
, , ,
$%
$%
∑ &' ∑ &'
, , ,
-+
,
, ,
+.
-+
,
, ,
(
&'
'
$%
&'
'
--+
&' &0
$% $%
+,4.
--+
&' &0
(1)
,
, , ,
, , 2, 3,, … . , 289 ",
3 ,
2, 3,, … . , 289 ",
|
|,
(3)
(4)
14 derived features from Haralick
aralick are written as follows:
Where p(i;j) is the element (i;j)th
)th of the normalized
coocurence matrix.
,
+,
$% $%
+,/.
GLCM can be measured as follows :
1) Create the GLCM symmetrical
2) Calculate the probability of each combination, the
probability is calculated :
, , ,
#
, , , (
∈
$%
1) Angular Second Moment (ASM)
ASM also known as uniformity or energy, measures
the image homogeneity. ASM is high when pixels are
very similar.
5'
(2)
If the co-occurrence
occurrence matrix is symmetric then p(i,j)
p(
=
(p(i,j) + p(i,j)T ) = 2 that T indicates the transpose matrix
and θ will be 0, 45, 90 and 135.
$% $%
--+
&' &'
,
,
:
(5)
2) Contrast
Direction of contrast calculation is to measure the
intensity or gray-level
level variations of reference pixel and
it’s neighbor.
3) Texture features calculations
In order to estimate the coocurence in texture
analysis using GLCM, Haralick proposed 14 statistical
E6-2
5:
-
$% 4'
;&0
:
<-
$%
&'
|
-
$%
&'
+
, =|
,
9) Entropy
By extract the entropy features of a grayscale
image, value of image which is needed for image
compressions.
(6)
3) Correlation
In co-occurene matrix, gray lavel values linear
dependency of a grayscale image is calculated in order
to obtain the correlation features. Which is represent the
relation between a pixel with pixel’s neighbor in eight
directions.
$%
$%
∑ &'
∑ &'
+ , ,
?, ?.
5>
(7)
@ @
, .
Where:
µ x = means of px
µ y = means of py
σx = standart deviations of px
σy = standart deviations of py
5A
--
? :+
&' &'
5B
-&' &'
13
1
,
,
:
(8)
+
,
,
- +,/.
&:
5F
5F : +,/.
- +,/.
&:
(14)
$% 4'
GH+,/.
- +,4.
&0
GH+,4.
13) Information Measures of Correlation 2
S'>
X
T 1 NR+U 2.0(OPQ@ OPQ W
Where
OPQ
$% $%
--+
,
&' &'
$% $%
--+
,
&' &'
OPQ2
(15)
,
GH +
(16)
,
,
(17)
(18)
,
GH +,
+.
(19)
(20)
$% $%
- - +,
+.
&' &'
GH +,
+.
(11)
14) Maximal Correlation Coefficients
5'A XT YNMG KLHNZ[N HN JK \NG5]
8) Sum of Entropy
:$%
(13)
12) Information Measures of Correlation 1
OPQ OPQ1
5':
KR OP, OQ
(10)
&:
-
,
,
10) Difference Variance
5'0 JKL K MNG5+,4.
7) Sum of Variance
5E
GH +
If high entropy obtained, it is mean that a grayscale
image has high contrast from one pixel with it is
neighbour and cannot be compressed as a low contrast
as a result low entropy
OPQ1
:$%
:$%
,
,
&' &'
HX and HY are entropies of px and py
(9)
6) Sum of Average
Sum of average calculate the average or mean of a
grayscale image.
5D
--+
5''
5) Inverse Difference Moment / Homogeneity
Inverse Difference Moment also sometimes called
homogeneity, measures the local homogeneity of a
grayscale image. Homogeneity returns the measures of
the closeness of the distribution of the GLCM elements
to the GLCM diagonal.
$% $%
5I
11) Difference Entropy
4) Sum of Squares (Variance)
Sum of Squares is statistic equation for extract the
variance of image gray tone in a grayscale image.
$% $%
$% $%
(21)
(12)
Where
In equation of sum of entropy, in some case the
probability is equal to zero. Causing the log(0) cannot be
defined, therefore to solve this problem, it is
recommended to use log(p+ε) ε is an fluctuative small
positive constant.
E6-3
]
,
^
+
,
, + , ,
+, +. ^
(22)
III. METHODOLOGY
A. Image Resizing
Resize the initial image is needed in image
preprocessing.Chest X-ray image is enlarged or
shrinked, depend on it’s amount of pixel by resizing the
resolution in one dimension, row or column. Then
another dimension (column or row). All images input
are resized in to 220 x 210 pixels, neglected their aspect
ratio. This algorithm can process all images input,
regardless their early resolution..
B. Image Grayscaling
Image grayscaling is needed to yield 8-bit images
from RGB images input. 8-bit image (or grayscale
image) stored as matrix data which it’s value represent
gray level or intensities of image in 0-255 range. Each
pixel in grayscale image is represented by value of the
matrix. Lower value of matrix represent dark colour, and
bright colour for higher one. Image convertion
Algorithm in MATLAB to convert chest X-ray image
from RGB to grayscale is written as :
I = rgb2gray(RGB)
D. Image Region of Interest Selection
It is It is neccessary to segmented lung region of
chest x-ray images for inspect the present of cancer
nodule in lung. Region of Interest is used to select subset
of chest image samples to identified the malignant cell.
In the current research, the equalized image is specified
it’s region of interest (ROI). Using the syntax :
I1=roipoly;
I2=roipoly;
Notice that I1 and I2 are used to select two ROI,
which is refer to the number of lung in human
respiratory systems. When the syntax is runned, it create
an interactive polygon tool, related with the image of
chest X-ray scan that displayed by using Imshow(I)
algorithm, equalized chest image is the targeted image.
This function also remove the hue and saturation
information of resized image. In matrix equation,
rgb2gray algorithm is written in equation
0,2989xb + 0,5870xe + 0,1140xg (30)
Figure 3 selected area of ROI in chest X-Ray images
Where I is matrix of grayscale image, R, G, and B are
matrixs of Red, Green, and Blue colour in RGB image.
C. Histogram Equalization
There are some methods to enhance an image, such
as image sharppening, deblurring, noise removing, and
histogram equalization. The last mentioned method is
used in current research. After convert the chest X-ray in
to grayscale image, histogram equalization is needed to
adjust the image intensities and enhance image contrast.
Repose the previous subsection. If grayscaled chest
image is written as I in MATLAB workspace and
represented as m x n matrix in distributed intensities
with magnitude from 0 to L-1, where L equals to 256. If
H is equalized chest image histogram of I, then :
KHNh [ℎ [N Z [j (31)
O [G[K \ kNLG5+ RN
As shown in figure 3, if cursor is moved to the
targeted image, it will be changed in to crosshair
symbol. Lung area is selected by selecting polygon’s
twist. It also can be moved or resized by moving the
crosshair symbol. Targeted area is rounded by blue line.
First region of lung declared as I1 in MATLAB
workspace, while I2 is read as second lung region. The
next objective is to separate lung region by eliminate
other organs and tissues appeared in chest X-ray. First
step is obtain the In from following equation :
;
(32)
Where c0 is the cumulative histogram, c1 is the
cumulative sum of histogram for all intensities k. The
equation is based on constrain that T should be
monotonic and c1(Ta)) always less than c0(a) by more
than half the distance between initial histogram. In
MATLAB algorithm, image histogram equalization of
chest image is written as : H = histeq(I);
'
3
:
R2
(33)
If value of single pixel in In matrix is less than 1, it
converted to 255, while for image with intensity 1< In<5
is converted to 0.
5
Image histogram in MATLAB choose the grayscale
transformation T to minimize :
│c1(T(k)0-c0(k) │
255,
< 1
l0,
1 < n < 5o
, G[ℎNLh ZN
(34)
Let R0 is the targeted image or equalized chest image,
after In calculated, next objective is to eliminate all
object except In. If Rn is the segmented image, then :
(35)
b; b0 ;
E6-4
5 b
l
0,
b ,
b < 0
o
G[ℎNLh ZN
Result of segmented image from targeted image is
shown in figure 4
The extracted features of chest X-Ray image are :
Energy, Correlation, Contrast, Entropy, Inverse
Difference Moment, Sum Average, Sum Variance, Sum
Entropy, and Difference Average 70 chest X-Ray image
are processed. Table 1 show the example of extraction
results from 10 samples
Table 1. Haralick feature extractions of chest X-Ray image (1)
Figure 4 Segmented chest X-ray image
E.
Image Matrix Reduction and Binarization
Grayscale image or 8-bit image can be
considered as one dimension matrix with n columns and
m rows with varies intensities in range 0-255. In order to
reduce the residual visual information after the it’s
segmented, the chest X-ray image is reduce their matrix
to remove the bone and other soft tissues and get the
visual representation of malignant cell in the lung. It is
also assumed that malignant cell has higher intensities
(which is represented with higher number in image
matrix).
Num
Energy
Correlation
Contrast
1
0.92877
0.71049
258.87782
2
0.90857
0.83260
211.87290
3
0.90045
0.70289
369.72523
4
0.94552
0.74382
179.60087
5
0.94495
0.69726
206.96194
6
0.95238
0.73182
162.76329
7
0.83631
0.87499
293.95611
8
0.92629
0.78637
209.76820
9
0.94926
0.76420
156.44920
10
0.92289
0.79031
216.08230
Table 2. Haralick feature extractions of chest X-Ray image (2)
F.
Image Feature Extraction
Feature extraction of segmented image is
neccesary to capturing visual content of chest images fo
indexing & retrieval. In the current research, Gray Level
Coocurrence Matrix (GLCM) is used as feature
extraction method. GLCM often used in image
recognition and compression by considering the
relationship between two neighborhood pixels. Based
on the hypothesis that in a texture configuration
reccurence is occurs, the first pixel is known as a
reference and the second is known as a neighbor pixel.
Then statistical features of the image is extracted from
Haralick equations which are extracted from texture
analysis
IV. RESULTS AND DISCUSSIONS
When processed image of chest X-Ray is obtained,
the next step is extract the image in order to get the
numerical information of the image and will be used as
data input in ANFIS. There are 9 extracted features for
each image which are derived from Haralick Features of
Gray Level Cooccurence Matrix. In Haralick Feature
Extraction Process, some paramaters should be
determined, i.e :
1.
2.
3.
4.
Image : The input image is gray scale image
Input bits : Gray level resolution of the image input
in this case, 256 gray level image is used, hence the
input bits is equal to 8.
The distance between pixel in feature calculations.
In the current research the distance of pixel is set to
1
The angle of feature calculations. It can be 0o, 45o,
90o, or 135o. Where 0o is selected, therefore the
gray level calculations will count in horizontal
directions (right and left side of reference pixel)
E6-5
0.27850
Inv. Diff.
Moment
0.98395
Sum
Average
7.24767
2
0.32912
0.98686
10.39091
3
0.36748
0.97708
10.20861
4
0.22134
0.98887
5.64567
5
0.22542
0.98717
5.50204
6
0.19839
0.98991
4.87229
7
0.51377
0.98178
20.10787
8
0.28090
0.98700
7.98238
9
0.20756
0.99030
5.33632
10
0.29111
0.98660
8.39117
Num
Entropy
1
May 30-31,
where
2 :
xu = Output Target
xwu 1= Output Training
n
= Number of training data
0
Table 3. Haralick feature extractions of chest X-Ray
X
image (3)
Num
Sum
Variance
Sum
Entropy
Diff.
Av.
1
1529.50234
0.26245
2.03841
2
2319.44702
0.31598
1.66829
3
2119.04656
0.34456
2.91122
4
1222.52621
0.21021
1.41418
5
1160.28484
0.21258
1.62962
6
1051.05972
0.18830
1.28160
7
4409.11707
0.49555
2.31462
8
1754.03879
0.26790
1.65172
9
1170.49981
0.19786
1.23188
10
1844.86317
0.27771
1.70144
1 3 5 7 9 11 13 15 17 19 21 23
Figure 6 Testing Results
1 is represent that there are malignant cell present
in the lung from processed Chest X-Ray
X
image, while 0
represent the normal lung. If the output results are less
than 0.5 then it classified as normal lung, in the other
hand, if the results are > 0.5 it classified as lung cancer
image. From the testing graphic plot in figure 6, the
validation results can be seen in table 4
anfisedit
1.30
Target
Output
0.30
-0.20 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43
Figure 5 Training Results
In order to obtain the training performance numerically,
Root Mean Square Error (RMSE) value of each training
is calculated by using the following equation
RMSE
∑vu&' xu
n
= 0.135026
t
X
xwu
:
Output
-1
Extracted features from Chest X-ray
X
image are set
as training and testing data which contain of the desired
input and output data pairs. Membership function
parameters can be estimated in training phase of ANFIS.
This project uses the ANFIS Editor GUI menu bar to
load a FIS
S training initialization, and then save the
trained FIS. To open the ANFIS GUI, following syntax
is typed in MATLAB command window :
0.80
Target
Table 4. Results of Validation
Classified
Misc.
Accuracy
23
2
92%
V. CONCLUSIONS AND FUTURE WORKS
A new method of lung cancer detection system has
been developed by using Haralick Features of Gray
Level Cooccurence Matrix as feature extraction method
to get the ANFIS input data from initial Chest X-Ray
X
image for cancer detection purpose in imaging test.
Before extracted, the image are processed through
several stages, i.e. (1) Resizing, (2) Grayscaling, (3)
Histogram equalization for contrast enhancement, (4)
Lung segmentation based on their region of interest, and
(5) Image matrix reduction by specific reduction
r
factor.
There are nine features (Energy, Correlation,
Entropy, Inverse Diff. Moment, Sum Average, Sum
Variance, Sum Entropy, Inf. measure of correlation 1,
and Inf. measure of correlation 2) appropriate as data
input in ANFIS as information of malignant
ma
cell present
in the lung Based on testing results,
results ANFIS parameters
with 9 inputs features can predict lung cancer in imaging
test with 92% accuracy with 2 missclasified from 25
images
REFERENCES
[1]
25 data are used in testing phase for validation. to load
the testing data, testing button is selected and data
da is
uploaded from MATLAB workspace. Testing phase or
model validation is required to test the performance of
model in data fitting process.
[2]
[3]
[4]
[5]
E6-6
Le Kim. Automated Detection of Early Lung Cancer and
Tuberculosis Based on X Ray Image Analysis.
Analysis International
Conference on signal, speech, and Image Processing WSEAS.
2006. 110
American Society of Clinical Oncology. Guide to Lung Cancer.
Alexandria. Conquer Cancer Foundation. 2011: 2.
Floche. Backgroundd information Non-small
Non
Lung Cancer.[pdf]
(URL:http://www.roche.co.id/fmfiles/re7229001/Indonesian/m
edia/background.library/oncology/lc/Lung.Cancer.Background
er.pdf accesed on July 30, 2011)
Anonymous. Kanker Paru Pedoman Diagnosis dan
Penatalaksanaan di Indonesia.
ndonesia. Perhimpunan Dokter Paru
Indonesia. 2003
Haralick, Robert M. Textural Features for Image Classification.
Classification
IEEE Transactions Onn Systems, Man, And Cybernetics. 1973,
3(6) : 612-619
Optimal EDR Methods for
Sleep Apnea Classification
Mungki Astiningrum1, Sani M. Isa2, Aniati Murni Arimuthy3
Faculty of Computer Science, University of Indonesia
Depok 16424/Jawa Barat, Indonesia
mungki.astiningrum@ui.ac.id, sani.muhammad@ui.ac.id, aniati@cs.ui.ac.id
Abstract – The purpose of this study is to compare
performance of the EDR methods for sleep apnea
classification. EDR methods used are two algorithm, R
Wave Detection and RR Interval Correction, and using
Naive-Bayes Classification. Data obtained from the
polysomnography recordings related to the 7 different
patients are used for evaluations. Data of each subject taken
at a specific time frame for 15 minutes. timescales are chosen
based on annotation data with apnea detection. Results of
both algorithm shows have generated signal similarity model
(wave form),and the classification results for the RRintervals correction method is slightly better than the Rwave detection method, for R Wave Detection, correctly
classified 0.5333 and kappa statistic 0.0395, for R-R Interval
correction , correctly classfied 0.5333 and kappa statistic
0.0676
Index terms – Apnea Classification, ECG, EDR, R-R
Interval.
I.
INTRODUCTION
The most important one of the sleep repiration
disorders is sleep apnea. Sleep apnea is a common
disorder in which there is a pause in breathing or shallow
breaths during sleep. There are three types of sleep apnea,
central,obtructive and mixed apnea. Obstructive sleep
apnea (OSA) has the highest prevalence. Together with
the absence of respiratory effort in the lungs, the absence
of air flow inside the mouth and nose is defined as central
sleep apnea. Despite the respiratory effort, the lack of air
flow in the nose and mouth is obstructive sleep apnea.
The situation starting with central sleep apnea and
continuing as obstructive sleep apnea is defined as mixed
sleep apnea. Mixed apnea subjects can be treated by the
methods applied to the subjects with obstructive sleep
apnea. Obstructive sleep apnea is the most common sleep
apnea syndrome. Obstructive sleep apnea is the state of
absence of oral and nasal air flow despite the respiratory
effort. Although the diaphragm and intercostal muscle
activity continued, exchange of air through the nose and
mouth stands [1]. In this case, it Is thought to be an
obstruction at the URT of patient. In order to prevent the
blockage, an intense activity in the chest and abdomen is
observed. Central sleep apnea (CSA) is the state in the
absence of both respiratory effort and air flow together.
Central apneas grow by the corruption of the central
regulation of respiration. Mixed sleep apnea is the state
starting with central sleep apnea and continuing the
absence of oral and nasal air flow when the respiratory
effort begins. How the respiratory effort after the central
sleep apnea starts is still a unresolved research topic [2].
The device used for measuring and recording
physiological signals during sleep is called as
polysomnograph and the signals retrieved from the device
are called as polysomnography (PSG). By the use of PSG,
it is possible to observe the physiological changes in
humans during sleep. Various physiological signals of
the subjects are recorded simultaneously by the PSG
device, which has an embedded multi-channel data
acquisition system. The recording process made as analog
recordings in the 90s has left its place to digital recorders
after the development of digital systems. Thus, the
prevention of errors caused by the hardware chaos of
analog systems is provided [3]. By the use of these
devices, Electroencephalogram (EEG), Electrocardiogram
(ECG), Electromyogram (EMG), Electrooculogram
(EOG), breathing, Pulseplethysmograph (PPG) and
various desired or necessary signals of subjects in sleep
are recorded. In this way, the subjects’ statuses are
determined during the night sleep and their diagnosis and
treatment outcomes can be delineated. The classification
of sleep apnea is also realized by the investigation of
these physiological signals obtained from the PSG device.
The following physiological activities in your body
are recorded during polysomnography [11].
• Brain
activity
is
measured
with
EEG
(electroencephalogram). Disruptions in sleep stages
may suggest narcolepsy or REM sleep behavior
disorder.
E7-1
• Eye movements are recorded with a EOG (electrooculargram). This helps to determine the presence of
sleep stages particularly REM sleep stage.
• Muscle activity such as face twitches, teeth grinding
and leg movements, is measured by EMG
(electromyogram). This helps to determine if the REM
sleep is present during sleep. Frequent leg movements
may indicate a periodic limb disorder such as restless
leg syndrome which is a sleep disorder because these
movements often disrupt your sleep.
• Airflow in and out of the lungs while you are asleep is
measured with a nasal airflow sensor.
• Snoring activity is measured with the help of snoring
microphone. Very loud snoring is indicative of sleep
apnea
• The percentage of oxygen in your blood is measured
(oxymetry) by a bandage like oxymeter probe or
sleeve that fits painlessly on one of your fingers. Low
oxygen levels may indicate a sleep apnea.
normally visible in 50-75% in the ECG. Fig. 1 shows a
typical example of an ECG signal. The duration of a
heartbeat is the time interval from one R wave to next R
wave, also known as RR-Interval [5]
In 2004, Chazal et al, suggested an obstructive sleep
apnea detection using ECG signal. Features used in these
studies are the statistical measurement of variables
derived from RR-intervals and ECG-derived respiratory
signal (EDR) [6].
Fig 2. Schematic representation of normal ECG
II. DATA AND METHODOLOGY
Fig 1. Polysomnography Process
The electrocardiogram (ECG) is a simple and low-cost
non-invasive recording that can be used to get respiratory
information. In consequence, different techniques have
been proposed to derive the respiratory signal from the
electrocardiogram [4]. ECG is a graph produced by an
electrocardiograph, which records the heart's electrical
activity within a certain time. Besides being used for the
diagnosis of heart disease, ECG also useful for diagnosing
pulmonary embolism, hypothermia, and sleep disorders.
ECG graph of the healthy subject cycle consists of a P
wave, a QRS complex and a T wave. A small U wave is
Data obtained from signals and derived from
simultaneously recorded ECG signals of the
polysomnography recordings related to the 16 different
subjects have been used for evaluations. They were
obtained from Polysomnographic Database in the
PhysioNet databank which is a web-based library of
physiologic data and analytic software sponsored by the
US National Institutes of Health [7]. The methods are gets
respiratory rate by measuring the number of ECG samples
in R-R interval and its advantage lies in its simplicity. The
other detects the rate by measuring the size of R wave in
QRS signal. This algorithm can detect the rate more
robustly but it is complicated and requires the ECG signal
base line to be stabilized.
A. Data
The data related to the 16 different subjects have been
obtained from MIT-BIH Polysomnographic Database.
The database is a collection of recordings of multiple
physiologic signals during sleep. The database contains
over 80 hours worth of four, six, and seven-channel
polysomnographic recordings, each with an ECG signal
annotated beat-by-beat, and EEG and respiration signals
annotated with respect to sleep stages and apnea for every
30s epoch [8] . In this database,all 16 subjects were
E7-2
May 30-31,
male,aged 32 to 56 (mean age 43), with weight ranging
from 89 to 152 kg (mean weight 119 kg). There were two
subjects that divided into 2 record, subject with code
slp01 and slp02, so total records are 18.
Data of each subject taken at a specific time frame for 15
minutes. timescales are chosen based on annotation data
with apnea detection. Spesific time frame of each subject
are shown in table3.
Table 1. Data for Types of Apnea
From Physionet Database.
Table 3. Specific Time Frame
Time
Subjects
Code
Apnea Types
No
Subject
Code
h
m
s
HA
OA
CA
Mix
1
slp01(a+b)
5
0
0
v
v
v
-
2
slp02(a+b)
5
15
0
v
v
-
-
3
slp03
6
0
0
v
v
v
v
4
slp04
6
0
0
v
v
-
-
5
slp14
6
0
0
v
v
-
-
6
slp16
6
0
0
v
v
v
v
7
slp32
5
20
0
v
v
-
-
8
slp37
5
50
0
-
v
-
-
9
slp41
6
30
0
-
-
-
-
10
slp45
6
20
0
v
v
v
-
11
slp48
6
20
0
v
v
-
-
12
slp59
4
0
0
v
v
v
-
13
slp60
5
50
0
v
v
v
-
14
slp61
6
10
0
v
v
v
-
15
slp66
3
40
0
v
v
-
-
16
slp67x
1
17
0
v
v
v
-
The data obtained consists of the types of sleep apnea
with patient information in accordance with the code and
the length of recording time are shown in Table 1.
ECG
Ann. Sleep Stage and Apnea
BP
EEG
RN
RA
EOG
EMG
Fig 3.. View of Polysomnography Recording (ECG, BP, EEG, RN, RA,
EOG, EMG Signals and Reference Sleep Stage and Apnea Annotations)
Seven of them have been used for evaluations,
subjects with code slp03, slp16, slp32, slp45, slp59,
slp60, slp67x. Chosen because they have a recording with
a diversity of types of apnea and respiratory available.
∆T
Start
Stop
slp03
[00:15:00.000]
[00:29:59.996]
slp16
[01:44:00.000]
[01:58:59.996]
slp32
[01:53:30.000]
[02:08:29.996]
slp45
[03:56:00.000]
[04:10:59.996]
slp59
[01:35:00.000]
[01:43:56.400]
slp60
[00:11:30.000]
[00:26:29.996]
slp67x
[01:57:00.000]
[02:11:59.996]
B. Methods
R Wave Detection [10]
Intrathoracic impedance increases with inspiration
and, as a result, the size of ECG on the vertical axis is
reduced. On the contrary, it decreases with expiration and,
as a result, the size of ECG on the vertical axis is
enlarged. A method of acquiring respiration
re
signal from
an ECG based on the physiological theory above.
1. QRS wave of ECG signal
2. Acquisition of ECG signal
3. Measuring the size of R wave
4. Respiration signall detection before filtering
5. First differentiation
ation of signal obtained in 4
6. Respirationn signal obtained through the band-pass
band
filtering of signal from
rom the first differentiation
R-R Interval correction [10]
The experiment is uses the respiratory pulse, which is
physiological interaction between the respiratory system
and the circulatory system. Because of change in heart
rate synchronized with respiration, the R-R
R
interval of
ECG is short during inspiration,
inspirat
and long during
expiration. The following is a method of acquiring
respiration signal from an ECG based on the
physiological theory above.
1. QRS wave of ECG signal
2. Acquisition of ECG signal
3. Measuring R-R interval
4. Respiration signal
al detection before filtering
5. First differentiation
ation of signal obtained in 4
6. Respiration signal obtained through the band-pass
band
fitering of signal from thee first differentiation
E7-3
Naive-Bayes Classification
A Naive-Bayes classifier is a probabilistic classifier based
on Bayes theorem with strong (naive) independence
assumptions. Naive Bayes classifier assumes that the
presence (or absence) of a particular feature of a class is
unrelated to the presence (or absence) of any other
feature. This classification method relies on transforming
the discrete decision labels output by the individual
matchers into continuous probability values.
EDR calculation using the above two methods
performed on 7 subjects who had been selected
previously. EDR signals from the two methods for all
subject shown in following figures.
1.2
1.0
0.8
0.6
RRin
Rwave
0.4
0.2
N
OA
N
OA
OA
N
N
N
N
N
N
N
N
MA
N
N
N
N
N
N
N
N
CA
N
N
N
N
N
N
N
0.0
Fig 4. Respiratory rate detection using the size of R wave
Fig 6. EDR Signals from RRinterval and Rwave with subject’s code
slp03
2.0
1.5
Rwave
RRin
1.0
0.5
N
N
MA
OA
OA
OA
OA
OA
OA
N
OA
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
OA
N
OA
OA
0.0
slp16
2.00
1.50
1.00
0.50
Rwave
RRin
0.00
N NOAOAN N NOAOAOAN N N N N NOAOAN NOANOANOANOANOAN
slp32
Fig 5. Respiratory rate detection using R-R interval
E7-4
calculation, and the x-axis shows the type of sleep apnea
that occurs.
Process with the naive-Bayes classification, performed
on each of the EDR calculation algorithm, R Wave
Detection and R-R Interval corection are, for R Wave
Detection, correctly classified 0.5333 and kappa statistic
0.0395, for R-R Interval correction , correctly classfied
0.5333 and kappa statistic 0.0676
1.50
1.00
Rwave
0.50
RRin
N
N
OA
N
N
OA
N
CA
N
OA
OA
N
OA
N
N
OA
N
OA
N
OA
N
OA
OA
N
N
N
OA
OA
N
N
0.00
Fig 9 EDR Signals from RRinterval and Rwave with subject’s code
slp45
IV. CONCLUSION
Based on the results of the experiment, the respiratory
signal can be obtained from the ECG signal processing.
method used is to use the RR-interval correction and Rwave detection. Results of both shows have generated
signal similarity model (wave form).
The signal generated from the two methods are used
for sleep apnea classification using Naive-Bayes
Classification. The classification results for the RRintervals correction method is slightly better than the Rwave detection method, with a kappa value of 0.0676.
In general, the classification of sleep apnea from the
ECG signal can still be obtained even better. For further
research can be used classifier method except NaiveBayes Classification, then to the process of acquiring the
respiratory signal from ECG signal processing can be
performed by methods other than the two algorithms are
used in this paper.
1.20
1.00
0.80
RRin
0.60
Rwave
0.40
0.20
N
N
CA
N
CA
CA
CA
N
N
CA
CA
CA
CA
CA
CA
OA
OA
N
OA
OA
N
OA
OA
N
OA
OA
N
CA
N
N
0.00
slp59
1.00
0.80
0.60
RRin
0.40
Rwave
REFERENCES
[1]
0.20
0.00
N N N N N CACACACAOACAOAOACAOACAOAOANOAN N N N NOAOAN N N
slp60
1.20
1.00
0.80
0.60
0.40
RRin
0.20
Rwave
0.00
N N CAOAN N CACA N CACACA N CACACA N CACA N N CA N N N CACA N CA N
slp67x
The figures above show the results of the acquisition
respiratory signal from RR-interval and Rwave
Aydin, H.; Ozgen, F.; Yetkin, S. & Sütcügil, L., “Sleep and
Respiratory Disorders in Sleep”, 2005
[2] Onur Kocak1, Tuncay Bayrak1, Aykut Erdamar1, Levent
Ozparlak2, Ziya Telatar3 and Osman Erogul, “Automated
Detection and Classificationof Sleep Apnea Types Using
Electrocardiogram (ECG) and Electroencephalogram (EEG)
Features”, 2010
[3] Erogul, O,”Engineering Approaches in Sleep Studies”,
Proceedings of 9th Sleep Medicine Congress, , 2008
[4] Lorena S. Correa, Eric Laciar, Vicente Mut, Abel Torres, and
Raimon Jané, “Sleep Apnea Detection based on Spectral Analysis
of Three ECG - Derived Respiratory Signals”, 2009
[5] Sani M. Isa, Mohamad Ivan Fanany, Wisnu Jatmiko,Aniati Murni
Arimuthy,” Optimal Features Selection and Cross Validation of
Classifiers for Apnea Detection”, ICACSIS 2010
[6] Chazal P, Penzel T, and Heneghan C, “Automated detection of
obstructive sleep apnoea at different time scales using the
electrocardiogram”. Physiological Measurement. 2004 July; 25:
967-983.
[7] www.physionet.org, visited at the date 20.02.2012.
[8] Y Ichimaru,GB Moody, “Development of the polysomnogaphic
database on CD-ROM, Psychiatric and Clinical Neurosciences
1999”, 53:175-177
[9] Boyle, J., Bidargaddi, N., Sarela, A. and Karunanithi, M.,
Automatic Detection of Respiration Rate From Ambulatory
Single-Lead ECG, 2009
[10] J.M. Kim, J.H. Hong, N.J. Kim, E.J. Cha, T.S. Lee, “Two
Algorithms for Detecting Respiratory Rate from ECG Signal”,
2007
[11] http://mediconweb.com/health-wellness/polysomnography-asleep-study/, access date of 16 March 2012
E7-5
Automated Detection of Congested Central Vein
Liver Histology of Mice Infected with
Plasmodium berghei Using CellProfiler 2.0
1
Tur Rahardjo1,Dwi Ramadhani1 and Siti Nurhayati1
Center for Technology of Radiation Safety and Metrology,
National Nuclear Energy Agency of Indonesia
dhani02@batan.go.id
Abstract— Malaria is initiated by Plasmodium sporozoites
infections, which are inoculated by mosquitoes.
Histopathologic lesions often described in the liver of
rodent with malaria are congested central vein with
neutrophils and eusinophils within the lumen. Detection of
congested central vein has a possibility to do automatically
using image analysis software. Here the used of
CellProfiler an open access cell image analysis software for
automated detection congested central vein liver histology
of mice infected with Plasmodium berghei is reported. The
results are compared to the manual detection. Wilcoxon
rank test was used for statistical analysis with H0
hypothesis that means there was no significant difference
between manual analysis and those with CellProfiler.
Totally 10 images were analysed for both manually and
using CellProfiler. Results showed that there were no
significant difference between manual and automatic
counting (p>0,05). Overall it appears that in our research
analyzes with CellProfiler are comparable but not better
than manual.
Keywords — CellProfiler, Central Vein, Congested,
Pipelines, Plasmodium berghei
I. INTRODUCTION
Malaria is the most serious and widespread
parasitic disease of humans. It affects at least 200 to 300
million people every year and causes an estimated 3
million deaths per year. Malaria is initiated by
Plasmodium sporozoites, which are inoculated by
mosquitoes. The disease is characterized by a range of
clinical features from asymptomatic infection to a fatal
disease [1]. There are four species of Plasmodium that
infect man and result in four kinds of malaria fever: P.
falciparum, P. vivax, P. ovale, and P. malariae [2].
Malarial involvement of liver is now a known entity
with it is specific histopathological lesions.
Histopathologic lesions often described in the liver of
rodent with malaria is congested central vein with
neutrophils and eusinophils within the lumen (Fig. 1)
[2,3]. Detection of congested central vein commonly
done manually under microscope. This process has a
possibility to do automatically using image analysis
software. With the availability of digital photography,
the congested central vein detection process can be done
on the image by marking the central vein first using an
image analysis performed with image analysis software
then detection the congested area inside central vein. The
results can be documented by saving the overlay image
with the marked target cells.
Fig 1. Congested central vein [3]
CellProfiler is freely available modular image analysis
software that is capable of handling hundreds of
thousands of images. The software contains
already-developed methods for many cell types and
assays and is also an open-source, flexible platform for
the sharing, testing, and development of new methods by
image analysis experts. CellProfiler uses the concept of a
'pipeline' of individual modules. Each module processes
the images in some manner, and the modules are placed
in sequential order to create a pipeline: usually image
processing,
then
object
identification,
then
measurement. Most modules are automatic, but
CellProfiler also allows interactive modules (for
example, the user clicks to outline a region of interest in
each image). Modules are mixed and matched for a
specific project and each module's settings are adjusted
appropriately. Upon starting the analysis, each image (or
group of images if multiple wavelengths are available)
travels through the pipeline and is processed by each
module in order [5].
Here the used of CellProfiler an open access cell
image analysis software for automated detection
congested central vein liver histology of mice infected
with Plasmodium berghei is reported. The results are
E8-1
compared to the manual detection of congested central
vein in liver histology. The advantages using Cellprofiler
is it could be instructed to process images in batches of
several hundred to automatically generate parasitemia
values without the need for supervision. This also
eliminates factors such as user fatigue and lack of
standardization that are often associated with manual
enumeration.
II. MATERIALS AND METHODS
2.1. Mice
2.6. Statistical analysis
Totals congested central vein obtained by manually or
with CellProfiler were compared using Wilcoxon test
with H0 hypothesis mean there are no significant
difference variation between was no significant
difference between manually analysis and with
CellProfiler. H1 hypothesis mean there was a significant
difference between manually analysis and with
CellProfiler. Significant level used in this research is
0.05 (5%).
Male Swiss mice ages 8 to 12 weeks were purchased
from Pusat Penelitian dan Pengembangan Gizi dan
Makanan, Kementerian Kesehatan Indonesia.
2.2. Parasites and infections
Mice were inoculated intraperitoneally with 106
erythrocytes infected by P. berghei. Mice were subjected
to euthanasia at one week after inoculation. Fragments of
the liver were fixed by immersion in 10% buffered
formalin during 24 hours. These samples were then
dehydrated, and processed for paraffin embedding. Five
µm sections were cut and stained with hematoxylin-eosin
(H&E).
2.3. Image acquisition
A Nikon Biophot microscope attached with Nikon
D3000 digital single lens reflects (DSLR) camera system
was used to capture images of the smears. The slides
were examined under 10× objective lens. Images were
captured at a resolution of 1936×1296 and saved as
JPEG files.
Fig 2. Pipelines for detected congested central vein.
III. RESULTS
3.1. Automated and Manual Detection
Ten images were collected and subjected to the
automated, as well as being analyzed manually by
pathologies. Scatter plots graph show linear relation
(r = 0.55; Fig 3) between analyzes using CellProfiler and
with manual counting. In our pipelines the time needed
for process one single image is approximately 17
seconds.
4.5
2.4. Manual detection congested central vein
y = 0.4419x + 1.1512
R2 = 0.5551
4
3.5
Manual Results
Ten images of liver histology section were analyzed
under personal computer using Microsoft Windows XP
SP 2 32-bit platform as operating system. Processor type
used inside the computer is AMD Athlon(tm) 64 X2
Dual Core 5000+ with memory (RAM) is 1.87 GB. .
3
2.5
2
1.5
1
0.5
2.5. Automated detection congested central vein
0
0
An open access cell image analysis software
CellProfiler 2.0 r10997 that developed by Broad
Institute was used for an automated detection congested
central vein. CellProfiler (CP) runs on Microsoft
Windows XP SP 2 32-bit platform. Processor type used
inside the computer is AMD Athlon(tm) 64 X2 Dual
Core 5000+ with memory (RAM) is 1.87 GB. A
pipelines was developed to doing automatic detection
congested central vein (Fig. 2).
1
2
3
4
5
6
7
CellProfiler Results
Fig 3. Scatter plots comparing congested central vein
defined by manually and with CellProfiler.
3.2. Statistical analysis compare between automated
and manual results
Statistical analysis using Wilcoxon Rank test show
that there are no significant different between manual
counting and automated counting (P = 1), because the
p-value is bigger than 0.05 it is mean H0 hypothesis is not
rejected.
E8-2
8
IV. DISCUSSION
Our pipelines consist several steps to detected the
congested central vein in liver histology. First we rescale
the image to speed up the time required for detected
congested central vein in the image, and then we convert
the images to grayscale images. Grayscale images show
bright color on objects and a dark color in background.
Because the central vein area is cover by a bright color
then it can be easy to identify using the thresholding
methods in Identify Primary Objects module.
Unfortunately because sometimes the material inside
central vein is attached to the border line of the central
vein, then the central vein area that detected by
thresholding method became very tight.
In order to refine the central vein area results we apply
several images morph processing to get better central
vein detection. Based on our experiment dilate the
images, fill holes after dilate the images and last erode
the images can be used for getting a better detection
central vein area. Interestingly CellProfiler provided all
image morph processing in Morph module. To define
which one is the congested central liver we must detect is
there any materials inside the central vein. Because
thresholding method is detected the bright area and the
materials are in gray color, first we must invert the
images so the material inside the central vein became
bright and can be detected by thresholding methods. In
our pipelines, the congested central vein detection is
based on “parent-child relationship” between the central
vein and the materials inside the central vein. The central
vein (parent) should define as the congested central vein
if it has at least one or more material (child) in it. It will
not define as congested central vein if the material inside
the central vein is not overlapping with the central vein.
Details of all steps of our pipelines are shown in Fig 4.
A significant variation between manually and
automatically detection of congested central vein
observed in one image. This happened because there are
large sinusoids areas and because the inside the sinusoid
areas there are many Kupffer cells then the pipelines also
detected as a congested central vein. A large sinusoid
area is also commonly histological lesions because
infection of Plasmodium, this phenomenon usually calls
a sinusoid dilatation. Baheti et al research show that
seventy five percent histological lesion in liver cause by
Plasmodium is sinusoid dilatation [7]. We are now in the
process of developing a better pipelines than can be used
for detected the congested central vein and can
differentiate between the congested central vein and
sinusoid dilatation.
Convert color image
to grayscale color
Detection Central Vein
Using Threshold methods
Morphology image
consist dilate, fill
holes and erode
image
Invert images so material inside
the central vein can be detected
using threshold methods
Detected material inside the
central vein using threshold
methods
Detection Central Vein that have
a material in the lumen
Fig 4. Flowchart automatic detection of congested
central vein defined by CellProfiler.
REFERENCES
[1]
[2]
[3]
[4]
[5]
V. CONCLUSION
We have developed pipelines for CellProfiler
software that can be used to detect the congested central
vein in liver histology section of mice infected with
Plasmodium berghei. Overall, our pipelines worked very
well to detection of congested central vein.
[6]
[7]
E8-3
SIO, S.W., SUN, W., KUMAR, S., BIN, W.Z., TAN, S.S., ONG,
S.H., KIKUCHI, H., OSHIMA, Y., and TAN, K.S.
MalariaCount: an image analysis-based program for the accurate
determination of parasitemia. J Microbiol Methods 2007,
68:11-18.
HISAEDA, H., YASUTOMO, K., and HIMENO, K. Malaria:
immune evasion by parasite. Int. J. Biochem. Cell Biol. 37,
700-706.
KHAN, Z.M., NG, C., and VANDERBERG, J.P. Early Hepatic
Stages of Plasmodium berghei: Release of Circumsporozoite
Protein and Host Cellular Inflammatory Response. Infection and
Immunity, 1992, 60(1), p. 264-270.
SILVA, A.P.C., RODRIGUES, S.C.O., MERLO, F.A.,
PAIXÃO, T.A., AND SANTOS, R.L. Acute and chronic
histopathologic changes in wild type or TLR-2-/-, TLR-4-/-,
TLR-6-/-, TLR-9-/-, CD14-/-, and MyD-88-/- mice experimentally
infected with Plasmodium chabaudi. Braz J Vet Pathol, 2011,
4(1), 5-12.
Carpenter, A.E., Jones, T.R., Lamprecht, M.R., Clarke, C.,
Kang, I.H., Friman, O., Guertin,D.A., Chang, J.H., Lindquist,
R.A., Moffat, J., Golland, P., and Sabatini, D.M. CellProfiler:
image analysis software for identifying and quantifying cell
phenotypes. Genome Biology, 2006, 7:R100.
IVANOVSKA, T., SCHENK, A., HOMEYER, A., DENG, M.,
DAHMEN, U., DIRSCH, O., HAHN, H.K., AND LINSEN, L. A
fast and robust hepatocyte quantification algorithm including
vein processing. BMC Bioinformatics 2010, 11:124
BAHETI, R., LADDHA, P., and GEHLOT, R.S. Liver
Involvement in Falciparum Malaria – A Histo-pathological
Analysis. JIACM CM, 2003; 4(1): 34-8
Telemonitoring Application in Health Safety
and Environment at PT. Pertamina Refinery
Unit IV Cilacap using Android Smartphone
1,2,3)
Budi Santosa1, Bambang Yuwono2, Mariza Feary3
Informatics Engineering Department, Universitas Pembangunan Nasional Babarsari
Tambakbayan , Yogyakarta, 55281, Indonesia Tel.:+62 274 485323
e-mail : dissan@upnyk.ac.id
Abstract— Health, Safety and Environment (HSE) is a
part of PT. Pertamina Refinery Unit IV Cilacap,
Indonesia. The main task of HSE is monitoring area and
situation of PT. Pertamina Refinery Unit IV Cilacap,
especially fire-prone areas. The situation and condition
must be monitored 24 hours / 7 days. This research
proposed the solution how to monitor the dangerous area
far away from the location using mobile device.
The methodology that used to develop this application is
Guidelines
for
Rapid
Application
Engineering
(GRAPPLE). Several tools that used to develop this
application are Eclipse IDE Helios, Text Editor and
ZoneMinder which is used as a server for collecting data.
The result of this research is a Telemonitoring Application
in Health, Safety and Environment at PT. Pertamina
Refinery Unit IV Cilacap using Android Smartphone. The
system has capability running in 2 ways. Firstly, from
Android Smartphone (through smartphone web browser
or telemonitoring application) and secondly, from web
browser which use zoneminder as a server. The
application is used to help the administrator monitoring
the condition and situation of area refinery unit from long
distance and also can be access from everywhere using
mobile device.
Index Terms— Android, Eclipse IDE Helios,
Telemonitoring, ZoneMinder.
Telemonitoring needs a certain media for transfer data
from data source to data processing center. At least the
media contains 3 criteria, data accuracy, long-distance
range and economically.
B. Android
Android is an operating system for mobile devices,
produced by Google. Google purchased the initial
developer of the software, Android Inc., in 2005. The
unveiling of the Android distribution in 2007 was
announced with the founding of the Open Handset
Alliance, a consortium of 84 hardware, software, and
telecommunication companies devoted to advancing
open standards for mobile devices. Google releases the
Android code as open-source, under the Apache
License. In this application, used Gingerbread Android
version (Android 2.3).
C. IPCamera
IPCamera is a type of digital video cameras used for
surveillance.
D. ZoneMinder
Application security and surveillance video cameras
on linux operating system, intended for single camera
and multi camera.
II. REQUIREMENT ANALYSYS AND DESIGN
I. INTRODUCTION
H
ealth Safety and Environment (HSE) is a part
division of PT. Pertamina Refinery Unit IV
Cilacap, Indonesia. HSE handled the problems in the
field of environmental health and worker safety at PT.
Pertamina RU IV Cilacap. But, the main focus of the
tasks is to monitor the area and situation of PT.
Pertamina Refinery Unit IV Cilacap, especially fireprone areas. Therefore many workers monitoring
directly into the field to monitor the situation and
conditions. The situation and conditions must be
monitored 24 hours / 7 days.
Telemonitoring is the activity to remote monitoring
of the situation by using communication equipments. In
this research, application can be used from android
smartphone and pc web browser. The application
already supported by 64 base to maintain its security.
A. Telemonitoring
Telemonitoring is the activity to remote monitoring
of the situation by using communication equipments.
The new developed system aimed to help
admininistrator monitored the condition and situation of
area refinery unit from the long distance. The
developed system have requirements below :
The software shall to monitored condition of
fire-prones area PT. Pertamina RU IV Cilacap.
The software shall be able to do video
streaming.
The software shall monitored anytime and
everywhere.
The software shall be able to do automatic
storage.
The software shall be able to download videos.
The design phase consist of architecture design and
interface design. The architectural design of the
software describes how the application work. the design
phase consist of architecture design and interface
design. The architectural design of the software
describes how the application work.
E9-1
Chromium Browser
Webmin (DNS Server)
ZoneMinder Linux CCTV Server
Java Android script programming
Fig. 1. Application Architecture using Android Smartphone
Figure 1 describes how the application running
through smartphone android. Administrator can
monitored the situation from smartphone android which
connect to WLAN PT. Pertamina RU IV Cilacap and
directly
access
hse.telemonitoring.com
/
192.168.100.10.
In other way, smartphone android connect via
WLAN PT. Pertamina RU IV Cilacap through the
server for automatic storage, download video data and
B. HARDWARE
Hardware that used to develop the system consist of :
Smartphone Android
o 2.3.3 Android Version (GingerBread
OS)
o 1 GHz Processor
o 2.6.32.9 Kernel Version
IP Camera
o VGA Camera Lens
o MJEPG Video Encoding Standard
o 270̊ Remote Pan, 120̊ Tilt Control
o Wifi IEEE 802.11 b/g
Wireless ADSL2 + Gateway
o Modem Linksys WAG54G2
o Wifi IEEE 802.11 b/g
Notebook
o Intel Core 2 Duo
o T5450 Processor 1.66 GHz, 667 MHz
FSB, 2MB L2
o Cache
o RAM 1,5 GB DDR3
C. Files that Used in the System
There are 3 packages contained in this application.
Where in the package there are 8 java class (*.java).
Table 1 indicates file that used in this system
No
1
File Name
Package akseshttp
Basis64.java
Fig. 2. Application Architecture using Personal Computer (PC)
GambarBerubahListene
r.java
monitored multiple camera at the same time.
Figure 2 describes how the application running
through Personal Computer (PC). PC connect via
WLAN PT. Pertamina RU IV Cilacap then access
telemonitoring.com through the server for automatic
storage, control the camera and monitored multiple
camera at the same time.
PemisahStream.java
StreamKamera.java
2
III. IMPLEMENTATION
This section explained the implementation of the
Telemonitoring Application in Health Safety and
Environment at PT. Pertamina Refinery Unit IV
Cilacap using Android Smartphone.
Package
hse.telemonitoring
IpCameraActivity.java
LoginActivity.java
3
Package Video
IImgData.java
ImgData.java
A. SOFTWARE
Software that used to develop the system consist of :
Ubuntu 11.04 Natty Narwhal
Apache Webserver
Eclipse IDE Helios + Android SDK and ADT
Plugin
E9-2
Remark
Package contains classes that
manage basis64 and streaming
File that used to password
encryption for security
Contains
change
image
detection method and error
notification.
File that used to separating the
image from http header
Implementation
from
GambarBerubahListener.java
interface and used to connect
to the camera and take pictures
repeatedly
Package contains application
menu
File that used to control the
camera
File that used to controlled
menu
Package to take byte data
Contains set bytes and get
bytes method.
Implementation
from
IImgData.java interface and
used to wrap the image data
Application Display from Android Smartphone
IV. CONCLUSION
The conclusion of this research is that has been built
the Telemonitoring Application in Health Safety and
Environment at PT. Pertamina Refinery Unit IV
Cilacap using Android Smartphone. The system has
been succesfully tested and have capability running in
2 ways. Firstly, from Android Smartphone (through
smartphone
web
browser
or
telemonitoring
application). Secondly, from web browser which uses
ZoneMinder as a server.
Fig 3. User Interface Login
REFERENCES
[1]
Fig 4. User Interface Camera
Application Display from Android Smartphone Web
Browser (telemonitoring.com/zm/skin?=mobile)
Fig 5. User Interface Application via Android Smartphone Web
Browser / Menu IPCAM MOBILE VIEW
Application Display
(telemonitoring.com)
from
PC
Web
Browser
Fig 6. User Interface Application via PC Web Browser
Implementation telemonitoring.com aimed to help
administrator to controlled and stored data. There are
three menus, IPCAM VIEW(hse.telemonitoring.com),
IPCAM
MONITOR
ZONEMINDER
(telemonitoring.com/zm/skin?=classic),
IPCAM
MOBILE VIEW (telemonitoring. com/zm/skin?=
mobile). IPCAM VIEW used to controlled the camera.
IPCAM MONITOR ZONEMINDER used to automatic
storage. IPCAM MOBILE VIEW used to store data in
smartphone.
Application Display from ZoneMinder CCTV Server
(telemonitoring.com/zm/skin?=classic)
_________, 2011, Apache HTTP Server.
http://en.wikipedia.org/wiki/Apache_HTTP_Server, (accessed
at Nov 20, 2011).
[2]
_________, 2011, FFmpeg. http://ffmpeg.org/index.html,
(accessed at Nov 20, 2011).
[3] Charibaldi, Novrido. (2010). Solusi Pemrograman Java
(Dilengkapi Contoh Soal dan Penyelesaian). Pyramida,
Yogyakarta.
[4] Eclipse Foundation., 2011, Eclipse.
http://www.eclipse.org/org/, (accessed at Nov 20, 2011).
[5] Fowler, Martin. 2005. UML Distilled 3th Ed, Yogyakarta:
Andi.
[6] Geoffrey, S., 2011, BIND.
http://en.wikipedia.org/wiki/BIND, (accessed at Des 12,
2011).
[7] Gombang, 2011, Wi-Fi.
http://id.wikipedia.org/wiki/Wi-Fi, (accessed at Sept 14,
2011).
[8] Graemel,
2011,
IP
camera.
http://en.wikipedia.org/wiki/IP_camera, (accessed at Sept 14,
2011).
[9]
Kadir, Abdul. 2002. “Pengenalan Sistem Informasi ”
Yogyakarta: Andi.
[10] Macks, D., 2011, Server (computing).
http://en.wikipedia.org/wiki/Server_
%28computing%29,
[11] Nicolas Gramlich, Andbook : Android Programming,
Download 10 Oktober 2009, http://andbook.anddev.org/
[12] Nugroho, Adi. 2005. Analisis dan Perancangan Sistem
Informasi
Dengan
Metodologi
Berorientasi
Objek.
Informatika. Bandung.
[13] OHA, Android.
http://www.openhandsetalliance.com/android_overview.html,
(accessed at Sept 19, 2011).
[14] Samulo,
A.,
2011,
Intranet.
http://id.wikipedia.org/wiki/Intranet, (accessed at Nov 2,
2011).
[15] Scott, C., 2011, Basic Access Authentication.
http://en.wikipedia.org/wiki/Basic_access_authentication,
[16] Setiawan, A., 2007, Perancangan Dan Implementasi Sistem
Monitoring Jarak Jauh Berbasis Protokol AX.25 Dengan
Menggunakan Mikrokontroler.
http://digilib.ittelkom.ac.id/index.php?option=com_repository
&Itemid=34&task=detail& nim=111020129, (accessed at Oct
13, 2011.
[17] Stachura, Max E. (2010). Telehomecare and Remote
Monitoring : An Outcomes Overview. Georgia : Advamed
[18] Triornis LTD, 2011, ZoneMinder.
http://www.zoneminder.com, (accessed at Nov 20, 2011.
[19] Webmin, 2006-2011, Webmin.
http://www.webmin.com/intro.html , (accessed at Dec 12,
2011).
Fig 6. User Interface Application Display from ZoneMinder CCTV
Server / Menu IPCAM MONITOR ZONEMINDER
E9-3
RIFASKES Geographic Information System
Based on Web
Istikmal, Yuliant S, Ratna M, Tody A W, Ridha M N, Kemas M L, Tengku A R
Electro and Commnucation Faculty, Telkom Institute Technology
itm@ittelkom.ac.id, ( yls,rmg,rmy,taw,tka)@ittelkom.ac.id, kemas.muslim@gmail.com
Abstract— Information access for medical care is a right
for every community and an obligation of the state to
provide health care as mandated by the Act of 2009. This is
a demanding health care facilities information can also be
accessed easily. RIFASKES (Health Research Facility) is a
program of the Ministry of Health of Indonesia through
the Balitbang Ministry of Health to monitor government
health facilities, including health centers (puskesmas).
Monitoring including the condition of the building health
centers, the type and level of service Categories, number of
health workers, provision of medical equipment.
Monitoring is done to optimize the role and functions of
government health facilities for public health, but is
constrained in getting the data that has been done
conventionally. In this study we created a web-based
geographic information system to present data
RIFASKES and can be updated in real time and named
SIGAPP KES. This research is a collaboration of IT
Telkom and Balitbang Ministry of Health. This
application is designed to enable people to find
information about the nearest puskesmas and to make
easier the government in monitoring the condition with
well updated. This application also making easier for
policy analysis in determining the appropriate authorities.
The application provides a graphical data analysis and
geographic (GIS, Geographic Information Systems). The
data has been inputted are the cities and counties in
western Java and made in the indonesian version. This
application has been presented to the health minister and
received a good reception for further development.
Index Terms— RIFASKES, GIS, SIGAPP KES.
I. INTRODUCTION
I.1 BACKGROUND
G
overnment of Indonesia has many various of health
facilities that are intended to provide health
services to the community. There currently are 693
RSUP (Government General Hospital) and the 9152
Puskesmas ( Community Health Center ) [1]. Certainly
not an easy task to monitor the condition of all health
facilitie, primarily puskesmas which located in
throughout Indonesia. Ministry of Health of Indonesia
through the Balitbang Ministry of Health implement the
program RIFASKES (Health Research Facility) which
aims to get the data and condition of government health
facilities. The data are needed to determine the measures
taken in increasing and optimizing health services
across the government health facilities.
The program in its implementation has obstacles in
the process of data retrieval and processing. Data
retrieval is done by spreading the field blank to all
government health facilities, and then collected again
after filled. These methods need a long time and great
cost considering its location spread throughout
Indonesia, especially puskesmas, in addition, not all
entries can be collected and blank filled completely.
This problem causes the data are incomplete and require
a long time, not to mention the data collected must be
verified. Finally, data processing was inhibited, and
takes a long time, this affects to the analysis and policies
to be taken because the data is too late and did not
update properly.
Application of geographic information systems in this
study intended to answer the above challenges in
RIFASKES program, that is research collaboration
between IT Telkom and Balitbang Ministry of Health.
For the first phase of this study is targeted puskesmas in
several cities and counties in western Java.
I.2 OBJECTIVES
Information and communication technology is
developing so rapidly, it should be used to facilitate the
utilization of any person in getting information anytime
and anywhere. The use of computers as a tool is very
common and its use for communication and information
sharing can be optimized as much as possible. It is
evident from the many puskesmas have computer
facilities both in urban and rural [1]. Figure 1. Show as
much as 97.7% in urban are available computer and
computer is not available 2.3%, while in the rural as
much as 79% of available computers and 21% are not
available.
E10-1
Figure 1. Availability of Computers in the
Community Health Center ( puskesmas ) [1]
The 6th – Electrical Power, Electronics, Communications, and Informatics
Informat International Seminar 2012
May 30-31,
Availability of computers in the puskesmas is
expected to support the use and implementation of these
applications later in the field. Today is also developing a
geographic information system more attractive to
present data in the form of geographically referenced
mapping or visualization.
In this study we built an application called SIGAPP
(Application of Geographic Information Systems)
which has the ability to build, store,
store manage and display
geographically referenced information that can be used
for scientific investigations,, resource management,
planning, monitoring or supervision,
supervision as well as the
analysis in any field . Because of this SIGAPP used in
the health field then this application is named SIGAPP
KES.
SIGAPP KES goal are to integrate RIFASKES
program into an application geographic information
system based on WEB. This application is intended data
RIFASKES program can be easily updated by each
puskesmas and the data can be processed,
processed analyzed, and
displayed either in the form of a graph and geographic
analysis.
METHODOLO
II. DESIGN AND METHODOLOGY
In making application SIGAPP KES conducted that
includes the stage of making the learning process
through references, data, geographic information
systems, databases, web design and methodology are
summarized as follows:
1. literature
Review of literature includes literature study of
information systems, database management, web,
mapping, researchh methodology, research kind ever
undertaken.
2. The methodology used is waterfall, in this
methodology the First Instance times is system need
analysis "requirements definition"
definition next is "system
and software design" and perform pre-processing the
data, then do the implementation and testing of
software design,, the final step is a software
development by performing the integration,
integration testing
and operation system maintenance.
maintenance This is done
related to the system's ability to offer that requires a
special study of the needs of the system,
system
pre-processing, design and development software.
• Applications
• data
• GIS Software
• Hardware
Application is a collection of procedures used to
process data into information.
information For example, the sum,
classification, rotation, geometry correction, query,
overlay, buffer, join table and so on.
4. Data retrieval
Data taken from the Balitbang ministries of health
from the result of RIFASKES 2011. From the
sugesstion of Balitbang the data entered into the
system is limited to first new refined. Here are
RIFASKES the data is entered into the system:
a. General information
Includes fiften information such as ID puskesmas,
puskesmas name, district code, clinic type, category
of health centers, and educational background of the
head clinic, condition and image building,
coordinates and address.
b. Human Resources (Health
Health workers)
Includes seven main health workers such as doctors,
nurses, midwives,, dentists, sanitarian, Promkes, and
nutrition.
c. Essential medical equipment,
equipment which consists of:
1. BP essential public health tool, such as
stethoscope, bed check,
check blood pressure meter
mercury, adult scales.
2. Essential medical equipment MCH (Maternal and
Child Health), which consists of the stethoscope, bed
check, blood pressure meter mercury, clinical
termometer, adult scales,, baby scales, dopler, and
hemoglobinnometer set (Sahli).
Sahli).
5. Data Preprocessing
The data obtained can not be directly used for this
application, initial data processing to be done to a
uniform data format before can be used.
6. Spatial Database Design and Software Design
Preparation of spatial databases based on spatial data
and attribute data that has been owned.
owned This spatial
database design determining the amount and type of
table / database that includes the attributes required
therein, and the relationship between a table with
other tables.
Figure 2. stages of waterfall methodology [8]
3. System Requirements Analysis
To be able to build this application requires at least
several components:
E10-2
Figure 3. HR database design
Table 1. Database tables and fields
NO
TABEL
1
Sdm_catagory
2
group_sdm
3
status_sdm
information such as address and type or category of
puskesmas, health workers held, and the essential public
health tools and equipment essential maternal and child
health.
FIELD
catagory_name
id_category_sdm
id_group_sdm
group_name
id_group_sdm
Status_name
id_status
7. Software Development
Software development is a coding process based
database and software design that has been created in
a programming language.
8. Testing and Repair
Software testing performed on every menu that is
created by using the blackbox methods.
These applications use the Google map, while for the
server and database using Xampp and AJAX are used
for web-based programming to create interactive
applications in GIS.
Figure 6. Search menu and tracking.
Users can utilize the tracking menu to look for a route
from where he was to the location of the targeted health
centers, as shown in figure 6.
III.1 SIGAPP KES for Public
This application is made into three main functions:
first to provide information to the public in finding
information nearest puskesmas. Community can look
directly through the map or menu search by city.
Figure 7. General information Health Center
After getting the location or community health center
sought to obtain general information by pressing the
menu centers on GIS or detailed information directly
from the search menu. Furthermore, by selecting the
menu of human resources and medical devices to obtain
information on health and medical devices are available
at the health center as shown in figure 8 and figure 9.
Figure 4. Initial view SIGAPP KES
Figure 4. An initial view web SIGAPP KES, after
entering the GIS map will be presented western Java as
in Figure 5, green tag indicates the location of
puskesmas, if we click it will display the name of the
puskesmas information.
Figure 8. Health personnel information
Figure 5. Location of health centers in the GIS
Furthermore the public can access information
directly from the puskesmas which consists of general
E10-3
Figure 9. Medical Device Information
III.2 SIGAPP KES for Goverment
The second major function of SIGAPP KES are for
government. This function is used to look at the health
center human resources information data, medical
devices, the type and category of health centers (
puskesmas ) from provincial to district level. Data are
presented in graphical form and GIS, this is to facilitate
the government in monitoring the development of health
centers as well as policy analysis function to take on the
latest conditions. Figure 10. display shows the login to
the government. There are two levels of logging that is a
government or puskesmas staff for the data update
function.
Next on the menu to choose the level of statistical
data to be seen whether the provincial, city, up to the
district level. We can also see the data of each clinic.
Figure 13 shows the doctor HR data statistics from the
provincial level (western Java), Tasikmalaya city to
kawalu district.
Figure 10. Login view of government
Once entered, the displayed menu is a list of all health
centers ( puskesmas ), looking at the data and statistical
analysis of the graphs and GIS. In the list of all our
health center can search by ID clinic, clinic name, or by
city.
Figure 11. Menu list and search the entire clinic
Figure 13. HR Data Doctor provincial, city and district.
From these data shows that the spread of doctor in
West Java has not been evenly distributed, at the city
level, there are at most 3 doctors in three health centers
and at least one doctor at seven puskesmas. At the
district level kawalu consists of one doctor, two doctors,
and three doctors for each health center.
There is a threshold field in the statistics menu that
are used to simplify the data to see if there are
established standards, as shown in Figure 14. If the
threshold is loaded first then the standard is a minimum
of one doctor for every clinic. Visible to all health
centers, Tasikmalaya availability of doctors meet the
standards of one person.
On the menu there is a choice satitstik HR, BP Tools
and equipment KIA. If the selected SDM, it will display
a menu of options that you want to see (doctors, nurses,
etc..), Whereas if the tool will display a menu of what
tools will be statistical (stethoscope, bed, Sahli, etc..),
As shown in Figure 12 .
Figure 14. Statistics with the threshold
The next menu is a look at the statistics of medical
equipment, medical devices by selecting the BP or KIA,
it will display a menu of choices of medical devices who
want to be seen. Then we specify the data that we will
see whether at the provincial or city.
Figure 15. Statistics indicate the availability of a
stethoscope in west Java province and city Majalengka.
Figure 12. Statistics and menu choices
E10-4
Seen from the graph there are 40 health centers do not
have a stethoscope and majelengka city, there are 9
health center does not have a clinical thermometer. With
these data the government can pay more attention to
health centers do not have requisite standard of medical
devices.
Figure 18. GIS information types of health centers
From the data shows that there are many health
centers in western Java are categorized as non PONED
and not the kind of care treatment centers. Government
seeks to improve health centers to PONED and care
catagory to improve health services to the community.
Figure 15. Health Statistics Tool
Next is the analysis of data in geographic form.
Figure 16. display shows the initial menu of statistical
analysis is also equipped SIG city search menu.
Functions are included in the statistics menu of GIS. In
the statistics menu functions are included GIS to view
HR data, type of health center consisting of categories of
care and non PONED or PONED. Then there is a menu
to see the physical condition of health centers.
Figure 19. GIS doctors Human Resources Information
To view the GIS information from HR we can select
the SDM. Figure 19 and 20 show images in Tasikmalaya
HR information. Filled with the limits for a direct look at
where the health centers that meet or not meet the
standard criteria. The green color indicates more than
the standard, meets standard is yellow color, red color is
less than the standard. Equitable distribution of health
workers can be seen that midwives better than doctors.
Figure 16. Statistics Display GIS
Categories puskesmas PONED or non PONED,
shown in the geographic centers of green to PONED and
red for non PONED. From the figure 17. Health centers
( puskesmas ) are generally seen in western Java is still a
lot of non PONED.
Figure 20. SIG Human Resources Information
Midwives
Figure 17. GIS information PONED category
Health Centers are also classified as care and non
care. Care health center means it can perform the service
road to inpatient care. From the figure 18. Seen the
number of care centers are still small.
In the availability of government health workers
should pay more attention again. Generally, there is still
a shortage of health workers, especially doctors,
promkes, and dentists. Not to mention unequal
distribution in the region. General health workers in the
rural difficulty in getting a good living facilities, and
lack of available health facilities.
III.2 SIGAPP KES for Administrator
The third main function is to administrator. This
function is used to regulate user access to applications
E10-5
SAGPP KES. Administrator managed user accounts and
government health centers authorized to view and
change data. Figure 21. Display shows the admin login
menu
Figure 21. Admin login menu display.
On the menu there is a list of admin users that are
currently active, there is a menu that user ca search by
name, code and name of the clinic group. Figure 22.
shows the admin menu.
IV. CONCLUSIONS AND RECOMMENDATIONS
IV.1 Conclusions
SIGAPP KES application is made to help RIFASKES
program to be more optimal utilization in results. This
application is expected to benefit the community and
government. The application can update the data more
quickly, so that both the government and the public can
obtain puskesmas information better. With this
application the Government may monitor and take a
strategic policy to develop and improve the quality of
health services and health center facilities ( puskesmas ).
IV.2 Recommendations
This application is still far from perfect, so we expect
to continued this riset for some improvement. For future
development, we can used map server and our own map
for easy processing of geography. Addition of data in
which if the critical parameters such as drugs, health
information, disease information. There should also be
developed GIS hospitals, laboratories, pharmacies. One
of the roadmap of this research is to make Indonesia GIS
Health System.
REFERENCES
[1]
[2]
[3]
[4]
[5]
Figure 22. Administrator menu display
[6]
Group code indicates that the user root as the main
admin, while the user admin as admin for puskesmas
and Government code indicates that users of the
government.
[7]
[8]
Balitbang Ministry of Health of the Republic Indonesia,
preliminary results of RIFASKES 2011.
Act ( UU ) No. 36 of 2009 on Health.
Kaswidianti Wilis, Budi Santosa, Rifky Satya, "Geographic
Information System Health facilities in the town of Magelang
web-based", National Seminar on Informatics 2008 UPN
yogyakartaUU No. 36 of 2009 on Health.
Software applications Healthmapper, WHO.
Women Research Institute,” Availability and Utilization of
Health Services For Maternal”, in 2008
Women Research Institute,” Utilization of Reproductive Health
Care for Poor Women”, in 2007
RI health minister's decision 1457/menkes/sk/x/2003 number of
minimum standards of health care in the district or city.
Fathul Wahid, Information Systems Research Methodology: an
overview, Media Informatics vol 2 no 1, June 2004.69 to 81.
Istikmal ST. MT. was born on 11 november 1979 in Kebumen.
Graduate degree in STT Telkom and continuing education for S2 at
ITB. currently a lecturer in Telecommunications Engineering
Program, Faculty of Electro and Communications, Telkom Institute of
Technology,
Bandung,
Indonesia.
E10-6
Fast and Accurate Interest Points Detection
Algorithm on Barycentric Coordinates using
Fitted Quadratic Surface Combined with Hilbert
Scanning Distance
1
Tibyani Tibyani and 2Sei-ichiro Kamata
Grad. School of Information, Production and System Waseda University
Kitakyushu, Fukuoka, Japan 808-0135
1
tibyani@akane.waseda.jp., 2kam@waseda.jp.
Abstract— The main purpose of interest points detection
algorithm on Barycentric coordinates for 3D objects based
on Harris operator is to find the fast computation of fitted
surface differences (FSD) of the derivative data for
different points. A FSD method is fitted quadratic surface
(FQS) approach combined with Hausdorff (HD) approach
as conventional method. In this paper, the extension of
Harris operator using FQS-HD (EHO-FQS-HD) and the
extension of Harris operator using FQS-Hilbert scanning
distance (EHO-FQS-HSD) as a proposed method to
interest points detection on Barycentric coordinates for 3D
meshes data is analyzed. The quality of this interest points
detection algorithm with EHO-FQS-HSD was measured
using the repeatability criterion. Experimental results
show that EHO-FQS-HSD is 5-10 faster than
EHO-FQS-HD. Moreover, it is a fast and accurate interest
points detector.
Index Terms—3D interest points detection, Harris
operator, Fitted Quadratic Surface, Hilbert scanning
distance,Hausdorff distance, 3D triangular mesh
I. INTRODUCTION
T
HE 3D interest points detection is a particular
processing step involved in computer vision and
pattern recognition algorithms. In the last decade,
interest points detection methods have been widely used
in various applications including 3D retrieval,
recognition, registration and matching. Because of they
are simple, flexible and excellent for many applications.
A method of the interest points detector for images
was introduced by Harris and Stephens [2], where their
method has an ability to respect to the information
contents and repeatability touchstones [8], robust to
noise [9], illumination change [10] and powerful
invariance to rotation and scale [8].
The interest points detection can be view as a
problem to determine the differences between two points
sets, i.e., to find the best transformation between a model
point set and image point set. In other worlds, given a
model point set, find the minimum or maximal value of
distance measures under the transformation in an image
point set. The points sets are usually feature points
extracted from the model and image. A well-known
measure called Hausdorff distance (HD) has been widely
used to this task However, HD is very sensitive to outlier
points and noise [12].
Glomb [6], in his seminal work proposed four
approaches of interest points detection algorithms using
extention Harris operator, i.e. Gaussian function (GF),
fitted quadratic surface (FQS), Hausdorff distance (HD)
and fitted surface differences (FSD) to define the
averaged derivatives to form matrix E [4]. From his
experimental results, the quality of FSD method showed
worser in computational cost for Harris operator values.
To address the noise and computational problems, a fast
and accurate measure is desired.
This paper is organized as follows. The conventional
2D Harris operator analysis, Extension of Harris
operator analysis on 3D triangular meshes and The
method of the conventional EHO-FQS-HD to compute
the E matrix, including mathematics descriptions are
described in Section 2. The proposed method of
EHO-FQS-HSD, which used to compute the E matrix is
described in Section 3. In Section 4, the experiments
result analyzed are designed to demonstrate the
performances of the proposed method. The experimental
results illustrate the fast and accurate of our proposed
method. It is about the experimental results using Hilbert
scanning distance and its comparison with Hausdorff
distance measures. Finally, Section 5 concludes this
study.
II. RELATED WORK
In this section, the conventional method of 2D Harris
operator analysis, extension of Harris operator analysis
on 3D manifold triangular meshes, fitted surface
differences method (FQS approach combined with HD
approach) as a conventional method to compute the E
matrix and Barycentric coordinates and manifold
triangular meshes are introduced.
E11-1
B. Extension of Harris Operator Analysis on 3D
Manifold Triangular Meshes
The interest points detection algorithm on
Barycentric coordinates for 3D manifold triangular
meshes is outlined in Figure 1. Given a vertex of a 3D
manifold triangular meshes object, a Harris operator
value associated with that points are calculated.
The steps of Harris operator extension analysis on
3D meshes are as follows [6]:
Fig. 1. Block diagram of the interest points detection algorithm on
Barycentric coordinates
1.
For manifold triangular meshes data, the input of
this research is a set of vertices V ⊂ ℜ3 and a given
v0 ∈ V. The camera digitizer generally produces a
surface of a 3D object to this input. Thus, this input
is reasonably to expect its neighborhood points to
sample a continuous surface of intrinsic 2-D space.
2.
Let a neighborhood of v0 as a subset V′′ ⊂ V
constructed of all points v ∈ V, obtained by starting
in v0 and following the edges from the meshes edge
graph, while the Euclidean distance || v0 - v || ≤ r,
where r is constant parameter.
(1)
3.
Define a preprocess neighborhood set of points V′′
and compute their centroid, then translate it to
[0,0,0]T.
where, q are the points in the Gaussian window function
W centered on (x,y), which defines the neighborhood
area in the analysis.
4.
Perform the regression to establish best fitting
plane.
5.
Rotate the set so that plane normal points to [0,0,1]T
to produce maximum spread (range of coordinates
values) is in the 0xy as plane. The set preprocessed
this way will be denoted with V′′′.
6.
Compute Harris operator from a given point
neighborhood indexed in 2-D space. The above
preprocessing allows us to approximate that
indexing on a point set V′′′ by using x and y
coordinates.
7.
Compute and define the derivatives which are
averaged to form matrix E.
A. Conventional 2D Harris Operator Analysis
The Harris operator has been implemented in huge
applications in pattern recognition and computer vision
because of its efficiency and simplicity. Given a 2D
image function f(x,y), a difference function e(MSE error)
of two neighborhoods including image points p=[x,y]T
and p+∆p can be calculated as follows:
e( p , ∆p ) =
∑ w (q )( f ( p + q + ∆ q ) − f ( p + q ))
2
q
The Taylor expansion truncated to the first order
approximation term shifted the image to obtain:
 ∑ w (q ) f x f x
e( p , ∆p ) = ∆p  q
w (q ) f y f x
 ∑ q
T
∑ w (q ) f
∑ w (q ) f
q
x
q
y
e( p , ∆p ) = ∆p T E∆p
fy
 ∆p
fy

(2)
where
fx =
∂f
(p + q)
∂x
fy =
∂f
( p + q ) (3)
∂y
with the directional first-order derivatives of image
function are fx and fy , respectively.
Harris and Stephens applied the eigenvalues to the
matrix E, which consists of enough local information in
conjunction with the neighborhood structure. In the
computational process, to prevent the highly eigenvalue
calculation, they designed to entrust with each pixel in
the approximation function image the subsequent value:
h ( p ) = det (E ) − k (tr (E ))
2
with k is a constant
C. Fitted surface differences as a conventional method
to compute the E matrix.
A FSD method is FQS approach combined with HD
approach [6]. This method in the state of art, namely the
FQS-HD.
In the FQS approach computation, a point set V′′′ can
fit a quadratic surface[16]. Derivatives calculation
perform to fit a quadratic surface to the set of
transformed points. Using least square approach to
discover a paraboloid of the form:
S (x , y ) =
(4)
a 2
c
x + bxy + y 2 + dx + ey + f
2
2
(5)
From (5), the directional first-order derivatives of
image function fx and fy are ready to compute.
E11-2
fx =
fy =
∂f (x , y )
∂x
x =0
∂ f (x , y )
∂y
(6)
(7)
y =0
geometric realization is denoted by specifying the
coordinates of the vertices xi ∈ R3 for all i ∈ V. The
barycentric coordinates can be represented as an N x 3
matrix. To acquire the meshes, a piece-wise planar
approximation by embedding the triangular faces
together,
Then to apply the integration of the derivatives with a
Gaussian function:
(
− x2 + y2
X = nv ∫ 2 e
R
(
2σ
− x2 + y2
Y = nv ∫ 2 e
R
(
2σ
R
nv =
2σ
)
2
− x2 + y2
Z = nv ∫ 2 e
)
2
2
)
. f x (x , y ) dx dy
(8)
. f y ( x , y ) dx dy
(9)
2
2
. f x (x , y ). f y ( x , y ) dx dy
1
(10)
(11)
2π σ
T (P ' ) =
U
NF
k =1
(
conv p t 1 , p t 2 , p t 3
k
k
k
)
(17)
The following Figure 2 depicts any point the meshes
T(P‘), a manifold can be represented including index k
of the triangle enclosing it and coefficients of the convex
(
)
x = u 1 p t1 + u 2 p t 2 + 1 − u 1 − u 2 p t 3 ;
k
k
(
k
x i ∈ [0 ,1] (18)
)
Vector u = u 1 , u 2 is called Barycentric coordinates.
where σ is a constant and nv is a normalization factor.
Using calculus theorem, the equations of (8) ,(9),
(10) can simplify the expressions to [7]
X = 2 a 2 + 2b 2 + d 2
Y = 2b 2 + 2 c 2 + e 2
Z = 2 ab + 2 bc + de
(12)
(13)
(14)
Fig. 2. Barycentric coordinates and manifold triangular meshes
Last, to calculate the matrix E associated with the
points :
X
E = 
Z
Z

Y 
III. FITTED QUADRATIC SURFACE COMBINED WITH
HILBERT SCANNING DISTANCE (FQS-HSD) AS A
PROPOSED METHOD TO COMPUTE THE E MATRIX
(15)
Because the proposed method use Barycentric
Coordinates [17-19], then if the object tessellation is
uniform, i.e., almost all triangles in the manifold
triangular meshes have the same size, this computation
can use a constant number of rings to all points, or use
the points contained in a ball of radius r and centered in
points vertices.
The HD approach can compute the E matrix using a
weighted average of the derivative 3D manifold
triangular meshes data for differents points. It can
perform the distance points between two points sets P
and Q using the difference fitted quadratic surface,
Given two finite points sets P ＝{p1…, pI} and Q＝
{q1…, qJ} such that each point p ∈ P and q ∈ Q, has
integer coordinates. Firstly, the Hilbert scanning is used
to convert them to new sets S ＝{s1…, sI} and T＝{t1…,
tJ} in the 1-D sequence, respectively. Then the directed
HSD from P to Q is computed by
d HSD (P , Q ) =
)
(17)
where || . || is the Euclidean norm distance in the 1-D
space and function is defined as:


d HD (P , Q ) = max sup inf p − q , sup inf p − q  (16)
q ∈Q p ∈ P
 p∈ P q∈Q

D. Barycentric Coordinates and Manifold Triangular
Meshes
The definition of a 3D data structure for boundary
representation, such triangular meshes, involves the
topological entities coding, with the linked up geometric
information and of a suitable subsets of topological
relationships between such entities. A data structure is
made up of vertices, edges and faces. The faces can be
represented as NF x 3 matrix of indices. The meshes
(
1 I
∑ i =1 ρ s i − t j
I
x
ρ (x ) = 
τ
(x ≤ τ )
(x > τ )
(18)
where, ρ is called the threshold elimination function and
τ is a the threshold prefined.
Then the directed HSD Q to P hhsd(Q,P) is obtained
similarly and HSD is defined by
E11-3
H HSD (P , Q ) = max (h HSD (P , Q ), h HSD (Q , P )) (19)
For reasons of efficiency running time computation,
a triangular meshes reduction algorithm is applied to
each meshes for Dragon, Rhino, Parasaurolophus,
Buddha, chicken, Chef and T-Rex data sets, resulting in
a reduced meshes with approximately 50000 vertices per
meshes, where Garland’s meshes is used for
simplification algorithm [13].
B. Comparison between the number of vertices and
interest points detection
Table I. Comparison between the number of vertices and
interest points detection using the proposed method
(EHO-FQS-HSD)
Fig. 3. 2-D Hilbert scanning (above) and 3-D Hilbert scanning (below)
IV. EXPERIMENTALS RESULTS
This section here describe the experiments
investigating performance of the extension of Harris
operator using EHO-FQS-HSD as a proposed method to
interest points detection on Barycentric coordinates for
3D meshes data. The presentation of results is divided
three parts. First, Comparison between the number of
vertices and interest points detection using the proposed
method of EHO-FQS-HSD. Second, The experiments
investigates the effect of the parameters on the
repeatability of interest points. Finally, Comparison the
proposed
method of EHO-FQS-HSD with
EHO-FQS-HD method in the state of art.
The comparison between the number of interest points
and the number of vertices in each model as shown in
Table I. It can be depicted in the table that the number of
interest points detection is significantly smaller than the
number of vertices in all 10 surfaces.
C. Evaluation methodology
A quantitative evaluation of the repeatability of
features extracted from the proposed method in different
noisy condition is shown in Figure 5.
A. The data set
This research work on ten data sets which scanned
with the Minolta Vivid 910 scanner. Their data sets are
visualized on figure 4 [14] [15].
Fig. 5. Repeatability of a interest points detection analysis on on
Barycentric coordinates using the proposed method (EHO-FQS-HSD)
The parameter k is used in (4) to calculate the Harris
operator value for a given vertex. This parameter
requires to be adjusted experimentally and modified the
parameter in the range [0.1,0.4].
Fig. 4. The Dataset
D. Comparison with other a method
The proposed method of EHO-FQS-HSD for
interest points detection algorithm on Barycentric
coordinates for 3D triangular meshes can be computed
faster than measure EHO-FQS-HD. Figure 6 shows the
E11-4
running time in the experiment for computing 250 Harris
operator values. The noise which is used in the
experiment is 1 to 7 and the has mean Gaussian noises
from σ ＝ 0 to σ ＝ 100.
[2]
[3]
[4]
[5]
[6]
Fig. 6. Comparison the proposed method (EHO-FQS-HSD) with
method in the state of art, namely the Extension of Harris
operator-Hausdorff distance (EHO-FQS-HD) for computing 250
Harris operator values
[7]
[8]
V. CONCLUSIONS
In this paper, the interest points detection analysis on 3D
triangular meshes using extension of Harris operator
combined with fitted quadratic surface (FQS) approach
and Hilbert scanning distance (EHO-FQS-HD) is
proposed.
Our major contribution is the Hilbert scanning
distance application on interest points detection
algorithm.
We demonstrated the speed and accuracy of this
algorithm in the presence of noise. That is to say, for
computing 250 Harris operator values the proposed fast
EHO-HSD is 5 – 10 times faster. Comparison with the
conventional algorithm EHO-FQS-HD [6], our proposed
algorithm of EHO-FQS-HSD is superior.
The future studies will aim at extending the
application fields, not just in matching objects, but also
in 3D registration and recognition objects.
[9]
[10]
[11]
[12]
[13]
[14]
[15]
ACKNOWLEDGMENT
We would like to acknowledge: A.S. Mian dataset
from the University of Western Australia and B. Taati
Queen’s Range Image for providing 3D model range
data. This research is sponsored and supported by
Indonesian Government Scholarship (Beasiswa Luar
Negeri
DIKTI-Kementerian
Pendidikan
dan
Kebudayaan Republik Indonesia)
[16]
[17]
[18]
REFERENCES
[1]
S. Kamata, R.O. Eason, and Y. Bandou, “A new algorithm for
N-dimenstional Hilbert scanning,” IEEE Transaction on Image
[19]
[20]
E11-5
Processing, volume 8, no. 7, pp. 964–973, July 1999.
(references)
Harris, C., and Stephens, M., “A combined corner and edge
detection,” In: Proceeding of The Fourth Alvey Vision
Conference, pp. 147–151, 1988.
A.S. Mian, M. Bennamoun and R. Owens, “A novel
representation and feature matching algorithm for automatic
pairwise registration of range image,” International Journal of
Computer Vision. Copyright Springer Verlag. vol. 66, no. 1, pp.
19-40, 2006.
A.S. Mian, M. Bennamoun and R. Owens, “3D model-based
object recognition and segmentation in cluttered scenes,” IEEE
Transaction in Pattern Analysis and Machine Intelligence, vol.
28, no. 10, pp. 1584–1601, 2006.
B. Taati, M. Bondy, P. Jasbedzki, and M. Greenspan, “Variable
dimensional local shape descriptors for object recognition in
range data,” In: Proceedings of International Conference on
Computer Vision 2007 – 3D Representation for Recognition
(3DRR), October 2007.
Glomb, P., “Detection of interest points on 3D data : extending
the Harris operator,” In: Computer Recognition System 3.
Advances in Soft Computing, vol. 57, pp. 103-111. Springer,
Berlin, 2009.
Sipiran, I. and Bustos, B., “Harris 3D: a robust extension of the
Harris operator for interest point detection on 3D meshes,” The
Visual Computer vol. 27, no. 11, pp. 963-976, 2011.
Schmid, C., Mohr, R. and Bauckhage, “Evaluation of interest
point detectors,” International Journal of Computer Vision.
Copyright Springer Verlag. vol. 32, no. 2, pp. 151-172, 2000.
Vincent, E., and Laganiere, R., “Detecting and matching feature
points,” Journal of Visual communication and Image
representation. volume 16, pp. 38-54, 2005.
Moreels, P. and Perona, P., “Evaluation of features detectors and
descriptors based on 3D objects,” International Journal of
Computer Vision. Copyright Springer Verlag. volume 73, no. 3,
pp. 263-284, 2007.
G. Rote, “Computing the minimum Hausdorff distance between
two points sets on a line under translation,” Inf. Process. Lett.,
volume 38, pp. 123-127, 1991.
D.P. Huttenlocher, G.A. Klanderman, and W.J. Rucklidge,
“Comparing images using the Hausdorff distance,” IEEE
Transaction in Pattern Analysis and Machine Intelligence, vol.
15, no. 9, pp. 850–863, 1993.
Garland, M., and Heckbert, P.S., “Surface simplification using
quadric error metrics,” In: SIGGRAPH, pp. 209–216, 1997.
(2010, Jan.) Queen’s Range Image and 3-D Model Database.
[Online]. Available:
http://rcvlab.ece.queensu.ca/qridb/QR3D/DatabasePagexyz.htm
l, 2009
(2010, Jan.) Database for 3D object recognition in cluttered
scenes. [Online]. Available: http://www.csse.uwa.edu.au/
ajmal/recognition.html
Z.L. Cheng and X.P. Zhang, “Estimating differential quantities
from point cloud based on a linear fitting of normal vectors,”
Journal of Science in China Series F: Information Sciences, vol.
52, no. 3, pp. 431–444, 2009.
T. Tibyani and S. Kamata, “Registering 3D objects triangular
meshes using an interest point detection on Barycentric
coordinates,” In: IEEE/OSA/IAPR International Conference on
Informatics, Electronics and Vision 2012 (ICIEV12), Dhaka,
Bangladesh, 18-19 May 2012.
Bottema, O. “On the area of a triangle in Barycentric
coordinates,” Crux. Math. 8, 228-231, 1982.
Coxeter, H. S. M. “Barycentric Coordinates.” in Introduction to
Geometry, 2nd ed, New York: Wiley, pp. 216-221, 1969.
th
The 6 – Electrical Power, Electronics, Communications, and Informatics International Seminar 2012
Generating Security Keys from Combination of
Multiple Biometric Sources
Primantara Hari Trisnawan
Computer Sciences, University of Science Malaysia
prima_at_ub@yahoo.com
Abstract— Biometric is a personal information and is
supposed to be unique. Biometric information can be
obtained generally from finger print or palmprint, iris,
voice, or face by using such instrument and methods with
appropriate algorithms to read their raw information.
Each of biometric sources typically results in biometric
information which are different from one person to
another. By combining multiple biometric information
with considering security aspects, it is convincingly able
to generate stronger key combinations. The key
generation is conducted by combining two or more
biometric information with an algorithm of Longest
Common Sequences. The complexity of the number of
new combinations resulting from various combination of
two or more biometric sources is O(n2). By applying to
some extent of the hamming distances between the
“database keys” and the combination keys, the
combination keys are acceptable and applicable in the use
for cryptography system.
Index Terms— biometric, cryptography, key
generations, longest common sequences, combination of
multiple biometric sources.
I.
I
INTRODUCTION
N recent years, the widely use of digital data and
data communication enabling people to deliver
digital data among them dramatically increase.
Obviously, digital data traversing communication media
are unsaved and easily eavesdropped. It is also supposed
to that private and personal digital data are frequently
misused by non authenticated people. For this reasons,
people require an equipment to save their data or an
authentication tools of which their data are surely their
own data. Hence, it is a need of cryptography which has
capability of providing these requirements.
Meanwhile, people are likely to use their different
passwords for accessing different systems. In general, the
passwords are relatively simple and short in order to
easily remember the passwords. For example, a user
employs one password (or key or account) to log into a
computer, and employs another password he/she has to
access building system. Thus, it would be difficult for
them to remember many passwords to corresponding
systems. It is more and more difficult to remember so
lengthy passwords. Otherwise, by keeping the same
password for all systems, their passwords are prone to be
vulnerable. Moreover, some systems may even prevent
people from keying the incorrect passwords in a certain
times, such as accessing a bank’s ATM.
Biometric is individual information and is considered
unique. It is unlikely that two people have the same
identical biometric, even for twin people. Biometric
information can be taken from such sources as finger
print, palm print, iris, voice, and face. Generating key not
only includes digitizing from a biometric source, but it
also minimizes errors [1][2]. A single biometric has
various key lengths [1][6][7] depending on the methods.
Never the less, in fact, some biometric information could
meet a theft problem. They could be duplicated easily. A
solution to this problem and still keeping people with
many keys is by using combination of biometric data.
People are still able to have many keys without the need
of memorizing various keys they might use. In order to
reduce, if not eliminate, this duplication problem by un
authorized people, combinations of biometric data are
applied to produce different but still secure unique keys.
Combining data from multiple sources operates methods
or algorithms and it results in various keys. By doing such
combination, it may lead to reducing percentage errors as
well.
The methods in combining biometric keys to produce
stronger unique keys are provided, such as concatenating
keys, adding keys and using Longest Common
Subsequences of keys. The first two methods will not
reduce the errors nor keeping the length the same. The
last method can be considered good, as by using longest
common sequences bits which are the same in the
sequence are preserved, whereas the not same bits in the
sequence are omitted.
There can be error bits existing in original biometric
keys, which but are kept as low as possible. These error
bits are possibly measured as input for the combination
process. The other thing is that there is a need to keep the
key length not more than the shortest original keys, as
quite short keys are easily broken by cryptanalysts using
E12-1
th
such brute force algorithm [3].
If new keys resulted from this combination is to be
applied in encryption and decryption which operates the
exactly same keys without any tolerances, it may lead to a
problem. It is because the combination key must be
exactly the same as the key in “data base”. Nevertheless,
if the keys are for digital signature and allowing to some
extent of hamming distance, it would be greatly
acceptable and applicable properly.
The main focus of this research is to compare methods
to combine various keys with different length. It also
includes how the algorithm is built with number of
generated biometric keys into new key combinations.
However, it should be kept in mind that newly combined
keys must be long enough in order to avoiding the keys
from being breakable.
[2]. Reducing error, in particular background noise as
proposed by [1] [2] was based on Hadamard transform
[10] (also called as Hadamard Code) which encodes
(k+1) bits into 2k bits. Meanwhile, reducing burst error as
used by [1][2] was based on Reed Solomon, whose
feature is that two information can be added together with
the length of n without reducing the distance [11]. Fig 2
shows the process of biometric binaries into biometric
information with applying both mechanisms.
Fig 2. Biometric Information Process.
However, based on some research works as in
[1][2][6][7] and [9], biometric information have been
obtained with various bit lengths. These bit length
comparisons are shown in following table.
II. PREVIOUS WORK
Biometric sources are taken from part of human body.
Table 1. Comparison of Biometric Information Sources
These sources include face [1], iris [7], fingerprint [6] or
FRR2
FAR1
palm print, and voice [9]. Examples of sources are shown Biometric Source Bit Lengths
Face [1]
240 bits
28%
in Fig1.
Iris [2][7]
140 bits
0,005%
0,235%
Fingerprint [6]
73 bits
1%
Voice [9]
46 bits
20%
Fig 1. Biometric Sources
Voice biometric is extracted from digitized voice
frequency [9]. Compared voice biometric to other three
kinds of biometric sources, the voice biometric is
considered not stable as the user’s voice has to remember
and maintain the same word and the same pace he/she
utters. The three kinds of biometric information do not
require the owner to do anything, but he/she just presents
his/her part of body for being scanned to obtain the
biometric information. However, for technical point of
view, the voice biometric generation is simpler than other
three biometric generations. The voice is ready in a form
of frequency signal, whereas the others are recognizable
in a certain pattern, such as circle, oval, bending line, etc.
Nevertheless, instead of four biometric information, there
are about 2 or 3 more biometric sources with few
references but not in profound explanation, including
gait, key stroke and signature.
After digitizing biometric sources into binaries, the
resulted signals are processed to remove some errors. The
modern tools the researchers use include reducing error
due to background noise, and reducing burst errors [1]
Table 1 shows that some researches result in various
key lengths with different error rates. In general, most of
those researchers relied on the FRR rather than FAR,
which meant that it was related to the percentage of
authentic person’s keys which should be accepted by the
system, but the keys were rejected. In practical
application, a prominent reference says that an accepted
value of FRR should be less then 20% [13], which is
argued by [1] that FRR at about 20-30% is still
acceptable.
Based on Table 1, biometric sources produce different
key lengths. Voice only generates 46 bits [9], which is
considered too short and breakable by about 70 seconds3
[3]. Whereas, the finger print with 73 bits [6] is breakable
in about 300 years. Therefore, biometric information with
the keys more than 73 bits can be considered secure.
In addition, an other research [8] applies entropy
analysis to reach the optimal security which is related to
1
FAR stands for false acceptance (error) rate, determine an
percentage of error rate at which an unauthentic person is accepted as
an authentic person. [http://www.cccure.org/Documents/HISM/039041.html]
2
FRR stands for false rejection (error) rate, which a percentage of
error rate to determine at which an authentic person is rejected as
invalid
person
[http://www.cccure.org/Documents/HISM/039041.html].
3
It is based on the time required to execute 106 decryptions per µs.
E12-2
th
biometric template [8]. This entropy, based on Shannon’s
entropy [12] which quantifies with regard to expected
value, measures the uncertainty regarding a random
sample. A biometric information source as a key
(comprising bits) is read many times to minimize the
uncertainty of the key.
To sum up, most biometric sources, except voice,
provide appropriate key length which is suitable for
modern cryptography system. With FRR is less than 30%,
the biometric key is still acceptable.
III. ALGORITHMS
III.1 Strings (Bits) Matching
There are some algorithms or methods about how two
groups of strings (or groups of bits) are to match and join
together. The algorithms include concatenation [4],
adding two value of strings (bits) and string matching in
particular using longest common subsequences [4]. The
concatenation of two group of string S1 and S2 with
length of L1 and L2 respectively, will result in a new
string of S1 adjacently followed by S2 with the total
length of an addition of L1 and L2. The second approach
is based on addition value, such as in ASCII4, of each
character (or bit) of first group with value of each
character (or bit) of the other group at corresponding
position. This will result in another string (represented by
the resulted value) with at most length of either (L1 + 1)
or (L2 + 1). The last method is to match sub sequences of
first group into sub sequence of second group, or vice
versa. The result is a sub sequence of string too, whose
length is at most the same as the length of the shorter
string.
Each of three methods has advantages and
disadvantages. The first method makes the key length is
longer. If the key is sufficiently long regarding to
security, the longer key will not be important. The second
method produces the key with highly possibility of not the
same (part of sub sequence of) the original keys.
However, by addition two groups of string (or bits), other
operation in addition of other groups possibly makes the
same result. The final method make the key length
relatively short and the key is part of string (or bit) sub
sequence of the original keys. Hence, the third method is
likely to be implemented in generating keys from multiple
biometric sources. With combination keys taken by using
this method will create much stronger combination key
than the previous method..
4
ASCII stands for American Standard Code for Information
Interchange, which is a character encoding based on the English
alphabet.
III.1 Longest Common Subsequence (LCS)
Longest Common Subsequence is to find common sub
sequences from (usually two) groups of sequences. Given
sequence in X and in Y [4], and will result in LCS5. The
rule is illustrated as follows.
This algorithm will show that for starting, cells in 0th
row and cells in 0th column are with value of 0. Value of
other cells depends on whether there is a commonness of
a character between two groups of sequences.
The algorithm for LCS is composed of two main
methods. The first method, namely LCSLength(), is to
memo the common sub sequence characters from string X
and string Y. It is precisely to mark the common character
in both sub sequences with the incremented value (in
integer). This method is as follows.
function LCSLength(X[1..m], Y[1..n])
C = array(0..m, 0..n)
for i := 0..m
C[i,0] = 0
for j := 0..n
C[0,j] = 0
for i := 1..m
for j := 1..n
if X[i] = Y[j]
C[i,j] := C[i-1,j-1] + 1
else:
C[i,j] := max(C[i,j-1], C[i-1,j])
return C[m,n]
The second method, namely backTrace(), is to read out
the content which has been saved by in a variable which
mark the same character in both groups in previous
method.
function backTrace(C[0..m,0..n], X[1..m], Y[1..n], i, j)
if i = 0 or j = 0
return ""
else if X[i] = Y[j]
return backTrace(C, X, Y, i-1, j-1) + X[i]
else
if C[i,j-1] > C[i-1,j]
return backTrace(C, X, Y, i, j-1)
else
return backTrace(C, X, Y, i-1, j)
III.2. Hamming Distance
Hamming distance in 1950 as stated in [5] describes
“The Hamming distance between two n-tuples X and Y,
5
Longest Common Subsequences is of dynamic programming.
Dynamic programming is not actually related to programming, but is
related to creating a table whose cells are filled with value according to
some rules[4].
E12-3
th
denoted by D(X, Y), is defined as the number of positions
in which the two words differ”. This algorithm is simply
by XOR-ing two groups of bits with the same length, and
then calculating the number of bit 1. The number of bit-1
means the number of bit difference in two group of bits.
This can evaluate the difference between key recorded in
“database” and key obtained from biometric information.
IV.2. Input Selection and LCS Output
With n input from biometric sources, two or more (or
up to n) combination can be performed. The chosen
combination is corresponding to such selector. Therefore,
the number of selector is the same as the number of input.
This selector with LCS output is shown in Figure 3.
IV. COMBINATIONS OF MULTIPLE BIOMETRIC
INFORMATION
IV.1. Combinations
In this research, multiple biometric information as keys
are combined in such a way to generate other keys. The
combination is merely based on mathematic formula; that
is combination. Combination formula is
C kn =
n!
k!(n − k )!
where :
n : number of a set of n biometric information
k : number of k biometric information of a set of n.
Combining a number of n biometric information, in
this research, does not implement k = 0 and k = 1. A
number of k = 0 means that there is no key being
combined and thus generated. Whereas k = 1 means that
there is only single member in combination. Hence, the
total number of combinations regardless k = 0 and k = 1,
are
n
Ctotal = ∑ C kn − C 0n − C1n
k =0
n
= ∑ C kn − 1 − n
k =0
= 2n − 1 − n
where : Ctotal : total number of all possible
combinations
Therefore, in the formula, it shows that the complexity
of Ctotal is still O(2n).
With regard to LCS which only performs combination
of two groups of string (or bit). Thus, for m input
combination, can be done in step by step combination, as
follows
LCS = I1 I2 I3 I4….. Im
= ((((I1 I2 )I3 )I4 ….) Im
With the regard to security, LCS should have
appropriate length. It is supposed that the difference
between two groups of keys are not wide.
Fig 3. Selector for n Input with Output from LCS
Assume that the selector operates on bit value. Each
selector is active when its bit value is 1. To provide
combination with at least two input selected, the number
of bit 1 in set of selector has to be at least two, as well.
The output of AND is as input if the selector is active (bit
1), whereas the output of AND is “open circuit” (not bit
0), if the the selector is in active (bit 0). All of
combination outputs from AND gates go to LCS
operation. In stead of processing LCS algorithm, LCS
block has a memory to hold all inputs coming from AND
gates, temporary result and final result. The final result is
the generated key from multiple biometric information.
IV.3. Design of Implementation
To implement this proposed design, it needs to
considered two main implementation designs. The first is
for generating keys by combining multiple biometric
sources then recording the biometric information keys,
while the other for application to which a user
authenticates his/her biometric sources with the recoded
biometric information. It is to be consider that the
biometric information for inputs to this design have been
E12-4
th
refined from what some past works have produced as in
[1][2][6][7] and [9].
V. DISCUSSION
IV.3.1. Recording Design
In this stage, key combinations generated from multiple
biometric information are recorded and stored in a
database with a specific tuple of parameter. Every tuple
consists of value of selectors, and value of combination
key corresponding to the selector.
Tuple = {selector value, combination key}
In combining of multiple biometric information, it is
assumed that the biometric information as input for
generating multiple keys have been committed in past
researches and have been valid with such keys’ bit
length, FAR and FRR. Thus, this research is merely
related to how biometric information are combined
together with consideration of security, which is key
length. The key resulted from combination will have its
length about the shorter input biometric information key.
The chosen combination can be selected using selector
value (in bit) to activate or pass the biometric input into
LCS block.
In recording implementation, it only needs to record
combination with related selector, then saves its result
together with the selector value into a tuple for the
database. In application, it needs additional block to
control which biometric information combination to be
selected based on a tuple, and compares the generated
combination with the combination key saved in this tuple.
The comparison of both generated combination key
taken from a user and combination key in a corresponding
tuple is likely not to precisely the same, but with applying
such hamming distance, the comparison of both keys can
be considered the same or valid.
The diagram for recording design as this Fig 4.
Fig 4. Recording Combination of Multiple Biometric
IV.3.1. Application Design
In application design, a user authenticates his/her
biometric. It needs his/her part of body be presented to
instrument for accessing his/her biometric information.
The system will activate the selector to read the
combination and then compare its value with those in
database. The illustration is as this.
Fig 5. Application for User Autentication.
The selector is activated by value a tuple from
database, then the output of the selector is compared with
the keys in the tuple. By using hamming distance, the
distance between these values are calculated. If the
distance is less than the defined distance (for tolerance),
then the biometric keys are considered to be valid.
VI. CONCLUSION
To conclude, the multiple key generation can be
implemented to the system which supports reading multi
biometric sources. The system can be implemented in
addition to systems which have been built by some
researches in the past works. Key generation from
multiple biometric sources can be implemented well. As
the key is a form of multiple biometric sources which
have to present at the same time, it is convinced that the
key is stronger than only single biometric source.
However, it may be suggestion to improve this system
with regard to keeping the length not less then the shortest
biometric keys when using LCS. This for example is by
repeating the bits or padding the bits.
ACKNOWLEDGMENT
This research is acknowledged by Assoc. Prof. Dr.
Azman Samsudin for Computer Security and
Cryptography, at Computer Science in USM.
E12-5
th
[8]
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
B. Chen and V. Chandran, “Biometric Based Cryptographic Key
Generation from Faces”, IEEE 2007
Hao, F., Anderson, R., Daugman, J., “Combining Cryptography
With Biometrics Effectively”, Technical Report UCAM-CL-TR640, University of Cambridge, 2005.
W. Stalling, “Cryptography and Network Security: Principles and
Practices”, 4th ed., Printice-Hall inc, 2006.
S. Dasgupta, C. H. Papadimitriou, and U. V. Vazirani,
“Algorithms”, July 18, 2006.
B. Bose, T.R.N. Rao, "Theory of Unidirectional Error
Correcting/Detecting Codes," IEEE Transactions on Computers,
vol. 31, no. 6, pp. 521-530, Jun., 1982
Ulrike Korte, et. Al, “A cryptographic biometric authentication
system
based
on
genetic
fingerprints”,
2007,
http://www.secunet.de/fileadmin/Downloads/Englisch/Sonstiges/s
ecunet_Fachartikel_BioKey.pdf.
Padma Polash Paul, and Md. Maruf Monwar, “Human Iris
Recognition for Biometric Identification”, IEEE, 2007.
[9]
[10]
[11]
[12]
[13]
E12-6
Jovan Dj. Golic´ and Madalina Baltatu, “Entropy Analysis and
New Constructions of Biometric Key Generation Systems”, IEEE
Transactions On Information Theory, Vol. 54, No. 5,, May 2008.
Fabian Monrose, et al., “Cryptographic Key Generation from
Voice”, In Proceedings of the 2001 IEEE Symposium on Security
and Privacy, May 2001.
M. Kunt, "On Computation of the Hadamard Transform and the R
Transform in Ordered Form," IEEE Transactions on Computers,
vol. 24, no. 11, pp. 1120-1121, Nov., 1975
Reed, I. S. and Solomon, G., “Polynomial Codes Over Certain
Finite Fields,” SIAM Journal of Applied Math., vol. 8, 1960, pp.
300-304.
I.J. Taneja, ” Generalized Information Measures and Their
Applications ”, 2001,
http://www.mtm.ufsc.br/~taneja/book/node1.html
U. Uludag, S. Pankanti, S. Prabhakar and A. K. Jain, “Biometric
cryptosystems: issues and challenges,” Proceedings of the IEEE,
Vol. 92, No. 6, pp. 948–960, 2004
GRATITUDE
WE WOULD LIKE TO THANKS TO ALL THAT DESCRIBED BELOW WHO HAVE GIVEN US INVALUABLE SUPPORTS
AND PARTICIPANTIONS
RECTOR OF BRAWIJAYA UNIVERSITY
COMPUTER SCIENCE, UNIVERSITY OF SCIENCE MALAYSIA
GRAD. SCHOOL OF INFORMATION, PRODUCTION AND SYSTEM, WASEDA UNIVERSITY, JAPAN
UNIVERSITAS PEMBANGUNAN NASIONAL BABARSARI TAMBAKBAYAN YOGYAKARTA
FACULTY OF ENGINEERING, UNIVERSITY OF WOLLONGONG, NSW, AUSTRALIA.
ELECTRICAL AND ELECTRONIC ENGINEERING OF UNIVERCITY OF MIYAZAKI
FAKULTAS TEKNIK, UNIVERSITAS BRAWIJAYA
POLITEKNIK NEGERI BENGKALIS
POLITEKNIK NEGERI AMBON
POLITEKNIK PEMBANGUNAN NASIONAL BANJARBARU
BADAN PENGKAJIAN DAN PENERAPAN TEKNOLOGI (BPPT)
UNIVERSITAS GAJAYANA MALANG
UNIVERSITAS SEMARANG
FAKULTAS SAINS DAN TEKNOLOGI UNIVERSITAS AL-AZHAR JAKARTA
UNIVERSITAS ISLAM NEGERI SUSKA RIAU
INSTITUT TEKNOLOGI SEPULUH NOPEMBER (ITS)
POLITEKNIK NEGERI SAMARINDA
PUSAT SAINS ANTARIKSA LEMBAGA PENERBANGAN DAN ANTARIKSA NASIONAL (LAPAN)
ELECTRICAL ENGINEERING DEPARTMENT, FACULTY OF ENGINEERING, MERCU BUANA UNIVERCITY
FAKULTAS TEKNIK UNIVERSITAS MURIA KUDUS
FAKULTAS TEKNOLOGI INFORMASI DAN KOMUNIKASI, UNIKOM, BANDUNG
PT. SUPRA ALUMINIUM INDUSTRI (SAI) PANDAAN-PASURUAN
UNIVERSITAS TRUNOJOYO MADURA
UNIVERSITAS JENDERAL AHMAD YANI, CIMAHI
POLITEKNIK NEGERI BATAM
FAKULTAS MIPA UNIVERSITAS INDONESIA
PETRA CHRISTIAN UNIVERSITY, SURABAYA
FACULTY OF MECHANICAL AND AEROSPACE ENGINEERING, INSTITUT TEKNOLOGI BANDUNG
SCHOOL OF ELECTRICAL ENGINEERING AND INFORMATICS, INSTITUT TEKNOLOGI BANDUNG
INSTITUT TEKNOLOGI TELKOM, BANDUNG
FAKULTAS TEKNIK, UIVERSITAS BRAWIJAYA
FAKULTAS TEKNIK, UNIVERSITAS DARMA PERSADA, JAKARTA
POLITEKNIK ELEKTRONIKA NEGERI SURABAYA – ITS
UNIVERSITAS ISLAM SULTAN AGUNG (UNISSULA), SEMARANG
FAKULTAS SAINS DAN TEKNOLOGI, UIN SYARIEF HIDAYATULLAH, JAKARTA
FAKULTAS TEKNIK, UNIVERSITAS BHAYANGKARA, SURABAYA
FAKULTAS TEKNIK UNIVERSITAS INDONESIA
POLITEKNIK NEGERI KUPANG
POLITEKNIK NEGERI MALANG
UNIVERSITAS ISLAM MALANG
UNIVERSITAS NEGERI MALANG
INSTITUT TEKNOLOGI BANDUNG
FAKULTAS TEKNIK UNIVERSITAS SULTAN AGENG TIRTAYASA CILEGON BANTEN
UNIVERSITAS GADJAH MADA YOGYAKARTA
CENTER FOR TECHNOLOGY OF RADIATION SAFETY AND METROLOGY,
NATIONAL NUCLEAR ENERGY AGENCY OF INDONESIA
LABORATORIUM SISTEM BERBASIS KOMPUTER TEKNIK ELEKTRO UNIVERSITAS HASANUDDIN
LABORATORIUM TEKNIK KOMPUTER DAN JARINGAN, POLITEKNIK NEGERI UJUNG PANDANG
POLITEKNIK NEGERI BALIKPAPAN
UNIVERSITAS MA CHUNG
STIMIK STIKOM BALI
FAKULTAS KEDOKTERAN UNIVERSITAS BRAWIJAYA
CENTER OF SATELLITE TECHNOLOGY, INDONESIAN INSTITUTE OF AERONAUTICS AND SPACE
AND OTHERS....

- Universitas Brawijaya

Transcription

Similar documents

Information of ＩＭＤＩＡ Auditor Change

EID MUBARAK!

Apprenticeship Program Selection in the Northern Territory Australia

20160105 Profil Anggota AC ENG Periode 2015

The TELKOM Assessment Center - International Congress on

Succeeding in Mobile Advertising

Indonesia Social Justice Network Brochure

Data Science - SourceMedia

Show BrochureJune 19, 2015 - 1:29 pm

something exciting is heading your way!